From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jani Nikula Subject: Re: [PATCH] drm/i915/vlv: Added/removed Render specific Hw Workarounds for VLV Date: Mon, 20 Jan 2014 10:19:48 +0200 Message-ID: <87r4836p5n.fsf@intel.com> References: <1389268309-13555-1-git-send-email-akash.goel@intel.com> <871u0hbatn.fsf@intel.com> <8BF5CF93467D8C498F250C96583BC09CC59765@BGSMSX103.gar.corp.intel.com> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Return-path: Received: from mga09.intel.com (mga09.intel.com [134.134.136.24]) by gabe.freedesktop.org (Postfix) with ESMTP id 0040CFACD4 for ; Mon, 20 Jan 2014 00:16:40 -0800 (PST) In-Reply-To: <8BF5CF93467D8C498F250C96583BC09CC59765@BGSMSX103.gar.corp.intel.com> List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: intel-gfx-bounces@lists.freedesktop.org Errors-To: intel-gfx-bounces@lists.freedesktop.org To: "Goel, Akash" , "intel-gfx@lists.freedesktop.org" List-Id: intel-gfx@lists.freedesktop.org On Sun, 19 Jan 2014, "Goel, Akash" wrote: >>> Frankly, I don't know much about these particular workarounds, but >>> for bisecting any regressions it would probably be a good idea to >>> split the patch per workaround touched. > Sorry for late response on this, actually we have done sufficient > testing for this patch. I'm happy to hear that. However no reasonable amount of testing will cover the astonishing variety of conditions this code will run under out there in the real world, and if something blows up, we might never be able reproduce it ourselves. > Moreover these workarounds are applicable to VLV only & will not > affect any other platforms. Really? WaReadAfterWriteHazard in intel_ring_flush_all_caches() and WaTlbInvalidateStoreDataBefore in intel_ring_invalidate_all_caches()? Btw our convention is to append the affected platforms to the w/a names in the code; please look around for examples. > So probably it can be pushed as it is without splitting. My opinion still stands. I think all workarounds are fragile by definition, and should be separate patches. Quoting Daniel, "if a regression would bisect to this, and the bisect is the only useful piece of evidence, would I stand a chance to understand it?" BR, Jani. -- Jani Nikula, Intel Open Source Technology Center