In complicated DMA pipelines such as graphics (multimedia, camera, gpu, display) a consumer of a buffer needs to know when the producer has finished producing it. Likewise the producer needs to know when the consumer is finished with the buffer so it can reuse it. A particular buffer may be consumed by multiple consumers which will retain the buffer for different amounts of time. In addition, a consumer may consume multiple buffers atomically. The sync framework adds an API which allows synchronization between the producers and consumers in a generic way while also allowing platforms which have shared hardware synchronization primitives to exploit them.

See Documentation/sync.txt for more information

Signed-off-by: Erik Gilling konkers@android.com --- Documentation/sync.txt | 75 +++++++ drivers/base/Kconfig | 8 + drivers/base/Makefile | 2 + drivers/base/sync.c | 525 ++++++++++++++++++++++++++++++++++++++++++++++++ include/linux/sync.h | 316 +++++++++++++++++++++++++++++ 5 files changed, 926 insertions(+), 0 deletions(-) create mode 100644 Documentation/sync.txt create mode 100644 drivers/base/sync.c create mode 100644 include/linux/sync.h

diff --git a/Documentation/sync.txt b/Documentation/sync.txt new file mode 100644 index 0000000..a2d05e7 --- /dev/null +++ b/Documentation/sync.txt @@ -0,0 +1,75 @@ +Motivation: + +In complicated DMA pipelines such as graphics (multimedia, camera, gpu, display) +a consumer of a buffer needs to know when the producer has finished producing +it. Likewise the producer needs to know when the consumer is finished with the +buffer so it can reuse it. A particular buffer may be consumed by multiple +consumers which will retain the buffer for different amounts of time. In +addition, a consumer may consume multiple buffers atomically. +The sync framework adds an API which allows synchronization between the +producers and consumers in a generic way while also allowing platforms which +have shared hardware synchronization primitives to exploit them. + +Goals: + * provide a generic API for expressing synchronization dependencies + * allow drivers to exploit hardware synchronization between hardware + blocks + * provide a userspace API that allows a compositor to manage + dependencies. + * provide rich telemetry data to allow debugging slowdowns and stalls of + the graphics pipeline. + +Objects: + * sync_timeline + * sync_pt + * sync_fence + +sync_timeline: + +A sync_timeline is an abstract monotonically increasing counter. In general, +each driver/hardware block context will have one of these. They can be backed +by the appropriate hardware or rely on the generic sw_sync implementation. +Timelines are only ever created through their specific implementations +(i.e. sw_sync.) + +sync_pt: + +A sync_pt is an abstract value which marks a point on a sync_timeline. Sync_pts +have a single timeline parent. They have 3 states: active, signaled, and error. +They start in active state and transition, once, to either signaled (when the +timeline counter advances beyond the sync_pt’s value) or error state. + +sync_fence: + +Sync_fences are the primary primitives used by drivers to coordinate +synchronization of their buffers. They are a collection of sync_pts which may +or may not have the same timeline parent. A sync_pt can only exist in one fence +and the fence's list of sync_pts is immutable once created. Fences can be +waited on synchronously or asynchronously. Two fences can also be merged to +create a third fence containing a copy of the two fences’ sync_pts. Fences are +backed by file descriptors to allow userspace to coordinate the display pipeline +dependencies. + +Use: + +A driver implementing sync support should have a work submission function which: + * takes a fence argument specifying when to begin work + * asynchronously queues that work to kick off when the fence is signaled + * returns a fence to indicate when its work will be done. + * signals the returned fence once the work is completed. + +Consider an imaginary display driver that has the following API: +/* + * assumes buf is ready to be displayed. + * blocks until the buffer is on screen. + */ + void display_buffer(struct dma_buf *buf); + +The new API will become: +/* + * will display buf when fence is signaled. + * returns immediately with a fence that will signal when buf + * is no longer displayed. + */ +struct sync_fence* display_buffer(struct dma_buf *buf, + struct sync_fence *fence); diff --git a/drivers/base/Kconfig b/drivers/base/Kconfig index 9aa618a..784d08b 100644 --- a/drivers/base/Kconfig +++ b/drivers/base/Kconfig @@ -192,4 +192,12 @@ config DMA_SHARED_BUFFER APIs extension; the file's descriptor can then be passed on to other driver.

