Revert "drm/i915: Avoid using PIPE_CONTROL on Ironlake"

Restore PIPE_CONTROL once again just for Ironlake, as it appears that
MI_USER_INTERRUPT does not have the same coherency guarantees, that is
on Ironlake the interrupt following a GPU write is not guaranteed to
arrive after the write is coherent from the CPU, as it does on the
other generations.

Reported-by: Zhenyu Wang <zhenyuw@linux.intel.com>
Reported-by: Shuang He <shuang.he@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=32402
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

drivers/gpu/drm/i915/i915_irq.c

@@ -349,7 +349,7 @@ static irqreturn_t ironlake_irq_handler(struct drm_device *dev)
 				READ_BREADCRUMB(dev_priv);
 	}
 
-	if (gt_iir & GT_USER_INTERRUPT)
+	if (gt_iir & (GT_USER_INTERRUPT | GT_PIPE_NOTIFY))
 		notify_ring(dev, &dev_priv->ring[RCS]);
 	if (gt_iir & bsd_usr_interrupt)
 		notify_ring(dev, &dev_priv->ring[VCS]);
@@ -1558,6 +1558,7 @@ static int ironlake_irq_postinstall(struct drm_device *dev)
 	else
 		render_irqs =
 			GT_USER_INTERRUPT |
+			GT_PIPE_NOTIFY |
 			GT_BSD_USER_INTERRUPT;
 	I915_WRITE(GTIER, render_irqs);
 	POSTING_READ(GTIER);

drivers/gpu/drm/i915/intel_ringbuffer.c

@@ -209,6 +209,78 @@ static int init_ring_common(struct intel_ring_buffer *ring)
 	return 0;
 }
 
+/*
+ * 965+ support PIPE_CONTROL commands, which provide finer grained control
+ * over cache flushing.
+ */
+struct pipe_control {
+	struct drm_i915_gem_object *obj;
+	volatile u32 *cpu_page;
+	u32 gtt_offset;
+};
+
+static int
+init_pipe_control(struct intel_ring_buffer *ring)
+{
+	struct pipe_control *pc;
+	struct drm_i915_gem_object *obj;
+	int ret;
+
+	if (ring->private)
+		return 0;
+
+	pc = kmalloc(sizeof(*pc), GFP_KERNEL);
+	if (!pc)
+		return -ENOMEM;
+
+	obj = i915_gem_alloc_object(ring->dev, 4096);
+	if (obj == NULL) {
+		DRM_ERROR("Failed to allocate seqno page\n");
+		ret = -ENOMEM;
+		goto err;
+	}
+	obj->agp_type = AGP_USER_CACHED_MEMORY;
+
+	ret = i915_gem_object_pin(obj, 4096, true);
+	if (ret)
+		goto err_unref;
+
+	pc->gtt_offset = obj->gtt_offset;
+	pc->cpu_page =  kmap(obj->pages[0]);
+	if (pc->cpu_page == NULL)
+		goto err_unpin;
+
+	pc->obj = obj;
+	ring->private = pc;
+	return 0;
+
+err_unpin:
+	i915_gem_object_unpin(obj);
+err_unref:
+	drm_gem_object_unreference(&obj->base);
+err:
+	kfree(pc);
+	return ret;
+}
+
+static void
+cleanup_pipe_control(struct intel_ring_buffer *ring)
+{
+	struct pipe_control *pc = ring->private;
+	struct drm_i915_gem_object *obj;
+
+	if (!ring->private)
+		return;
+
+	obj = pc->obj;
+	kunmap(obj->pages[0]);
+	i915_gem_object_unpin(obj);
+	drm_gem_object_unreference(&obj->base);
+
+	kfree(pc);
+	ring->private = NULL;
+}
+
 static int init_render_ring(struct intel_ring_buffer *ring)
 {
 	struct drm_device *dev = ring->dev;
@@ -222,9 +294,24 @@ static int init_render_ring(struct intel_ring_buffer *ring)
 		I915_WRITE(MI_MODE, mode);
 	}
 
+	if (INTEL_INFO(dev)->gen >= 6) {
+	} else if (IS_GEN5(dev)) {
+		ret = init_pipe_control(ring);
+		if (ret)
+			return ret;
+	}
+
 	return ret;
 }
 
+static void render_ring_cleanup(struct intel_ring_buffer *ring)
+{
+	if (!ring->private)
+		return;
+
+	cleanup_pipe_control(ring);
+}
+
 static void
 update_semaphore(struct intel_ring_buffer *ring, int i, u32 seqno)
 {
@@ -299,6 +386,65 @@ intel_ring_sync(struct intel_ring_buffer *ring,
 	return 0;
 }
 
