* [PATCH 1/6] drm: execution context for GEM buffers v6
2023-06-28 10:44 AMDGPU patches for the drm_exec context Christian König
@ 2023-06-28 10:44 ` Christian König
2023-06-28 10:44 ` [PATCH 2/6] drm: add drm_exec selftests v4 Christian König
` (4 subsequent siblings)
5 siblings, 0 replies; 10+ messages in thread
From: Christian König @ 2023-06-28 10:44 UTC (permalink / raw)
To: dri-devel, amd-gfx, alexander.deucher
This adds the infrastructure for an execution context for GEM buffers
which is similar to the existing TTMs execbuf util and intended to replace
it in the long term.
The basic functionality is that we abstracts the necessary loop to lock
many different GEM buffers with automated deadlock and duplicate handling.
v2: drop xarray and use dynamic resized array instead, the locking
overhead is unecessary and measurable.
v3: drop duplicate tracking, radeon is really the only one needing that.
v4: fixes issues pointed out by Danilo, some typos in comments and a
helper for lock arrays of GEM objects.
v5: some suggestions by Boris Brezillon, especially just use one retry
macro, drop loop in prepare_array, use flags instead of bool
v6: minor changes suggested by Thomas, Boris and Danilo
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Danilo Krummrich <dakr@redhat.com>
Tested-by: Danilo Krummrich <dakr@redhat.com>
---
Documentation/gpu/drm-mm.rst | 12 ++
drivers/gpu/drm/Kconfig | 6 +
drivers/gpu/drm/Makefile | 2 +
drivers/gpu/drm/drm_exec.c | 333 +++++++++++++++++++++++++++++++++++
include/drm/drm_exec.h | 121 +++++++++++++
5 files changed, 474 insertions(+)
create mode 100644 drivers/gpu/drm/drm_exec.c
create mode 100644 include/drm/drm_exec.h
diff --git a/Documentation/gpu/drm-mm.rst b/Documentation/gpu/drm-mm.rst
index a79fd3549ff8..a52e6f4117d6 100644
--- a/Documentation/gpu/drm-mm.rst
+++ b/Documentation/gpu/drm-mm.rst
@@ -493,6 +493,18 @@ DRM Sync Objects
.. kernel-doc:: drivers/gpu/drm/drm_syncobj.c
:export:
+DRM Execution context
+=====================
+
+.. kernel-doc:: drivers/gpu/drm/drm_exec.c
+ :doc: Overview
+
+.. kernel-doc:: include/drm/drm_exec.h
+ :internal:
+
+.. kernel-doc:: drivers/gpu/drm/drm_exec.c
+ :export:
+
GPU Scheduler
=============
diff --git a/drivers/gpu/drm/Kconfig b/drivers/gpu/drm/Kconfig
index afb3b2f5f425..c2f3d234c89e 100644
--- a/drivers/gpu/drm/Kconfig
+++ b/drivers/gpu/drm/Kconfig
@@ -194,6 +194,12 @@ config DRM_TTM
GPU memory types. Will be enabled automatically if a device driver
uses it.
+config DRM_EXEC
+ tristate
+ depends on DRM
+ help
+ Execution context for command submissions
+
config DRM_BUDDY
tristate
depends on DRM
diff --git a/drivers/gpu/drm/Makefile b/drivers/gpu/drm/Makefile
index 7a09a89b493b..414855e2a463 100644
--- a/drivers/gpu/drm/Makefile
+++ b/drivers/gpu/drm/Makefile
@@ -78,6 +78,8 @@ obj-$(CONFIG_DRM_PANEL_ORIENTATION_QUIRKS) += drm_panel_orientation_quirks.o
#
# Memory-management helpers
#
+#
+obj-$(CONFIG_DRM_EXEC) += drm_exec.o
obj-$(CONFIG_DRM_BUDDY) += drm_buddy.o
diff --git a/drivers/gpu/drm/drm_exec.c b/drivers/gpu/drm/drm_exec.c
new file mode 100644
index 000000000000..a240dd194b3b
--- /dev/null
+++ b/drivers/gpu/drm/drm_exec.c
@@ -0,0 +1,333 @@
+/* SPDX-License-Identifier: GPL-2.0 OR MIT */
+
+#include <drm/drm_exec.h>
+#include <drm/drm_gem.h>
+#include <linux/dma-resv.h>
+
+/**
+ * DOC: Overview
+ *
+ * This component mainly abstracts the retry loop necessary for locking
+ * multiple GEM objects while preparing hardware operations (e.g. command
+ * submissions, page table updates etc..).
+ *
+ * If a contention is detected while locking a GEM object the cleanup procedure
+ * unlocks all previously locked GEM objects and locks the contended one first
+ * before locking any further objects.
+ *
+ * After an object is locked fences slots can optionally be reserved on the
+ * dma_resv object inside the GEM object.
+ *
+ * A typical usage pattern should look like this::
+ *
+ * struct drm_gem_object *obj;
+ * struct drm_exec exec;
+ * unsigned long index;
+ * int ret;
+ *
+ * drm_exec_init(&exec, DRM_EXEC_INTERRUPTIBLE_WAIT);
+ * drm_exec_until_all_locked(&exec) {
+ * ret = drm_exec_prepare_obj(&exec, boA, 1);
+ * drm_exec_retry_on_contention(&exec);
+ * if (ret)
+ * goto error;
+ *
+ * ret = drm_exec_prepare_obj(&exec, boB, 1);
+ * drm_exec_retry_on_contention(&exec);
+ * if (ret)
+ * goto error;
+ * }
+ *
+ * drm_exec_for_each_locked_object(&exec, index, obj) {
+ * dma_resv_add_fence(obj->resv, fence, DMA_RESV_USAGE_READ);
+ * ...
+ * }
+ * drm_exec_fini(&exec);
+ *
+ * See struct dma_exec for more details.
+ */
+
+/* Dummy value used to initially enter the retry loop */
+#define DRM_EXEC_DUMMY (void*)~0
+
+/* Unlock all objects and drop references */
+static void drm_exec_unlock_all(struct drm_exec *exec)
+{
+ struct drm_gem_object *obj;
+ unsigned long index;
+
+ drm_exec_for_each_locked_object(exec, index, obj) {
+ dma_resv_unlock(obj->resv);
+ drm_gem_object_put(obj);
+ }
+
+ drm_gem_object_put(exec->prelocked);
+ exec->prelocked = NULL;
+}
+
+/**
+ * drm_exec_init - initialize a drm_exec object
+ * @exec: the drm_exec object to initialize
+ * @flags: controls locking behavior, see DRM_EXEC_* defines
+ *
+ * Initialize the object and make sure that we can track locked objects.
+ */
+void drm_exec_init(struct drm_exec *exec, uint32_t flags)
+{
+ exec->flags = flags;
+ exec->objects = kmalloc(PAGE_SIZE, GFP_KERNEL);
+
+ /* If allocation here fails, just delay that till the first use */
+ exec->max_objects = exec->objects ? PAGE_SIZE / sizeof(void *) : 0;
+ exec->num_objects = 0;
+ exec->contended = DRM_EXEC_DUMMY;
+ exec->prelocked = NULL;
+}
+EXPORT_SYMBOL(drm_exec_init);
+
+/**
+ * drm_exec_fini - finalize a drm_exec object
+ * @exec: the drm_exec object to finalize
+ *
+ * Unlock all locked objects, drop the references to objects and free all memory
+ * used for tracking the state.
+ */
+void drm_exec_fini(struct drm_exec *exec)
+{
+ drm_exec_unlock_all(exec);
+ kvfree(exec->objects);
+ if (exec->contended != DRM_EXEC_DUMMY) {
+ drm_gem_object_put(exec->contended);
+ ww_acquire_fini(&exec->ticket);
+ }
+}
+EXPORT_SYMBOL(drm_exec_fini);
+
+/**
+ * drm_exec_cleanup - cleanup when contention is detected
+ * @exec: the drm_exec object to cleanup
+ *
+ * Cleanup the current state and return true if we should stay inside the retry
+ * loop, false if there wasn't any contention detected and we can keep the
+ * objects locked.
+ */
+bool drm_exec_cleanup(struct drm_exec *exec)
+{
+ if (likely(!exec->contended)) {
+ ww_acquire_done(&exec->ticket);
+ return false;
+ }
+
+ if (likely(exec->contended == DRM_EXEC_DUMMY)) {
+ exec->contended = NULL;
+ ww_acquire_init(&exec->ticket, &reservation_ww_class);
+ return true;
+ }
+
+ drm_exec_unlock_all(exec);
+ exec->num_objects = 0;
+ return true;
+}
+EXPORT_SYMBOL(drm_exec_cleanup);
+
+/* Track the locked object in the array */
+static int drm_exec_obj_locked(struct drm_exec *exec,
+ struct drm_gem_object *obj)
+{
+ if (unlikely(exec->num_objects == exec->max_objects)) {
+ size_t size = exec->max_objects * sizeof(void *);
+ void *tmp;
+
+ tmp = kvrealloc(exec->objects, size, size + PAGE_SIZE,
+ GFP_KERNEL);
+ if (!tmp)
+ return -ENOMEM;
+
+ exec->objects = tmp;
+ exec->max_objects += PAGE_SIZE / sizeof(void *);
+ }
+ drm_gem_object_get(obj);
+ exec->objects[exec->num_objects++] = obj;
+
+ return 0;
+}
+
+/* Make sure the contended object is locked first */
+static int drm_exec_lock_contended(struct drm_exec *exec)
+{
+ struct drm_gem_object *obj = exec->contended;
+ int ret;
+
+ if (likely(!obj))
+ return 0;
+
+ /* Always cleanup the contention so that error handling can kick in */
+ exec->contended = NULL;
+ if (exec->flags & DRM_EXEC_INTERRUPTIBLE_WAIT) {
+ ret = dma_resv_lock_slow_interruptible(obj->resv,
+ &exec->ticket);
+ if (unlikely(ret))
+ goto error_dropref;
+ } else {
+ dma_resv_lock_slow(obj->resv, &exec->ticket);
+ }
+
+ ret = drm_exec_obj_locked(exec, obj);
+ if (unlikely(ret))
+ goto error_unlock;
+
+ exec->prelocked = obj;
+ return 0;
+
+error_unlock:
+ dma_resv_unlock(obj->resv);
+
+error_dropref:
+ drm_gem_object_put(obj);
+ return ret;
+}
+
+/**
+ * drm_exec_lock_obj - lock a GEM object for use
+ * @exec: the drm_exec object with the state
+ * @obj: the GEM object to lock
+ *
+ * Lock a GEM object for use and grab a reference to it.
+ *
+ * Returns: -EDEADLK if a contention is detected, -EALREADY when object is
+ * already locked (can be suppressed by setting the DRM_EXEC_IGNORE_DUPLICATES
+ * flag), -ENOMEM when memory allocation failed and zero for success.
+ */
+int drm_exec_lock_obj(struct drm_exec *exec, struct drm_gem_object *obj)
+{
+ int ret;
+
+ ret = drm_exec_lock_contended(exec);
+ if (unlikely(ret))
+ return ret;
+
+ if (exec->prelocked == obj) {
+ drm_gem_object_put(exec->prelocked);
+ exec->prelocked = NULL;
+ return 0;
+ }
+
+ if (exec->flags & DRM_EXEC_INTERRUPTIBLE_WAIT)
+ ret = dma_resv_lock_interruptible(obj->resv, &exec->ticket);
+ else
+ ret = dma_resv_lock(obj->resv, &exec->ticket);
+
+ if (unlikely(ret == -EDEADLK)) {
+ drm_gem_object_get(obj);
+ exec->contended = obj;
+ return -EDEADLK;
+ }
+
+ if (unlikely(ret == -EALREADY) &&
+ exec->flags & DRM_EXEC_IGNORE_DUPLICATES)
+ return 0;
+
+ if (unlikely(ret))
+ return ret;
+
+ ret = drm_exec_obj_locked(exec, obj);
+ if (ret)
+ goto error_unlock;
+
+ return 0;
+
+error_unlock:
+ dma_resv_unlock(obj->resv);
+ return ret;
+}
+EXPORT_SYMBOL(drm_exec_lock_obj);
+
+/**
+ * drm_exec_unlock_obj - unlock a GEM object in this exec context
+ * @exec: the drm_exec object with the state
+ * @obj: the GEM object to unlock
+ *
+ * Unlock the GEM object and remove it from the collection of locked objects.
+ * Should only be used to unlock the most recently locked objects. It's not time
+ * efficient to unlock objects locked long ago.
+ */
+void drm_exec_unlock_obj(struct drm_exec *exec, struct drm_gem_object *obj)
+{
+ unsigned int i;
+
+ for (i = exec->num_objects; i--;) {
+ if (exec->objects[i] == obj) {
+ dma_resv_unlock(obj->resv);
+ for (++i; i < exec->num_objects; ++i)
+ exec->objects[i - 1] = exec->objects[i];
+ --exec->num_objects;
+ drm_gem_object_put(obj);
+ return;
+ }
+
+ }
+}
+EXPORT_SYMBOL(drm_exec_unlock_obj);
+
+/**
+ * drm_exec_prepare_obj - prepare a GEM object for use
+ * @exec: the drm_exec object with the state
+ * @obj: the GEM object to prepare
+ * @num_fences: how many fences to reserve
+ *
+ * Prepare a GEM object for use by locking it and reserving fence slots.
+ *
+ * Returns: -EDEADLK if a contention is detected, -EALREADY when object is
+ * already locked, -ENOMEM when memory allocation failed and zero for success.