+config SYNC + bool "Synchronization framework" + default n + select ANON_INODES + help + This option enables the framework for synchronization between multiple + drivers. Sync implementations can take advantage of hardware + synchronization built into devices like GPUs. endmenu diff --git a/drivers/base/Makefile b/drivers/base/Makefile index b6d1b9c..a43e369 100644 --- a/drivers/base/Makefile +++ b/drivers/base/Makefile @@ -21,5 +21,7 @@ obj-$(CONFIG_SYS_HYPERVISOR) += hypervisor.o obj-$(CONFIG_REGMAP) += regmap/ obj-$(CONFIG_SOC_BUS) += soc.o

+obj-$(CONFIG_SYNC) += sync.o + ccflags-$(CONFIG_DEBUG_DRIVER) := -DDEBUG

diff --git a/drivers/base/sync.c b/drivers/base/sync.c new file mode 100644 index 0000000..4a8604b --- /dev/null +++ b/drivers/base/sync.c @@ -0,0 +1,525 @@ +/* + * drivers/base/sync.c + * + * Copyright (C) 2012 Google, Inc. + * + * This software is licensed under the terms of the GNU General Public + * License version 2, as published by the Free Software Foundation, and + * may be copied, distributed, and modified under those terms. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + */ + +#include <linux/file.h> +#include <linux/fs.h> +#include <linux/kernel.h> +#include <linux/sched.h> +#include <linux/slab.h> +#include <linux/sync.h> +#include <linux/uaccess.h> + +#include <linux/anon_inodes.h> + +static void sync_fence_signal_pt(struct sync_pt *pt); +static int _sync_pt_has_signaled(struct sync_pt *pt); + +struct sync_timeline *sync_timeline_create(const struct sync_timeline_ops *ops, + int size, const char *name) +{ + struct sync_timeline *obj; + + if (size < sizeof(struct sync_timeline)) + return NULL; + + obj = kzalloc(size, GFP_KERNEL); + if (obj == NULL) + return NULL; + + obj->ops = ops; + strlcpy(obj->name, name, sizeof(obj->name)); + + INIT_LIST_HEAD(&obj->child_list_head); + spin_lock_init(&obj->child_list_lock); + + INIT_LIST_HEAD(&obj->active_list_head); + spin_lock_init(&obj->active_list_lock); + + return obj; +} + +void sync_timeline_destroy(struct sync_timeline *obj) +{ + unsigned long flags; + bool needs_freeing; + + spin_lock_irqsave(&obj->child_list_lock, flags); + obj->destroyed = true; + needs_freeing = list_empty(&obj->child_list_head); + spin_unlock_irqrestore(&obj->child_list_lock, flags); + + if (needs_freeing) + kfree(obj); + else + sync_timeline_signal(obj); +} + +static void sync_timeline_add_pt(struct sync_timeline *obj, struct sync_pt *pt) +{ + unsigned long flags; + + pt->parent = obj; + + spin_lock_irqsave(&obj->child_list_lock, flags); + list_add_tail(&pt->child_list, &obj->child_list_head); + spin_unlock_irqrestore(&obj->child_list_lock, flags); +} + +static void sync_timeline_remove_pt(struct sync_pt *pt) +{ + struct sync_timeline *obj = pt->parent; + unsigned long flags; + bool needs_freeing; + + spin_lock_irqsave(&obj->active_list_lock, flags); + if (!list_empty(&pt->active_list)) + list_del_init(&pt->active_list); + spin_unlock_irqrestore(&obj->active_list_lock, flags); + + spin_lock_irqsave(&obj->child_list_lock, flags); + list_del(&pt->child_list); + needs_freeing = obj->destroyed && list_empty(&obj->child_list_head); + spin_unlock_irqrestore(&obj->child_list_lock, flags); + + if (needs_freeing) + kfree(obj); +} + +void sync_timeline_signal(struct sync_timeline *obj) +{ + unsigned long flags; + LIST_HEAD(signaled_pts); + struct list_head *pos, *n; + + spin_lock_irqsave(&obj->active_list_lock, flags); + + list_for_each_safe(pos, n, &obj->active_list_head) { + struct sync_pt *pt = + container_of(pos, struct sync_pt, active_list); + + if (_sync_pt_has_signaled(pt)) + list_move(pos, &signaled_pts); + } + + spin_unlock_irqrestore(&obj->active_list_lock, flags); + + list_for_each_safe(pos, n, &signaled_pts) { + struct sync_pt *pt = + container_of(pos, struct sync_pt, active_list); + + list_del_init(pos); + sync_fence_signal_pt(pt); + } +} + +struct sync_pt *sync_pt_create(struct sync_timeline *parent, int size) +{ + struct sync_pt *pt; + + if (size < sizeof(struct sync_pt)) + return NULL; + + pt = kzalloc(size, GFP_KERNEL); + if (pt == NULL) + return NULL; + + INIT_LIST_HEAD(&pt->active_list); + sync_timeline_add_pt(parent, pt); + + return pt; +} + +void sync_pt_free(struct sync_pt *pt) +{ + if (pt->parent->ops->free_pt) + pt->parent->ops->free_pt(pt); + + sync_timeline_remove_pt(pt); + + kfree(pt); +} + +/* call with pt->parent->active_list_lock held */ +static int _sync_pt_has_signaled(struct sync_pt *pt) +{ + if (!pt->status) + pt->status = pt->parent->ops->has_signaled(pt); + + if (!pt->status && pt->parent->destroyed) + pt->status = -ENOENT; + + return pt->status; +} + +static struct sync_pt *sync_pt_dup(struct sync_pt *pt) +{ + return pt->parent->ops->dup(pt); +} + +/* Adds a sync pt to the active queue. Called when added to a fence */ +static void sync_pt_activate(struct sync_pt *pt) +{ + struct sync_timeline *obj = pt->parent; + unsigned long flags; + int err; + + spin_lock_irqsave(&obj->active_list_lock, flags); + + err = _sync_pt_has_signaled(pt); + if (err != 0) + goto out; + + list_add_tail(&pt->active_list, &obj->active_list_head); + +out: + spin_unlock_irqrestore(&obj->active_list_lock, flags); +} + +static int sync_fence_release(struct inode *inode, struct file *file); +static long sync_fence_ioctl(struct file *file, unsigned int