On Wed, May 04, 2022 at 03:34:03PM -0500, Jason Ekstrand wrote:
> Modern userspace APIs like Vulkan are built on an explicit
> synchronization model.  This doesn't always play nicely with the
> implicit synchronization used in the kernel and assumed by X11 and
> Wayland.  The client -> compositor half of the synchronization isn't too
> bad, at least on intel, because we can control whether or not i915
> synchronizes on the buffer and whether or not it's considered written.
>
> The harder part is the compositor -> client synchronization when we get
> the buffer back from the compositor.  We're required to be able to
> provide the client with a VkSemaphore and VkFence representing the point
> in time where the window system (compositor and/or display) finished
> using the buffer.  With current APIs, it's very hard to do this in such
> a way that we don't get confused by the Vulkan driver's access of the
> buffer.  In particular, once we tell the kernel that we're rendering to
> the buffer again, any CPU waits on the buffer or GPU dependencies will
> wait on some of the client rendering and not just the compositor.
>
> This new IOCTL solves this problem by allowing us to get a snapshot of
> the implicit synchronization state of a given dma-buf in the form of a
> sync file.  It's effectively the same as a poll() or I915_GEM_WAIT only,
> instead of CPU waiting directly, it encapsulates the wait operation, at
> the current moment in time, in a sync_file so we can check/wait on it
> later.  As long as the Vulkan driver does the sync_file export from the
> dma-buf before we re-introduce it for rendering, it will only contain
> fences from the compositor or display.  This allows to accurately turn
> it into a VkFence or VkSemaphore without any over-synchronization.
>
> By making this an ioctl on the dma-buf itself, it allows this new
> functionality to be used in an entirely driver-agnostic way without
> having access to a DRM fd. This makes it ideal for use in driver-generic
> code in Mesa or in a client such as a compositor where the DRM fd may be
> hard to reach.
>
> v2 (Jason Ekstrand):
>  - Use a wrapper dma_fence_array of all fences including the new one
>    when importing an exclusive fence.
>
> v3 (Jason Ekstrand):
>  - Lock around setting shared fences as well as exclusive
>  - Mark SIGNAL_SYNC_FILE as a read-write ioctl.
>  - Initialize ret to 0 in dma_buf_wait_sync_file
>
> v4 (Jason Ekstrand):
>  - Use the new dma_resv_get_singleton helper
>
> v5 (Jason Ekstrand):
>  - Rename the IOCTLs to import/export rather than wait/signal
>  - Drop the WRITE flag and always get/set the exclusive fence
>
> v6 (Jason Ekstrand):
>  - Drop the sync_file import as it was all-around sketchy and not nearly
>    as useful as import.
>  - Re-introduce READ/WRITE flag support for export
>  - Rework the commit message
>
> v7 (Jason Ekstrand):
>  - Require at least one sync flag
>  - Fix a refcounting bug: dma_resv_get_excl() doesn't take a reference
>  - Use _rcu helpers since we're accessing the dma_resv read-only
>
> v8 (Jason Ekstrand):
>  - Return -ENOMEM if the sync_file_create fails
>  - Predicate support on IS_ENABLED(CONFIG_SYNC_FILE)
>
> v9 (Jason Ekstrand):
>  - Add documentation for the new ioctl
>
> v10 (Jason Ekstrand):
>  - Go back to dma_buf_sync_file as the ioctl struct name
>
> v11 (Daniel Vetter):
>  - Go back to dma_buf_export_sync_file as the ioctl struct name
>  - Better kerneldoc describing what the read/write flags do
>
> v12 (Christian König):
>  - Document why we chose to make it an ioctl on dma-buf
>
> v12 (Jason Ekstrand):
>  - Rebase on Christian König's fence rework
>
> Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
> Acked-by: Simon Ser <contact@emersion.fr>
> Acked-by: Christian König <christian.koenig@amd.com>
> Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>

Not sure which version it was that I reviewed, but with dma_resv_usage
this all looks neat and tidy. One nit below.

> Cc: Sumit Semwal <sumit.semwal@linaro.org>
> Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
> ---
>  drivers/dma-buf/dma-buf.c    | 64 ++++++++++++++++++++++++++++++++++++
>  include/uapi/linux/dma-buf.h | 35 ++++++++++++++++++++
>  2 files changed, 99 insertions(+)
>
> diff --git a/drivers/dma-buf/dma-buf.c b/drivers/dma-buf/dma-buf.c
> index 79795857be3e..529e0611e53b 100644
> --- a/drivers/dma-buf/dma-buf.c
> +++ b/drivers/dma-buf/dma-buf.c
> @@ -20,6 +20,7 @@
>  #include <linux/debugfs.h>
>  #include <linux/module.h>
>  #include <linux/seq_file.h>
> +#include <linux/sync_file.h>
>  #include <linux/poll.h>
>  #include <linux/dma-resv.h>
>  #include <linux/mm.h>
> @@ -192,6 +193,9 @@ static loff_t dma_buf_llseek(struct file *file, loff_t offset, int whence)
>   * Note that this only signals the completion of the respective fences, i.e. the
>   * DMA transfers are complete. Cache flushing and any other necessary
>   * preparations before CPU access can begin still need to happen.
> + *
> + * As an alternative to poll(), the set of fences on DMA buffer can be
> + * exported as a &sync_file using &dma_buf_sync_file_export.
>   */
> 
>  static void dma_buf_poll_cb(struct dma_fence *fence, struct dma_fence_cb *cb)
> @@ -326,6 +330,61 @@ static long dma_buf_set_name(struct dma_buf *dmabuf, const char __user *buf)
>       return 0;
>  }
> 
> +#if IS_ENABLED(CONFIG_SYNC_FILE)
> +static long dma_buf_export_sync_file(struct dma_buf *dmabuf,
> +                                  void __user *user_data)
> +{
> +     struct dma_buf_export_sync_file arg;
> +     enum dma_resv_usage usage;
> +     struct dma_fence *fence = NULL;
> +     struct sync_file *sync_file;
> +     int fd, ret;
> +
> +     if (copy_from_user(&arg, user_data, sizeof(arg)))
> +             return -EFAULT;
> +
> +     if (arg.flags & ~DMA_BUF_SYNC_RW)
> +             return -EINVAL;
> +
> +     if ((arg.flags & DMA_BUF_SYNC_RW) == 0)
> +             return -EINVAL;

We allow userspace to set both SYNC_READ and SYNC_WRITE here, I think

        if ((arg.flags & DMA_BUF_SYNC_RW) == DMA_BUF_SYNC_RW)
                return -EINVAL;

is missing?