+ */
+int drm_exec_prepare_obj(struct drm_exec *exec, struct drm_gem_object *obj,
+ unsigned int num_fences)
+{
+ int ret;
+
+ ret = drm_exec_lock_obj(exec, obj);
+ if (ret)
+ return ret;
+
+ ret = dma_resv_reserve_fences(obj->resv, num_fences);
+ if (ret) {
+ drm_exec_unlock_obj(exec, obj);
+ return ret;
+ }
+
+ return 0;
+}
+EXPORT_SYMBOL(drm_exec_prepare_obj);
+
+/**
+ * drm_exec_prepare_array - helper to prepare an array of objects
+ * @exec: the drm_exec object with the state
+ * @objects: array of GEM object to prepare
+ * @num_objects: number of GEM objects in the array
+ * @num_fences: number of fences to reserve on each GEM object
+ *
+ * Prepares all GEM objects in an array, aborts on first error.
+ * Reserves @num_fences on each GEM object after locking it.
+ *
+ * Returns: -EDEADLOCK on contention, -EALREADY when object is already locked,
+ * -ENOMEM when memory allocation failed and zero for success.
+ */
+int drm_exec_prepare_array(struct drm_exec *exec,
+ struct drm_gem_object **objects,
+ unsigned int num_objects,
+ unsigned int num_fences)
+{
+ int ret;
+
+ for (unsigned int i = 0; i < num_objects; ++i) {
+ ret = drm_exec_prepare_obj(exec, objects[i], num_fences);
+ if (unlikely(ret))
+ return ret;
+ }
+
+ return 0;
+}
+EXPORT_SYMBOL(drm_exec_prepare_array);
+
+MODULE_DESCRIPTION("DRM execution context");
+MODULE_LICENSE("Dual MIT/GPL");
diff --git a/include/drm/drm_exec.h b/include/drm/drm_exec.h
new file mode 100644
index 000000000000..e8d1e26ad1e4
--- /dev/null
+++ b/include/drm/drm_exec.h
@@ -0,0 +1,121 @@
+/* SPDX-License-Identifier: GPL-2.0 OR MIT */
+
+#ifndef __DRM_EXEC_H__
+#define __DRM_EXEC_H__
+
+#include <linux/ww_mutex.h>
+
+#define DRM_EXEC_INTERRUPTIBLE_WAIT BIT(0)
+#define DRM_EXEC_IGNORE_DUPLICATES BIT(1)
+
+struct drm_gem_object;
+
+/**
+ * struct drm_exec - Execution context
+ */
+struct drm_exec {
+ /**
+ * @flags: Flags to control locking behavior
+ */
+ uint32_t flags;
+
+ /**
+ * @ticket: WW ticket used for acquiring locks
+ */
+ struct ww_acquire_ctx ticket;
+
+ /**
+ * @num_objects: number of objects locked
+ */
+ unsigned int num_objects;
+
+ /**
+ * @max_objects: maximum objects in array
+ */
+ unsigned int max_objects;
+
+ /**
+ * @objects: array of the locked objects
+ */
+ struct drm_gem_object **objects;
+
+ /**
+ * @contended: contended GEM object we backed off for
+ */
+ struct drm_gem_object *contended;
+
+ /**
+ * @prelocked: already locked GEM object due to contention
+ */
+ struct drm_gem_object *prelocked;
+};
+
+/**
+ * drm_exec_for_each_locked_object - iterate over all the locked objects
+ * @exec: drm_exec object
+ * @index: unsigned long index for the iteration
+ * @obj: the current GEM object
+ *
+ * Iterate over all the locked GEM objects inside the drm_exec object.
+ */
+#define drm_exec_for_each_locked_object(exec, index, obj) \
+ for (index = 0, obj = (exec)->objects[0]; \
+ index < (exec)->num_objects; \
+ ++index, obj = (exec)->objects[index])
+
+/**
+ * drm_exec_until_all_locked - loop until all GEM objects are locked
+ * @exec: drm_exec object
+ *
+ * Core functionality of the drm_exec object. Loops until all GEM objects are
+ * locked and no more contention exists. At the beginning of the loop it is
+ * guaranteed that no GEM object is locked.
+ *
+ * Since labels can't be defined local to the loops body we use a jump pointer
+ * to make sure that the retry is only used from within the loops body.
+ */
+#define drm_exec_until_all_locked(exec) \
+ for (void *__drm_exec_retry_ptr; ({ \
+ __label__ __drm_exec_retry; \
+__drm_exec_retry: \
+ __drm_exec_retry_ptr = &&__drm_exec_retry; \
+ (void)__drm_exec_retry_ptr; \
+ drm_exec_cleanup(exec); \
+ });)
+
+/**
+ * drm_exec_retry_on_contention - restart the loop to grap all locks
+ * @exec: drm_exec object
+ *
+ * Control flow helper to continue when a contention was detected and we need to
+ * clean up and re-start the loop to prepare all GEM objects.
+ */
+#define drm_exec_retry_on_contention(exec) \
+ if (unlikely(drm_exec_is_contended(exec))) \
+ goto *__drm_exec_retry_ptr
+
+/**
+ * drm_exec_is_contended - check for contention
+ * @exec: drm_exec object
+ *
+ * Returns true if the drm_exec object has run into some contention while
+ * locking a GEM object and needs to clean up.
+ */
+static inline bool drm_exec_is_contended(struct drm_exec *exec)
+{
+ return !!exec->contended;
+}
+
+void drm_exec_init(struct drm_exec *exec, uint32_t flags);
+void drm_exec_fini(struct drm_exec *exec);
+bool drm_exec_cleanup(struct drm_exec *exec);
+int drm_exec_lock_obj(struct drm_exec *exec, struct drm_gem_object *obj);
+void drm_exec_unlock_obj(struct drm_exec *exec, struct drm_gem_object *obj);
+int drm_exec_prepare_obj(struct drm_exec *exec, struct drm_gem_object *obj,
+ unsigned int num_fences);
+int drm_exec_prepare_array(struct drm_exec *exec,
+ struct drm_gem_object **objects,
+ unsigned int num_objects,
+ unsigned int num_fences);
+
+#endif
--
2.34.1
^ permalink raw reply related [flat|nested] 10+ messages in thread* [PATCH 2/6] drm: add drm_exec selftests v4
2023-06-28 10:44 AMDGPU patches for the drm_exec context Christian König
2023-06-28 10:44 ` [PATCH 1/6] drm: execution context for GEM buffers v6 Christian König
@ 2023-06-28 10:44 ` Christian König
2023-06-28 10:44 ` [PATCH 3/6] drm/amdkfd: switch over to using drm_exec v2 Christian König
` (3 subsequent siblings)
5 siblings, 0 replies; 10+ messages in thread
From: Christian König @ 2023-06-28 10:44 UTC (permalink / raw)
To: dri-devel, amd-gfx, alexander.deucher
Exercise at least all driver facing functions of this new component.
v2: add array test as well
v3: some kunit cleanups
v4: more tests and cleanups
Signed-off-by: Christian König <christian.koenig@amd.com>
---
drivers/gpu/drm/Kconfig | 1 +
drivers/gpu/drm/tests/Makefile | 3 +-
drivers/gpu/drm/tests/drm_exec_test.c | 159 ++++++++++++++++++++++++++
3 files changed, 162 insertions(+), 1 deletion(-)
create mode 100644 drivers/gpu/drm/tests/drm_exec_test.c
diff --git a/drivers/gpu/drm/Kconfig b/drivers/gpu/drm/Kconfig
index c2f3d234c89e..47e0bfe19757 100644
--- a/drivers/gpu/drm/Kconfig
+++ b/drivers/gpu/drm/Kconfig
@@ -80,6 +80,7 @@ config DRM_KUNIT_TEST
select DRM_BUDDY
select DRM_EXPORT_FOR_TESTS if m
select DRM_KUNIT_TEST_HELPERS
+ select DRM_EXEC
default KUNIT_ALL_TESTS
help
This builds unit tests for DRM. This option is not useful for
diff --git a/drivers/gpu/drm/tests/Makefile b/drivers/gpu/drm/tests/Makefile
index bca726a8f483..ba7baa622675 100644
--- a/drivers/gpu/drm/tests/Makefile
+++ b/drivers/gpu/drm/tests/Makefile
@@ -17,6 +17,7 @@ obj-$(CONFIG_DRM_KUNIT_TEST) += \
drm_modes_test.o \
drm_plane_helper_test.o \
drm_probe_helper_test.o \
- drm_rect_test.o
+ drm_rect_test.o \
+ drm_exec_test.o
CFLAGS_drm_mm_test.o := $(DISABLE_STRUCTLEAK_PLUGIN)
diff --git a/drivers/gpu/drm/tests/drm_exec_test.c b/drivers/gpu/drm/tests/drm_exec_test.c
new file mode 100644
index 000000000000..727ac267682e
--- /dev/null
+++ b/drivers/gpu/drm/tests/drm_exec_test.c
@@ -0,0 +1,159 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright 2022 Advanced Micro Devices, Inc.
+ */
+
+#define pr_fmt(fmt) "drm_exec: " fmt
+
+#include <kunit/test.h>
+
+#include <linux/module.h>
+#include <linux/prime_numbers.h>
+
+#include <drm/drm_exec.h>
+#include <drm/drm_device.h>
+#include <drm/drm_gem.h>
+
+#include "../lib/drm_random.h"
+
+static struct drm_device dev;
+
+static void sanitycheck(struct kunit *test)
+{
+ struct drm_exec exec;
+
+ drm_exec_init(&exec, DRM_EXEC_INTERRUPTIBLE_WAIT);
+ drm_exec_fini(&exec);
+ KUNIT_SUCCEED(test);
+}
+
+static void test_lock(struct kunit *test)
+{
+ struct drm_gem_object gobj = { };
+ struct drm_exec exec;
+ int ret;
+
+ drm_gem_private_object_init(&dev, &gobj, PAGE_SIZE);
+
+ drm_exec_init(&exec, DRM_EXEC_INTERRUPTIBLE_WAIT);
+ drm_exec_until_all_locked(&exec) {
+ ret = drm_exec_lock_obj(&exec, &gobj);
+ drm_exec_retry_on_contention(&exec);
+ KUNIT_EXPECT_EQ(test, ret, 0);
+ if (ret)
+ break;
+ }
+ drm_exec_fini(&exec);
+}
+
+static void test_lock_unlock(struct kunit *test)
+{
+ struct drm_gem_object gobj = { };
+ struct drm_exec exec;
+ int ret;
+
+ drm_gem_private_object_init(&dev, &gobj, PAGE_SIZE);
+
+ drm_exec_init(&exec, DRM_EXEC_INTERRUPTIBLE_WAIT);
+ drm_exec_until_all_locked(&exec) {
+ ret = drm_exec_lock_obj(&exec, &gobj);
+ drm_exec_retry_on_contention(&exec);
+ KUNIT_EXPECT_EQ(test, ret, 0);
+ if (ret)
+ break;
+
+ drm_exec_unlock_obj(&exec, &gobj);
+ ret = drm_exec_lock_obj(&exec, &gobj);
+ drm_exec_retry_on_contention(&exec);
+ KUNIT_EXPECT_EQ(test, ret, 0);
+ if (ret)
+ break;
+ }
+ drm_exec_fini(&exec);
+}
+
+static void test_duplicates(struct kunit *test)
+{
+ struct drm_gem_object gobj = { };
+ struct drm_exec exec;
+ int ret;
+
+ drm_gem_private_object_init(&dev, &gobj, PAGE_SIZE);
+
+ drm_exec_init(&exec, DRM_EXEC_IGNORE_DUPLICATES);
+ drm_exec_until_all_locked(&exec) {
+ ret = drm_exec_lock_obj(&exec, &gobj);
+ drm_exec_retry_on_contention(&exec);
+ KUNIT_EXPECT_EQ(test, ret, 0);
+ if (ret)
+ break;
+
+ ret = drm_exec_lock_obj(&exec, &gobj);
+ drm_exec_retry_on_contention(&exec);
+ KUNIT_EXPECT_EQ(test, ret, 0);
+ if (ret)
+ break;
+ }
+ drm_exec_unlock_obj(&exec, &gobj);
+ drm_exec_fini(&exec);
+}
+
+
+
+static void test_prepare(struct kunit *test)
+{
+ struct drm_gem_object gobj = { };
+ struct drm_exec exec;
+ int ret;
+
+ drm_gem_private_object_init(&dev, &gobj, PAGE_SIZE);
+
+ drm_exec_init(&exec, DRM_EXEC_INTERRUPTIBLE_WAIT);
+ drm_exec_until_all_locked(&exec) {
+ ret = drm_exec_prepare_obj(&exec, &gobj, 1);
+ drm_exec_retry_on_contention(&exec);
+ KUNIT_EXPECT_EQ(test, ret, 0);
+ if (ret)
+ break;
+ }
+ drm_exec_fini(&exec);
+}
+
+static void test_prepare_array(struct kunit *test)
+{
+ struct drm_gem_object gobj1 = { };
+ struct drm_gem_object gobj2 = { };
+ struct drm_gem_object *array[] = { &gobj1, &gobj2 };
+ struct drm_exec exec;
+ int ret;
+
+ drm_gem_private_object_init(&dev, &gobj1, PAGE_SIZE);
+ drm_gem_private_object_init(&dev, &gobj2, PAGE_SIZE);
+
+ drm_exec_init(&exec, DRM_EXEC_INTERRUPTIBLE_WAIT);
+ drm_exec_until_all_locked(&exec)
+ ret = drm_exec_prepare_array(&exec, array, ARRAY_SIZE(array),
+ 1);
+ KUNIT_EXPECT_EQ(test, ret, 0);
+ drm_exec_fini(&exec);
+}
+
+static struct kunit_case drm_exec_tests[] = {
+ KUNIT_CASE(sanitycheck),
+ KUNIT_CASE(test_lock),
+ KUNIT_CASE(test_lock_unlock),
+ KUNIT_CASE(test_duplicates),
+ KUNIT_CASE(test_prepare),
+ KUNIT_CASE(test_prepare_array),
+ {}
+};
+
+static struct kunit_suite drm_exec_test_suite = {
+ .name = "drm_exec",
+ .test_cases = drm_exec_tests,
+};
+
+kunit_test_suite(drm_exec_test_suite);
+
+MODULE_AUTHOR("AMD");
+MODULE_LICENSE("GPL and additional rights");
--
2.34.1
^ permalink raw reply related [flat|nested] 10+ messages in thread* [PATCH 3/6] drm/amdkfd: switch over to using drm_exec v2
2023-06-28 10:44 AMDGPU patches for the drm_exec context Christian König
2023-06-28 10:44 ` [PATCH 1/6] drm: execution context for GEM buffers v6 Christian König
2023-06-28 10:44 ` [PATCH 2/6] drm: add drm_exec selftests v4 Christian König
@ 2023-06-28 10:44 ` Christian König
2023-06-28 10:44 ` [PATCH 4/6] drm/amdgpu: use drm_exec for GEM and CSA handling Christian König
` (2 subsequent siblings)
5 siblings, 0 replies; 10+ messages in thread
From: Christian König @ 2023-06-28 10:44 UTC (permalink / raw)
To: dri-devel, amd-gfx, alexander.deucher
Avoids quite a bit of logic and kmalloc overhead.