cmd, + unsigned long arg); + + +static const struct file_operations sync_fence_fops = { + .release = sync_fence_release, + .unlocked_ioctl = sync_fence_ioctl, +}; + +static struct sync_fence *sync_fence_alloc(const char *name) +{ + struct sync_fence *fence; + + fence = kzalloc(sizeof(struct sync_fence), GFP_KERNEL); + if (fence == NULL) + return NULL; + + fence->file = anon_inode_getfile("sync_fence", &sync_fence_fops, + fence, 0); + if (fence->file == NULL) + goto err; + + strlcpy(fence->name, name, sizeof(fence->name)); + + INIT_LIST_HEAD(&fence->pt_list_head); + INIT_LIST_HEAD(&fence->waiter_list_head); + spin_lock_init(&fence->waiter_list_lock); + + init_waitqueue_head(&fence->wq); + return fence; + +err: + kfree(fence); + return NULL; +} + +/* TODO: implement a create which takes more that one sync_pt */ +struct sync_fence *sync_fence_create(const char *name, struct sync_pt *pt) +{ + struct sync_fence *fence; + + if (pt->fence) + return NULL; + + fence = sync_fence_alloc(name); + if (fence == NULL) + return NULL; + + pt->fence = fence; + list_add(&pt->pt_list, &fence->pt_list_head); + sync_pt_activate(pt); + + return fence; +} + +static int sync_fence_copy_pts(struct sync_fence *dst, struct sync_fence *src) +{ + struct list_head *pos; + + list_for_each(pos, &src->pt_list_head) { + struct sync_pt *orig_pt = + container_of(pos, struct sync_pt, pt_list); + struct sync_pt *new_pt = sync_pt_dup(orig_pt); + + if (new_pt == NULL) + return -ENOMEM; + + new_pt->fence = dst; + list_add(&new_pt->pt_list, &dst->pt_list_head); + sync_pt_activate(new_pt); + } + + return 0; +} + +static void sync_fence_free_pts(struct sync_fence *fence) +{ + struct list_head *pos, *n; + + list_for_each_safe(pos, n, &fence->pt_list_head) { + struct sync_pt *pt = container_of(pos, struct sync_pt, pt_list); + sync_pt_free(pt); + } +} + +struct sync_fence *sync_fence_fdget(int fd) +{ + struct file *file = fget(fd); + + if (file == NULL) + return NULL; + + if (file->f_op != &sync_fence_fops) + goto err; + + return file->private_data; + +err: + fput(file); + return NULL; +} + +void sync_fence_put(struct sync_fence *fence) +{ + fput(fence->file); +} + +void sync_fence_install(struct sync_fence *fence, int fd) +{ + fd_install(fd, fence->file); +} + +static int sync_fence_get_status(struct sync_fence *fence) +{ + struct list_head *pos; + int status = 1; + + list_for_each(pos, &fence->pt_list_head) { + struct sync_pt *pt = container_of(pos, struct sync_pt, pt_list); + int pt_status = pt->status; + + if (pt_status < 0) { + status = pt_status; + break; + } else if (status == 1) { + status = pt_status; + } + } + + return status; +} + +struct sync_fence *sync_fence_merge(const char *name, + struct sync_fence *a, struct sync_fence *b) +{ + struct sync_fence *fence; + int err; + + fence = sync_fence_alloc(name); + if (fence == NULL) + return NULL; + + err = sync_fence_copy_pts(fence, a); + if (err < 0) + goto err; + + err = sync_fence_copy_pts(fence, b); + if (err < 0) + goto err; + + fence->status = sync_fence_get_status(fence); + + return fence; +err: + sync_fence_free_pts(fence); + kfree(fence); + return NULL; +} + +static void sync_fence_signal_pt(struct sync_pt *pt) +{ + LIST_HEAD(signaled_waiters); + struct sync_fence *fence = pt->fence; + struct list_head *pos; + struct list_head *n; + unsigned long flags; + int status; + + status = sync_fence_get_status(fence); + + spin_lock_irqsave(&fence->waiter_list_lock, flags); + /* + * this should protect against two threads racing on the signaled + * false -> true transition + */ + if (status && !fence->status) { + list_for_each_safe(pos, n, &fence->waiter_list_head) + list_move(pos, &signaled_waiters); + + fence->status = status; + } else { + status = 0; + } + spin_unlock_irqrestore(&fence->waiter_list_lock, flags); + + if (status) { + list_for_each_safe(pos, n, &signaled_waiters) { + struct sync_fence_waiter *waiter = + container_of(pos, struct sync_fence_waiter, + waiter_list); + + waiter->callback(fence, waiter->callback_data); + list_del(pos); + kfree(waiter); + } + wake_up(&fence->wq); + } +} + +int sync_fence_wait_async(struct sync_fence *fence, + void (*callback)(struct sync_fence *, void *data), + void *callback_data) +{ + struct sync_fence_waiter *waiter; + unsigned long flags; + int err = 0; + + waiter = kzalloc(sizeof(struct sync_fence_waiter), GFP_KERNEL); + if (waiter == NULL) + return -ENOMEM; + + waiter->callback = callback; + waiter->callback_data = callback_data; + + spin_lock_irqsave(&fence->waiter_list_lock, flags); + + if (fence->status) { + kfree(waiter); + err = fence->status; + goto out; + } + + list_add_tail(&waiter->waiter_list, &fence->waiter_list_head); +out: + spin_unlock_irqrestore(&fence->waiter_list_lock, flags); + + return err; +} + +int sync_fence_wait(struct sync_fence *fence, long timeout) +{ + int err; + + if (timeout) { + timeout = msecs_to_jiffies(timeout); + err = wait_event_interruptible_timeout(fence->wq, + fence->status != 0, + timeout); + } else { + err = wait_event_interruptible(fence->wq, fence->status != 0); + } + + if (err < 0) + return err; + + if (fence->status < 0) + return fence->status; + + if (fence->status == 0) + return -ETIME; + + return 0; +} + +static int sync_fence_release(struct inode *inode, struct file *file) +{ + struct sync_fence *fence = file->private_data; + + sync_fence_free_pts(fence); + kfree(fence); + + return 0; +} + +static long sync_fence_ioctl_wait(struct sync_fence *fence, unsigned long arg) +{ + __u32 value; + + if (copy_from_user(&value, (void __user *)arg, sizeof(value))) + return -EFAULT; + + return sync_fence_wait(fence, value); +} + +static long sync_fence_ioctl_merge(struct sync_fence *fence, unsigned long arg) +{ + int fd = get_unused_fd(); + int err; + struct sync_fence *fence2, *fence3; + struct sync_merge_data data; + + if (copy_from_user(&data, (void __user *)arg, sizeof(data))) + return -EFAULT; + + fence2 = sync_fence_fdget(data.fd2); + if (fence2 == NULL) { + err = -ENOENT; + goto err_put_fd; + } + + data.name[sizeof(data.name) - 1] = '\0'; + fence3 = sync_fence_merge(data.name, fence, fence2); + if (fence3 == NULL) { + err = -ENOMEM; + goto err_put_fence2; + } + + data.fence = fd; + if (copy_to_user((void __user *)arg, &data, sizeof(data))) { + err = -EFAULT; + goto err_put_fence3; + } + + sync_fence_install(fence3, fd); + sync_fence_put(fence2); + return 0; + +err_put_fence3: + sync_fence_put(fence3); + +err_put_fence2: + sync_fence_put(fence2); + +err_put_fd: + put_unused_fd(fd); + return err; +} + + +static long sync_fence_ioctl(struct file *file, unsigned int cmd, + unsigned long arg) +{ + struct sync_fence *fence = file->private_data; + switch (cmd) { + case SYNC_IOC_WAIT: + return sync_fence_ioctl_wait(fence, arg); + + case SYNC_IOC_MERGE: + return sync_fence_ioctl_merge(fence, arg); + default: + return -ENOTTY; + } +} + diff --git a/include/linux/sync.h b/include/linux/sync.h new file mode 100644 index 0000000..2e7ee55 --- /dev/null +++ b/include/linux/sync.h @@ -0,0 +1,316 @@ +/* + * include/linux/sync.h + * + * Copyright (C) 2012 Google, Inc. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + */ + +#ifndef _LINUX_SYNC_H +#define _LINUX_SYNC_H + +#include <linux/types.h> +#ifdef __KERNEL__ + +#include <linux/list.h> +#include <linux/spinlock.h> +#include <linux/wait.h> + +struct sync_timeline; +struct sync_pt; +struct sync_fence; + +/** + * struct sync_timeline_ops - sync object implementation ops + * @driver_name: name of the implentation + * @dup: duplicate a sync_pt + * @has_signaled: returns: + * 1 if pt has signaled + * 0 if pt has not signaled + * <0 on error + * must be callable from atomic context + * @compare: returns: + * 1 if b will signal before a + * 0 if a and b will signal at the same time + * -1 if a will signabl before b + * must be callable from atomic context + * @free_pt: called before sync_pt is freed + * @release_obj: called before sync_timeline is freed + */ +struct sync_timeline_ops { + const char *driver_name; + + /* required */ + struct sync_pt *(*dup)(struct sync_pt *pt); + + /* required */ + int (*has_signaled)(struct sync_pt *pt); + + /* required */ + int (*compare)(struct sync_pt *a, struct sync_pt *b); + + /* optional */ + void (*free_pt)(struct sync_pt *sync_pt); + + /* optional */ + void (*release_obj)(struct sync_timeline *sync_timeline); +}; + +/** + * struct sync_timeline - sync object + * @ops: ops that define the implementaiton of the sync_timeline + * @name: name of the sync_timeline. Useful for debugging + * @destoryed: set when sync_timeline is destroyed + * @child_list_head: list of children sync_pts for this sync_timeline + * @child_list_lock: lock protecting @child_list_head, destroyed, and + * sync_pt.status + * @active_list_head: list of active (unsignaled/errored) sync_pts + */ +struct sync_timeline { + const struct sync_timeline_ops *ops; + char name[32]; + + /* protected by child_list_lock */ + bool destroyed; + + struct list_head child_list_head; + spinlock_t child_list_lock; + + struct list_head active_list_head; + spinlock_t active_list_lock; +}; + +/** + * struct sync_pt - sync point + * @parent: sync_timeline to which this sync_pt belongs + * @child_list: membership in sync_timeline.child_list_head + * @active_list: membership in sync_timeline.active_list_head + * @fence: sync_fence to which the sync_pt belongs + * @pt_list: membership in sync_fence.pt_list_head + * @status: 1: signaled, 0:active, <0: error + */ +struct sync_pt { + struct sync_timeline *parent; + struct list_head child_list; + + struct list_head active_list; + + struct sync_fence *fence; + struct list_head pt_list; + + /* protected by parent->active_list_lock */ + int status; +}; + +/** + * struct sync_fence - sync fence + * @file: file representing this fence + * @name: name of sync_fence. Useful for debugging + * @pt_list_head: list of sync_pts in ths fence. immutable once fence + * is created + * @waiter_list_head: list of asynchronous waiters on this fence + * @waiter_list_lock: lock protecting @waiter_list_head and @status + * @status: 1: signaled, 0:active, <0: error + * + * @wq: wait queue for fence signaling + */ +struct sync_fence { + struct file *file; + char name[32]; + + /* this list is immutable once the fence is created */ + struct list_head pt_list_head; + + struct list_head waiter_list_head; + spinlock_t waiter_list_lock; /* also protects status */ + int status; + + wait_queue_head_t wq; +}; + +/** + * struct sync_fence_waiter - metadata for asynchronous waiter on a fence + * @waiter_list: membership in sync_fence.waiter_list_head + * @callback: function pointer to call when fence signals + * @callback_data: pointer to pass to @callback + */ +struct sync_fence_waiter { + struct list_head waiter_list; + + void (*callback)(struct sync_fence *fence, void *data); + void *callback_data; +}; + +/* + * API for sync_timeline implementers + */ + +/** + * sync_timeline_create() - creates a sync object + * @ops: specifies the implemention ops for the object + * @size: size to allocate for this obj + * @name: sync_timeline name + * + * Creates a new sync_timeline which will use the implemetation specified by + * @ops. @size bytes will be allocated allowing for implemntation specific + * data to be kept after the generic sync_timeline stuct. + */ +struct sync_timeline *sync_timeline_create(const struct sync_timeline_ops *ops, + int size, const char *name); + +/** + * sync_timeline_destory() - destorys a sync object + * @obj: sync_timeline to destroy + * + * A sync implemntation should call this when the @obj is going away + * (i.e. module unload.) @obj won't actually be freed until all its childern + * sync_pts are freed. + */ +void sync_timeline_destroy(struct sync_timeline *obj); + +/** + * sync_timeline_signal() - signal a status change on a sync_timeline + * @obj: sync_timeline to signal + * + * A sync implemntation should call this any time one of it's sync_pts + * has signaled or has an error condition. + */ +void sync_timeline_signal(struct sync_timeline *obj); + +/** + * sync_pt_create() - creates a sync pt + * @parent: sync_pt's parent sync_timeline + * @size: size to allocate for this pt + * + * Creates a new sync_pt as a chiled of @parent. @size bytes will be + * allocated allowing for implemntation specific data to be kept after + * the generic sync_timeline struct. + */ +struct sync_pt *sync_pt_create(struct sync_timeline *parent, int size); + +/** + * sync_pt_free() - frees a sync pt + * @pt: sync_pt to free + * + * This should only be called on sync_pts which have been created but + * not added to a fence. + */ +void sync_pt_free(struct sync_pt *pt); + +/** + * sync_fence_create() - creates a sync fence + * @name: name of fence to create + * @pt: sync_pt to add to the fence + * + * Creates a fence containg @pt. Once this is called, the fence takes + * ownership of @pt. + */ +struct sync_fence *sync_fence_create(const char *name, struct sync_pt *pt); + +/* + * API for sync_fence consumers + */ + +/** + * sync_fence_merge() - merge two fences + * @name: name of new fence + * @a: fence a + * @b: fence b + * + * Creates a new fence which contains copies of all the sync_pts in both + * @a and @b. @a and @b remain valid, independent fences. + */ +struct sync_fence *sync_fence_merge(const char *name, + struct sync_fence *a, struct sync_fence *b); + +/** + * sync_fence_fdget() - get a fence from an fd + * @fd: fd referencing a fence + * + * Ensures @fd references a valid fence, increments the refcount of the backing + * file, and returns the fence. + */ +struct sync_fence *sync_fence_fdget(int fd); + +/** + * sync_fence_put() - puts a refernnce of a sync fence + * @fence: fence to put + * + * Puts a reference on @fence. If this is the last reference, the fence and + * all it's sync_pts will be freed + */ +void sync_fence_put(struct sync_fence *fence); + +/** + * sync_fence_install() - installs a fence into a file descriptor + * @fence: fence to instal + * @fd: file descriptor in which to install the fence + * + * Installs @fence into @fd. @fd's should be acquired through get_unused_fd(). + */ +void sync_fence_install(struct sync_fence *fence, int fd); + +/** + * sync_fence_wait_async() - registers and async wait on the fence + * @fence: fence to wait on + * @callback: callback + * @callback_data data to pass to the callback + * + * Returns 1 if @fence has already signaled. + * + * Registers a callback to be called when @fence signals or has an error + */ +int sync_fence_wait_async(struct sync_fence *fence, + void (*callback)(struct sync_fence *, void *data), + void *callback_data); + +/** + * sync_fence_wait() - wait on fence + * @fence: fence to wait on + * @tiemout: timeout in ms + * + * Wait for @fence to be signaled or have an error. Waits indefintly + * if @timeout = 0 + */ +int sync_fence_wait(struct sync_fence *fence, long timeout); + +/* useful for sync driver's debug print handlers */ +const char *sync_status_str(int status); + +#endif /* __KERNEL__ */ + +/** + * struct sync_merge_data - data passed to merge ioctl + * @fd2: file descriptor of second fence + * @name: name of new fence + * @fence: returns the fd of the new fence to userspace + */ +struct sync_merge_data { + __s32 fd2; /* fd of second fence */ + char name[32]; /* name of new fence */ + __s32 fence; /* fd on newly created fence */ +}; + +#define SYNC_IOC_MAGIC '>' + +/** + * DOC: SYNC_IOC_WAIT - wait for a fence to signal + * + * pass timeout in milliseconds. + */ +#define SYNC_IOC_WAIT _IOW(SYNC_IOC_MAGIC, 0, __u32) + +/** + * DOC: SYNC_IOC_MERGE - merge two fences + * + * Takes a struct sync_merge_data. Creates a new fence containing copies of + * the sync_pts in both the calling fd and sync_merge_data.fd2. Returns the + * new fence's fd in sync_merge_data.fence + */ +#define SYNC_IOC_MERGE _IOWR(SYNC_IOC_MAGIC, 1, struct sync_merge_data) + +#endif /* _LINUX_SYNC_H */

