[Intel-gfx] [PATCH v3] kcmp: Support selection of SYS_kcmp without CHECKPOINT_RESTORE

[Emil Velikov]

On Fri, 12 Feb 2021 at 13:14, Simon Ser <contact at emersion.fr> wrote:
>
> On Friday, February 12th, 2021 at 1:57 PM, Emil Velikov <emil.l.velikov at gmail.com> wrote:
>
> > On Fri, 5 Feb 2021 at 22:01, Chris Wilson <chris at chris-wilson.co.uk> wrote:
> > >
> > > Userspace has discovered the functionality offered by SYS_kcmp and has
> > > started to depend upon it. In particular, Mesa uses SYS_kcmp for
> > > os_same_file_description() in order to identify when two fd (e.g. device
> > > or dmabuf)
> >
> > As you rightfully point out, SYS_kcmp is a bit of a two edged sword.
> > While you mention the CONFIG issue, there is also a portability aspect
> > (mesa runs on more than just linux) and as well as sandbox filtering
> > of the extra syscall.
> >
> > Last time I looked, the latter was still an issue and mesa was using
> > SYS_kcmp to compare device node fds.
> > A far shorter and more portable solution is possible, so let me
> > prepare a Mesa patch.
>
> Comparing two DMA-BUFs can be done with their inode number, I think.
>
> Comparing two device FDs is more subtle, because of GEM handle
> ref'counting. You sometimes really want to check whether two FDs are
> backed by the same file *description*. See [1] for details.
>
Thanks for the correction and the reference.
Seems like I've short circuited file description table vs file descriptor.

Emil