v2: fix multiple problems pointed out by Felix
Signed-off-by: Christian König <christian.koenig@amd.com>
---
drivers/gpu/drm/amd/amdgpu/Kconfig | 1 +
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h | 5 +-
.../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 299 +++++++-----------
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 18 ++
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 4 +
drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 45 ++-
6 files changed, 162 insertions(+), 210 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/Kconfig b/drivers/gpu/drm/amd/amdgpu/Kconfig
index b91e79c721e2..22d88f8ef527 100644
--- a/drivers/gpu/drm/amd/amdgpu/Kconfig
+++ b/drivers/gpu/drm/amd/amdgpu/Kconfig
@@ -21,6 +21,7 @@ config DRM_AMDGPU
select INTERVAL_TREE
select DRM_BUDDY
select DRM_SUBALLOC_HELPER
+ select DRM_EXEC
# amdgpu depends on ACPI_VIDEO when ACPI is enabled, for select to work
# ACPI_VIDEO's dependencies must also be selected.
select INPUT if ACPI
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
index 2d0406bff84e..1e4cc1fe88fc 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
@@ -25,6 +25,7 @@
#ifndef AMDGPU_AMDKFD_H_INCLUDED
#define AMDGPU_AMDKFD_H_INCLUDED
+#include <linux/list.h>
#include <linux/types.h>
#include <linux/mm.h>
#include <linux/kthread.h>
@@ -32,7 +33,6 @@
#include <linux/mmu_notifier.h>
#include <linux/memremap.h>
#include <kgd_kfd_interface.h>
-#include <drm/ttm/ttm_execbuf_util.h>
#include "amdgpu_sync.h"
#include "amdgpu_vm.h"
#include "amdgpu_xcp.h"
@@ -71,8 +71,7 @@ struct kgd_mem {
struct hmm_range *range;
struct list_head attachments;
/* protected by amdkfd_process_info.lock */
- struct ttm_validate_buffer validate_list;
- struct ttm_validate_buffer resv_list;
+ struct list_head validate_list;
uint32_t domain;
unsigned int mapped_to_gpu_memory;
uint64_t va;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index f61527b800e6..27d331e9419d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -27,6 +27,8 @@
#include <linux/sched/task.h>
#include <drm/ttm/ttm_tt.h>
+#include <drm/drm_exec.h>
+
#include "amdgpu_object.h"
#include "amdgpu_gem.h"
#include "amdgpu_vm.h"
@@ -964,28 +966,20 @@ static void add_kgd_mem_to_kfd_bo_list(struct kgd_mem *mem,
struct amdkfd_process_info *process_info,
bool userptr)
{
- struct ttm_validate_buffer *entry = &mem->validate_list;
- struct amdgpu_bo *bo = mem->bo;
-
- INIT_LIST_HEAD(&entry->head);
- entry->num_shared = 1;
- entry->bo = &bo->tbo;
mutex_lock(&process_info->lock);
if (userptr)
- list_add_tail(&entry->head, &process_info->userptr_valid_list);
+ list_add_tail(&mem->validate_list,
+ &process_info->userptr_valid_list);
else
- list_add_tail(&entry->head, &process_info->kfd_bo_list);
+ list_add_tail(&mem->validate_list, &process_info->kfd_bo_list);
mutex_unlock(&process_info->lock);
}
static void remove_kgd_mem_from_kfd_bo_list(struct kgd_mem *mem,
struct amdkfd_process_info *process_info)
{
- struct ttm_validate_buffer *bo_list_entry;
-
- bo_list_entry = &mem->validate_list;
mutex_lock(&process_info->lock);
- list_del(&bo_list_entry->head);
+ list_del(&mem->validate_list);
mutex_unlock(&process_info->lock);
}
@@ -1072,13 +1066,12 @@ static int init_user_pages(struct kgd_mem *mem, uint64_t user_addr,
* object can track VM updates.
*/
struct bo_vm_reservation_context {
- struct amdgpu_bo_list_entry kfd_bo; /* BO list entry for the KFD BO */
- unsigned int n_vms; /* Number of VMs reserved */
- struct amdgpu_bo_list_entry *vm_pd; /* Array of VM BO list entries */
- struct ww_acquire_ctx ticket; /* Reservation ticket */
- struct list_head list, duplicates; /* BO lists */
- struct amdgpu_sync *sync; /* Pointer to sync object */
- bool reserved; /* Whether BOs are reserved */
+ /* DRM execution context for the reservation */
+ struct drm_exec exec;
+ /* Number of VMs reserved */
+ unsigned int n_vms;
+ /* Pointer to sync object */
+ struct amdgpu_sync *sync;
};
enum bo_vm_match {
@@ -1102,35 +1095,26 @@ static int reserve_bo_and_vm(struct kgd_mem *mem,
WARN_ON(!vm);
- ctx->reserved = false;
ctx->n_vms = 1;
ctx->sync = &mem->sync;
-
- INIT_LIST_HEAD(&ctx->list);
- INIT_LIST_HEAD(&ctx->duplicates);
-
- ctx->vm_pd = kcalloc(ctx->n_vms, sizeof(*ctx->vm_pd), GFP_KERNEL);
- if (!ctx->vm_pd)
- return -ENOMEM;
-
- ctx->kfd_bo.priority = 0;
- ctx->kfd_bo.tv.bo = &bo->tbo;
- ctx->kfd_bo.tv.num_shared = 1;
- list_add(&ctx->kfd_bo.tv.head, &ctx->list);
-
- amdgpu_vm_get_pd_bo(vm, &ctx->list, &ctx->vm_pd[0]);
-
- ret = ttm_eu_reserve_buffers(&ctx->ticket, &ctx->list,
- false, &ctx->duplicates);
- if (ret) {
- pr_err("Failed to reserve buffers in ttm.\n");
- kfree(ctx->vm_pd);
- ctx->vm_pd = NULL;
- return ret;
+ drm_exec_init(&ctx->exec, DRM_EXEC_INTERRUPTIBLE_WAIT);
+ drm_exec_until_all_locked(&ctx->exec) {
+ ret = amdgpu_vm_lock_pd(vm, &ctx->exec, 2);
+ drm_exec_retry_on_contention(&ctx->exec);
+ if (unlikely(ret))
+ goto error;
+
+ ret = drm_exec_lock_obj(&ctx->exec, &bo->tbo.base);
+ drm_exec_retry_on_contention(&ctx->exec);
+ if (unlikely(ret))
+ goto error;
}
-
- ctx->reserved = true;
return 0;
+
+error:
+ pr_err("Failed to reserve buffers in ttm.\n");
+ drm_exec_fini(&ctx->exec);
+ return ret;
}
/**
@@ -1147,63 +1131,39 @@ static int reserve_bo_and_cond_vms(struct kgd_mem *mem,
struct amdgpu_vm *vm, enum bo_vm_match map_type,
struct bo_vm_reservation_context *ctx)
{
- struct amdgpu_bo *bo = mem->bo;
struct kfd_mem_attachment *entry;
- unsigned int i;
+ struct amdgpu_bo *bo = mem->bo;
int ret;
- ctx->reserved = false;
- ctx->n_vms = 0;
- ctx->vm_pd = NULL;
ctx->sync = &mem->sync;
+ drm_exec_init(&ctx->exec, DRM_EXEC_INTERRUPTIBLE_WAIT);
+ drm_exec_until_all_locked(&ctx->exec) {
+ ctx->n_vms = 0;
+ list_for_each_entry(entry, &mem->attachments, list) {
+ if ((vm && vm != entry->bo_va->base.vm) ||
+ (entry->is_mapped != map_type
+ && map_type != BO_VM_ALL))
+ continue;
- INIT_LIST_HEAD(&ctx->list);
- INIT_LIST_HEAD(&ctx->duplicates);
-
- list_for_each_entry(entry, &mem->attachments, list) {
- if ((vm && vm != entry->bo_va->base.vm) ||
- (entry->is_mapped != map_type
- && map_type != BO_VM_ALL))
- continue;
-
- ctx->n_vms++;
- }
-
- if (ctx->n_vms != 0) {
- ctx->vm_pd = kcalloc(ctx->n_vms, sizeof(*ctx->vm_pd),
- GFP_KERNEL);
- if (!ctx->vm_pd)
- return -ENOMEM;
- }
-
- ctx->kfd_bo.priority = 0;
- ctx->kfd_bo.tv.bo = &bo->tbo;
- ctx->kfd_bo.tv.num_shared = 1;
- list_add(&ctx->kfd_bo.tv.head, &ctx->list);
-
- i = 0;
- list_for_each_entry(entry, &mem->attachments, list) {
- if ((vm && vm != entry->bo_va->base.vm) ||
- (entry->is_mapped != map_type
- && map_type != BO_VM_ALL))
- continue;
-
- amdgpu_vm_get_pd_bo(entry->bo_va->base.vm, &ctx->list,
- &ctx->vm_pd[i]);
- i++;
- }
+ ret = amdgpu_vm_lock_pd(entry->bo_va->base.vm,
+ &ctx->exec, 2);
+ drm_exec_retry_on_contention(&ctx->exec);
+ if (unlikely(ret))
+ goto error;
+ ++ctx->n_vms;
+ }
- ret = ttm_eu_reserve_buffers(&ctx->ticket, &ctx->list,
- false, &ctx->duplicates);
- if (ret) {
- pr_err("Failed to reserve buffers in ttm.\n");
- kfree(ctx->vm_pd);
- ctx->vm_pd = NULL;
- return ret;
+ ret = drm_exec_prepare_obj(&ctx->exec, &bo->tbo.base, 1);
+ drm_exec_retry_on_contention(&ctx->exec);
+ if (unlikely(ret))
+ goto error;
}
-
- ctx->reserved = true;
return 0;
+
+error:
+ pr_err("Failed to reserve buffers in ttm.\n");
+ drm_exec_fini(&ctx->exec);
+ return ret;
}
/**
@@ -1224,15 +1184,8 @@ static int unreserve_bo_and_vms(struct bo_vm_reservation_context *ctx,
if (wait)
ret = amdgpu_sync_wait(ctx->sync, intr);
- if (ctx->reserved)
- ttm_eu_backoff_reservation(&ctx->ticket, &ctx->list);
- kfree(ctx->vm_pd);
-
+ drm_exec_fini(&ctx->exec);
ctx->sync = NULL;
-
- ctx->reserved = false;
- ctx->vm_pd = NULL;
-
return ret;
}
@@ -1854,7 +1807,6 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu(
bool use_release_notifier = (mem->bo->kfd_bo == mem);
struct kfd_mem_attachment *entry, *tmp;
struct bo_vm_reservation_context ctx;
- struct ttm_validate_buffer *bo_list_entry;
unsigned int mapped_to_gpu_memory;
int ret;
bool is_imported = false;
@@ -1882,9 +1834,8 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu(
}
/* Make sure restore workers don't access the BO any more */
- bo_list_entry = &mem->validate_list;
mutex_lock(&process_info->lock);
- list_del(&bo_list_entry->head);
+ list_del(&mem->validate_list);
mutex_unlock(&process_info->lock);
/* Cleanup user pages and MMU notifiers */
@@ -2451,14 +2402,14 @@ static int update_invalid_user_pages(struct amdkfd_process_info *process_info,
/* Move all invalidated BOs to the userptr_inval_list */
list_for_each_entry_safe(mem, tmp_mem,
&process_info->userptr_valid_list,
- validate_list.head)
+ validate_list)
if (mem->invalid)
- list_move_tail(&mem->validate_list.head,
+ list_move_tail(&mem->validate_list,
&process_info->userptr_inval_list);
/* Go through userptr_inval_list and update any invalid user_pages */
list_for_each_entry(mem, &process_info->userptr_inval_list,
- validate_list.head) {
+ validate_list) {
invalid = mem->invalid;
if (!invalid)
/* BO hasn't been invalidated since the last
@@ -2538,50 +2489,41 @@ static int update_invalid_user_pages(struct amdkfd_process_info *process_info,
*/
static int validate_invalid_user_pages(struct amdkfd_process_info *process_info)
{
- struct amdgpu_bo_list_entry *pd_bo_list_entries;
- struct list_head resv_list, duplicates;
- struct ww_acquire_ctx ticket;
+ struct ttm_operation_ctx ctx = { false, false };
struct amdgpu_sync sync;
+ struct drm_exec exec;
struct amdgpu_vm *peer_vm;
struct kgd_mem *mem, *tmp_mem;
struct amdgpu_bo *bo;
- struct ttm_operation_ctx ctx = { false, false };
- int i, ret;
-
- pd_bo_list_entries = kcalloc(process_info->n_vms,
- sizeof(struct amdgpu_bo_list_entry),
- GFP_KERNEL);
- if (!pd_bo_list_entries) {
- pr_err("%s: Failed to allocate PD BO list entries\n", __func__);
- ret = -ENOMEM;
- goto out_no_mem;
- }
-
- INIT_LIST_HEAD(&resv_list);
- INIT_LIST_HEAD(&duplicates);
+ int ret;
- /* Get all the page directory BOs that need to be reserved */
- i = 0;
- list_for_each_entry(peer_vm, &process_info->vm_list_head,
- vm_list_node)
- amdgpu_vm_get_pd_bo(peer_vm, &resv_list,
- &pd_bo_list_entries[i++]);
- /* Add the userptr_inval_list entries to resv_list */
- list_for_each_entry(mem, &process_info->userptr_inval_list,
- validate_list.head) {
- list_add_tail(&mem->resv_list.head, &resv_list);
- mem->resv_list.bo = mem->validate_list.bo;
- mem->resv_list.num_shared = mem->validate_list.num_shared;
- }
+ amdgpu_sync_create(&sync);
+ drm_exec_init(&exec, DRM_EXEC_INTERRUPTIBLE_WAIT);
/* Reserve all BOs and page tables for validation */
- ret = ttm_eu_reserve_buffers(&ticket, &resv_list, false, &duplicates);
- WARN(!list_empty(&duplicates), "Duplicates should be empty");
- if (ret)
- goto out_free;
+ drm_exec_until_all_locked(&exec) {
+ /* Reserve all the page directories */
+ list_for_each_entry(peer_vm, &process_info->vm_list_head,
+ vm_list_node) {
+ ret = amdgpu_vm_lock_pd(peer_vm, &exec, 2);
+ drm_exec_retry_on_contention(&exec);
+ if (unlikely(ret))
+ goto unreserve_out;
+ }
- amdgpu_sync_create(&sync);
+ /* Reserve the userptr_inval_list entries to resv_list */
+ list_for_each_entry(mem, &process_info->userptr_inval_list,
+ validate_list) {
+ struct drm_gem_object *gobj;
+
+ gobj = &mem->bo->tbo.base;
+ ret = drm_exec_prepare_obj(&exec, gobj, 1);
+ drm_exec_retry_on_contention(&exec);
+ if (unlikely(ret))
+ goto unreserve_out;
+ }
+ }
ret = process_validate_vms(process_info);
if (ret)
@@ -2590,7 +2532,7 @@ static int validate_invalid_user_pages(struct amdkfd_process_info *process_info)
/* Validate BOs and update GPUVM page tables */
list_for_each_entry_safe(mem, tmp_mem,
&process_info->userptr_inval_list,
- validate_list.head) {
+ validate_list) {
struct kfd_mem_attachment *attachment;
bo = mem->bo;
@@ -2632,12 +2574,9 @@ static int validate_invalid_user_pages(struct amdkfd_process_info *process_info)
ret = process_update_pds(process_info, &sync);
unreserve_out:
- ttm_eu_backoff_reservation(&ticket, &resv_list);
+ drm_exec_fini(&exec);
amdgpu_sync_wait(&sync, false);
amdgpu_sync_free(&sync);
-out_free:
- kfree(pd_bo_list_entries);
-out_no_mem:
return ret;
}
@@ -2653,7 +2592,7 @@ static int confirm_valid_user_pages_locked(struct amdkfd_process_info *process_i
list_for_each_entry_safe(mem, tmp_mem,
&process_info->userptr_inval_list,
- validate_list.head) {
+ validate_list) {
bool valid;
/* keep mem without hmm range at userptr_inval_list */
@@ -2677,7 +2616,7 @@ static int confirm_valid_user_pages_locked(struct amdkfd_process_info *process_i
continue;
}
- list_move_tail(&mem->validate_list.head,
+ list_move_tail(&mem->validate_list,
&process_info->userptr_valid_list);
}
@@ -2787,50 +2726,44 @@ static void amdgpu_amdkfd_restore_userptr_worker(struct work_struct *work)
*/
int amdgpu_amdkfd_gpuvm_restore_process_bos(void *info, struct dma_fence **ef)
{
- struct amdgpu_bo_list_entry *pd_bo_list;
struct amdkfd_process_info *process_info = info;
struct amdgpu_vm *peer_vm;
struct kgd_mem *mem;
- struct bo_vm_reservation_context ctx;
struct amdgpu_amdkfd_fence *new_fence;
- int ret = 0, i;
struct list_head duplicate_save;
struct amdgpu_sync sync_obj;
unsigned long failed_size = 0;
unsigned long total_size = 0;
+ struct drm_exec exec;
+ int ret;
INIT_LIST_HEAD(&duplicate_save);
- INIT_LIST_HEAD(&ctx.list);
- INIT_LIST_HEAD(&ctx.duplicates);
-
- pd_bo_list = kcalloc(process_info->n_vms,
- sizeof(struct amdgpu_bo_list_entry),
- GFP_KERNEL);
- if (!pd_bo_list)
- return -ENOMEM;
- i = 0;
mutex_lock(&process_info->lock);
- list_for_each_entry(peer_vm, &process_info->vm_list_head,
- vm_list_node)
- amdgpu_vm_get_pd_bo(peer_vm, &ctx.list, &pd_bo_list[i++]);
- /* Reserve all BOs and page tables/directory. Add all BOs from
- * kfd_bo_list to ctx.list
- */
- list_for_each_entry(mem, &process_info->kfd_bo_list,
- validate_list.head) {
-
- list_add_tail(&mem->resv_list.head, &ctx.list);
- mem->resv_list.bo = mem->validate_list.bo;
- mem->resv_list.num_shared = mem->validate_list.num_shared;
- }
+ drm_exec_init(&exec, 0);
+ drm_exec_until_all_locked(&exec) {
+ list_for_each_entry(peer_vm, &process_info->vm_list_head,
+ vm_list_node) {
+ ret = amdgpu_vm_lock_pd(peer_vm, &exec, 2);
+ drm_exec_retry_on_contention(&exec);
+ if (unlikely(ret))
+ goto ttm_reserve_fail;
+ }
- ret = ttm_eu_reserve_buffers(&ctx.ticket, &ctx.list,
- false, &duplicate_save);
- if (ret) {
- pr_debug("Memory eviction: TTM Reserve Failed. Try again\n");
- goto ttm_reserve_fail;
+ /* Reserve all BOs and page tables/directory. Add all BOs from
+ * kfd_bo_list to ctx.list
+ */
+ list_for_each_entry(mem, &process_info->kfd_bo_list,
+ validate_list) {
+ struct drm_gem_object *gobj;
+
+ gobj = &mem->bo->tbo.base;
+ ret = drm_exec_prepare_obj(&exec, gobj, 1);
+ drm_exec_retry_on_contention(&exec);
+ if (unlikely(ret))
+ goto ttm_reserve_fail;
+ }
}
amdgpu_sync_create(&sync_obj);
@@ -2848,7 +2781,7 @@ int amdgpu_amdkfd_gpuvm_restore_process_bos(void *info, struct dma_fence **ef)
/* Validate BOs and map them to GPUVM (update VM page tables). */
list_for_each_entry(mem, &process_info->kfd_bo_list,
- validate_list.head) {
+ validate_list) {
struct amdgpu_bo *bo = mem->bo;
uint32_t domain = mem->domain;
@@ -2921,8 +2854,7 @@ int amdgpu_amdkfd_gpuvm_restore_process_bos(void *info, struct dma_fence **ef)
*ef = dma_fence_get(&new_fence->base);
/* Attach new eviction fence to all BOs except pinned ones */
- list_for_each_entry(mem, &process_info->kfd_bo_list,
- validate_list.head) {
+ list_for_each_entry(mem, &process_info->kfd_bo_list, validate_list) {
if (mem->bo->tbo.pin_count)
continue;
@@ -2941,11 +2873,10 @@ int amdgpu_amdkfd_gpuvm_restore_process_bos(void *info, struct dma_fence **ef)
}
validate_map_fail:
- ttm_eu_backoff_reservation(&ctx.ticket, &ctx.list);
amdgpu_sync_free(&sync_obj);
ttm_reserve_fail:
+ drm_exec_fini(&exec);
mutex_unlock(&process_info->lock);
- kfree(pd_bo_list);
return ret;
}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 143d11afe0e5..c5c5f2eb76c5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -34,6 +34,7 @@
#include <drm/amdgpu_drm.h>
#include <drm/drm_drv.h>
#include <drm/ttm/ttm_tt.h>
+#include <drm/drm_exec.h>
#include "amdgpu.h"
#include "amdgpu_trace.h"
#include "amdgpu_amdkfd.h"
@@ -360,6 +361,23 @@ void amdgpu_vm_get_pd_bo(struct amdgpu_vm *vm,
list_add(&entry->tv.head, validated);
}
+/**
+ * amdgpu_vm_lock_pd - lock PD in drm_exec
+ *
+ * @vm: vm providing the BOs
+ * @exec: drm execution context
+ * @num_fences: number of extra fences to reserve
+ *
+ * Lock the VM root PD in the DRM execution context.
+ */
+int amdgpu_vm_lock_pd(struct amdgpu_vm *vm, struct drm_exec *exec,
+ unsigned int num_fences)
+{
+ /* We need at least two fences for the VM PD/PT updates */
+ return drm_exec_prepare_obj(exec, &vm->root.bo->tbo.base,
+ 2 + num_fences);
+}
+
/**
* amdgpu_vm_move_to_lru_tail - move all BOs to the end of LRU
*
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
index 9c85d494f2a2..3e29886e5535 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
@@ -36,6 +36,8 @@
#include "amdgpu_ring.h"
#include "amdgpu_ids.h"
+struct drm_exec;
+
struct amdgpu_bo_va;
struct amdgpu_job;
struct amdgpu_bo_list_entry;
@@ -399,6 +401,8 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm);
void amdgpu_vm_get_pd_bo(struct amdgpu_vm *vm,
struct list_head *validated,
struct amdgpu_bo_list_entry *entry);
+int amdgpu_vm_lock_pd(struct amdgpu_vm *vm, struct drm_exec *exec,
+ unsigned int num_fences);
bool amdgpu_vm_ready(struct amdgpu_vm *vm);
uint64_t amdgpu_vm_generation(struct amdgpu_device *adev, struct amdgpu_vm *vm);
int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm,
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
index 5ff1a5a89d96..93c7a8c87113 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
@@ -24,6 +24,8 @@
#include <linux/types.h>
#include <linux/sched/task.h>
#include <drm/ttm/ttm_tt.h>
+#include <drm/drm_exec.h>
+
#include "amdgpu_sync.h"
#include "amdgpu_object.h"
#include "amdgpu_vm.h"
@@ -1455,9 +1457,7 @@ struct svm_validate_context {
struct svm_range *prange;
bool intr;
DECLARE_BITMAP(bitmap, MAX_GPU_INSTANCE);
- struct ttm_validate_buffer tv[MAX_GPU_INSTANCE];
- struct list_head validate_list;
- struct ww_acquire_ctx ticket;
+ struct drm_exec exec;
};
static int svm_range_reserve_bos(struct svm_validate_context *ctx)
@@ -1467,25 +1467,24 @@ static int svm_range_reserve_bos(struct svm_validate_context *ctx)
uint32_t gpuidx;
int r;
- INIT_LIST_HEAD(&ctx->validate_list);
- for_each_set_bit(gpuidx, ctx->bitmap, MAX_GPU_INSTANCE) {
- pdd = kfd_process_device_from_gpuidx(ctx->process, gpuidx);
- if (!pdd) {
- pr_debug("failed to find device idx %d\n", gpuidx);
- return -EINVAL;
- }
- vm = drm_priv_to_vm(pdd->drm_priv);
-
- ctx->tv[gpuidx].bo = &vm->root.bo->tbo;
- ctx->tv[gpuidx].num_shared = 4;
- list_add(&ctx->tv[gpuidx].head, &ctx->validate_list);
- }
+ drm_exec_init(&ctx->exec, DRM_EXEC_INTERRUPTIBLE_WAIT);
+ drm_exec_until_all_locked(&ctx->exec) {
+ for_each_set_bit(gpuidx, ctx->bitmap, MAX_GPU_INSTANCE) {
+ pdd = kfd_process_device_from_gpuidx(ctx->process, gpuidx);
+ if (!pdd) {
+ pr_debug("failed to find device idx %d\n", gpuidx);
+ r = -EINVAL;
+ goto unreserve_out;
+ }
+ vm = drm_priv_to_vm(pdd->drm_priv);
- r = ttm_eu_reserve_buffers(&ctx->ticket, &ctx->validate_list,
- ctx->intr, NULL);
- if (r) {
- pr_debug("failed %d to reserve bo\n", r);
- return r;
+ r = amdgpu_vm_lock_pd(vm, &ctx->exec, 2);
+ drm_exec_retry_on_contention(&ctx->exec);
+ if (unlikely(r)) {
+ pr_debug("failed %d to reserve bo\n", r);
+ goto unreserve_out;
+ }
+ }
}
for_each_set_bit(gpuidx, ctx->bitmap, MAX_GPU_INSTANCE) {
@@ -1508,13 +1507,13 @@ static int svm_range_reserve_bos(struct svm_validate_context *ctx)
return 0;
unreserve_out:
- ttm_eu_backoff_reservation(&ctx->ticket, &ctx->validate_list);
+ drm_exec_fini(&ctx->exec);
return r;
}
static void svm_range_unreserve_bos(struct svm_validate_context *ctx)
{
- ttm_eu_backoff_reservation(&ctx->ticket, &ctx->validate_list);
+ drm_exec_fini(&ctx->exec);
}
static void *kfd_svm_page_owner(struct kfd_process *p, int32_t gpuidx)
--
2.34.1
^ permalink raw reply related [flat|nested] 10+ messages in thread* [PATCH 4/6] drm/amdgpu: use drm_exec for GEM and CSA handling
2023-06-28 10:44 AMDGPU patches for the drm_exec context Christian König
` (2 preceding siblings ...)
2023-06-28 10:44 ` [PATCH 3/6] drm/amdkfd: switch over to using drm_exec v2 Christian König
@ 2023-06-28 10:44 ` Christian König
2023-06-29 8:20 ` Tatsuyuki Ishi
2023-06-28 10:44 ` [PATCH 5/6] drm/amdgpu: use drm_exec for MES testing Christian König
2023-06-28 10:44 ` [PATCH 6/6] drm/amdgpu: use the new drm_exec object for CS v2 Christian König
5 siblings, 1 reply; 10+ messages in thread
From: Christian König @ 2023-06-28 10:44 UTC (permalink / raw)
To: dri-devel, amd-gfx, alexander.deucher
Start using the new component here as well.
Signed-off-by: Christian König <christian.koenig@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_csa.c | 78 +++++++++++-------------
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 80 ++++++++++++-------------
2 files changed, 71 insertions(+), 87 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_csa.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_csa.c
index 23d054526e7c..720011019741 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_csa.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_csa.c
@@ -22,6 +22,8 @@
* * Author: Monk.liu@amd.com
*/
+#include <drm/drm_exec.h>
+
#include "amdgpu.h"
uint64_t amdgpu_csa_vaddr(struct amdgpu_device *adev)
@@ -65,31 +67,25 @@ int amdgpu_map_static_csa(struct amdgpu_device *adev, struct amdgpu_vm *vm,
struct amdgpu_bo *bo, struct amdgpu_bo_va **bo_va,
uint64_t csa_addr, uint32_t size)
{
- struct ww_acquire_ctx ticket;
- struct list_head list;
- struct amdgpu_bo_list_entry pd;
- struct ttm_validate_buffer csa_tv;
+ struct drm_exec exec;
int r;
- INIT_LIST_HEAD(&list);
- INIT_LIST_HEAD(&csa_tv.head);
- csa_tv.bo = &bo->tbo;
- csa_tv.num_shared = 1;
-
- list_add(&csa_tv.head, &list);
- amdgpu_vm_get_pd_bo(vm, &list, &pd);
-
- r = ttm_eu_reserve_buffers(&ticket, &list, true, NULL);
- if (r) {
- DRM_ERROR("failed to reserve CSA,PD BOs: err=%d\n", r);
- return r;
+ drm_exec_init(&exec, DRM_EXEC_INTERRUPTIBLE_WAIT);
+ drm_exec_until_all_locked(&exec) {
+ r = amdgpu_vm_lock_pd(vm, &exec, 0);
+ if (likely(!r))
+ r = drm_exec_lock_obj(&exec, &bo->tbo.base);
+ drm_exec_retry_on_contention(&exec);
+ if (unlikely(r)) {
+ DRM_ERROR("failed to reserve CSA,PD BOs: err=%d\n", r);
+ goto error;
+ }
}
*bo_va = amdgpu_vm_bo_add(adev, vm, bo);
if (!*bo_va) {
- ttm_eu_backoff_reservation(&ticket, &list);
- DRM_ERROR("failed to create bo_va for static CSA\n");
- return -ENOMEM;
+ r = -ENOMEM;
+ goto error;
}
r = amdgpu_vm_bo_map(adev, *bo_va, csa_addr, 0, size,
@@ -99,48 +95,42 @@ int amdgpu_map_static_csa(struct amdgpu_device *adev, struct amdgpu_vm *vm,
if (r) {
DRM_ERROR("failed to do bo_map on static CSA, err=%d\n", r);
amdgpu_vm_bo_del(adev, *bo_va);
- ttm_eu_backoff_reservation(&ticket, &list);
- return r;
+ goto error;
}
- ttm_eu_backoff_reservation(&ticket, &list);
- return 0;
+error:
+ drm_exec_fini(&exec);
+ return r;
}
int amdgpu_unmap_static_csa(struct amdgpu_device *adev, struct amdgpu_vm *vm,
struct amdgpu_bo *bo, struct amdgpu_bo_va *bo_va,
uint64_t csa_addr)
{
- struct ww_acquire_ctx ticket;
- struct list_head list;
- struct amdgpu_bo_list_entry pd;
- struct ttm_validate_buffer csa_tv;
+ struct drm_exec exec;
int r;
- INIT_LIST_HEAD(&list);
- INIT_LIST_HEAD(&csa_tv.head);
- csa_tv.bo = &bo->tbo;
- csa_tv.num_shared = 1;
-
- list_add(&csa_tv.head, &list);
- amdgpu_vm_get_pd_bo(vm, &list, &pd);
-
- r = ttm_eu_reserve_buffers(&ticket, &list, true, NULL);
- if (r) {
- DRM_ERROR("failed to reserve CSA,PD BOs: err=%d\n", r);
- return r;
+ drm_exec_init(&exec, DRM_EXEC_INTERRUPTIBLE_WAIT);
+ drm_exec_until_all_locked(&exec) {
+ r = amdgpu_vm_lock_pd(vm, &exec, 0);
+ if (likely(!r))
+ r = drm_exec_lock_obj(&exec, &bo->tbo.base);
+ drm_exec_retry_on_contention(&exec);
+ if (unlikely(r)) {
+ DRM_ERROR("failed to reserve CSA,PD BOs: err=%d\n", r);
+ goto error;
+ }
}
r = amdgpu_vm_bo_unmap(adev, bo_va, csa_addr);
if (r) {
DRM_ERROR("failed to do bo_unmap on static CSA, err=%d\n", r);
- ttm_eu_backoff_reservation(&ticket, &list);
- return r;
+ goto error;
}
amdgpu_vm_bo_del(adev, bo_va);
- ttm_eu_backoff_reservation(&ticket, &list);
-
- return 0;
+error:
+ drm_exec_fini(&exec);
+ return r;
}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
index 74055cba3dc9..6811fc866494 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
@@ -33,6 +33,7 @@
#include <drm/amdgpu_drm.h>
#include <drm/drm_drv.h>
+#include <drm/drm_exec.h>
#include <drm/drm_gem_ttm_helper.h>
#include <drm/ttm/ttm_tt.h>
@@ -198,29 +199,24 @@ static void amdgpu_gem_object_close(struct drm_gem_object *obj,
struct amdgpu_fpriv *fpriv = file_priv->driver_priv;
struct amdgpu_vm *vm = &fpriv->vm;
- struct amdgpu_bo_list_entry vm_pd;
- struct list_head list, duplicates;
struct dma_fence *fence = NULL;
- struct ttm_validate_buffer tv;
- struct ww_acquire_ctx ticket;
struct amdgpu_bo_va *bo_va;
+ struct drm_exec exec;
long r;
- INIT_LIST_HEAD(&list);
- INIT_LIST_HEAD(&duplicates);
-
- tv.bo = &bo->tbo;
- tv.num_shared = 2;
- list_add(&tv.head, &list);
-
- amdgpu_vm_get_pd_bo(vm, &list, &vm_pd);
-
- r = ttm_eu_reserve_buffers(&ticket, &list, false, &duplicates);
- if (r) {
- dev_err(adev->dev, "leaking bo va because "
- "we fail to reserve bo (%ld)\n", r);
- return;
+ drm_exec_init(&exec, DRM_EXEC_IGNORE_DUPLICATES);
+ drm_exec_until_all_locked(&exec) {
+ r = drm_exec_prepare_obj(&exec, &bo->tbo.base, 1);
+ drm_exec_retry_on_contention(&exec);
+ if (unlikely(r))
+ goto out_unlock;
+
+ r = amdgpu_vm_lock_pd(vm, &exec, 0);
+ drm_exec_retry_on_contention(&exec);
+ if (unlikely(r))
+ goto out_unlock;
}
+
bo_va = amdgpu_vm_bo_find(vm, bo);
if (!bo_va || --bo_va->ref_count)
goto out_unlock;
@@ -230,6 +226,9 @@ static void amdgpu_gem_object_close(struct drm_gem_object *obj,
goto out_unlock;
r = amdgpu_vm_clear_freed(adev, vm, &fence);
+ if (unlikely(r < 0))
+ dev_err(adev->dev, "failed to clear page "
+ "tables on GEM object close (%ld)\n", r);
if (r || !fence)
goto out_unlock;
@@ -237,10 +236,9 @@ static void amdgpu_gem_object_close(struct drm_gem_object *obj,
dma_fence_put(fence);
out_unlock:
- if (unlikely(r < 0))
- dev_err(adev->dev, "failed to clear page "
- "tables on GEM object close (%ld)\n", r);
- ttm_eu_backoff_reservation(&ticket, &list);
+ if (r)
+ dev_err(adev->dev, "leaking bo va (%ld)\n", r);
+ drm_exec_fini(&exec);
}
static int amdgpu_gem_object_mmap(struct drm_gem_object *obj, struct vm_area_struct *vma)
@@ -675,10 +673,7 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
struct amdgpu_fpriv *fpriv = filp->driver_priv;
struct amdgpu_bo *abo;
struct amdgpu_bo_va *bo_va;
- struct amdgpu_bo_list_entry vm_pd;
- struct ttm_validate_buffer tv;
- struct ww_acquire_ctx ticket;
- struct list_head list, duplicates;
+ struct drm_exec exec;
uint64_t va_flags;
uint64_t vm_size;
int r = 0;
@@ -728,36 +723,37 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
return -EINVAL;
}
- INIT_LIST_HEAD(&list);
- INIT_LIST_HEAD(&duplicates);
if ((args->operation != AMDGPU_VA_OP_CLEAR) &&
!(args->flags & AMDGPU_VM_PAGE_PRT)) {
gobj = drm_gem_object_lookup(filp, args->handle);
if (gobj == NULL)
return -ENOENT;
abo = gem_to_amdgpu_bo(gobj);
- tv.bo = &abo->tbo;
- if (abo->flags & AMDGPU_GEM_CREATE_VM_ALWAYS_VALID)
- tv.num_shared = 1;
- else
- tv.num_shared = 0;
- list_add(&tv.head, &list);
} else {
gobj = NULL;
abo = NULL;
}
- amdgpu_vm_get_pd_bo(&fpriv->vm, &list, &vm_pd);
+ drm_exec_init(&exec, DRM_EXEC_INTERRUPTIBLE_WAIT);
+ drm_exec_until_all_locked(&exec) {
+ if (gobj) {
+ r = drm_exec_lock_obj(&exec, gobj);
+ drm_exec_retry_on_contention(&exec);
+ if (unlikely(r))
+ goto error;
+ }
- r = ttm_eu_reserve_buffers(&ticket, &list, true, &duplicates);
- if (r)
- goto error_unref;
+ r = amdgpu_vm_lock_pd(&fpriv->vm, &exec, 2);
+ drm_exec_retry_on_contention(&exec);
+ if (unlikely(r))
+ goto error;
+ }
if (abo) {
bo_va = amdgpu_vm_bo_find(&fpriv->vm, abo);
if (!bo_va) {
r = -ENOENT;
- goto error_backoff;
+ goto error;
}
} else if (args->operation != AMDGPU_VA_OP_CLEAR) {
bo_va = fpriv->prt_va;
@@ -794,10 +790,8 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
amdgpu_gem_va_update_vm(adev, &fpriv->vm, bo_va,
args->operation);
-error_backoff:
- ttm_eu_backoff_reservation(&ticket, &list);
-
-error_unref:
+error:
+ drm_exec_fini(&exec);
drm_gem_object_put(gobj);
return r;
}
--
2.34.1
^ permalink raw reply related [flat|nested] 10+ messages in thread* Re: [PATCH 4/6] drm/amdgpu: use drm_exec for GEM and CSA handling
2023-06-28 10:44 ` [PATCH 4/6] drm/amdgpu: use drm_exec for GEM and CSA handling Christian König
@ 2023-06-29 8:20 ` Tatsuyuki Ishi
0 siblings, 0 replies; 10+ messages in thread
From: Tatsuyuki Ishi @ 2023-06-29 8:20 UTC (permalink / raw)
To: Christian König, dri-devel, amd-gfx, alexander.deucher
On 6/28/23 19:44, Christian König wrote:
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
> index 74055cba3dc9..6811fc866494 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
> @@ -728,36 +723,37 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
> return -EINVAL;
> }
>
> - INIT_LIST_HEAD(&list);
> - INIT_LIST_HEAD(&duplicates);
> if ((args->operation != AMDGPU_VA_OP_CLEAR) &&
> !(args->flags & AMDGPU_VM_PAGE_PRT)) {
> gobj = drm_gem_object_lookup(filp, args->handle);
> if (gobj == NULL)
> return -ENOENT;
> abo = gem_to_amdgpu_bo(gobj);
> - tv.bo = &abo->tbo;
> - if (abo->flags & AMDGPU_GEM_CREATE_VM_ALWAYS_VALID)
> - tv.num_shared = 1;
> - else
> - tv.num_shared = 0;
> - list_add(&tv.head, &list);
> } else {
> gobj = NULL;
> abo = NULL;
> }
>
> - amdgpu_vm_get_pd_bo(&fpriv->vm, &list, &vm_pd);
> + drm_exec_init(&exec, DRM_EXEC_INTERRUPTIBLE_WAIT);
Sorry, I missed this last time, but this needs to allow duplicates as well or mapping
always_valid BOs doesn't work.
