Hook the vma itself into the i915_gem_request_retire() so that we can
accurately track when a solitary vma is inactive (as opposed to having
to wait for the entire object to be idle). This improves the interaction
when using multiple contexts (with full-ppgtt) and eliminates some
frequent list walking when retiring objects after a completed request.
A side-effect is that we get an active vma reference for free. The
consequence of this is shown in the next patch...
Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
---
drivers/gpu/drm/i915/i915_debugfs.c | 2 +-
drivers/gpu/drm/i915/i915_gem.c | 50 +++++++++++++++++-------------
drivers/gpu/drm/i915/i915_gem_execbuffer.c | 10 +++++-
drivers/gpu/drm/i915/i915_gem_gtt.c | 20 ++++++++++++
drivers/gpu/drm/i915/i915_gem_gtt.h | 28 +++++++++++++++++
5 files changed, 86 insertions(+), 24 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
index 6693dfbca4f1..3870eaeeddb4 100644
--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -367,7 +367,7 @@ static int per_file_stats(int id, void *ptr, void *data)
continue;
}
- if (obj->active) /* XXX per-vma statistic */
+ if (i915_vma_is_active(vma))
stats->active += vma->node.size;
else
stats->inactive += vma->node.size;
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 2e0b54fa03f9..61fa8049076e 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -2354,7 +2354,6 @@ i915_gem_object_retire__read(struct i915_gem_active *active,
int idx = request->engine->id;
struct drm_i915_gem_object *obj =
container_of(active, struct drm_i915_gem_object, last_read[idx]);
- struct i915_vma *vma;
GEM_BUG_ON((obj->active & (1 << idx)) == 0);
@@ -2366,12 +2365,9 @@ i915_gem_object_retire__read(struct i915_gem_active *active,
* so that we don't steal from recently used but inactive objects
* (unless we are forced to ofc!)
*/
- list_move_tail(&obj->global_list, &request->i915->mm.bound_list);
-
- list_for_each_entry(vma, &obj->vma_list, obj_link) {
- if (!list_empty(&vma->vm_link))
- list_move_tail(&vma->vm_link, &vma->vm->inactive_list);
- }
+ if (obj->bind_count)
+ list_move_tail(&obj->global_list,
+ &request->i915->mm.bound_list);
i915_gem_object_put(obj);
}
@@ -2804,8 +2800,29 @@ static void __i915_vma_iounmap(struct i915_vma *vma)
static int __i915_vma_unbind(struct i915_vma *vma, bool wait)
{
struct drm_i915_gem_object *obj = vma->obj;
+ unsigned long active;
int ret;
+ /* First wait upon any activity as retiring the request may
+ * have side-effects such as unpinning or even unbinding this vma.
+ */
+ active = vma->active;
+ if (active && wait) {
+ int idx;
+
+ for_each_active(active, idx) {
+ ret = i915_gem_active_retire(&vma->last_read[idx],
+ &vma->vm->dev->struct_mutex);
+ if (ret)
+ return ret;
+ }
+
+ GEM_BUG_ON(i915_vma_is_active(vma));
+ }
+
+ if (vma->pin_count)
+ return -EBUSY;
+
if (list_empty(&vma->obj_link))
return 0;
@@ -2814,18 +2831,9 @@ static int __i915_vma_unbind(struct i915_vma *vma, bool wait)
return 0;
}
- if (vma->pin_count)
- return -EBUSY;
-
GEM_BUG_ON(obj->bind_count == 0);
GEM_BUG_ON(!obj->pages);
- if (wait) {
- ret = i915_gem_object_wait_rendering(obj, false);
- if (ret)
- return ret;
- }
-
if (vma->is_ggtt && vma->ggtt_view.type == I915_GGTT_VIEW_NORMAL) {
i915_gem_object_finish_gtt(obj);
@@ -3208,9 +3216,6 @@ i915_gem_object_flush_cpu_write_domain(struct drm_i915_gem_object *obj)
int
i915_gem_object_set_to_gtt_domain(struct drm_i915_gem_object *obj, bool write)
{
- struct drm_device *dev = obj->base.dev;
- struct drm_i915_private *dev_priv = to_i915(dev);
- struct i915_ggtt *ggtt = &dev_priv->ggtt;
uint32_t old_write_domain, old_read_domains;
struct i915_vma *vma;
int ret;
@@ -3263,9 +3268,10 @@ i915_gem_object_set_to_gtt_domain(struct drm_i915_gem_object *obj, bool write)
/* And bump the LRU for this access */
vma = i915_gem_obj_to_ggtt(obj);
- if (vma && drm_mm_node_allocated(&vma->node) && !obj->active)
- list_move_tail(&vma->vm_link,
- &ggtt->base.inactive_list);
+ if (vma &&
+ drm_mm_node_allocated(&vma->node) &&
+ !i915_vma_is_active(vma))
+ list_move_tail(&vma->vm_link, &vma->vm->inactive_list);
return 0;
}
diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
index d0f1da2863e4..366ae72d479e 100644
--- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
@@ -1154,7 +1154,13 @@ void i915_vma_move_to_active(struct i915_vma *vma,
obj->dirty = 1; /* be paranoid */
- /* Add a reference if we're newly entering the active list. */
+ /* Add a reference if we're newly entering the active list.
