[Intel-xe] [PATCH 25/30] drm: execution context for GEM buffers v3
Christian König
christian.koenig at amd.com
Mon May 1 14:57:41 UTC 2023
Just FYI I found time to work a bit on this last week and will probably
send out the next version this week.
Christian.
Am 01.05.23 um 09:51 schrieb Matthew Brost:
> From: Christian König <ckoenig.leichtzumerken at gmail.com>
>
> This adds the infrastructure for an execution context for GEM buffers
> which is similar to the existinc 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 measureable.
> v3: drop duplicate tracking, radeon is really the only one needing that.
>
> Signed-off-by: Christian König <christian.koenig at amd.com>
> ---
> Documentation/gpu/drm-mm.rst | 12 ++
> drivers/gpu/drm/Kconfig | 6 +
> drivers/gpu/drm/Makefile | 2 +
> drivers/gpu/drm/drm_exec.c | 248 +++++++++++++++++++++++++++++++++++
> include/drm/drm_exec.h | 115 ++++++++++++++++
> 5 files changed, 383 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 fe40ee686f6e..c9f120cfe730 100644
> --- a/Documentation/gpu/drm-mm.rst
> +++ b/Documentation/gpu/drm-mm.rst
> @@ -524,6 +524,18 @@ 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 e928284b4357..39c9d079d52a 100644
> --- a/drivers/gpu/drm/Kconfig
> +++ b/drivers/gpu/drm/Kconfig
> @@ -201,6 +201,12 @@ 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 ad6267273503..ab728632d8a2 100644
> --- a/drivers/gpu/drm/Makefile
> +++ b/drivers/gpu/drm/Makefile
> @@ -80,6 +80,8 @@ 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..f645d22a0863
> --- /dev/null
> +++ b/drivers/gpu/drm/drm_exec.c
> @@ -0,0 +1,248 @@
> +/* 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, true);
> + * drm_exec_while_not_all_locked(&exec) {
> + * ret = drm_exec_prepare_obj(&exec, boA, 1);
> + * drm_exec_continue_on_contention(&exec);
> + * if (ret)
> + * goto error;
> + *
> + * ret = drm_exec_lock(&exec, boB, 1);
> + * drm_exec_continue_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);
> + }
> +
> + if (exec->prelocked) {
> + drm_gem_object_put(exec->prelocked);
> + exec->prelocked = NULL;
> + }
> +}
> +
> +/**
> + * drm_exec_init - initialize a drm_exec object
> + * @exec: the drm_exec object to initialize
> + * @interruptible: if locks should be acquired interruptible
> + *
> + * Initialize the object and make sure that we can track locked and duplicate
> + * objects.
> + */
> +void drm_exec_init(struct drm_exec *exec, bool interruptible)
> +{
> + exec->interruptible = interruptible;
> + 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 finilize
> + *
> + * 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 xa and reserve fences */
> +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;
> +
> + if (exec->interruptible) {
> + 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)) {
> + dma_resv_unlock(obj->resv);
> + goto error_dropref;
> + }
> +
> + swap(exec->prelocked, obj);
> +
> +error_dropref:
> + /* Always cleanup the contention so that error handling can kick in */
> + drm_gem_object_put(obj);
> + exec->contended = NULL;
> + return ret;
> +}
> +
> +/**
> + * 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. All
> + * succesfully locked objects are put into the locked container. Duplicates
> + * detected as well and automatically moved into the duplicates container.
> + *
> + * Returns: -EDEADLK if a contention is detected, -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_contended(exec);
> + if (unlikely(ret))
> + return ret;
> +
> + if (exec->prelocked == obj) {
> + drm_gem_object_put(exec->prelocked);
> + exec->prelocked = NULL;
> +
> + return dma_resv_reserve_fences(obj->resv, num_fences);
> + }
> +
> + if (exec->interruptible)
> + 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))
> + return ret;
> +
> + ret = drm_exec_obj_locked(exec, obj);
> + if (ret)
> + goto error_unlock;
> +
> + /* Keep locked when reserving fences fails */
> + return dma_resv_reserve_fences(obj->resv, num_fences);
> +
> +error_unlock:
> + dma_resv_unlock(obj->resv);
> + return ret;
> +}
> +EXPORT_SYMBOL(drm_exec_prepare_obj);
> +
> +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..65e518c01db3
> --- /dev/null
> +++ b/include/drm/drm_exec.h
> @@ -0,0 +1,115 @@
> +/* SPDX-License-Identifier: GPL-2.0 OR MIT */
> +
> +#ifndef __DRM_EXEC_H__
> +#define __DRM_EXEC_H__
> +
> +#include <linux/ww_mutex.h>
> +
> +struct drm_gem_object;
> +
> +/**
> + * struct drm_exec - Execution context
> + */
> +struct drm_exec {
> + /**
> + * @interruptible: If locks should be taken interruptible
> + */
> + bool interruptible;
> +
> + /**
> + * @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 backet of for
> + */
> + struct drm_gem_object *contended;
> +
> + /**
> + * @prelocked: already locked GEM object because of 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_while_not_all_locked - loop until all GEM objects are prepared
> + * @exec: drm_exec object
> + *
> + * Core functionality of the drm_exec object. Loops until all GEM objects are
> + * prepared and no more contention exists.
> + *
> + * At the beginning of the loop it is guaranteed that no GEM object is locked.
> + */
> +#define drm_exec_while_not_all_locked(exec) \
> + while (drm_exec_cleanup(exec))
> +
> +/**
> + * drm_exec_continue_on_contention - continue the loop when we need to cleanup
> + * @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_continue_on_contention(exec) \
> + if (unlikely(drm_exec_is_contended(exec))) \
> + continue
> +
> +/**
> + * drm_exec_break_on_contention - break a subordinal loop on contention
> + * @exec: drm_exec object
> + *
> + * Control flow helper to break a subordinal loop when a contention was detected
> + * and we need to clean up and re-start the loop to prepare all GEM objects.
> + */
> +#define drm_exec_break_on_contention(exec) \
> + if (unlikely(drm_exec_is_contended(exec))) \
> + break
> +
> +/**
> + * 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, bool interruptible);
> +void drm_exec_fini(struct drm_exec *exec);
> +bool drm_exec_cleanup(struct drm_exec *exec);
> +int drm_exec_prepare_obj(struct drm_exec *exec, struct drm_gem_object *obj,
> + unsigned int num_fences);
> +
> +#endif
More information about the Intel-xe
mailing list