[PATCH 21/26] drm/i915: Add support for asynchronous moving fence waiting

[Maarten Lankhorst]

For now, we will only allow async migration when TTM is used,
so the paths we care about are related to TTM.

The mmap path is handled by having the fence in ttm_bo->moving,
when pinning, the binding only becomes available after the moving
fence is signaled, and pinning a cpu map will only work after
the moving fence signals.

This should close all holes where userspace can read a buffer
before it's fully migrated.

Signed-off-by: Maarten Lankhorst <maarten.lankhorst at linux.intel.com>
---
 drivers/gpu/drm/i915/gem/i915_gem_pages.c | 10 ++++++++++
 drivers/gpu/drm/i915/i915_vma.c           | 10 ++++++++--
 2 files changed, 18 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/i915/gem/i915_gem_pages.c b/drivers/gpu/drm/i915/gem/i915_gem_pages.c
index cec1d0438ab3..482d5cfaae69 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_pages.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_pages.c
@@ -349,6 +349,7 @@ void *i915_gem_object_pin_map(struct drm_i915_gem_object *obj,
 			      enum i915_map_type type)
 {
 	enum i915_map_type has_type;
+	struct dma_fence *moving;
 	bool pinned;
 	void *ptr;
 	int err;
@@ -359,6 +360,15 @@ void *i915_gem_object_pin_map(struct drm_i915_gem_object *obj,
 
 	assert_object_held(obj);
 
+	moving = i915_gem_object_get_moving_fence(obj);
+	if (moving) {
+		err = dma_fence_wait(moving, true);
+		dma_fence_put(moving);
+
+		if (err)
+			return ERR_PTR(err);
+	}
+
 	pinned = !(type & I915_MAP_OVERRIDE);
 	type &= ~I915_MAP_OVERRIDE;
 
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index bed2667b592e..ab8e24922287 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -395,7 +395,7 @@ int i915_vma_bind(struct i915_vma *vma,
 	GEM_BUG_ON(!vma->pages);
 
 	trace_i915_vma_bind(vma, bind_flags);
-	if (work && bind_flags & vma->vm->bind_async_flags) {
+	if (work) {
 		struct dma_fence *prev;
 
 		work->vma = vma;
@@ -1225,6 +1225,7 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
 		    u64 size, u64 alignment, u64 flags)
 {
 	struct i915_vma_work *work = NULL;
+	struct dma_fence *moving = NULL;
 	intel_wakeref_t wakeref = 0;
 	unsigned int bound;
 	int err;
@@ -1248,7 +1249,8 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
 	if (flags & PIN_GLOBAL)
 		wakeref = intel_runtime_pm_get(&vma->vm->i915->runtime_pm);
 
-	if (flags & vma->vm->bind_async_flags) {
+	moving = i915_gem_object_get_moving_fence(vma->obj);
+	if (flags & vma->vm->bind_async_flags || moving) {
 		/* lock VM */
 		err = i915_vm_lock_objects(vma->vm, ww);
 		if (err)
@@ -1262,6 +1264,8 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
 
 		work->vm = i915_vm_get(vma->vm);
 
+		dma_fence_work_chain(&work->base, moving);
+
 		/* Allocate enough page directories to used PTE */
 		if (vma->vm->allocate_va_range) {
 			err = i915_vm_alloc_pt_stash(vma->vm,
@@ -1369,6 +1373,8 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
 err_rpm:
 	if (wakeref)
 		intel_runtime_pm_put(&vma->vm->i915->runtime_pm, wakeref);
+	if (moving)
+		dma_fence_put(moving);
 	i915_vma_put_pages(vma);
 	return err;
 }
-- 
2.34.0