[PATCH 05/16] drm/i915: Generalise the clflush dma-worker

Chris Wilson chris at chris-wilson.co.uk
Thu Aug 22 00:23:15 UTC 2019


Extract the dma-fence worker used by clflush for wider use, as we
anticipate using workers coupled to dma-fences more frequently.

Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld at intel.com>
---
 drivers/gpu/drm/i915/Makefile               |   1 +
 drivers/gpu/drm/i915/gem/i915_gem_clflush.c | 123 +++++++-------------
 drivers/gpu/drm/i915/i915_sw_fence_work.c   |  95 +++++++++++++++
 drivers/gpu/drm/i915/i915_sw_fence_work.h   |  44 +++++++
 4 files changed, 183 insertions(+), 80 deletions(-)
 create mode 100644 drivers/gpu/drm/i915/i915_sw_fence_work.c
 create mode 100644 drivers/gpu/drm/i915/i915_sw_fence_work.h

diff --git a/drivers/gpu/drm/i915/Makefile b/drivers/gpu/drm/i915/Makefile
index 45add812048b..658b930d34a8 100644
--- a/drivers/gpu/drm/i915/Makefile
+++ b/drivers/gpu/drm/i915/Makefile
@@ -62,6 +62,7 @@ i915-y += \
 	i915_memcpy.o \
 	i915_mm.o \
 	i915_sw_fence.o \
+	i915_sw_fence_work.o \
 	i915_syncmap.o \
 	i915_user_extensions.o
 
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_clflush.c b/drivers/gpu/drm/i915/gem/i915_gem_clflush.c
index fb0ef176ba5b..b9f504ba3b32 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_clflush.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_clflush.c
@@ -8,88 +8,67 @@
 
 #include "i915_drv.h"
 #include "i915_gem_clflush.h"
+#include "i915_sw_fence_work.h"
 #include "i915_trace.h"
 
-static DEFINE_SPINLOCK(clflush_lock);
-
 struct clflush {
-	struct dma_fence dma; /* Must be first for dma_fence_free() */
-	struct i915_sw_fence wait;
-	struct work_struct work;
+	struct dma_fence_work base;
 	struct drm_i915_gem_object *obj;
 };
 
-static const char *i915_clflush_get_driver_name(struct dma_fence *fence)
-{
-	return DRIVER_NAME;
-}
-
-static const char *i915_clflush_get_timeline_name(struct dma_fence *fence)
-{
-	return "clflush";
-}
-
-static void i915_clflush_release(struct dma_fence *fence)
-{
-	struct clflush *clflush = container_of(fence, typeof(*clflush), dma);
-
-	i915_sw_fence_fini(&clflush->wait);
-
-	BUILD_BUG_ON(offsetof(typeof(*clflush), dma));
-	dma_fence_free(&clflush->dma);
-}
-
-static const struct dma_fence_ops i915_clflush_ops = {
-	.get_driver_name = i915_clflush_get_driver_name,
-	.get_timeline_name = i915_clflush_get_timeline_name,
-	.release = i915_clflush_release,
-};
-
-static void __i915_do_clflush(struct drm_i915_gem_object *obj)
+static void __do_clflush(struct drm_i915_gem_object *obj)
 {
 	GEM_BUG_ON(!i915_gem_object_has_pages(obj));
 	drm_clflush_sg(obj->mm.pages);
 	intel_frontbuffer_flush(obj->frontbuffer, ORIGIN_CPU);
 }
 
-static void i915_clflush_work(struct work_struct *work)
+static int clflush_work(struct dma_fence_work *base)
 {
-	struct clflush *clflush = container_of(work, typeof(*clflush), work);
-	struct drm_i915_gem_object *obj = clflush->obj;
-
-	if (i915_gem_object_pin_pages(obj)) {
-		DRM_ERROR("Failed to acquire obj->pages for clflushing\n");
-		goto out;
-	}
+	struct clflush *clflush = container_of(base, typeof(*clflush), base);
+	struct drm_i915_gem_object *obj = fetch_and_zero(&clflush->obj);
+	int err;
 
-	__i915_do_clflush(obj);
+	err = i915_gem_object_pin_pages(obj);
+	if (err)
+		goto put;
 
+	__do_clflush(obj);
 	i915_gem_object_unpin_pages(obj);
 
-out:
+put:
 	i915_gem_object_put(obj);
+	return err;
+}
+
+static void clflush_release(struct dma_fence_work *base)
+{
+	struct clflush *clflush = container_of(base, typeof(*clflush), base);
 
-	dma_fence_signal(&clflush->dma);
-	dma_fence_put(&clflush->dma);
+	if (clflush->obj)
+		i915_gem_object_put(clflush->obj);
 }
 
-static int __i915_sw_fence_call
-i915_clflush_notify(struct i915_sw_fence *fence,
-		    enum i915_sw_fence_notify state)
+static const struct dma_fence_work_ops clflush_ops = {
+	.name = "clflush",
+	.work = clflush_work,
+	.release = clflush_release,
+};
+
+static struct clflush *clflush_work_create(struct drm_i915_gem_object *obj)
 {
-	struct clflush *clflush = container_of(fence, typeof(*clflush), wait);
+	struct clflush *clflush;
 
-	switch (state) {
-	case FENCE_COMPLETE:
-		schedule_work(&clflush->work);
-		break;
+	GEM_BUG_ON(!obj->cache_dirty);
 
-	case FENCE_FREE:
-		dma_fence_put(&clflush->dma);
-		break;
-	}
+	clflush = kmalloc(sizeof(*clflush), GFP_KERNEL);
+	if (!clflush)
+		return NULL;
 
-	return NOTIFY_DONE;
+	dma_fence_work_init(&clflush->base, &clflush_ops);
+	clflush->obj = i915_gem_object_get(obj); /* obj <-> clflush cycle */
+
+	return clflush;
 }
 
 bool i915_gem_clflush_object(struct drm_i915_gem_object *obj,
@@ -127,32 +106,16 @@ bool i915_gem_clflush_object(struct drm_i915_gem_object *obj,
 
 	clflush = NULL;
 	if (!(flags & I915_CLFLUSH_SYNC))
-		clflush = kmalloc(sizeof(*clflush), GFP_KERNEL);
+		clflush = clflush_work_create(obj);
 	if (clflush) {
-		GEM_BUG_ON(!obj->cache_dirty);
-
-		dma_fence_init(&clflush->dma,
-			       &i915_clflush_ops,
-			       &clflush_lock,
-			       0, 0);
-		i915_sw_fence_init(&clflush->wait, i915_clflush_notify);
-
-		clflush->obj = i915_gem_object_get(obj);
-		INIT_WORK(&clflush->work, i915_clflush_work);
-
-		dma_fence_get(&clflush->dma);
-
-		i915_sw_fence_await_reservation(&clflush->wait,
-						obj->base.resv, NULL,
-						true, I915_FENCE_TIMEOUT,
+		i915_sw_fence_await_reservation(&clflush->base.chain,
+						obj->base.resv, NULL, true,
+						I915_FENCE_TIMEOUT,
 						I915_FENCE_GFP);
-
-		dma_resv_add_excl_fence(obj->base.resv,
-						  &clflush->dma);
-
-		i915_sw_fence_commit(&clflush->wait);
+		dma_resv_add_excl_fence(obj->base.resv, &clflush->base.dma);
+		dma_fence_work_commit(&clflush->base);
 	} else if (obj->mm.pages) {
-		__i915_do_clflush(obj);
+		__do_clflush(obj);
 	} else {
 		GEM_BUG_ON(obj->write_domain != I915_GEM_DOMAIN_CPU);
 	}
diff --git a/drivers/gpu/drm/i915/i915_sw_fence_work.c b/drivers/gpu/drm/i915/i915_sw_fence_work.c
new file mode 100644
index 000000000000..07552cd544f2
--- /dev/null
+++ b/drivers/gpu/drm/i915/i915_sw_fence_work.c
@@ -0,0 +1,95 @@
+// SPDX-License-Identifier: MIT
+
+/*
+ * Copyright © 2019 Intel Corporation
+ */
+
+#include "i915_sw_fence_work.h"
+
+static void fence_work(struct work_struct *work)
+{
+	struct dma_fence_work *f = container_of(work, typeof(*f), work);
+	int err;
+
+	err = f->ops->work(f);
+	if (err)
+		dma_fence_set_error(&f->dma, err);
+	dma_fence_signal(&f->dma);
+	dma_fence_put(&f->dma);
+}
+
+static int __i915_sw_fence_call
+fence_notify(struct i915_sw_fence *fence, enum i915_sw_fence_notify state)
+{
+	struct dma_fence_work *f = container_of(fence, typeof(*f), chain);
+
+	switch (state) {
+	case FENCE_COMPLETE:
+		if (fence->error)
+			dma_fence_set_error(&f->dma, fence->error);
+
+		if (!f->dma.error) {
+			dma_fence_get(&f->dma);
+			queue_work(system_unbound_wq, &f->work);
+		} else {
+			dma_fence_signal(&f->dma);
+		}
+		break;
+
+	case FENCE_FREE:
+		dma_fence_put(&f->dma);
+		break;
+	}
+
+	return NOTIFY_DONE;
+}
+
+static const char *get_driver_name(struct dma_fence *fence)
+{
+	return "dma-fence";
+}
+
+static const char *get_timeline_name(struct dma_fence *fence)
+{
+	struct dma_fence_work *f = container_of(fence, typeof(*f), dma);
+
+	return f->ops->name ?: "work";
+}
+
+static void fence_release(struct dma_fence *fence)
+{
+	struct dma_fence_work *f = container_of(fence, typeof(*f), dma);
+
+	if (f->ops->release)
+		f->ops->release(f);
+
+	i915_sw_fence_fini(&f->chain);
+
+	BUILD_BUG_ON(offsetof(typeof(*f), dma));
+	dma_fence_free(&f->dma);
+}
+
+static const struct dma_fence_ops fence_ops = {
+	.get_driver_name = get_driver_name,
+	.get_timeline_name = get_timeline_name,
+	.release = fence_release,
+};
+
+void dma_fence_work_init(struct dma_fence_work *f,
+			 const struct dma_fence_work_ops *ops)
+{
+	spin_lock_init(&f->lock);
+	dma_fence_init(&f->dma, &fence_ops, &f->lock, 0, 0);
+	i915_sw_fence_init(&f->chain, fence_notify);
+	INIT_WORK(&f->work, fence_work);
+
+	f->ops = ops;
+}
+
+int dma_fence_work_chain(struct dma_fence_work *f, struct dma_fence *signal)
+{
+	if (!signal)
+		return 0;
+
+	return __i915_sw_fence_await_dma_fence(&f->chain, signal, &f->cb);
+}
diff --git a/drivers/gpu/drm/i915/i915_sw_fence_work.h b/drivers/gpu/drm/i915/i915_sw_fence_work.h
new file mode 100644
index 000000000000..3a22b287e201
--- /dev/null
+++ b/drivers/gpu/drm/i915/i915_sw_fence_work.h
@@ -0,0 +1,44 @@
+/* SPDX-License-Identifier: MIT */
+
+/*
+ * Copyright © 2019 Intel Corporation
+ */
+
+#ifndef I915_SW_FENCE_WORK_H
+#define I915_SW_FENCE_WORK_H
+
+#include <linux/dma-fence.h>
+#include <linux/spinlock.h>
+#include <linux/workqueue.h>
+
+#include "i915_sw_fence.h"
+
+struct dma_fence_work;
+
+struct dma_fence_work_ops {
+	const char *name;
+	int (*work)(struct dma_fence_work *f);
+	void (*release)(struct dma_fence_work *f);
+};
+
+struct dma_fence_work {
+	struct dma_fence dma;
+	spinlock_t lock;
+
+	struct i915_sw_fence chain;
+	struct i915_sw_dma_fence_cb cb;
+
+	struct work_struct work;
+	const struct dma_fence_work_ops *ops;
+};
+
+void dma_fence_work_init(struct dma_fence_work *f,
+			 const struct dma_fence_work_ops *ops);
+int dma_fence_work_chain(struct dma_fence_work *f, struct dma_fence *signal);
+
+static inline void dma_fence_work_commit(struct dma_fence_work *f)
+{
+	i915_sw_fence_commit(&f->chain);
+}
+
+#endif /* I915_SW_FENCE_WORK_H */
-- 
2.23.0



More information about the Intel-gfx-trybot mailing list