From patchwork Tue Oct 9 08:38:58 2012 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Chris Wilson X-Patchwork-Id: 1569171 Return-Path: X-Original-To: patchwork-intel-gfx@patchwork.kernel.org Delivered-To: patchwork-process-083081@patchwork2.kernel.org Received: from gabe.freedesktop.org (gabe.freedesktop.org [131.252.210.177]) by patchwork2.kernel.org (Postfix) with ESMTP id 91700DFFAD for ; Tue, 9 Oct 2012 08:39:30 +0000 (UTC) Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id B03A6A0951 for ; Tue, 9 Oct 2012 01:39:29 -0700 (PDT) X-Original-To: intel-gfx@lists.freedesktop.org Delivered-To: intel-gfx@lists.freedesktop.org Received: from fireflyinternet.com (smtp.fireflyinternet.com [109.228.6.236]) by gabe.freedesktop.org (Postfix) with ESMTP id 7523CA0895 for ; Tue, 9 Oct 2012 01:39:18 -0700 (PDT) X-Default-Received-SPF: pass (skip=forwardok (res=PASS)) x-ip-name=78.156.73.22; Received: from arrandale.alporthouse.com (unverified [78.156.73.22]) by fireflyinternet.com (Firefly Internet SMTP) with ESMTP id 123739015-1500050 for multiple; Tue, 09 Oct 2012 09:39:06 +0100 From: Chris Wilson To: intel-gfx@lists.freedesktop.org Date: Tue, 9 Oct 2012 09:38:58 +0100 Message-Id: <1349771938-18887-1-git-send-email-chris@chris-wilson.co.uk> X-Mailer: git-send-email 1.7.10.4 X-Originating-IP: 78.156.73.22 Subject: [Intel-gfx] [PATCH] drm/i915: Only insert the mb() before updating the fence parameter X-BeenThere: intel-gfx@lists.freedesktop.org X-Mailman-Version: 2.1.13 Precedence: list List-Id: Intel graphics driver community testing & development List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , MIME-Version: 1.0 Sender: intel-gfx-bounces+patchwork-intel-gfx=patchwork.kernel.org@lists.freedesktop.org Errors-To: intel-gfx-bounces+patchwork-intel-gfx=patchwork.kernel.org@lists.freedesktop.org With a fence, we only need to insert a memory barrier around the actual fence alteration for CPU accesses through the GTT. Performing the barrier in flush-fence was inserting unnecessary and expensive barriers for never fenced objects. Note removing the barriers from flush-fence, which was effectively a barrier before every direct access through the GTT, revealed that we where missing a barrier before the first access through the GTT. Lack of that barrier was sufficient to cause GPU hangs. Signed-off-by: Chris Wilson Reviewed-by: Daniel Vetter --- drivers/gpu/drm/i915/i915_gem.c | 33 +++++++++++++++++++++++---------- 1 file changed, 23 insertions(+), 10 deletions(-) diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c index 05ff790..c1ef0a8 100644 --- a/drivers/gpu/drm/i915/i915_gem.c +++ b/drivers/gpu/drm/i915/i915_gem.c @@ -2771,9 +2771,22 @@ static void i830_write_fence_reg(struct drm_device *dev, int reg, POSTING_READ(FENCE_REG_830_0 + reg * 4); } +inline static bool i915_gem_object_needs_mb(struct drm_i915_gem_object *obj) +{ + return obj && obj->base.read_domains & I915_GEM_DOMAIN_GTT; +} + static void i915_gem_write_fence(struct drm_device *dev, int reg, struct drm_i915_gem_object *obj) { + struct drm_i915_private *dev_priv = dev->dev_private; + + /* Ensure that all CPU reads are completed before installing a fence + * and all writes before removing the fence. + */ + if (i915_gem_object_needs_mb(dev_priv->fence_regs[reg].obj)) + mb(); + switch (INTEL_INFO(dev)->gen) { case 7: case 6: sandybridge_write_fence_reg(dev, reg, obj); break; @@ -2783,6 +2796,9 @@ static void i915_gem_write_fence(struct drm_device *dev, int reg, case 2: i830_write_fence_reg(dev, reg, obj); break; default: break; } + + if (i915_gem_object_needs_mb(obj)) + mb(); } static inline int fence_number(struct drm_i915_private *dev_priv, @@ -2812,7 +2828,7 @@ static void i915_gem_object_update_fence(struct drm_i915_gem_object *obj, } static int -i915_gem_object_flush_fence(struct drm_i915_gem_object *obj) +i915_gem_object_wait_fence(struct drm_i915_gem_object *obj) { if (obj->last_fenced_seqno) { int ret = i915_wait_seqno(obj->ring, obj->last_fenced_seqno); @@ -2822,12 +2838,6 @@ i915_gem_object_flush_fence(struct drm_i915_gem_object *obj) obj->last_fenced_seqno = 0; } - /* Ensure that all CPU reads are completed before installing a fence - * and all writes before removing the fence. - */ - if (obj->base.read_domains & I915_GEM_DOMAIN_GTT) - mb(); - obj->fenced_gpu_access = false; return 0; } @@ -2838,7 +2848,7 @@ i915_gem_object_put_fence(struct drm_i915_gem_object *obj) struct drm_i915_private *dev_priv = obj->base.dev->dev_private; int ret; - ret = i915_gem_object_flush_fence(obj); + ret = i915_gem_object_wait_fence(obj); if (ret) return ret; @@ -2912,7 +2922,7 @@ i915_gem_object_get_fence(struct drm_i915_gem_object *obj) * will need to serialise the write to the associated fence register? */ if (obj->fence_dirty) { - ret = i915_gem_object_flush_fence(obj); + ret = i915_gem_object_wait_fence(obj); if (ret) return ret; } @@ -2933,7 +2943,7 @@ i915_gem_object_get_fence(struct drm_i915_gem_object *obj) if (reg->obj) { struct drm_i915_gem_object *old = reg->obj; - ret = i915_gem_object_flush_fence(old); + ret = i915_gem_object_wait_fence(old); if (ret) return ret; @@ -3244,6 +3254,9 @@ i915_gem_object_set_to_gtt_domain(struct drm_i915_gem_object *obj, bool write) i915_gem_object_flush_cpu_write_domain(obj); + if ((obj->base.read_domains & I915_GEM_DOMAIN_GTT) == 0) + mb(); + old_write_domain = obj->base.write_domain; old_read_domains = obj->base.read_domains;