drm/amdgpu: Optimize mutex usage (v4)

In original function amdgpu_bo_list_get, the waiting
for result->lock can be quite long while mutex
bo_list_lock was holding. It can make other tasks
waiting for bo_list_lock for long period.

Secondly, this patch allows several tasks(readers of idr)
to proceed at the same time.

v2: use rcu and kref (Dave Airlie and Christian König)
v3: update v1 commit message (Michel Dänzer)
v4: rebase on upstream (Alex Deucher)

Signed-off-by: Alex Xie <AlexBin.Xie@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 9f827aca90ee308745072220cedc963a12ff5593..12d61edb3597f42f338bb51f5282268b0a053ed2 100644 (file)
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -869,6 +869,8 @@ struct amdgpu_fpriv {
 
 struct amdgpu_bo_list {
        struct mutex lock;
+       struct rcu_head rhead;
+       struct kref refcount;
        struct amdgpu_bo *gds_obj;
        struct amdgpu_bo *gws_obj;
        struct amdgpu_bo *oa_obj;

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
index 7b5c3bb2142ed2d87e2fdddafc5ed885d5562d22..f621ee115c98d4e5d5a4faa0d845864dc65a0b8c 100644 (file)
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
@@ -41,6 +41,20 @@ static int amdgpu_bo_list_set(struct amdgpu_device *adev,
                                     struct drm_amdgpu_bo_list_entry *info,
                                     unsigned num_entries);
 
+static void amdgpu_bo_list_release_rcu(struct kref *ref)
+{
+       unsigned i;
+       struct amdgpu_bo_list *list = container_of(ref, struct amdgpu_bo_list,
+                                                  refcount);
+
+       for (i = 0; i < list->num_entries; ++i)
+               amdgpu_bo_unref(&list->array[i].robj);
+
+       mutex_destroy(&list->lock);
+       kvfree(list->array);
+       kfree_rcu(list, rhead);
+}
+
 static int amdgpu_bo_list_create(struct amdgpu_device *adev,
                                 struct drm_file *filp,
                                 struct drm_amdgpu_bo_list_entry *info,
@@ -57,7 +71,7 @@ static int amdgpu_bo_list_create(struct amdgpu_device *adev,
 
        /* initialize bo list*/
        mutex_init(&list->lock);
-
+       kref_init(&list->refcount);
        r = amdgpu_bo_list_set(adev, filp, list, info, num_entries);
        if (r) {
                kfree(list);
@@ -83,14 +97,9 @@ static void amdgpu_bo_list_destroy(struct amdgpu_fpriv *fpriv, int id)
 
        mutex_lock(&fpriv->bo_list_lock);
        list = idr_remove(&fpriv->bo_list_handles, id);
-       if (list) {
-               /* Another user may have a reference to this list still */
-               mutex_lock(&list->lock);
-               mutex_unlock(&list->lock);
-               amdgpu_bo_list_free(list);
-       }
-
        mutex_unlock(&fpriv->bo_list_lock);
+       if (list)
+               kref_put(&list->refcount, amdgpu_bo_list_release_rcu);
 }
 
 static int amdgpu_bo_list_set(struct amdgpu_device *adev,
@@ -185,11 +194,17 @@ amdgpu_bo_list_get(struct amdgpu_fpriv *fpriv, int id)
 {
        struct amdgpu_bo_list *result;
 
-       mutex_lock(&fpriv->bo_list_lock);
+       rcu_read_lock();
        result = idr_find(&fpriv->bo_list_handles, id);
-       if (result)
-               mutex_lock(&result->lock);
-       mutex_unlock(&fpriv->bo_list_lock);
+
+       if (result) {
+               if (kref_get_unless_zero(&result->refcount))
+                       mutex_lock(&result->lock);
+               else
+                       result = NULL;
+       }
+       rcu_read_unlock();
+
        return result;
 }
 
@@ -227,6 +242,7 @@ void amdgpu_bo_list_get_list(struct amdgpu_bo_list *list,
 void amdgpu_bo_list_put(struct amdgpu_bo_list *list)
 {
        mutex_unlock(&list->lock);
+       kref_put(&list->refcount, amdgpu_bo_list_release_rcu);
 }
 
 void amdgpu_bo_list_free(struct amdgpu_bo_list *list)