+#define PIPE_CONTROL_FLUSH(ring__, addr__)					\
+do {									\
+	intel_ring_emit(ring__, GFX_OP_PIPE_CONTROL | PIPE_CONTROL_QW_WRITE |		\
+		 PIPE_CONTROL_DEPTH_STALL | 2);				\
+	intel_ring_emit(ring__, (addr__) | PIPE_CONTROL_GLOBAL_GTT);			\
+	intel_ring_emit(ring__, 0);							\
+	intel_ring_emit(ring__, 0);							\
+} while (0)
+
+static int
+pc_render_add_request(struct intel_ring_buffer *ring,
+		      u32 *result)
+{
+	struct drm_device *dev = ring->dev;
+	u32 seqno = i915_gem_get_seqno(dev);
+	struct pipe_control *pc = ring->private;
+	u32 scratch_addr = pc->gtt_offset + 128;
+	int ret;
+
+	/* For Ironlake, MI_USER_INTERRUPT was deprecated and apparently
+	 * incoherent with writes to memory, i.e. completely fubar,
+	 * so we need to use PIPE_NOTIFY instead.
+	 *
+	 * However, we also need to workaround the qword write
+	 * incoherence by flushing the 6 PIPE_NOTIFY buffers out to
+	 * memory before requesting an interrupt.
+	 */
+	ret = intel_ring_begin(ring, 32);
+	if (ret)
+		return ret;
+
+	intel_ring_emit(ring, GFX_OP_PIPE_CONTROL | PIPE_CONTROL_QW_WRITE |
+			PIPE_CONTROL_WC_FLUSH | PIPE_CONTROL_TC_FLUSH);
+	intel_ring_emit(ring, pc->gtt_offset | PIPE_CONTROL_GLOBAL_GTT);
+	intel_ring_emit(ring, seqno);
+	intel_ring_emit(ring, 0);
+	PIPE_CONTROL_FLUSH(ring, scratch_addr);
+	scratch_addr += 128; /* write to separate cachelines */
+	PIPE_CONTROL_FLUSH(ring, scratch_addr);
+	scratch_addr += 128;
+	PIPE_CONTROL_FLUSH(ring, scratch_addr);
+	scratch_addr += 128;
+	PIPE_CONTROL_FLUSH(ring, scratch_addr);
+	scratch_addr += 128;
+	PIPE_CONTROL_FLUSH(ring, scratch_addr);
+	scratch_addr += 128;
+	PIPE_CONTROL_FLUSH(ring, scratch_addr);
+	intel_ring_emit(ring, GFX_OP_PIPE_CONTROL | PIPE_CONTROL_QW_WRITE |
+			PIPE_CONTROL_WC_FLUSH | PIPE_CONTROL_TC_FLUSH |
+			PIPE_CONTROL_NOTIFY);
+	intel_ring_emit(ring, pc->gtt_offset | PIPE_CONTROL_GLOBAL_GTT);
+	intel_ring_emit(ring, seqno);
+	intel_ring_emit(ring, 0);
+	intel_ring_advance(ring);
+
+	*result = seqno;
+	return 0;
+}
+
 static int
 render_ring_add_request(struct intel_ring_buffer *ring,
 			u32 *result)
@@ -327,6 +473,13 @@ ring_get_seqno(struct intel_ring_buffer *ring)
 	return intel_read_status_page(ring, I915_GEM_HWS_INDEX);
 }
 
+static u32
+pc_render_get_seqno(struct intel_ring_buffer *ring)
+{
+	struct pipe_control *pc = ring->private;
+	return pc->cpu_page[0];
+}
+
 static bool
 render_ring_get_irq(struct intel_ring_buffer *ring)
 {
@@ -342,7 +495,7 @@ render_ring_get_irq(struct intel_ring_buffer *ring)
 		spin_lock_irqsave(&dev_priv->irq_lock, irqflags);
 		if (HAS_PCH_SPLIT(dev))
 			ironlake_enable_graphics_irq(dev_priv,
-						     GT_USER_INTERRUPT);
+						     GT_PIPE_NOTIFY | GT_USER_INTERRUPT);
 		else
 			i915_enable_irq(dev_priv, I915_USER_INTERRUPT);
 		spin_unlock_irqrestore(&dev_priv->irq_lock, irqflags);
@@ -363,7 +516,8 @@ render_ring_put_irq(struct intel_ring_buffer *ring)
 		spin_lock_irqsave(&dev_priv->irq_lock, irqflags);
 		if (HAS_PCH_SPLIT(dev))
 			ironlake_disable_graphics_irq(dev_priv,
-						      GT_USER_INTERRUPT);
+						      GT_USER_INTERRUPT |
+						      GT_PIPE_NOTIFY);
 		else
 			i915_disable_irq(dev_priv, I915_USER_INTERRUPT);
 		spin_unlock_irqrestore(&dev_priv->irq_lock, irqflags);
@@ -776,6 +930,7 @@ static const struct intel_ring_buffer render_ring = {
 	.irq_get		= render_ring_get_irq,
 	.irq_put		= render_ring_put_irq,
 	.dispatch_execbuffer	= render_ring_dispatch_execbuffer,
+       .cleanup			= render_ring_cleanup,
 };
 
 /* ring buffer for bit-stream decoder */
@@ -1010,6 +1165,9 @@ int intel_init_render_ring_buffer(struct drm_device *dev)
 	*ring = render_ring;
 	if (INTEL_INFO(dev)->gen >= 6) {
 		ring->add_request = gen6_add_request;
+	} else if (IS_GEN5(dev)) {
+		ring->add_request = pc_render_add_request;
+		ring->get_seqno = pc_render_get_seqno;
 	}
 
 	if (!I915_NEED_GFX_HWS(dev)) {