2019-04-10 08:03:03

by Gerd Hoffmann

[permalink] [raw]
Subject: [PATCH] drm: add DRM_PRIME_CAP_LOCAL

Some drivers (for example qxl) support neither import nor export of
dma-bufs. But you can still use dma-bufs to pass buffer references
from one process to another; drm_gem_prime_import() will figure the
dma-buf came from the same driver and just takes a reference in that
case instead of doing a full export/import.

Right now there is no way for userspace to figure it can do this. Add
the new prime capability DRM_PRIME_CAP_LOCAL to indicate that. Set the
bit for every driver which has the DRIVER_PRIME feature bit set.

Cc: Dave Airlie <[email protected]>
Signed-off-by: Gerd Hoffmann <[email protected]>
---
include/uapi/drm/drm.h | 1 +
drivers/gpu/drm/drm_ioctl.c | 2 ++
2 files changed, 3 insertions(+)

diff --git a/include/uapi/drm/drm.h b/include/uapi/drm/drm.h
index 236b01a1fabf..b6157d48dc1a 100644
--- a/include/uapi/drm/drm.h
+++ b/include/uapi/drm/drm.h
@@ -632,6 +632,7 @@ struct drm_gem_open {
#define DRM_CAP_PRIME 0x5
#define DRM_PRIME_CAP_IMPORT 0x1
#define DRM_PRIME_CAP_EXPORT 0x2
+#define DRM_PRIME_CAP_LOCAL 0x4
#define DRM_CAP_TIMESTAMP_MONOTONIC 0x6
#define DRM_CAP_ASYNC_PAGE_FLIP 0x7
/*
diff --git a/drivers/gpu/drm/drm_ioctl.c b/drivers/gpu/drm/drm_ioctl.c
index d337f161909c..00599758783b 100644
--- a/drivers/gpu/drm/drm_ioctl.c
+++ b/drivers/gpu/drm/drm_ioctl.c
@@ -241,6 +241,8 @@ static int drm_getcap(struct drm_device *dev, void *data, struct drm_file *file_
case DRM_CAP_PRIME:
req->value |= dev->driver->prime_fd_to_handle ? DRM_PRIME_CAP_IMPORT : 0;
req->value |= dev->driver->prime_handle_to_fd ? DRM_PRIME_CAP_EXPORT : 0;
+ req->value |= (dev->driver->driver_features & DRIVER_PRIME)
+ ? DRM_PRIME_CAP_LOCAL : 0;
return 0;
case DRM_CAP_SYNCOBJ:
req->value = drm_core_check_feature(dev, DRIVER_SYNCOBJ);
--
2.18.1


2019-04-16 07:20:35

by Daniel Vetter

[permalink] [raw]
Subject: Re: [PATCH] drm: add DRM_PRIME_CAP_LOCAL

On Wed, Apr 10, 2019 at 09:47:12AM +0200, Gerd Hoffmann wrote:
> Some drivers (for example qxl) support neither import nor export of
> dma-bufs. But you can still use dma-bufs to pass buffer references
> from one process to another; drm_gem_prime_import() will figure the
> dma-buf came from the same driver and just takes a reference in that
> case instead of doing a full export/import.
>
> Right now there is no way for userspace to figure it can do this. Add
> the new prime capability DRM_PRIME_CAP_LOCAL to indicate that. Set the
> bit for every driver which has the DRIVER_PRIME feature bit set.

It can just try? Any driver that supports dma-buf (even if only for
self-import) should set DRIVER_PRIME. It's perfectly fine to reject any
other sharing.
-Daniel

>
> Cc: Dave Airlie <[email protected]>
> Signed-off-by: Gerd Hoffmann <[email protected]>
> ---
> include/uapi/drm/drm.h | 1 +
> drivers/gpu/drm/drm_ioctl.c | 2 ++
> 2 files changed, 3 insertions(+)
>
> diff --git a/include/uapi/drm/drm.h b/include/uapi/drm/drm.h
> index 236b01a1fabf..b6157d48dc1a 100644
> --- a/include/uapi/drm/drm.h
> +++ b/include/uapi/drm/drm.h
> @@ -632,6 +632,7 @@ struct drm_gem_open {
> #define DRM_CAP_PRIME 0x5
> #define DRM_PRIME_CAP_IMPORT 0x1
> #define DRM_PRIME_CAP_EXPORT 0x2
> +#define DRM_PRIME_CAP_LOCAL 0x4
> #define DRM_CAP_TIMESTAMP_MONOTONIC 0x6
> #define DRM_CAP_ASYNC_PAGE_FLIP 0x7
> /*
> diff --git a/drivers/gpu/drm/drm_ioctl.c b/drivers/gpu/drm/drm_ioctl.c
> index d337f161909c..00599758783b 100644
> --- a/drivers/gpu/drm/drm_ioctl.c
> +++ b/drivers/gpu/drm/drm_ioctl.c
> @@ -241,6 +241,8 @@ static int drm_getcap(struct drm_device *dev, void *data, struct drm_file *file_
> case DRM_CAP_PRIME:
> req->value |= dev->driver->prime_fd_to_handle ? DRM_PRIME_CAP_IMPORT : 0;
> req->value |= dev->driver->prime_handle_to_fd ? DRM_PRIME_CAP_EXPORT : 0;
> + req->value |= (dev->driver->driver_features & DRIVER_PRIME)
> + ? DRM_PRIME_CAP_LOCAL : 0;
> return 0;
> case DRM_CAP_SYNCOBJ:
> req->value = drm_core_check_feature(dev, DRIVER_SYNCOBJ);
> --
> 2.18.1
>

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