[Mesa-dev] [PATCH 0/2] anv/i965: drop libdrm dependency completely

[Daniel Vetter]

On Thu, Apr 27, 2017 at 10:58:42AM -0700, Lionel Landwerlin wrote:
> On 27/04/17 08:20, Eric Anholt wrote:
> > Emil Velikov <emil.l.velikov at gmail.com> writes:
> > 
> > > On 25 April 2017 at 23:56, Lionel Landwerlin
> > > <lionel.g.landwerlin at intel.com> wrote:
> > > > Hi,
> > > > 
> > > > While working with changes that span from kernel to user space, I've
> > > > been wondering whether we need to depend on libdrm at all for the anv
> > > > & i965 drivers. Indeed with Ken's recent changes, we only depend on
> > > > libdrm for its kernel header files which we could just embed
> > > > ourselves.
> > > > 
> > > > I've only included the minimal set of header files we need from the
> > > > kernel for anv & i965. Maybe other drivers would be interested and
> > > > maybe we should put all the kernel drm uapi headers into include?
> > > > 
> > > AFAICT the goal behind the libdrm_intel fold was to allow rapid and
> > > seamless rework of the interface.
> > > With a potential goal to make it a shared one, as it gets stabilised.
> > > 
> > > Currently ANV uses more than just the UAPI headers. But if we ignore
> > > that, coping them is a bad idea.
> 
> What else is anv using from libdrm? (maybe I missed something)
> 
> > > 
> > > Why - adds a, yet another, copy and making synchronisation more
> > > annoying. First example - you blindly copied the files rather than
> > > using `make headers_install' first ;-)
> > > Not to mention that it makes the chicken and egg problem* even more
> > > confusing and error prone.
> 
> It doesn't feel like that to me. Actually instead of modifying 3 different
> repos, you end up modifying just 2.
> Sounds less error prone.
> 
> > > 
> > > Emil
> > > *  Which patches land first - kernel or userspace
> > I don't see how it does that at all.  It simplifies the process: Now you
> > send out your proposed new userspace to one repo, instead of two.
> > 
> > And, after the first commit, it'll be obvious when you screw up using
> > make headers_install because there will be surprising diff.
> 
> Right now we have to update libdrm, then update the mesa to depend on the
> right libdrm to actually get the header files we want.
> If you depend on libdrm from more than just uapi headers, it might now be
> too much of a burden. But it seems we're clearly moving away from that in
> anv/i965.
> As a developer it feels a lot easier to have just the update mesa with both
> the new kernel API you depend on & the changes to the user space driver
> using that API.

As long as the headers are never installed into the system I'm in
principle ok with pulling all the i915.ko specific stuff into mesa. Seems
like a reasonable thing to do.

Of course still the same rules apply for merging new uapi: All parts must
be reviewed, then we merge the kernel, and only afterwards userspace. The
headers process in libdrm (see libdrm/include/drm/README) is imo the best
model for that.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch