From mboxrd@z Thu Jan 1 00:00:00 1970 From: Daniel Vetter Subject: Re: [PATCH 9/9] drm/i915: swizzling support for snb/ivb Date: Fri, 11 Nov 2011 18:22:25 +0100 Message-ID: <20111111172225.GB7295@phenom.ffwll.local> References: <1320931087-1557-1-git-send-email-daniel.vetter@ffwll.ch> <1320931087-1557-10-git-send-email-daniel.vetter@ffwll.ch> <87sjluob6x.fsf@eliezer.anholt.net> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Return-path: Received: from mail-ww0-f43.google.com (mail-ww0-f43.google.com [74.125.82.43]) by gabe.freedesktop.org (Postfix) with ESMTP id D2209A0EA9 for ; Fri, 11 Nov 2011 09:21:22 -0800 (PST) Received: by wwp14 with SMTP id 14so1952896wwp.12 for ; Fri, 11 Nov 2011 09:21:22 -0800 (PST) Content-Disposition: inline In-Reply-To: <87sjluob6x.fsf@eliezer.anholt.net> List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: intel-gfx-bounces+gcfxdi-intel-gfx=m.gmane.org@lists.freedesktop.org Errors-To: intel-gfx-bounces+gcfxdi-intel-gfx=m.gmane.org@lists.freedesktop.org To: Eric Anholt Cc: Daniel Vetter , intel-gfx List-Id: intel-gfx@lists.freedesktop.org On Fri, Nov 11, 2011 at 08:50:30AM -0800, Eric Anholt wrote: > On Thu, 10 Nov 2011 14:18:07 +0100, Daniel Vetter wrote: > > We have to do this manually. Somebody had a Great Idea. > > > > Signed-Off-by: Daniel Vetter > > People playing with this when not strictly required is scary to me. > Manually swizzling was a world of hurt. I got to play with things like > "when the management engine is enabled, it carves out the top N MB of > one of the dimms, and the corresponding N MB of the other dimm doesn't > get swizzled, and you lose". Looks like yet another patch series of mine that scares away people ... Would this patch be less scary when we have a test that slurps in the entire ram to quickly diagnose such issues? We can then either revert this or fix up the detection to not enable swizzling in such cases. Also the manually swizzling is a world of hurt argument is pretty void: Up to very recent kernels we've advertised bit9 swizzling on snb+ without any swizzling actually going on. So userspace clearly doesn't rely on this anymore (the issue was caught by running the pread tests in i-g-t). Also we already have a bug for gm45 which looks like a portion of ram isn't swizzled: https://bugs.freedesktop.org/show_bug.cgi?id=28813 -Daniel -- Daniel Vetter Mail: daniel@ffwll.ch Mobile: +41 (0)79 365 57 48