> + drm_exec_until_all_locked(&exec) {
> + if (gobj) {
> + r = drm_exec_lock_obj(&exec, gobj);
> + drm_exec_retry_on_contention(&exec);
> + if (unlikely(r))
> + goto error;
> + }
>
> - r = ttm_eu_reserve_buffers(&ticket, &list, true, &duplicates);
> - if (r)
> - goto error_unref;
> + r = amdgpu_vm_lock_pd(&fpriv->vm, &exec, 2);
> + drm_exec_retry_on_contention(&exec);
> + if (unlikely(r))
> + goto error;
> + }
>
> if (abo) {
> bo_va = amdgpu_vm_bo_find(&fpriv->vm, abo);
> if (!bo_va) {
> r = -ENOENT;
> - goto error_backoff;
> + goto error;
> }
> } else if (args->operation != AMDGPU_VA_OP_CLEAR) {
> bo_va = fpriv->prt_va;
> @@ -794,10 +790,8 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
> amdgpu_gem_va_update_vm(adev, &fpriv->vm, bo_va,
> args->operation);
>
> -error_backoff:
> - ttm_eu_backoff_reservation(&ticket, &list);
> -
> -error_unref:
> +error:
> + drm_exec_fini(&exec);
> drm_gem_object_put(gobj);
> return r;
> }
^ permalink raw reply [flat|nested] 10+ messages in thread
* [PATCH 5/6] drm/amdgpu: use drm_exec for MES testing
2023-06-28 10:44 AMDGPU patches for the drm_exec context Christian König
` (3 preceding siblings ...)
2023-06-28 10:44 ` [PATCH 4/6] drm/amdgpu: use drm_exec for GEM and CSA handling Christian König
@ 2023-06-28 10:44 ` Christian König
2023-06-28 10:44 ` [PATCH 6/6] drm/amdgpu: use the new drm_exec object for CS v2 Christian König
5 siblings, 0 replies; 10+ messages in thread
From: Christian König @ 2023-06-28 10:44 UTC (permalink / raw)
To: dri-devel, amd-gfx, alexander.deucher
Start using the new component here as well.
Signed-off-by: Christian König <christian.koenig@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c | 90 ++++++++++++-------------
1 file changed, 43 insertions(+), 47 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c
index e9091ebfe230..bc28ce15b70d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c
@@ -22,6 +22,7 @@
*/
#include <linux/firmware.h>
+#include <drm/drm_exec.h>
#include "amdgpu_mes.h"
#include "amdgpu.h"
@@ -1168,34 +1169,31 @@ int amdgpu_mes_ctx_map_meta_data(struct amdgpu_device *adev,
struct amdgpu_mes_ctx_data *ctx_data)
{
struct amdgpu_bo_va *bo_va;
- struct ww_acquire_ctx ticket;
- struct list_head list;
- struct amdgpu_bo_list_entry pd;
- struct ttm_validate_buffer csa_tv;
struct amdgpu_sync sync;
+ struct drm_exec exec;
int r;
amdgpu_sync_create(&sync);
- INIT_LIST_HEAD(&list);
- INIT_LIST_HEAD(&csa_tv.head);
- csa_tv.bo = &ctx_data->meta_data_obj->tbo;
- csa_tv.num_shared = 1;
-
- list_add(&csa_tv.head, &list);
- amdgpu_vm_get_pd_bo(vm, &list, &pd);
-
- r = ttm_eu_reserve_buffers(&ticket, &list, true, NULL);
- if (r) {
- DRM_ERROR("failed to reserve meta data BO: err=%d\n", r);
- return r;
+ drm_exec_init(&exec, 0);
+ drm_exec_until_all_locked(&exec) {
+ r = drm_exec_lock_obj(&exec,
+ &ctx_data->meta_data_obj->tbo.base);
+ drm_exec_retry_on_contention(&exec);
+ if (unlikely(r))
+ goto error_fini_exec;
+
+ r = amdgpu_vm_lock_pd(vm, &exec, 0);
+ drm_exec_retry_on_contention(&exec);
+ if (unlikely(r))
+ goto error_fini_exec;
}
bo_va = amdgpu_vm_bo_add(adev, vm, ctx_data->meta_data_obj);
if (!bo_va) {
- ttm_eu_backoff_reservation(&ticket, &list);
DRM_ERROR("failed to create bo_va for meta data BO\n");
- return -ENOMEM;
+ r = -ENOMEM;
+ goto error_fini_exec;
}
r = amdgpu_vm_bo_map(adev, bo_va, ctx_data->meta_data_gpu_addr, 0,
@@ -1205,33 +1203,35 @@ int amdgpu_mes_ctx_map_meta_data(struct amdgpu_device *adev,
if (r) {
DRM_ERROR("failed to do bo_map on meta data, err=%d\n", r);
- goto error;
+ goto error_del_bo_va;
}
r = amdgpu_vm_bo_update(adev, bo_va, false);
if (r) {
DRM_ERROR("failed to do vm_bo_update on meta data\n");
- goto error;
+ goto error_del_bo_va;
}
amdgpu_sync_fence(&sync, bo_va->last_pt_update);
r = amdgpu_vm_update_pdes(adev, vm, false);
if (r) {
DRM_ERROR("failed to update pdes on meta data\n");
- goto error;
+ goto error_del_bo_va;
}
amdgpu_sync_fence(&sync, vm->last_update);
amdgpu_sync_wait(&sync, false);
- ttm_eu_backoff_reservation(&ticket, &list);
+ drm_exec_fini(&exec);
amdgpu_sync_free(&sync);
ctx_data->meta_data_va = bo_va;
return 0;
-error:
+error_del_bo_va:
amdgpu_vm_bo_del(adev, bo_va);
- ttm_eu_backoff_reservation(&ticket, &list);
+
+error_fini_exec:
+ drm_exec_fini(&exec);
amdgpu_sync_free(&sync);
return r;
}
@@ -1242,34 +1242,30 @@ int amdgpu_mes_ctx_unmap_meta_data(struct amdgpu_device *adev,
struct amdgpu_bo_va *bo_va = ctx_data->meta_data_va;
struct amdgpu_bo *bo = ctx_data->meta_data_obj;
struct amdgpu_vm *vm = bo_va->base.vm;
- struct amdgpu_bo_list_entry vm_pd;
- struct list_head list, duplicates;
- struct dma_fence *fence = NULL;
- struct ttm_validate_buffer tv;
- struct ww_acquire_ctx ticket;
- long r = 0;
-
- INIT_LIST_HEAD(&list);
- INIT_LIST_HEAD(&duplicates);
-
- tv.bo = &bo->tbo;
- tv.num_shared = 2;
- list_add(&tv.head, &list);
-
- amdgpu_vm_get_pd_bo(vm, &list, &vm_pd);
-
- r = ttm_eu_reserve_buffers(&ticket, &list, false, &duplicates);
- if (r) {
- dev_err(adev->dev, "leaking bo va because "
- "we fail to reserve bo (%ld)\n", r);
- return r;
+ struct dma_fence *fence;
+ struct drm_exec exec;
+ long r;
+
+ drm_exec_init(&exec, 0);
+ drm_exec_until_all_locked(&exec) {
+ r = drm_exec_lock_obj(&exec,
+ &ctx_data->meta_data_obj->tbo.base);
+ drm_exec_retry_on_contention(&exec);
+ if (unlikely(r))
+ goto out_unlock;
+
+ r = amdgpu_vm_lock_pd(vm, &exec, 0);
+ drm_exec_retry_on_contention(&exec);
+ if (unlikely(r))
+ goto out_unlock;
}
amdgpu_vm_bo_del(adev, bo_va);
if (!amdgpu_vm_ready(vm))
goto out_unlock;
- r = dma_resv_get_singleton(bo->tbo.base.resv, DMA_RESV_USAGE_BOOKKEEP, &fence);
+ r = dma_resv_get_singleton(bo->tbo.base.resv, DMA_RESV_USAGE_BOOKKEEP,
+ &fence);
if (r)
goto out_unlock;
if (fence) {
@@ -1288,7 +1284,7 @@ int amdgpu_mes_ctx_unmap_meta_data(struct amdgpu_device *adev,
out_unlock:
if (unlikely(r < 0))
dev_err(adev->dev, "failed to clear page tables (%ld)\n", r);
- ttm_eu_backoff_reservation(&ticket, &list);
+ drm_exec_fini(&exec);
return r;
}
--
2.34.1
^ permalink raw reply related [flat|nested] 10+ messages in thread* [PATCH 6/6] drm/amdgpu: use the new drm_exec object for CS v2
2023-06-28 10:44 AMDGPU patches for the drm_exec context Christian König
` (4 preceding siblings ...)
2023-06-28 10:44 ` [PATCH 5/6] drm/amdgpu: use drm_exec for MES testing Christian König
@ 2023-06-28 10:44 ` Christian König
2023-06-29 8:20 ` Tatsuyuki Ishi
5 siblings, 1 reply; 10+ messages in thread
From: Christian König @ 2023-06-28 10:44 UTC (permalink / raw)
To: dri-devel, amd-gfx, alexander.deucher
Use the new component here as well and remove the old handling.
v2: drop dupplicate handling
Signed-off-by: Christian König <christian.koenig@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 -
drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c | 71 ++-----
drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.h | 5 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 211 +++++++++-----------
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.h | 7 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 22 --
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 3 -
7 files changed, 116 insertions(+), 204 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index a84bd4a0c421..b49c34a5c04b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -53,7 +53,6 @@
#include <drm/ttm/ttm_bo.h>
#include <drm/ttm/ttm_placement.h>
-#include <drm/ttm/ttm_execbuf_util.h>
#include <drm/amdgpu_drm.h>
#include <drm/drm_gem.h>
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
index 252a876b0725..b6298e901cbd 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
@@ -28,6 +28,7 @@
* Christian König <deathsimple@vodafone.de>
*/
+#include <linux/sort.h>
#include <linux/uaccess.h>
#include "amdgpu.h"
@@ -50,13 +51,20 @@ static void amdgpu_bo_list_free(struct kref *ref)
refcount);
struct amdgpu_bo_list_entry *e;
- amdgpu_bo_list_for_each_entry(e, list) {
- struct amdgpu_bo *bo = ttm_to_amdgpu_bo(e->tv.bo);
+ amdgpu_bo_list_for_each_entry(e, list)
+ amdgpu_bo_unref(&e->bo);
+ call_rcu(&list->rhead, amdgpu_bo_list_free_rcu);
+}
- amdgpu_bo_unref(&bo);
- }
+static int amdgpu_bo_list_entry_cmp(const void *_a, const void *_b)
+{
+ const struct amdgpu_bo_list_entry *a = _a, *b = _b;
- call_rcu(&list->rhead, amdgpu_bo_list_free_rcu);
+ if (a->priority > b->priority)
+ return 1;
+ if (a->priority < b->priority)
+ return -1;
+ return 0;
}
int amdgpu_bo_list_create(struct amdgpu_device *adev, struct drm_file *filp,
@@ -118,7 +126,7 @@ int amdgpu_bo_list_create(struct amdgpu_device *adev, struct drm_file *filp,
entry->priority = min(info[i].bo_priority,
AMDGPU_BO_LIST_MAX_PRIORITY);
- entry->tv.bo = &bo->tbo;
+ entry->bo = bo;
if (bo->preferred_domains == AMDGPU_GEM_DOMAIN_GDS)
list->gds_obj = bo;
@@ -133,6 +141,8 @@ int amdgpu_bo_list_create(struct amdgpu_device *adev, struct drm_file *filp,
list->first_userptr = first_userptr;
list->num_entries = num_entries;
+ sort(array, last_entry, sizeof(struct amdgpu_bo_list_entry),
+ amdgpu_bo_list_entry_cmp, NULL);
trace_amdgpu_cs_bo_status(list->num_entries, total_size);
@@ -141,16 +151,10 @@ int amdgpu_bo_list_create(struct amdgpu_device *adev, struct drm_file *filp,
return 0;
error_free:
- for (i = 0; i < last_entry; ++i) {
- struct amdgpu_bo *bo = ttm_to_amdgpu_bo(array[i].tv.bo);
-
- amdgpu_bo_unref(&bo);
- }
- for (i = first_userptr; i < num_entries; ++i) {
- struct amdgpu_bo *bo = ttm_to_amdgpu_bo(array[i].tv.bo);
-
- amdgpu_bo_unref(&bo);
- }
+ for (i = 0; i < last_entry; ++i)
+ amdgpu_bo_unref(&array[i].bo);
+ for (i = first_userptr; i < num_entries; ++i)
+ amdgpu_bo_unref(&array[i].bo);
kvfree(list);
return r;
@@ -182,41 +186,6 @@ int amdgpu_bo_list_get(struct amdgpu_fpriv *fpriv, int id,
return -ENOENT;
}
-void amdgpu_bo_list_get_list(struct amdgpu_bo_list *list,
- struct list_head *validated)
-{
- /* This is based on the bucket sort with O(n) time complexity.
