[Mesa-dev] [PATCH] i965: Solve Android native fence fd double close issue

[Xu, Randy]

> -----Original Message-----
> From: Emil Velikov [mailto:emil.l.velikov at gmail.com]
> Sent: Saturday, April 29, 2017 8:36 PM
> To: Chad Versace <chadversary at chromium.org>; Emil Velikov
> <emil.l.velikov at gmail.com>; Xu, Randy <randy.xu at intel.com>; mesa-
> dev at lists.freedesktop.org
> Subject: Re: [Mesa-dev] [PATCH] i965: Solve Android native fence fd double
> close issue
> 
> On 28 April 2017 at 17:18, Chad Versace <chadversary at chromium.org>
> wrote:
> > On Thu 27 Apr 2017, Emil Velikov wrote:
> >> On 27 April 2017 at 12:14, Xu, Randy <randy.xu at intel.com> wrote:
> >> > Hi, Chad
> >> >
> >> > Please review this patch, we need it to solve some instability
> >> > issues
> >
> > Randy and Tapani, could you provide a few dEQP test names that this
> > patch fixes? I'd like to mention at least one EGL and one Vulkan test
> > in the commit message.
> >
> >> The patch is correct, although the commit message can be improved upon.
> >> Read through the following example and consider the alternative
> >> solution mentioned within.
> >
> > Yes, this patch is correct. It makes brw_dri_create_fence_fd() behave
> > like all the other drivers' create_fence_fd funcs, which call dup().
> > Since this is an easy one-liner that can backport to stable, let's
> > take it.
> >
> > However, I believe the fully correct solution is Emil's plan B:
> > __DRI2fenceExtensionRec::create_fence_fd should transfer fd ownership
> > to the driver, and therefore no dup is needed. But that's a slightly
> > more invasive change that's not as easily backported to stable.
> >
> > Reviewed-by: Chad Versace <chadversary at chromium.org>
> > Cc: mesa-stable at lists.freedesktop.org
> >
> > Emil, how about one of us appends your extended commit message to
> > Randy's, and then pushes?
> >
> Feel free to polish and push Chad.
> 
> Randy please don't forget to send a patch for dri_interface.h.

Thanks, Emil,  if create_fence_fd just transfer fd ownership to the driver, 
not dup, we also need change dri2_dup_native_fence_fd, which should not
keep the fd after calls get_fence_fd. It should be workable, but I need do
more test.

Chad, please help to polish my comments, or give me some suggestions.

> 
> Thanks
> Emil