Re: [PATCH 1/3] drm/i915: Use pagecache write to prepopulate shmemfs from pwrite-ioctl

From: Chris Wilson <chris@chris-wilson.co.uk>
To: Petri Latvala <petri.latvala@intel.com>
Cc: intel-gfx@lists.freedesktop.org
Subject: Re: [PATCH 1/3] drm/i915: Use pagecache write to prepopulate shmemfs from pwrite-ioctl
Date: Tue, 7 Mar 2017 10:22:09 +0000	[thread overview]
Message-ID: <20170307102209.GO3201@nuc-i3427.alporthouse.com> (raw)
In-Reply-To: <20170307095520.s7e6aodeircwbbyo@platvala-desk.ger.corp.intel.com>

On Tue, Mar 07, 2017 at 11:55:20AM +0200, Petri Latvala wrote:
> 
> For the record, manually launched extended test run on SKL
> resulted in:
> 
> commit f88cf1067cc499f4c9236c4e3ff7e410f9cc76d7
> Author: Chris Wilson <chris@chris-wilson.co.uk>
> Date:   Sat Mar 4 10:35:32 2017 +0000
> 
>     drm/i915: Use pagecache write to prepopulate shmemfs from pwrite-ioctl
> 
> 
> igt@gem_madvise@dontneed-before-exec: pass -> {'fail', 'incomplete'}
That test is expected to fail, that it ever passed is a fluke.
incomplete there should be a failure in the runner? That should quite
happily spot the purged object in execbuf.

> igt@gem_madvise@dontneed-before-pwrite: pass -> fail
Hmm. Here we are relying on shmemfs reporting the write to the truncated
object as invalid. Looks like we need to do the filtering ourselves. It
is not a huge problem as we still prevent any later instantiation of
pages so just a temporary slip. Setting obj->mm.pages = -EFAULT; is
likely a fun solution.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx