From 0201f1ecf4b81f08799b1fb9c8cdf1125b9b78a6 Mon Sep 17 00:00:00 2001
From: Chris Wilson <chris@chris-wilson.co.uk>
Date: Fri, 20 Jul 2012 12:41:01 +0100
Subject: drm/i915: Replace the pending_gpu_write flag with an explicit seqno
Git-commit: 0201f1ecf4b81f08799b1fb9c8cdf1125b9b78a6
Patch-mainline: v3.7-rc1

As we always flush the GPU cache prior to emitting the breadcrumb, we no
longer have to worry about the deferred flush causing the
pending_gpu_write to be delayed. So we can instead utilize the known
last_write_seqno to hopefully minimise the wait times.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Acked-by: Michal Srb <msrb@suse.com>
---
 drivers/gpu/drm/i915/i915_debugfs.c        |    9 ++-
 drivers/gpu/drm/i915/i915_drv.h            |   12 +----
 drivers/gpu/drm/i915/i915_gem.c            |   66 +++++++++++++++++------------
 drivers/gpu/drm/i915/i915_gem_execbuffer.c |    2 
 drivers/gpu/drm/i915/i915_irq.c            |    5 +-
 5 files changed, 51 insertions(+), 43 deletions(-)

--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -120,14 +120,15 @@ static const char *cache_level_str(int t
 static void
 describe_obj(struct seq_file *m, struct drm_i915_gem_object *obj)
 {
-	seq_printf(m, "%pK: %s%s %8zdKiB %04x %04x %d %d%s%s%s",
+	seq_printf(m, "%pK: %s%s %8zdKiB %04x %04x %d %d %d%s%s%s",
 		   &obj->base,
 		   get_pin_flag(obj),
 		   get_tiling_flag(obj),
 		   obj->base.size / 1024,
 		   obj->base.read_domains,
 		   obj->base.write_domain,
-		   obj->last_rendering_seqno,
+		   obj->last_read_seqno,
+		   obj->last_write_seqno,
 		   obj->last_fenced_seqno,
 		   cache_level_str(obj->cache_level),
 		   obj->dirty ? " dirty" : "",
@@ -629,12 +630,12 @@ static void print_error_buffers(struct s
 	seq_printf(m, "%s [%d]:\n", name, count);
 
 	while (count--) {
-		seq_printf(m, "  %08x %8u %04x %04x %08x%s%s%s%s%s%s%s",
+		seq_printf(m, "  %08x %8u %04x %04x %x %x%s%s%s%s%s%s%s",
 			   err->gtt_offset,
 			   err->size,
 			   err->read_domains,
 			   err->write_domain,
-			   err->seqno,
+			   err->rseqno, err->wseqno,
 			   pin_flag(err->pinned),
 			   tiling_flag(err->tiling),
 			   dirty_flag(err->dirty),
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -221,7 +221,7 @@ struct drm_i915_error_state {
 	struct drm_i915_error_buffer {
 		u32 size;
 		u32 name;
-		u32 seqno;
+		u32 rseqno, wseqno;
 		u32 gtt_offset;
 		u32 read_domains;
 		u32 write_domain;
@@ -895,12 +895,6 @@ struct drm_i915_gem_object {
 	unsigned int dirty:1;
 
 	/**
-	 * This is set if the object has been written to since the last
-	 * GPU flush.
-	 */
-	unsigned int pending_gpu_write:1;
-
-	/**
 	 * Fence register bits (if any) for this object.  Will be set
 	 * as needed when mapped into the GTT.
 	 * Protected by dev->struct_mutex.
@@ -992,7 +986,8 @@ struct drm_i915_gem_object {
 	struct intel_ring_buffer *ring;
 
 	/** Breadcrumb of last rendering to the buffer. */
-	uint32_t last_rendering_seqno;
+	uint32_t last_read_seqno;
+	uint32_t last_write_seqno;
 	/** Breadcrumb of last fenced GPU access to the buffer. */
 	uint32_t last_fenced_seqno;
 
@@ -1291,7 +1286,6 @@ void i915_gem_lastclose(struct drm_devic
 int i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj,
 				  gfp_t gfpmask);
 int __must_check i915_mutex_lock_interruptible(struct drm_device *dev);
-int __must_check i915_gem_object_wait_rendering(struct drm_i915_gem_object *obj);
 int i915_gem_object_sync(struct drm_i915_gem_object *obj,
 			 struct intel_ring_buffer *to);
 void i915_gem_object_move_to_active(struct drm_i915_gem_object *obj,
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -1448,7 +1448,7 @@ i915_gem_object_move_to_active(struct dr
 	list_move_tail(&obj->mm_list, &dev_priv->mm.active_list);
 	list_move_tail(&obj->ring_list, &ring->active_list);
 
-	obj->last_rendering_seqno = seqno;
+	obj->last_read_seqno = seqno;
 
 	if (obj->fenced_gpu_access) {
 		obj->last_fenced_seqno = seqno;
@@ -1468,7 +1468,8 @@ static void
 i915_gem_object_move_off_active(struct drm_i915_gem_object *obj)
 {
 	list_del_init(&obj->ring_list);
-	obj->last_rendering_seqno = 0;
+	obj->last_read_seqno = 0;
+	obj->last_write_seqno = 0;
 	obj->last_fenced_seqno = 0;
 }
 
@@ -1500,7 +1501,6 @@ i915_gem_object_move_to_inactive(struct
 	obj->fenced_gpu_access = false;
 
 	obj->active = 0;
-	obj->pending_gpu_write = false;
 	drm_gem_object_unreference(&obj->base);
 
 	WARN_ON(i915_verify_lists(dev));
@@ -1819,7 +1819,7 @@ i915_gem_retire_requests_ring(struct int
 				      struct drm_i915_gem_object,
 				      ring_list);
 
-		if (!i915_seqno_passed(seqno, obj->last_rendering_seqno))
+		if (!i915_seqno_passed(seqno, obj->last_read_seqno))
 			break;
 
 		if (obj->base.write_domain != 0)
@@ -2043,9 +2043,11 @@ i915_wait_seqno(struct intel_ring_buffer
  * Ensures that all rendering to the object has completed and the object is
  * safe to unbind from the GTT or access from the CPU.
  */
-int
-i915_gem_object_wait_rendering(struct drm_i915_gem_object *obj)
+static __must_check int
+i915_gem_object_wait_rendering(struct drm_i915_gem_object *obj,
+			       bool readonly)
 {
+	u32 seqno;
 	int ret;
 
 	/* This function only exists to support waiting for existing rendering,
@@ -2056,13 +2058,27 @@ i915_gem_object_wait_rendering(struct dr
 	/* If there is rendering queued on the buffer being evicted, wait for
 	 * it.
 	 */
-	if (obj->active) {
-		ret = i915_wait_seqno(obj->ring, obj->last_rendering_seqno);
-		if (ret)
-			return ret;
-		i915_gem_retire_requests_ring(obj->ring);
+	if (readonly)
+		seqno = obj->last_write_seqno;
+	else
+		seqno = obj->last_read_seqno;
+	if (seqno == 0)
+		return 0;
+
+	ret = i915_wait_seqno(obj->ring, seqno);
+	if (ret)
+		return ret;
+
+	/* Manually manage the write flush as we may have not yet retired
+	 * the buffer.
+	 */
+	if (obj->last_write_seqno &&
+	    i915_seqno_passed(seqno, obj->last_write_seqno)) {
+		obj->last_write_seqno = 0;
+		obj->base.write_domain &= ~I915_GEM_GPU_DOMAINS;
 	}
 
+	i915_gem_retire_requests_ring(obj->ring);
 	return 0;
 }
 
@@ -2081,10 +2097,10 @@ i915_gem_object_flush_active(struct drm_
 		if (ret)
 			return ret;
 
-		ret = i915_gem_check_olr(obj->ring,
-					 obj->last_rendering_seqno);
+		ret = i915_gem_check_olr(obj->ring, obj->last_read_seqno);
 		if (ret)
 			return ret;
+
 		i915_gem_retire_requests_ring(obj->ring);
 	}
 
@@ -2144,7 +2160,7 @@ i915_gem_wait_ioctl(struct drm_device *d
 		goto out;
 
 	if (obj->active) {
-		seqno = obj->last_rendering_seqno;
+		seqno = obj->last_read_seqno;
 		ring = obj->ring;
 	}
 
@@ -2199,11 +2215,11 @@ i915_gem_object_sync(struct drm_i915_gem
 		return 0;
 
 	if (to == NULL || !i915_semaphore_is_enabled(obj->base.dev))
-		return i915_gem_object_wait_rendering(obj);
+		return i915_gem_object_wait_rendering(obj, false);
 
 	idx = intel_ring_sync_index(from, to);
 
-	seqno = obj->last_rendering_seqno;
+	seqno = obj->last_read_seqno;
 	if (seqno <= from->sync_seqno[idx])
 		return 0;
 
@@ -2947,11 +2963,9 @@ i915_gem_object_set_to_gtt_domain(struct
 	if (ret)
 		return ret;
 
-	if (obj->pending_gpu_write || write) {
-		ret = i915_gem_object_wait_rendering(obj);
-		if (ret)
-			return ret;
-	}
+	ret = i915_gem_object_wait_rendering(obj, !write);
+	if (ret)
+		return ret;
 
 	i915_gem_object_flush_cpu_write_domain(obj);
 
@@ -3122,7 +3136,7 @@ i915_gem_object_finish_gpu(struct drm_i9
 			return ret;
 	}
 
-	ret = i915_gem_object_wait_rendering(obj);
+	ret = i915_gem_object_wait_rendering(obj, false);
 	if (ret)
 		return ret;
 
@@ -3150,11 +3164,9 @@ i915_gem_object_set_to_cpu_domain(struct
 	if (ret)
 		return ret;
 
-	if (write || obj->pending_gpu_write) {
-		ret = i915_gem_object_wait_rendering(obj);
-		if (ret)
-			return ret;
-	}
+	ret = i915_gem_object_wait_rendering(obj, !write);
+	if (ret)
+		return ret;
 
 	i915_gem_object_flush_gtt_write_domain(obj);
 
--- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
@@ -980,7 +980,7 @@ i915_gem_execbuffer_move_to_active(struc
 		i915_gem_object_move_to_active(obj, ring, seqno);
 		if (obj->base.write_domain) {
 			obj->dirty = 1;
-			obj->pending_gpu_write = true;
+			obj->last_write_seqno = seqno;
 			list_move_tail(&obj->gpu_write_list,
 				       &ring->gpu_write_list);
 			if (obj->pin_count) /* check for potential scanout */
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -949,7 +949,8 @@ static void capture_bo(struct drm_i915_e
 {
 	err->size = obj->base.size;
 	err->name = obj->base.name;
-	err->seqno = obj->last_rendering_seqno;
+	err->rseqno = obj->last_read_seqno;
+	err->wseqno = obj->last_write_seqno;
 	err->gtt_offset = obj->gtt_offset;
 	err->read_domains = obj->base.read_domains;
 	err->write_domain = obj->base.write_domain;
@@ -1044,7 +1045,7 @@ i915_error_first_batchbuffer(struct drm_
 		if (obj->ring != ring)
 			continue;
 
-		if (i915_seqno_passed(seqno, obj->last_rendering_seqno))
+		if (i915_seqno_passed(seqno, obj->last_read_seqno))
 			continue;
 
 		if ((obj->base.read_domains & I915_GEM_DOMAIN_COMMAND) == 0)