- * An item with priority "i" is added to bucket[i]. The lists are then
- * concatenated in descending order.
- */
- struct list_head bucket[AMDGPU_BO_LIST_NUM_BUCKETS];
- struct amdgpu_bo_list_entry *e;
- unsigned i;
-
- for (i = 0; i < AMDGPU_BO_LIST_NUM_BUCKETS; i++)
- INIT_LIST_HEAD(&bucket[i]);
-
- /* Since buffers which appear sooner in the relocation list are
- * likely to be used more often than buffers which appear later
- * in the list, the sort mustn't change the ordering of buffers
- * with the same priority, i.e. it must be stable.
- */
- amdgpu_bo_list_for_each_entry(e, list) {
- struct amdgpu_bo *bo = ttm_to_amdgpu_bo(e->tv.bo);
- unsigned priority = e->priority;
-
- if (!bo->parent)
- list_add_tail(&e->tv.head, &bucket[priority]);
-
- e->user_pages = NULL;
- e->range = NULL;
- }
-
- /* Connect the sorted buckets in the output list. */
- for (i = 0; i < AMDGPU_BO_LIST_NUM_BUCKETS; i++)
- list_splice(&bucket[i], validated);
-}
-
void amdgpu_bo_list_put(struct amdgpu_bo_list *list)
{
kref_put(&list->refcount, amdgpu_bo_list_free);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.h
index ededdc01ca28..26c01cb131f2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.h
@@ -23,7 +23,6 @@
#ifndef __AMDGPU_BO_LIST_H__
#define __AMDGPU_BO_LIST_H__
-#include <drm/ttm/ttm_execbuf_util.h>
#include <drm/amdgpu_drm.h>
struct hmm_range;
@@ -36,7 +35,7 @@ struct amdgpu_bo_va;
struct amdgpu_fpriv;
struct amdgpu_bo_list_entry {
- struct ttm_validate_buffer tv;
+ struct amdgpu_bo *bo;
struct amdgpu_bo_va *bo_va;
uint32_t priority;
struct page **user_pages;
@@ -60,8 +59,6 @@ struct amdgpu_bo_list {
int amdgpu_bo_list_get(struct amdgpu_fpriv *fpriv, int id,
struct amdgpu_bo_list **result);
-void amdgpu_bo_list_get_list(struct amdgpu_bo_list *list,
- struct list_head *validated);
void amdgpu_bo_list_put(struct amdgpu_bo_list *list);
int amdgpu_bo_create_list_entry_array(struct drm_amdgpu_bo_list_in *in,
struct drm_amdgpu_bo_list_entry **info_param);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index d9503882ea97..76eb3afa8085 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -65,6 +65,7 @@ static int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p,
}
amdgpu_sync_create(&p->sync);
+ drm_exec_init(&p->exec, DRM_EXEC_INTERRUPTIBLE_WAIT);
return 0;
}
@@ -125,7 +126,6 @@ static int amdgpu_cs_p1_user_fence(struct amdgpu_cs_parser *p,
uint32_t *offset)
{
struct drm_gem_object *gobj;
- struct amdgpu_bo *bo;
unsigned long size;
int r;
@@ -133,21 +133,16 @@ static int amdgpu_cs_p1_user_fence(struct amdgpu_cs_parser *p,
if (gobj == NULL)
return -EINVAL;
- bo = amdgpu_bo_ref(gem_to_amdgpu_bo(gobj));
- p->uf_entry.priority = 0;
- p->uf_entry.tv.bo = &bo->tbo;
- /* One for TTM and two for the CS job */
- p->uf_entry.tv.num_shared = 3;
-
+ p->uf_bo = amdgpu_bo_ref(gem_to_amdgpu_bo(gobj));
drm_gem_object_put(gobj);
- size = amdgpu_bo_size(bo);
+ size = amdgpu_bo_size(p->uf_bo);
if (size != PAGE_SIZE || (data->offset + 8) > size) {
r = -EINVAL;
goto error_unref;
}
- if (amdgpu_ttm_tt_get_usermm(bo->tbo.ttm)) {
+ if (amdgpu_ttm_tt_get_usermm(p->uf_bo->tbo.ttm)) {
r = -EINVAL;
goto error_unref;
}
@@ -157,7 +152,7 @@ static int amdgpu_cs_p1_user_fence(struct amdgpu_cs_parser *p,
return 0;
error_unref:
- amdgpu_bo_unref(&bo);
+ amdgpu_bo_unref(&p->uf_bo);
return r;
}
@@ -314,7 +309,7 @@ static int amdgpu_cs_pass1(struct amdgpu_cs_parser *p,
goto free_all_kdata;
}
- if (p->uf_entry.tv.bo)
+ if (p->uf_bo)
p->gang_leader->uf_addr = uf_offset;
kvfree(chunk_array);
@@ -359,7 +354,7 @@ static int amdgpu_cs_p2_ib(struct amdgpu_cs_parser *p,
ib = &job->ibs[job->num_ibs++];
/* MM engine doesn't support user fences */
- if (p->uf_entry.tv.bo && ring->funcs->no_user_fence)
+ if (p->uf_bo && ring->funcs->no_user_fence)
return -EINVAL;
if (chunk_ib->ip_type == AMDGPU_HW_IP_GFX &&
@@ -844,55 +839,18 @@ static int amdgpu_cs_bo_validate(void *param, struct amdgpu_bo *bo)
return r;
}
-static int amdgpu_cs_list_validate(struct amdgpu_cs_parser *p,
- struct list_head *validated)
-{
- struct ttm_operation_ctx ctx = { true, false };
- struct amdgpu_bo_list_entry *lobj;
- int r;
-
- list_for_each_entry(lobj, validated, tv.head) {
- struct amdgpu_bo *bo = ttm_to_amdgpu_bo(lobj->tv.bo);
- struct mm_struct *usermm;
-
- usermm = amdgpu_ttm_tt_get_usermm(bo->tbo.ttm);
- if (usermm && usermm != current->mm)
- return -EPERM;
-
- if (amdgpu_ttm_tt_is_userptr(bo->tbo.ttm) &&
- lobj->user_invalidated && lobj->user_pages) {
- amdgpu_bo_placement_from_domain(bo,
- AMDGPU_GEM_DOMAIN_CPU);
- r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
- if (r)
- return r;
-
- amdgpu_ttm_tt_set_user_pages(bo->tbo.ttm,
- lobj->user_pages);
- }
-
- r = amdgpu_cs_bo_validate(p, bo);
- if (r)
- return r;
-
- kvfree(lobj->user_pages);
- lobj->user_pages = NULL;
- }
- return 0;
-}
-
static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
union drm_amdgpu_cs *cs)
{
struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
+ struct ttm_operation_ctx ctx = { true, false };
struct amdgpu_vm *vm = &fpriv->vm;
struct amdgpu_bo_list_entry *e;
- struct list_head duplicates;
+ struct drm_gem_object *obj;
+ unsigned long index;
unsigned int i;
int r;
- INIT_LIST_HEAD(&p->validated);
-
/* p->bo_list could already be assigned if AMDGPU_CHUNK_ID_BO_HANDLES is present */
if (cs->in.bo_list_handle) {
if (p->bo_list)
@@ -912,25 +870,13 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
mutex_lock(&p->bo_list->bo_list_mutex);
- /* One for TTM and one for the CS job */
- amdgpu_bo_list_for_each_entry(e, p->bo_list)
- e->tv.num_shared = 2;
-
- amdgpu_bo_list_get_list(p->bo_list, &p->validated);
-
- INIT_LIST_HEAD(&duplicates);
- amdgpu_vm_get_pd_bo(&fpriv->vm, &p->validated, &p->vm_pd);
-
- if (p->uf_entry.tv.bo && !ttm_to_amdgpu_bo(p->uf_entry.tv.bo)->parent)
- list_add(&p->uf_entry.tv.head, &p->validated);
-
/* Get userptr backing pages. If pages are updated after registered
* in amdgpu_gem_userptr_ioctl(), amdgpu_cs_list_validate() will do
* amdgpu_ttm_backend_bind() to flush and invalidate new pages
*/
amdgpu_bo_list_for_each_userptr_entry(e, p->bo_list) {
- struct amdgpu_bo *bo = ttm_to_amdgpu_bo(e->tv.bo);
bool userpage_invalidated = false;
+ struct amdgpu_bo *bo = e->bo;
int i;
e->user_pages = kvmalloc_array(bo->tbo.ttm->num_pages,
@@ -958,18 +904,57 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
e->user_invalidated = userpage_invalidated;
}
- r = ttm_eu_reserve_buffers(&p->ticket, &p->validated, true,
- &duplicates);
- if (unlikely(r != 0)) {
- if (r != -ERESTARTSYS)
- DRM_ERROR("ttm_eu_reserve_buffers failed.\n");
- goto out_free_user_pages;
+ drm_exec_until_all_locked(&p->exec) {
+ r = amdgpu_vm_lock_pd(&fpriv->vm, &p->exec, 1 + p->gang_size);
+ drm_exec_retry_on_contention(&p->exec);
+ if (unlikely(r))
+ goto out_free_user_pages;
+
+ amdgpu_bo_list_for_each_entry(e, p->bo_list) {
+ /* One fence for TTM and one for each CS job */
+ r = drm_exec_prepare_obj(&p->exec, &e->bo->tbo.base,
+ 1 + p->gang_size);
+ drm_exec_retry_on_contention(&p->exec);
+ if (unlikely(r))
+ goto out_free_user_pages;
+
+ e->bo_va = amdgpu_vm_bo_find(vm, e->bo);
+ e->range = NULL;
+ }
+
+ if (p->uf_bo) {
+ r = drm_exec_prepare_obj(&p->exec, &p->uf_bo->tbo.base,
+ 1 + p->gang_size);
+ drm_exec_retry_on_contention(&p->exec);
+ if (unlikely(r))
+ goto out_free_user_pages;
+ }
}
- amdgpu_bo_list_for_each_entry(e, p->bo_list) {
- struct amdgpu_bo *bo = ttm_to_amdgpu_bo(e->tv.bo);
+ amdgpu_bo_list_for_each_userptr_entry(e, p->bo_list) {
+ struct mm_struct *usermm;
- e->bo_va = amdgpu_vm_bo_find(vm, bo);
+ usermm = amdgpu_ttm_tt_get_usermm(e->bo->tbo.ttm);
+ if (usermm && usermm != current->mm) {
+ r = -EPERM;
+ goto out_free_user_pages;
+ }
+
+ if (amdgpu_ttm_tt_is_userptr(e->bo->tbo.ttm) &&
+ e->user_invalidated && e->user_pages) {
+ amdgpu_bo_placement_from_domain(e->bo,
+ AMDGPU_GEM_DOMAIN_CPU);
+ r = ttm_bo_validate(&e->bo->tbo, &e->bo->placement,
+ &ctx);
+ if (r)
+ goto out_free_user_pages;
+
+ amdgpu_ttm_tt_set_user_pages(e->bo->tbo.ttm,
+ e->user_pages);
+ }
+
+ kvfree(e->user_pages);
+ e->user_pages = NULL;
}
amdgpu_cs_get_threshold_for_moves(p->adev, &p->bytes_moved_threshold,
@@ -981,25 +966,21 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
amdgpu_cs_bo_validate, p);
if (r) {
DRM_ERROR("amdgpu_vm_validate_pt_bos() failed.\n");
- goto error_validate;
+ goto out_free_user_pages;
}
- r = amdgpu_cs_list_validate(p, &duplicates);
- if (r)
- goto error_validate;
-
- r = amdgpu_cs_list_validate(p, &p->validated);
- if (r)
- goto error_validate;
-
- if (p->uf_entry.tv.bo) {
- struct amdgpu_bo *uf = ttm_to_amdgpu_bo(p->uf_entry.tv.bo);
+ drm_exec_for_each_locked_object(&p->exec, index, obj) {
+ r = amdgpu_cs_bo_validate(p, gem_to_amdgpu_bo(obj));
+ if (unlikely(r))
+ goto out_free_user_pages;
+ }
- r = amdgpu_ttm_alloc_gart(&uf->tbo);
- if (r)
- goto error_validate;
+ if (p->uf_bo) {
+ r = amdgpu_ttm_alloc_gart(&p->uf_bo->tbo);
+ if (unlikely(r))
+ goto out_free_user_pages;
- p->gang_leader->uf_addr += amdgpu_bo_gpu_offset(uf);
+ p->gang_leader->uf_addr += amdgpu_bo_gpu_offset(p->uf_bo);
}
amdgpu_cs_report_moved_bytes(p->adev, p->bytes_moved,
@@ -1011,12 +992,9 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
p->bo_list->oa_obj);
return 0;
-error_validate:
- ttm_eu_backoff_reservation(&p->ticket, &p->validated);
-
out_free_user_pages:
amdgpu_bo_list_for_each_userptr_entry(e, p->bo_list) {
- struct amdgpu_bo *bo = ttm_to_amdgpu_bo(e->tv.bo);
+ struct amdgpu_bo *bo = e->bo;
if (!e->user_pages)
continue;
@@ -1122,7 +1100,6 @@ static int amdgpu_cs_vm_handling(struct amdgpu_cs_parser *p)
struct amdgpu_vm *vm = &fpriv->vm;
struct amdgpu_bo_list_entry *e;
struct amdgpu_bo_va *bo_va;
- struct amdgpu_bo *bo;
unsigned int i;
int r;
@@ -1151,11 +1128,6 @@ static int amdgpu_cs_vm_handling(struct amdgpu_cs_parser *p)
}
amdgpu_bo_list_for_each_entry(e, p->bo_list) {
- /* ignore duplicates */
- bo = ttm_to_amdgpu_bo(e->tv.bo);
- if (!bo)
- continue;
-
bo_va = e->bo_va;
if (bo_va == NULL)
continue;
@@ -1193,7 +1165,7 @@ static int amdgpu_cs_vm_handling(struct amdgpu_cs_parser *p)
if (amdgpu_vm_debug) {
