[PATCH 19/22] drm/i915: Allows contexts to be unreferenced locklessly
Chris Wilson
chris at chris-wilson.co.uk
Fri Mar 31 00:25:43 UTC 2017
If we move the actual cleanup of the context to a worker, we can allow
the final free to be called from any context and avoid undue latency in
the caller.
v2: Negotiate handling the delayed contexts free during fini
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
---
drivers/gpu/drm/i915/gvt/scheduler.c | 2 +-
drivers/gpu/drm/i915/i915_drv.h | 23 +---------
drivers/gpu/drm/i915/i915_gem_context.c | 53 ++++++++++++++++++++++--
drivers/gpu/drm/i915/i915_gem_context.h | 15 ++++++-
drivers/gpu/drm/i915/i915_perf.c | 4 +-
drivers/gpu/drm/i915/selftests/mock_context.c | 9 ++++
drivers/gpu/drm/i915/selftests/mock_context.h | 2 +
drivers/gpu/drm/i915/selftests/mock_gem_device.c | 2 +-
8 files changed, 81 insertions(+), 29 deletions(-)
diff --git a/drivers/gpu/drm/i915/gvt/scheduler.c b/drivers/gpu/drm/i915/gvt/scheduler.c
index a77db2332e68..4436ab84f44f 100644
--- a/drivers/gpu/drm/i915/gvt/scheduler.c
+++ b/drivers/gpu/drm/i915/gvt/scheduler.c
@@ -598,7 +598,7 @@ int intel_gvt_init_workload_scheduler(struct intel_gvt *gvt)
void intel_vgpu_clean_gvt_context(struct intel_vgpu *vgpu)
{
- i915_gem_context_put_unlocked(vgpu->shadow_ctx);
+ i915_gem_context_put(vgpu->shadow_ctx);
}
int intel_vgpu_init_gvt_context(struct intel_vgpu *vgpu)
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 45bc9a65ec53..aa5b7c8969ab 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -2367,6 +2367,8 @@ struct drm_i915_private {
struct {
struct list_head list;
+ struct llist_head free_list;
+ struct work_struct free_work;
u32 hw_size;
/* The hw wants to have a stable context identifier for the
@@ -3516,27 +3518,6 @@ i915_gem_context_lookup(struct drm_i915_file_private *file_priv, u32 id)
return ctx;
}
-static inline struct i915_gem_context *
-i915_gem_context_get(struct i915_gem_context *ctx)
-{
- kref_get(&ctx->ref);
- return ctx;
-}
-
-static inline void i915_gem_context_put(struct i915_gem_context *ctx)
-{
- lockdep_assert_held(&ctx->i915->drm.struct_mutex);
- kref_put(&ctx->ref, i915_gem_context_free);
-}
-
-static inline void i915_gem_context_put_unlocked(struct i915_gem_context *ctx)
-{
- struct mutex *lock = &ctx->i915->drm.struct_mutex;
-
- if (kref_put_mutex(&ctx->ref, i915_gem_context_free, lock))
- mutex_unlock(lock);
-}
-
static inline struct intel_timeline *
i915_gem_context_lookup_timeline(struct i915_gem_context *ctx,
struct intel_engine_cs *engine)
diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c
index c0f3acece66a..18953916df58 100644
--- a/drivers/gpu/drm/i915/i915_gem_context.c
+++ b/drivers/gpu/drm/i915/i915_gem_context.c
@@ -184,13 +184,11 @@ static void vma_lut_free(struct i915_gem_context *ctx)
kvfree(lut->ht);
}
-void i915_gem_context_free(struct kref *ctx_ref)
+static void i915_gem_context_free(struct i915_gem_context *ctx)
{
- struct i915_gem_context *ctx = container_of(ctx_ref, typeof(*ctx), ref);
int i;
lockdep_assert_held(&ctx->i915->drm.struct_mutex);
- trace_i915_context_free(ctx);
GEM_BUG_ON(!i915_gem_context_is_closed(ctx));
vma_lut_free(ctx);
@@ -218,6 +216,41 @@ void i915_gem_context_free(struct kref *ctx_ref)
kfree(ctx);
}
+static void contexts_free(struct drm_i915_private *i915)
+{
+ struct i915_gem_context *ctx, *cn;
+ struct llist_node *freed;
+
+ freed = llist_del_all(&i915->contexts.free_list);
+ llist_for_each_entry_safe(ctx, cn, freed, free_link)
+ i915_gem_context_free(ctx);
+}
+
+static void contexts_free_worker(struct work_struct *work)
+{
+ struct drm_i915_private *i915 =
+ container_of(work, typeof(*i915), contexts.free_work);
+
+ /* We need to the worker with i915_gem_context_fini() and so we only
+ * take the freed list whilst under the struct_mutex. This allows us
+ * to manually free each context from the within fini under the
+ * struct_mutex and when the worker wakes it will find nothing to do.
+ */
+ mutex_lock(&i915->drm.struct_mutex);
+ contexts_free(i915);
+ mutex_unlock(&i915->drm.struct_mutex);
+}
+
+void i915_gem_context_release(struct kref *ctx_ref)
+{
+ struct i915_gem_context *ctx = container_of(ctx_ref, typeof(*ctx), ref);
+ struct drm_i915_private *i915 = ctx->i915;
+
+ trace_i915_context_free(ctx);
+ if (llist_add(&ctx->free_link, &i915->contexts.free_list))
+ queue_work(i915->wq, &i915->contexts.free_work);
+}
+
static struct drm_i915_gem_object *
alloc_context_obj(struct drm_i915_private *dev_priv, u64 size)
{
@@ -512,6 +545,8 @@ int i915_gem_context_init(struct drm_i915_private *dev_priv)
return 0;
INIT_LIST_HEAD(&dev_priv->contexts.list);
+ INIT_WORK(&dev_priv->contexts.free_work, contexts_free_worker);
+ init_llist_head(&dev_priv->contexts.free_list);
if (intel_vgpu_active(dev_priv) &&
HAS_LOGICAL_RING_CONTEXTS(dev_priv)) {
@@ -604,6 +639,16 @@ void i915_gem_context_lost(struct drm_i915_private *dev_priv)
}
}
+static void contexts_flush_work(struct drm_i915_private *i915)
+{
+ /* As we hold struct_mutex, and i915_gem_context_free itself requires
+ * struct_mutex, we cannot simply call flush_workqueue() and instead
+ * must free our contexts manually.
+ */
+ lockdep_assert_held(&i915->drm.struct_mutex);
+ contexts_free(i915);
+}
+
void i915_gem_context_fini(struct drm_i915_private *dev_priv)
{
struct i915_gem_context *dctx = dev_priv->kernel_context;
@@ -615,6 +660,8 @@ void i915_gem_context_fini(struct drm_i915_private *dev_priv)
context_close(dctx);
dev_priv->kernel_context = NULL;
+ /* Must flush context frees before destroying the hw_ida */
+ contexts_flush_work(dev_priv);
ida_destroy(&dev_priv->contexts.hw_ida);
}
diff --git a/drivers/gpu/drm/i915/i915_gem_context.h b/drivers/gpu/drm/i915/i915_gem_context.h
index db5b28a28d75..526a4a87d23c 100644
--- a/drivers/gpu/drm/i915/i915_gem_context.h
+++ b/drivers/gpu/drm/i915/i915_gem_context.h
@@ -86,6 +86,7 @@ struct i915_gem_context {
/** link: place with &drm_i915_private.context_list */
struct list_head link;
+ struct llist_node free_link;
/**
* @ref: reference count
@@ -279,7 +280,7 @@ int i915_gem_context_open(struct drm_device *dev, struct drm_file *file);
void i915_gem_context_close(struct drm_device *dev, struct drm_file *file);
int i915_switch_context(struct drm_i915_gem_request *req);
int i915_gem_switch_to_kernel_context(struct drm_i915_private *dev_priv);
-void i915_gem_context_free(struct kref *ctx_ref);
+void i915_gem_context_release(struct kref *ctx_ref);
struct i915_gem_context *
i915_gem_context_create_gvt(struct drm_device *dev);
@@ -294,4 +295,16 @@ int i915_gem_context_setparam_ioctl(struct drm_device *dev, void *data,
int i915_gem_context_reset_stats_ioctl(struct drm_device *dev, void *data,
struct drm_file *file);
+static inline struct i915_gem_context *
+i915_gem_context_get(struct i915_gem_context *ctx)
+{
+ kref_get(&ctx->ref);
+ return ctx;
+}
+
+static inline void i915_gem_context_put(struct i915_gem_context *ctx)
+{
+ kref_put(&ctx->ref, i915_gem_context_release);
+}
+
#endif /* !__I915_GEM_CONTEXT_H__ */
diff --git a/drivers/gpu/drm/i915/i915_perf.c b/drivers/gpu/drm/i915/i915_perf.c
index 060b171480d5..ce83c3ea10ca 100644
--- a/drivers/gpu/drm/i915/i915_perf.c
+++ b/drivers/gpu/drm/i915/i915_perf.c
@@ -1564,7 +1564,7 @@ static void i915_perf_destroy_locked(struct i915_perf_stream *stream)
list_del(&stream->link);
if (stream->ctx)
- i915_gem_context_put_unlocked(stream->ctx);
+ i915_gem_context_put(stream->ctx);
kfree(stream);
}
@@ -1735,7 +1735,7 @@ i915_perf_open_ioctl_locked(struct drm_i915_private *dev_priv,
kfree(stream);
err_ctx:
if (specific_ctx)
- i915_gem_context_put_unlocked(specific_ctx);
+ i915_gem_context_put(specific_ctx);
err:
return ret;
}
diff --git a/drivers/gpu/drm/i915/selftests/mock_context.c b/drivers/gpu/drm/i915/selftests/mock_context.c
index 243325b97d4c..9c7c68181f82 100644
--- a/drivers/gpu/drm/i915/selftests/mock_context.c
+++ b/drivers/gpu/drm/i915/selftests/mock_context.c
@@ -86,3 +86,12 @@ void mock_context_close(struct i915_gem_context *ctx)
i915_gem_context_put(ctx);
}
+
+void mock_init_contexts(struct drm_i915_private *i915)
+{
+ INIT_LIST_HEAD(&i915->contexts.list);
+ ida_init(&i915->contexts.hw_ida);
+
+ INIT_WORK(&i915->contexts.free_work, contexts_free_worker);
+ init_llist_head(&i915->contexts.free_list);
+}
diff --git a/drivers/gpu/drm/i915/selftests/mock_context.h b/drivers/gpu/drm/i915/selftests/mock_context.h
index 2427e5c0916a..383941a61124 100644
--- a/drivers/gpu/drm/i915/selftests/mock_context.h
+++ b/drivers/gpu/drm/i915/selftests/mock_context.h
@@ -25,6 +25,8 @@
#ifndef __MOCK_CONTEXT_H
#define __MOCK_CONTEXT_H
+void mock_init_contexts(struct drm_i915_private *i915);
+
struct i915_gem_context *
mock_context(struct drm_i915_private *i915,
const char *name);
diff --git a/drivers/gpu/drm/i915/selftests/mock_gem_device.c b/drivers/gpu/drm/i915/selftests/mock_gem_device.c
index a356db346ac7..42f9c4f7d565 100644
--- a/drivers/gpu/drm/i915/selftests/mock_gem_device.c
+++ b/drivers/gpu/drm/i915/selftests/mock_gem_device.c
@@ -156,7 +156,7 @@ struct drm_i915_private *mock_gem_device(void)
INIT_LIST_HEAD(&i915->mm.unbound_list);
INIT_LIST_HEAD(&i915->mm.bound_list);
- ida_init(&i915->contexts.hw_ida);
+ mock_init_contexts(i915);
INIT_DELAYED_WORK(&i915->gt.retire_work, mock_retire_work_handler);
INIT_DELAYED_WORK(&i915->gt.idle_work, mock_idle_work_handler);
--
2.11.0
More information about the Intel-gfx-trybot
mailing list