+ * The order in which we add operations to the retirement queue is
+ * vital here: mark_active adds to the start of the callback list,
+ * such that subsequent callbacks are called first. Therefore we
+ * add the active reference first and queue for it to be dropped
+ * *last*.
+ */
if (obj->active == 0)
i915_gem_object_get(obj);
obj->active |= 1 << idx;
@@ -1179,6 +1185,8 @@ void i915_vma_move_to_active(struct i915_vma *vma,
}
}
+ i915_vma_set_active(vma, idx);
+ i915_gem_active_set(&vma->last_read[idx], req);
list_move_tail(&vma->vm_link, &vma->vm->active_list);
}
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index 9d493b8e3bfc..ef0dc7131808 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -3324,12 +3324,30 @@ void i915_gem_restore_gtt_mappings(struct drm_device *dev)
i915_ggtt_flush(dev_priv);
}
+static void
+i915_vma_retire(struct i915_gem_active *active,
+ struct drm_i915_gem_request *rq)
+{
+ const unsigned int idx = rq->engine->id;
+ struct i915_vma *vma =
+ container_of(active, struct i915_vma, last_read[idx]);
+
+ GEM_BUG_ON(!i915_vma_has_active_engine(vma, idx));
+
+ i915_vma_unset_active(vma, idx);
+ if (i915_vma_is_active(vma))
+ return;
+
+ list_move_tail(&vma->vm_link, &vma->vm->inactive_list);
+}
+
static struct i915_vma *
__i915_gem_vma_create(struct drm_i915_gem_object *obj,
struct i915_address_space *vm,
const struct i915_ggtt_view *ggtt_view)
{
struct i915_vma *vma;
+ int i;
if (WARN_ON(i915_is_ggtt(vm) != !!ggtt_view))
return ERR_PTR(-EINVAL);
@@ -3341,6 +3359,8 @@ __i915_gem_vma_create(struct drm_i915_gem_object *obj,
INIT_LIST_HEAD(&vma->vm_link);
INIT_LIST_HEAD(&vma->obj_link);
INIT_LIST_HEAD(&vma->exec_list);
+ for (i = 0; i < ARRAY_SIZE(vma->last_read); i++)
+ init_request_active(&vma->last_read[i], i915_vma_retire);
vma->vm = vm;
vma->obj = obj;
vma->is_ggtt = i915_is_ggtt(vm);
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.h b/drivers/gpu/drm/i915/i915_gem_gtt.h
index 529fb483afc8..d2206f40f7b2 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.h
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.h
@@ -36,6 +36,8 @@
#include <linux/io-mapping.h>
+#include "i915_gem_request.h"
+
struct drm_i915_file_private;
typedef uint32_t gen6_pte_t;
@@ -179,6 +181,9 @@ struct i915_vma {
struct i915_address_space *vm;
void __iomem *iomap;
+ unsigned int active;
+ struct i915_gem_active last_read[I915_NUM_ENGINES];
+
/** Flags and address space this VMA is bound to */
#define GLOBAL_BIND (1<<0)
#define LOCAL_BIND (1<<1)
@@ -222,6 +227,29 @@ struct i915_vma {
#define DRM_I915_GEM_OBJECT_MAX_PIN_COUNT 0xf
};
+static inline bool i915_vma_is_active(const struct i915_vma *vma)
+{
+ return vma->active;
+}
+
+static inline void i915_vma_set_active(struct i915_vma *vma,
+ unsigned int engine)
+{
+ vma->active |= BIT(engine);
+}
+
+static inline void i915_vma_unset_active(struct i915_vma *vma,
+ unsigned int engine)
+{
+ vma->active &= ~BIT(engine);
+}
+
+static inline bool i915_vma_has_active_engine(const struct i915_vma *vma,
+ unsigned int engine)
+{
+ return vma->active & BIT(engine);
+}
+
struct i915_page_dma {
struct page *page;
union {
--
2.8.1
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