When profiling performance, knowing when various points in the graphics pipeline finished is very useful.

Signed-off-by: Erik Gilling konkers@android.com --- drivers/base/sync.c | 5 +++++ include/linux/sync.h | 5 +++++ 2 files changed, 10 insertions(+), 0 deletions(-)

diff --git a/drivers/base/sync.c b/drivers/base/sync.c index 4a8604b..ea0b664 100644 --- a/drivers/base/sync.c +++ b/drivers/base/sync.c @@ -155,12 +155,17 @@ void sync_pt_free(struct sync_pt *pt) /* call with pt->parent->active_list_lock held */ static int _sync_pt_has_signaled(struct sync_pt *pt) { + int old_status = pt->status; + if (!pt->status) pt->status = pt->parent->ops->has_signaled(pt);

if (!pt->status && pt->parent->destroyed) pt->status = -ENOENT;

+ if (pt->status != old_status) + pt->timestamp = ktime_get(); + return pt->status; }

diff --git a/include/linux/sync.h b/include/linux/sync.h index 2e7ee55..b9ee821 100644 --- a/include/linux/sync.h +++ b/include/linux/sync.h @@ -16,6 +16,7 @@ #include <linux/types.h> #ifdef __KERNEL__

+#include <linux/ktime.h> #include <linux/list.h> #include <linux/spinlock.h> #include <linux/wait.h> @@ -92,6 +93,8 @@ struct sync_timeline { * @fence: sync_fence to which the sync_pt belongs * @pt_list: membership in sync_fence.pt_list_head * @status: 1: signaled, 0:active, <0: error + * @timestamp: time which sync_pt status transitioned from active to + * singaled or error. */ struct sync_pt { struct sync_timeline *parent; @@ -104,6 +107,8 @@ struct sync_pt {

/* protected by parent->active_list_lock */ int status; + + ktime_t timestamp; };

/**

Allows sw_sync specific information to be included in debugfs.

Signed-off-by: Erik Gilling konkers@android.com --- drivers/base/sw_sync.c | 19 +++++++++++++++++++ 1 files changed, 19 insertions(+), 0 deletions(-)

diff --git a/drivers/base/sw_sync.c b/drivers/base/sw_sync.c index e94ecc7..41beb9e 100644 --- a/drivers/base/sw_sync.c +++ b/drivers/base/sw_sync.c @@ -69,11 +69,30 @@ static int sw_sync_pt_compare(struct sync_pt *a, struct sync_pt *b) return sw_sync_cmp(pt_a->value, pt_b->value); }

+static void sw_sync_print_obj(struct seq_file *s, + struct sync_timeline *sync_timeline) +{ + struct sw_sync_timeline *obj = (struct sw_sync_timeline *)sync_timeline; + + seq_printf(s, "%d", obj->value); +} + +static void sw_sync_print_pt(struct seq_file *s, struct sync_pt *sync_pt) +{ + struct sw_sync_pt *pt = (struct sw_sync_pt *)sync_pt; + struct sw_sync_timeline *obj = + (struct sw_sync_timeline *)sync_pt->parent; + + seq_printf(s, "%d / %d", pt->value, obj->value); +} + struct sync_timeline_ops sw_sync_timeline_ops = { .driver_name = "sw_sync", .dup = sw_sync_pt_dup, .has_signaled = sw_sync_pt_has_signaled, .compare = sw_sync_pt_compare, + .print_obj = sw_sync_print_obj, + .print_pt = sw_sync_print_pt, };

Allows sw_sync private data to be available in get fence data ioctl.

Signed-off-by: Erik Gilling konkers@android.com --- drivers/base/sw_sync.c | 14 ++++++++++++++ 1 files changed, 14 insertions(+), 0 deletions(-)

diff --git a/drivers/base/sw_sync.c b/drivers/base/sw_sync.c index 41beb9e..21ddf4f 100644 --- a/drivers/base/sw_sync.c +++ b/drivers/base/sw_sync.c @@ -86,6 +86,19 @@ static void sw_sync_print_pt(struct seq_file *s, struct sync_pt *sync_pt) seq_printf(s, "%d / %d", pt->value, obj->value); }

+static int sw_sync_fill_driver_data(struct sync_pt *sync_pt, + void *data, int size) +{ + struct sw_sync_pt *pt = (struct sw_sync_pt *)sync_pt; + + if (size < sizeof(pt->value)) + return -ENOMEM; + + memcpy(data, &pt->value, sizeof(pt->value)); + + return sizeof(pt->value); +} + struct sync_timeline_ops sw_sync_timeline_ops = { .driver_name = "sw_sync", .dup = sw_sync_pt_dup, @@ -93,6 +106,7 @@ struct sync_timeline_ops sw_sync_timeline_ops = { .compare = sw_sync_pt_compare, .print_obj = sw_sync_print_obj, .print_pt = sw_sync_print_pt, + .fill_driver_data = sw_sync_fill_driver_data, };

In order to allow drivers to cleanly handled teardown we need to allow them to cancel pending async waits. To do this cleanly, we move allocation of sync_fence_waiter to the driver calling sync_async_wait().

Signed-off-by: Erik Gilling konkers@android.com --- drivers/base/sync.c | 46 +++++++++++++++++++++++++++++++--------------- include/linux/sync.h | 36 +++++++++++++++++++++++++++++------- 2 files changed, 60 insertions(+), 22 deletions(-)

diff --git a/drivers/base/sync.c b/drivers/base/sync.c index cc40e4c..5e89195 100644 --- a/drivers/base/sync.c +++ b/drivers/base/sync.c @@ -421,33 +421,22 @@ static void sync_fence_signal_pt(struct sync_pt *pt) container_of(pos, struct sync_fence_waiter, waiter_list);

- waiter->callback(fence, waiter->callback_data); list_del(pos); - kfree(waiter); + waiter->callback(fence, waiter); } wake_up(&fence->wq); } }

int sync_fence_wait_async(struct sync_fence *fence, - void (*callback)(struct sync_fence *, void *data), - void *callback_data) + struct sync_fence_waiter *waiter) { - struct sync_fence_waiter *waiter; unsigned long flags; int err = 0;

- waiter = kzalloc(sizeof(struct sync_fence_waiter), GFP_KERNEL); - if (waiter == NULL) - return -ENOMEM; - - waiter->callback = callback; - waiter->callback_data = callback_data; - spin_lock_irqsave(&fence->waiter_list_lock, flags);

if (fence->status) { - kfree(waiter); err = fence->status; goto out; } @@ -459,6 +448,34 @@ out: return err; }

