linux-kernel - Re: [PATCH 3/3] dma-buf/fence: add signal_on

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <20160523140522.GD27098@phenom.ffwll.local>
Date:	Mon, 23 May 2016 16:05:22 +0200
From:	Daniel Vetter <daniel@...ll.ch>
To:	Christian König <deathsimple@...afone.de>
Cc:	dri-devel@...ts.freedesktop.org, linux-kernel@...r.kernel.org,
	gustavo@...ovan.org, chris@...is-wilson.co.uk
Subject: Re: [PATCH 3/3] dma-buf/fence: add signal_on_any to the fence array

On Mon, May 23, 2016 at 03:58:41PM +0200, Christian König wrote:
> Am 23.05.2016 um 15:50 schrieb Daniel Vetter:
> >On Mon, May 23, 2016 at 01:25:22PM +0200, Christian König wrote:
> >>From: Christian König <christian.koenig@....com>
> >>
> >>If @signal_on_any is true the fence array signals if any fence in the array
> >>signals, otherwise it signals when all fences in the array signal.
> >What do you need this for? Sounds neat, but I can't come up with a use
> >case. And that use case imo should be both in the commit message and
> >kerneldoc.
> 
> Good point, going to fix this when if we really use the patch. Just wanted
> to get opinions on this.

For getting opinions on this I get a bit the impression we're slightly
talking past each another. Having the actual user of these functions handy
should help a lot for this discussion.

> I'm currently working on making some of the resource management in amdgpu
> more friendly to our GPU scheduler.
> 
> Waiting for resources in the GPU scheduler is implemented as waiting for a
> fence to fire. So when you want to wait for any resource to become available
> (like in our case the VMIDs) you could make good use of something like this.

Hm, ok, if you want a fence that fires when any resource becomes available
of a given type becomes available that might be useful. But how do you
handle the next work item that needs such a resource? Add _all_ previous
fences that use it (there's not much ordering with the scheduler) and let
them race, and retry if they fail?
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch