drm/i915: Initial introduction of vma resources

Introduce vma resources, sort of similar to TTM resources,  needed for
asynchronous bind management. Initially we will use them to hold
completion of unbinding when we capture data from a vma, but they will
be used extensively in upcoming patches for asynchronous vma unbinding.

v6:
- Some documentation updates

Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220110172219.107131-2-thomas.hellstrom@linux.intel.com

1 files changed, 49 insertions, 6 deletions

diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 29a858c53bdd..b3a10c003dca 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -37,6 +37,7 @@
 #include "i915_sw_fence_work.h"
 #include "i915_trace.h"
 #include "i915_vma.h"
+#include "i915_vma_resource.h"
 
 static struct kmem_cache *slab_vmas;
 
@@ -380,6 +381,8 @@ static int i915_vma_verify_bind_complete(struct i915_vma *vma)
  * @cache_level: mapping cache level
  * @flags: flags like global or local mapping
  * @work: preallocated worker for allocating and binding the PTE
+ * @vma_res: pointer to a preallocated vma resource. The resource is either
+ * consumed or freed.
  *
  * DMA addresses are taken from the scatter-gather table of this object (or of
  * this VMA in case of non-default GGTT views) and PTE entries set up.
@@ -388,7 +391,8 @@ static int i915_vma_verify_bind_complete(struct i915_vma *vma)
 int i915_vma_bind(struct i915_vma *vma,
 		  enum i915_cache_level cache_level,
 		  u32 flags,
-		  struct i915_vma_work *work)
+		  struct i915_vma_work *work,
+		  struct i915_vma_resource *vma_res)
 {
 	u32 bind_flags;
 	u32 vma_flags;
@@ -399,11 +403,15 @@ int i915_vma_bind(struct i915_vma *vma,
 
 	if (GEM_DEBUG_WARN_ON(range_overflows(vma->node.start,
 					      vma->node.size,
-					      vma->vm->total)))
+					      vma->vm->total))) {
+		kfree(vma_res);
 		return -ENODEV;
+	}
 
-	if (GEM_DEBUG_WARN_ON(!flags))
+	if (GEM_DEBUG_WARN_ON(!flags)) {
+		kfree(vma_res);
 		return -EINVAL;
+	}
 
 	bind_flags = flags;
 	bind_flags &= I915_VMA_GLOBAL_BIND | I915_VMA_LOCAL_BIND;
@@ -412,11 +420,21 @@ int i915_vma_bind(struct i915_vma *vma,
 	vma_flags &= I915_VMA_GLOBAL_BIND | I915_VMA_LOCAL_BIND;
 
 	bind_flags &= ~vma_flags;
-	if (bind_flags == 0)
+	if (bind_flags == 0) {
+		kfree(vma_res);
 		return 0;
+	}
 
 	GEM_BUG_ON(!atomic_read(&vma->pages_count));
 
+	if (vma->resource || !vma_res) {
+		/* Rebinding with an additional I915_VMA_*_BIND */
+		GEM_WARN_ON(!vma_flags);
+		kfree(vma_res);
+	} else {
+		i915_vma_resource_init(vma_res);
+		vma->resource = vma_res;
+	}
 	trace_i915_vma_bind(vma, bind_flags);
 	if (work && bind_flags & vma->vm->bind_async_flags) {
 		struct dma_fence *prev;
@@ -1224,6 +1242,7 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
 {
 	struct i915_vma_work *work = NULL;
 	struct dma_fence *moving = NULL;
+	struct i915_vma_resource *vma_res = NULL;
 	intel_wakeref_t wakeref = 0;
 	unsigned int bound;
 	int err;
@@ -1278,6 +1297,12 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
 		}
 	}
 
+	vma_res = i915_vma_resource_alloc();
+	if (IS_ERR(vma_res)) {
+		err = PTR_ERR(vma_res);
+		goto err_fence;
+	}
+
 	/*
 	 * Differentiate between user/kernel vma inside the aliasing-ppgtt.
 	 *
@@ -1298,7 +1323,7 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
 	err = mutex_lock_interruptible_nested(&vma->vm->mutex,
 					      !(flags & PIN_GLOBAL));
 	if (err)
-		goto err_fence;
+		goto err_vma_res;
 
 	/* No more allocations allowed now we hold vm->mutex */
 
@@ -1339,7 +1364,8 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
 	GEM_BUG_ON(!vma->pages);
 	err = i915_vma_bind(vma,
 			    vma->obj->cache_level,
-			    flags, work);
+			    flags, work, vma_res);
+	vma_res = NULL;
 	if (err)
 		goto err_remove;
 
@@ -1362,6 +1388,8 @@ err_active:
 	i915_active_release(&vma->active);
 err_unlock:
 	mutex_unlock(&vma->vm->mutex);
+err_vma_res:
+	kfree(vma_res);
 err_fence:
 	if (work)
 		dma_fence_work_commit_imm(&work->base);
@@ -1512,6 +1540,7 @@ void i915_vma_release(struct kref *ref)
 	i915_vm_put(vma->vm);
 
 	i915_active_fini(&vma->active);
+	GEM_WARN_ON(vma->resource);
 	i915_vma_free(vma);
 }
 
@@ -1660,6 +1689,8 @@ int _i915_vma_move_to_active(struct i915_vma *vma,
 
 void __i915_vma_evict(struct i915_vma *vma)
 {
+	struct dma_fence *unbind_fence;
+
 	GEM_BUG_ON(i915_vma_is_pinned(vma));
 
 	if (i915_vma_is_map_and_fenceable(vma)) {
@@ -1697,8 +1728,20 @@ void __i915_vma_evict(struct i915_vma *vma)
 	atomic_and(~(I915_VMA_BIND_MASK | I915_VMA_ERROR | I915_VMA_GGTT_WRITE),
 		   &vma->flags);
 
+	unbind_fence = i915_vma_resource_unbind(vma->resource);
+	i915_vma_resource_put(vma->resource);
+	vma->resource = NULL;
+
 	i915_vma_detach(vma);
 	vma_unbind_pages(vma);
+
+	/*
+	 * This uninterruptible wait under the vm mutex is currently
+	 * only ever blocking while the vma is being captured from.
+	 * With async unbinding, this wait here will be removed.
+	 */
+	dma_fence_wait(unbind_fence, false);
+	dma_fence_put(unbind_fence);
 }
 
 int __i915_vma_unbind(struct i915_vma *vma)