Re: [PATCH 4/4] drm/i915: Get rid of acthd based guilty batch search

From: Chris Wilson <chris@chris-wilson.co.uk>
To: Ben Widawsky <ben@bwidawsk.net>
Cc: intel-gfx@lists.freedesktop.org, miku@iki.fi
Subject: Re: [PATCH 4/4] drm/i915: Get rid of acthd based guilty batch search
Date: Thu, 30 Jan 2014 12:59:32 +0000	[thread overview]
Message-ID: <20140130125932.GH29091@nuc-i3427.alporthouse.com> (raw)
In-Reply-To: <20140129214428.GD32231@bwidawsk.net>

On Wed, Jan 29, 2014 at 01:44:34PM -0800, Ben Widawsky wrote:
> Last word: As I've discussed with Chris too, I am overall a bit wary of
> removing any use upon hardware for doing a lot of these error triage,
> detection and collection functions. I really like that no matter how
> bonghits our driver gets, we can read certain registers to try to figure
> things out. I say this now since I think after this series I will no
> longer have a leg to stand on in the, we shouldn't use requests for
> error collection, discussion. Thanks for reading my rant.

I know exactly what you mean. But without a complete snapshot of the
hardware and memory (including old contents of overwritten buffers) we
will always be missing something when we come to post-mortem debugging.

I trust tracking activity by seqno though, and in the error states where
that itself has been erroneous the fault has stood out (and had been the
root cause of the hang anyway). Hence my feeling that it is exactly what
I want for port-mortem debugging.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre