[<prev] [next>] [day] [month] [year] [list]
Message-ID: <YBp5qzLBBMJE0Yhn@phenom.ffwll.local>
Date: Wed, 3 Feb 2021 11:23:39 +0100
From: Daniel Vetter <daniel@...ll.ch>
To: Christian König <christian.koenig@....com>
Cc: Simon Ser <contact@...rsion.fr>,
Pekka Paalanen <ppaalanen@...il.com>,
Michal Hocko <mhocko@...e.com>,
"open list:DOCUMENTATION" <linux-doc@...r.kernel.org>,
Szabolcs Nagy <szabolcs.nagy@....com>,
dri-devel@...ts.freedesktop.org, Andrei Vagin <avagin@...il.com>,
Kalesh Singh <kaleshsingh@...gle.com>, Hui Su <sh_def@....com>,
Michel Lespinasse <walken@...gle.com>,
Jonathan Corbet <corbet@....net>,
Mauro Carvalho Chehab <mchehab+huawei@...nel.org>,
Jeffrey Vander Stoep <jeffv@...gle.com>,
Daniel Jordan <daniel.m.jordan@...cle.com>,
kernel-team <kernel-team@...roid.com>,
Alexey Dobriyan <adobriyan@...il.com>,
Linux Media Mailing List <linux-media@...r.kernel.org>,
Kees Cook <keescook@...omium.org>,
Jann Horn <jannh@...gle.com>, linaro-mm-sig@...ts.linaro.org,
linux-fsdevel <linux-fsdevel@...r.kernel.org>,
Bernd Edlinger <bernd.edlinger@...mail.de>,
Suren Baghdasaryan <surenb@...gle.com>,
Alexey Gladkov <gladkov.alexey@...il.com>,
kernel list <linux-kernel@...r.kernel.org>,
Minchan Kim <minchan@...nel.org>,
Yafang Shao <laoar.shao@...il.com>,
"Eric W. Biederman" <ebiederm@...ssion.com>,
Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
Hridya Valsaraju <hridya@...gle.com>,
Andrew Morton <akpm@...ux-foundation.org>,
Linux API <linux-api@...r.kernel.org>
Subject: Re: [PATCH] procfs/dmabuf: Add /proc/<pid>/task/<tid>/dmabuf_fds
On Fri, Jan 29, 2021 at 03:22:06PM +0100, Christian König wrote:
> Am 29.01.21 um 15:17 schrieb Simon Ser:
> > On Friday, January 29th, 2021 at 3:13 PM, Pekka Paalanen <ppaalanen@...il.com> wrote:
> >
> > > > Re-importing it adds quite a huge CPU overhead to both userspace as well
> > > > as the kernel.
> > > Perhaps, but so far it seems no-one has noticed the overhead, with Mesa
> > > at least.
> > >
> > > I happily stand corrected.
> > Note, all of this doesn't mean that compositors will stop keeping
> > DMA-BUF FDs around. They may want to keep them open for other purposes
> > like importing them into KMS or other EGL displays as needed.
>
> Correct and that's a perfectly valid use case. Just re-importing it on every
> frame is something we should really try to avoid.
>
> At least with debugging enabled it's massive overhead and maybe even
> performance penalty when we have to re-create device page tables all the
> time.
>
> But thinking more about that it is possible that we short-cut this step as
> long as the original import was still referenced. Otherwise we probably
> would have noticed this much earlier.
Yeah kernel keeps lots of caches around and just gives you back the
previous buffer if it's still around. Still probably not the smartest
idea.
-Daniel
--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
Powered by blists - more mailing lists