/* Invalidate all BOs to test for userspace bugs */
amdgpu_bo_list_for_each_entry(e, p->bo_list) {
- struct amdgpu_bo *bo = ttm_to_amdgpu_bo(e->tv.bo);
+ struct amdgpu_bo *bo = e->bo;
/* ignore duplicates */
if (!bo)
@@ -1210,8 +1182,9 @@ static int amdgpu_cs_sync_rings(struct amdgpu_cs_parser *p)
{
struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
struct drm_gpu_scheduler *sched;
- struct amdgpu_bo_list_entry *e;
+ struct drm_gem_object *obj;
struct dma_fence *fence;
+ unsigned long index;
unsigned int i;
int r;
@@ -1222,8 +1195,9 @@ static int amdgpu_cs_sync_rings(struct amdgpu_cs_parser *p)
return r;
}
- list_for_each_entry(e, &p->validated, tv.head) {
- struct amdgpu_bo *bo = ttm_to_amdgpu_bo(e->tv.bo);
+ drm_exec_for_each_locked_object(&p->exec, index, obj) {
+ struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj);
+
struct dma_resv *resv = bo->tbo.base.resv;
enum amdgpu_sync_mode sync_mode;
@@ -1287,6 +1261,8 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
struct amdgpu_job *leader = p->gang_leader;
struct amdgpu_bo_list_entry *e;
+ struct drm_gem_object *gobj;
+ unsigned long index;
unsigned int i;
uint64_t seq;
int r;
@@ -1325,9 +1301,8 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
*/
r = 0;
amdgpu_bo_list_for_each_userptr_entry(e, p->bo_list) {
- struct amdgpu_bo *bo = ttm_to_amdgpu_bo(e->tv.bo);
-
- r |= !amdgpu_ttm_tt_get_user_pages_done(bo->tbo.ttm, e->range);
+ r |= !amdgpu_ttm_tt_get_user_pages_done(e->bo->tbo.ttm,
+ e->range);
e->range = NULL;
}
if (r) {
@@ -1337,20 +1312,22 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
}
p->fence = dma_fence_get(&leader->base.s_fence->finished);
- list_for_each_entry(e, &p->validated, tv.head) {
+ drm_exec_for_each_locked_object(&p->exec, index, gobj) {
+
+ ttm_bo_move_to_lru_tail_unlocked(&gem_to_amdgpu_bo(gobj)->tbo);
/* Everybody except for the gang leader uses READ */
for (i = 0; i < p->gang_size; ++i) {
if (p->jobs[i] == leader)
continue;
- dma_resv_add_fence(e->tv.bo->base.resv,
+ dma_resv_add_fence(gobj->resv,
&p->jobs[i]->base.s_fence->finished,
DMA_RESV_USAGE_READ);
}
- /* The gang leader is remembered as writer */
- e->tv.num_shared = 0;
+ /* The gang leader as remembered as writer */
+ dma_resv_add_fence(gobj->resv, p->fence, DMA_RESV_USAGE_WRITE);
}
seq = amdgpu_ctx_add_fence(p->ctx, p->entities[p->gang_leader_idx],
@@ -1366,7 +1343,7 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
cs->out.handle = seq;
leader->uf_sequence = seq;
- amdgpu_vm_bo_trace_cs(&fpriv->vm, &p->ticket);
+ amdgpu_vm_bo_trace_cs(&fpriv->vm, &p->exec.ticket);
for (i = 0; i < p->gang_size; ++i) {
amdgpu_job_free_resources(p->jobs[i]);
trace_amdgpu_cs_ioctl(p->jobs[i]);
@@ -1375,7 +1352,6 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
}
amdgpu_vm_move_to_lru_tail(p->adev, &fpriv->vm);
- ttm_eu_fence_buffer_objects(&p->ticket, &p->validated, p->fence);
mutex_unlock(&p->adev->notifier_lock);
mutex_unlock(&p->bo_list->bo_list_mutex);
@@ -1388,6 +1364,8 @@ static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser)
unsigned int i;
amdgpu_sync_free(&parser->sync);
+ drm_exec_fini(&parser->exec);
+
for (i = 0; i < parser->num_post_deps; i++) {
drm_syncobj_put(parser->post_deps[i].syncobj);
kfree(parser->post_deps[i].chain);
@@ -1408,11 +1386,7 @@ static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser)
if (parser->jobs[i])
amdgpu_job_free(parser->jobs[i]);
}
- if (parser->uf_entry.tv.bo) {
- struct amdgpu_bo *uf = ttm_to_amdgpu_bo(parser->uf_entry.tv.bo);
-
- amdgpu_bo_unref(&uf);
- }
+ amdgpu_bo_unref(&parser->uf_bo);
}
int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
@@ -1473,7 +1447,6 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
return 0;
error_backoff:
- ttm_eu_backoff_reservation(&parser.ticket, &parser.validated);
mutex_unlock(&parser.bo_list->bo_list_mutex);
error_fini:
@@ -1808,7 +1781,7 @@ int amdgpu_cs_find_mapping(struct amdgpu_cs_parser *parser,
*map = mapping;
/* Double check that the BO is reserved by this CS */
- if (dma_resv_locking_ctx((*bo)->tbo.base.resv) != &parser->ticket)
+ if (dma_resv_locking_ctx((*bo)->tbo.base.resv) != &parser->exec.ticket)
return -EINVAL;
if (!((*bo)->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS)) {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.h
index fb3e3d56d427..39c33ad100cb 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.h
@@ -24,6 +24,7 @@
#define __AMDGPU_CS_H__
#include <linux/ww_mutex.h>
+#include <drm/drm_exec.h>
#include "amdgpu_job.h"
#include "amdgpu_bo_list.h"
@@ -62,11 +63,9 @@ struct amdgpu_cs_parser {
struct amdgpu_job *gang_leader;
/* buffer objects */
- struct ww_acquire_ctx ticket;
+ struct drm_exec exec;
struct amdgpu_bo_list *bo_list;
struct amdgpu_mn *mn;
- struct amdgpu_bo_list_entry vm_pd;
- struct list_head validated;
struct dma_fence *fence;
uint64_t bytes_moved_threshold;
uint64_t bytes_moved_vis_threshold;
@@ -74,7 +73,7 @@ struct amdgpu_cs_parser {
uint64_t bytes_moved_vis;
/* user fence */
- struct amdgpu_bo_list_entry uf_entry;
+ struct amdgpu_bo *uf_bo;
unsigned num_post_deps;
struct amdgpu_cs_post_dep *post_deps;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index c5c5f2eb76c5..854260171dfd 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -339,28 +339,6 @@ void amdgpu_vm_bo_base_init(struct amdgpu_vm_bo_base *base,
amdgpu_vm_bo_evicted(base);
}
-/**
- * amdgpu_vm_get_pd_bo - add the VM PD to a validation list
- *
- * @vm: vm providing the BOs
- * @validated: head of validation list
- * @entry: entry to add
- *
- * Add the page directory to the list of BOs to
- * validate for command submission.
- */
-void amdgpu_vm_get_pd_bo(struct amdgpu_vm *vm,
- struct list_head *validated,
- struct amdgpu_bo_list_entry *entry)
-{
- entry->priority = 0;
- entry->tv.bo = &vm->root.bo->tbo;
- /* Two for VM updates, one for TTM and one for the CS job */
- entry->tv.num_shared = 4;
- entry->user_pages = NULL;
- list_add(&entry->tv.head, validated);
-}
-
/**
* amdgpu_vm_lock_pd - lock PD in drm_exec
*
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
index 3e29886e5535..45bcf46c6f59 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
@@ -398,9 +398,6 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm);
int amdgpu_vm_make_compute(struct amdgpu_device *adev, struct amdgpu_vm *vm);
void amdgpu_vm_release_compute(struct amdgpu_device *adev, struct amdgpu_vm *vm);
void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm);
-void amdgpu_vm_get_pd_bo(struct amdgpu_vm *vm,
- struct list_head *validated,
- struct amdgpu_bo_list_entry *entry);
int amdgpu_vm_lock_pd(struct amdgpu_vm *vm, struct drm_exec *exec,
unsigned int num_fences);
bool amdgpu_vm_ready(struct amdgpu_vm *vm);
--
2.34.1
^ permalink raw reply related [flat|nested] 10+ messages in thread* Re: [PATCH 6/6] drm/amdgpu: use the new drm_exec object for CS v2
2023-06-28 10:44 ` [PATCH 6/6] drm/amdgpu: use the new drm_exec object for CS v2 Christian König
@ 2023-06-29 8:20 ` Tatsuyuki Ishi
2023-06-29 13:58 ` Christian König
0 siblings, 1 reply; 10+ messages in thread
From: Tatsuyuki Ishi @ 2023-06-29 8:20 UTC (permalink / raw)
To: Christian König, dri-devel, amd-gfx, alexander.deucher
On 6/28/23 19:44, Christian König wrote:
> @@ -958,18 +904,57 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
> e->user_invalidated = userpage_invalidated;
> }
>
> - r = ttm_eu_reserve_buffers(&p->ticket, &p->validated, true,
> - &duplicates);
> - if (unlikely(r != 0)) {
> - if (r != -ERESTARTSYS)
> - DRM_ERROR("ttm_eu_reserve_buffers failed.\n");
> - goto out_free_user_pages;
> + drm_exec_until_all_locked(&p->exec) {
> + r = amdgpu_vm_lock_pd(&fpriv->vm, &p->exec, 1 + p->gang_size);
> + drm_exec_retry_on_contention(&p->exec);
> + if (unlikely(r))
> + goto out_free_user_pages;
> +
> + amdgpu_bo_list_for_each_entry(e, p->bo_list) {
> + /* One fence for TTM and one for each CS job */
> + r = drm_exec_prepare_obj(&p->exec, &e->bo->tbo.base,
> + 1 + p->gang_size);
> + drm_exec_retry_on_contention(&p->exec);
> + if (unlikely(r))
> + goto out_free_user_pages;
> +
> + e->bo_va = amdgpu_vm_bo_find(vm, e->bo);
> + e->range = NULL;
Still leaking.
> + }
> +
> + if (p->uf_bo) {
> + r = drm_exec_prepare_obj(&p->exec, &p->uf_bo->tbo.base,
> + 1 + p->gang_size);
> + drm_exec_retry_on_contention(&p->exec);
> + if (unlikely(r))
> + goto out_free_user_pages;
> + }
> }
^ permalink raw reply [flat|nested] 10+ messages in thread* Re: [PATCH 6/6] drm/amdgpu: use the new drm_exec object for CS v2
2023-06-29 8:20 ` Tatsuyuki Ishi
@ 2023-06-29 13:58 ` Christian König
0 siblings, 0 replies; 10+ messages in thread
From: Christian König @ 2023-06-29 13:58 UTC (permalink / raw)
To: Tatsuyuki Ishi, dri-devel, amd-gfx, alexander.deucher
Am 29.06.23 um 10:20 schrieb Tatsuyuki Ishi:
> On 6/28/23 19:44, Christian König wrote:
>> @@ -958,18 +904,57 @@ static int amdgpu_cs_parser_bos(struct
>> amdgpu_cs_parser *p,
>> e->user_invalidated = userpage_invalidated;
>> }
>> - r = ttm_eu_reserve_buffers(&p->ticket, &p->validated, true,
>> - &duplicates);
>> - if (unlikely(r != 0)) {
>> - if (r != -ERESTARTSYS)
>> - DRM_ERROR("ttm_eu_reserve_buffers failed.\n");
>> - goto out_free_user_pages;
>> + drm_exec_until_all_locked(&p->exec) {
>> + r = amdgpu_vm_lock_pd(&fpriv->vm, &p->exec, 1 + p->gang_size);
>> + drm_exec_retry_on_contention(&p->exec);
>> + if (unlikely(r))
>> + goto out_free_user_pages;
>> +
>> + amdgpu_bo_list_for_each_entry(e, p->bo_list) {
>> + /* One fence for TTM and one for each CS job */
>> + r = drm_exec_prepare_obj(&p->exec, &e->bo->tbo.base,
>> + 1 + p->gang_size);
>> + drm_exec_retry_on_contention(&p->exec);
>> + if (unlikely(r))
>> + goto out_free_user_pages;
>> +
>> + e->bo_va = amdgpu_vm_bo_find(vm, e->bo);
>> + e->range = NULL;
> Still leaking.
Scratching my head, I though I fixed this.
Thanks for pointing this out,
Christian.
>> + }
>> +
>> + if (p->uf_bo) {
>> + r = drm_exec_prepare_obj(&p->exec, &p->uf_bo->tbo.base,
>> + 1 + p->gang_size);
>> + drm_exec_retry_on_contention(&p->exec);
>> + if (unlikely(r))
>> + goto out_free_user_pages;
>> + }
>> }
>
^ permalink raw reply [flat|nested] 10+ messages in thread