Also maybe a case to add to your igt.

> +
> +     fd = get_unused_fd_flags(O_CLOEXEC);
> +     if (fd < 0)
> +             return fd;
> +
> +     usage = (arg.flags & DMA_BUF_SYNC_WRITE) ? DMA_RESV_USAGE_WRITE :
> +                                                DMA_RESV_USAGE_READ;
> +     ret = dma_resv_get_singleton(dmabuf->resv, usage, &fence);
> +     if (ret)
> +             goto err_put_fd;
> +
> +     if (!fence)
> +             fence = dma_fence_get_stub();
> +
> +     sync_file = sync_file_create(fence);
> +
> +     dma_fence_put(fence);
> +
> +     if (!sync_file) {
> +             ret = -ENOMEM;
> +             goto err_put_fd;
> +     }
> +
> +     fd_install(fd, sync_file->file);
> +
> +     arg.fd = fd;
> +     if (copy_to_user(user_data, &arg, sizeof(arg)))
> +             return -EFAULT;
> +
> +     return 0;
> +
> +err_put_fd:
> +     put_unused_fd(fd);
> +     return ret;
> +}
> +#endif
> +
>  static long dma_buf_ioctl(struct file *file,
>                         unsigned int cmd, unsigned long arg)
>  {
> @@ -369,6 +428,11 @@ static long dma_buf_ioctl(struct file *file,
>       case DMA_BUF_SET_NAME_B:
>               return dma_buf_set_name(dmabuf, (const char __user *)arg);
> 
> +#if IS_ENABLED(CONFIG_SYNC_FILE)
> +     case DMA_BUF_IOCTL_EXPORT_SYNC_FILE:
> +             return dma_buf_export_sync_file(dmabuf, (void __user *)arg);
> +#endif
> +
>       default:
>               return -ENOTTY;
>       }
> diff --git a/include/uapi/linux/dma-buf.h b/include/uapi/linux/dma-buf.h
> index 8e4a2ca0bcbf..46f1e3e98b02 100644
> --- a/include/uapi/linux/dma-buf.h
> +++ b/include/uapi/linux/dma-buf.h
> @@ -85,6 +85,40 @@ struct dma_buf_sync {
> 
>  #define DMA_BUF_NAME_LEN     32
> 
> +/**
> + * struct dma_buf_export_sync_file - Get a sync_file from a dma-buf
> + *
> + * Userspace can perform a DMA_BUF_IOCTL_EXPORT_SYNC_FILE to retrieve the
> + * current set of fences on a dma-buf file descriptor as a sync_file.  CPU
> + * waits via poll() or other driver-specific mechanisms typically wait on
> + * whatever fences are on the dma-buf at the time the wait begins.  This
> + * is similar except that it takes a snapshot of the current fences on the
> + * dma-buf for waiting later instead of waiting immediately.  This is
> + * useful for modern graphics APIs such as Vulkan which assume an explicit
> + * synchronization model but still need to inter-operate with dma-buf.
> + */
> +struct dma_buf_export_sync_file {
> +     /**
> +      * @flags: Read/write flags
> +      *
> +      * Must be DMA_BUF_SYNC_READ, DMA_BUF_SYNC_WRITE, or both.
> +      *
> +      * If DMA_BUF_SYNC_READ is set and DMA_BUF_SYNC_WRITE is not set,
> +      * the returned sync file waits on any writers of the dma-buf to
> +      * complete.  Waiting on the returned sync file is equivalent to
> +      * poll() with POLLIN.
> +      *
> +      * If DMA_BUF_SYNC_WRITE is set, the returned sync file waits on
> +      * any users of the dma-buf (read or write) to complete.  Waiting
> +      * on the returned sync file is equivalent to poll() with POLLOUT.
> +      * If both DMA_BUF_SYNC_WRITE and DMA_BUF_SYNC_READ are set, this
> +      * is equivalent to just DMA_BUF_SYNC_WRITE.
> +      */
> +     __u32 flags;
> +     /** @fd: Returned sync file descriptor */
> +     __s32 fd;
> +};
> +
>  #define DMA_BUF_BASE         'b'
>  #define DMA_BUF_IOCTL_SYNC   _IOW(DMA_BUF_BASE, 0, struct dma_buf_sync)
> 
> @@ -94,5 +128,6 @@ struct dma_buf_sync {
>  #define DMA_BUF_SET_NAME     _IOW(DMA_BUF_BASE, 1, const char *)
>  #define DMA_BUF_SET_NAME_A   _IOW(DMA_BUF_BASE, 1, u32)
>  #define DMA_BUF_SET_NAME_B   _IOW(DMA_BUF_BASE, 1, u64)
> +#define DMA_BUF_IOCTL_EXPORT_SYNC_FILE       _IOWR(DMA_BUF_BASE, 2, struct dma_buf_export_sync_file)

With the one nit fixed for this version:

Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>

> 
>  #endif
> --
> 2.36.0
>

--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch