From: "Gupta, Sourab" Subject: Re: [PATCH 1/6] drm/i915/vlv: Added a rendering specific Hw WA 'WaTlbInvalidateStoreDataBefore' Date: Mon, 24 Mar 2014 11:20:40 +0000 Message-ID: <1395660076.19358.75.camel@sourabgu-desktop> References: <1395643764-24353-1-git-send-email-sourab.gupta@intel.com> <1395643764-24353-2-git-send-email-sourab.gupta@intel.com> <20140324093201.GK4366@nuc-i3427.alporthouse.com> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Return-path: Received: from mga02.intel.com (mga02.intel.com [134.134.136.20]) by gabe.freedesktop.org (Postfix) with ESMTP id C97656E97B for ; Mon, 24 Mar 2014 04:20:44 -0700 (PDT) In-Reply-To: <20140324093201.GK4366@nuc-i3427.alporthouse.com> Content-Language: en-US Content-ID: <9D52026DBE1C184DA58440F5A1677427@intel.com> List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: intel-gfx-bounces@lists.freedesktop.org Sender: "Intel-gfx" To: Chris Wilson Cc: "Goel, Akash" , "intel-gfx@lists.freedesktop.org" List-Id: intel-gfx@lists.freedesktop.org On Mon, 2014-03-24 at 09:32 +0000, Chris Wilson wrote: > On Mon, Mar 24, 2014 at 12:19:19PM +0530, sourab.gupta@intel.com wrote: > > From: Akash Goel > > > > Added a new rendering specific Workaround 'WaTlbInvalidateStoreDataBefore'. > > In this WA, before pipecontrol with TLB invalidate set, need to add 2 MI > > Store data commands. > > > > Signed-off-by: Akash Goel > > Signed-off-by: Sourab Gupta > > --- > > drivers/gpu/drm/i915/intel_ringbuffer.c | 22 ++++++++++++++++++++++ > > 1 file changed, 22 insertions(+) > > > > diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c > > index 87d1a2d..2812384 100644 > > --- a/drivers/gpu/drm/i915/intel_ringbuffer.c > > +++ b/drivers/gpu/drm/i915/intel_ringbuffer.c > > @@ -2207,6 +2207,28 @@ intel_ring_invalidate_all_caches(struct intel_ring_buffer *ring) > > uint32_t flush_domains; > > int ret; > > > > + if (IS_VALLEYVIEW(ring->dev)) { > The ring flushes are vfuncs, so why is this here and not in a special > vlv ring flush? Yes, we can as well put it in the platform specific vlv flush. Since we apply this WA only for invalidate_all_caches function, we have to differentiate in the vlv flush function regarding where the flush originated from. For this we plan to check the 'invalidate_domains' field of flush function. (This field will be non-zero in case the call originated from invalidate_all_caches function). So, we'll have a vlv_render_ring_flush something like this: if(invalidate_domains) apply_our_wa; gen7_render_ring_flush(); Does this look okay? Regards, Sourab > > > + /* > > + * WaTlbInvalidateStoreDataBefore:vlv > > + * Before pipecontrol with TLB invalidate set, need 2 store > > + * data commands (such as MI_STORE_DATA_IMM or MI_STORE_DATA_INDEX) > > + * Without this, hardware cannot guarantee the command after the > > + * PIPE_CONTROL with TLB inv will not use the old TLB values. > > Crumbs, it sounds like our i-g-t are not sensitive enough. This bug > crops up in many disguises over the years, do you have any suggestion on > how we can improve our tests? > We'll think of how to capture the scenario in the i-g-t testcases and come back with suggestions. > > + */ > > + int i; > > + ret = intel_ring_begin(ring, 4 * 2); > > This can be we written to use 6 dwords. > Agreed. We'll have this in our next version > > + if (ret) > > + return ret; > > + for (i = 0; i < 2; i++) { > > + intel_ring_emit(ring, MI_STORE_DWORD_INDEX); > > + intel_ring_emit(ring, I915_GEM_HWS_SCRATCH_INDEX << > > + MI_STORE_DWORD_INDEX_SHIFT); > > This is I915_GEM_HWS_SCRATCH_ADDR Agreed. We'll have this in our next version > > > + intel_ring_emit(ring, 0); > > + intel_ring_emit(ring, MI_NOOP); > > + } > > + intel_ring_advance(ring); > > + } > > + > > flush_domains = 0; > > if (ring->gpu_caches_dirty) > > flush_domains = I915_GEM_GPU_DOMAINS; >