+int sync_fence_cancel_async(struct sync_fence *fence, + struct sync_fence_waiter *waiter) +{ + struct list_head *pos; + struct list_head *n; + unsigned long flags; + int ret = -ENOENT; + + spin_lock_irqsave(&fence->waiter_list_lock, flags); + /* + * Make sure waiter is still in waiter_list because it is possible for + * the waiter to be removed from the list while the callback is still + * pending. + */ + list_for_each_safe(pos, n, &fence->waiter_list_head) { + struct sync_fence_waiter *list_waiter = + container_of(pos, struct sync_fence_waiter, + waiter_list); + if (list_waiter == waiter) { + list_del(pos); + ret = 0; + break; + } + } + spin_unlock_irqrestore(&fence->waiter_list_lock, flags); + return ret; +} + int sync_fence_wait(struct sync_fence *fence, long timeout) { int err; @@ -739,8 +756,7 @@ static void sync_print_fence(struct seq_file *s, struct sync_fence *fence) container_of(pos, struct sync_fence_waiter, waiter_list);

- seq_printf(s, "waiter %pF %p\n", waiter->callback, - waiter->callback_data); + seq_printf(s, "waiter %pF\n", waiter->callback); } spin_unlock_irqrestore(&fence->waiter_list_lock, flags); } diff --git a/include/linux/sync.h b/include/linux/sync.h index 6db0528..1beaa53 100644 --- a/include/linux/sync.h +++ b/include/linux/sync.h @@ -161,6 +161,10 @@ struct sync_fence { struct list_head sync_fence_list; };

+struct sync_fence_waiter; +typedef void (*sync_callback_t)(struct sync_fence *fence, + struct sync_fence_waiter *waiter); + /** * struct sync_fence_waiter - metadata for asynchronous waiter on a fence * @waiter_list: membership in sync_fence.waiter_list_head @@ -170,10 +174,15 @@ struct sync_fence { struct sync_fence_waiter { struct list_head waiter_list;

- void (*callback)(struct sync_fence *fence, void *data); - void *callback_data; + sync_callback_t callback; };

+static inline void sync_fence_waiter_init(struct sync_fence_waiter *waiter, + sync_callback_t callback) +{ + waiter->callback = callback; +} + /* * API for sync_timeline implementers */ @@ -286,16 +295,29 @@ void sync_fence_install(struct sync_fence *fence, int fd); /** * sync_fence_wait_async() - registers and async wait on the fence * @fence: fence to wait on - * @callback: callback - * @callback_data data to pass to the callback + * @waiter: waiter callback struck * * Returns 1 if @fence has already signaled. * - * Registers a callback to be called when @fence signals or has an error + * Registers a callback to be called when @fence signals or has an error. + * @waiter should be initialized with sync_fence_waiter_init(). */ int sync_fence_wait_async(struct sync_fence *fence, - void (*callback)(struct sync_fence *, void *data), - void *callback_data); + struct sync_fence_waiter *waiter); + +/** + * sync_fence_cancel_async() - cancels an async wait + * @fence: fence to wait on + * @waiter: waiter callback struck + * + * returns 0 if waiter was removed from fence's async waiter list. + * returns -ENOENT if waiter was not found on fence's async waiter list. + * + * Cancels a previously registered async wait. Will fail gracefully if + * @waiter was never registered or if @fence has already signaled @waiter. + */ +int sync_fence_cancel_async(struct sync_fence *fence, + struct sync_fence_waiter *waiter);

/** * sync_fence_wait() - wait on fence

Needed to let modules link against sw_sync.

Signed-off-by: Erik Gilling konkers@android.com --- drivers/base/sw_sync.c | 5 ++++- 1 files changed, 4 insertions(+), 1 deletions(-)

diff --git a/drivers/base/sw_sync.c b/drivers/base/sw_sync.c index 21ddf4f..935769c 100644 --- a/drivers/base/sw_sync.c +++ b/drivers/base/sw_sync.c @@ -15,6 +15,7 @@ */

#include <linux/kernel.h> +#include <linux/export.h> #include <linux/file.h> #include <linux/fs.h> #include <linux/miscdevice.h> @@ -42,6 +43,7 @@ struct sync_pt *sw_sync_pt_create(struct sw_sync_timeline *obj, u32 value)

return (struct sync_pt *)pt; } +EXPORT_SYMBOL(sw_sync_pt_create);

static struct sync_pt *sw_sync_pt_dup(struct sync_pt *sync_pt) { @@ -119,6 +121,7 @@ struct sw_sync_timeline *sw_sync_timeline_create(const char *name)

return obj; } +EXPORT_SYMBOL(sw_sync_timeline_create);

void sw_sync_timeline_inc(struct sw_sync_timeline *obj, u32 inc) { @@ -126,7 +129,7 @@ void sw_sync_timeline_inc(struct sw_sync_timeline *obj, u32 inc)

sync_timeline_signal(&obj->obj); } - +EXPORT_SYMBOL(sw_sync_timeline_inc);

#ifdef CONFIG_SW_SYNC_USER /* *WARNING*