drm: Allow CAP_DUMB_BUFFERS on !MODESET
Subsystems (2)
PATCH 1/1 drm: Allow CAP_DUMB_BUFFERS on !MODESET
It's perfectly possible to get dumb buffers out of drivers that don't support modeset. This is the case of vgem, which can be used to export dmabuf to run various tests. Therefore, move the dumb buffer capability check before the mode set check.
Inspired by commit f3f4c4d68a28 ("drm: Allow CAP_PRIME on !MODESET").
--- drivers/gpu/drm/drm_ioctl.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-)
drivers/gpu/drm/drm_ioctl.c +3 -4
@@ -244,6 +244,9 @@ static int drm_getcap(struct drm_device *dev, void *data, struct drm_file *file_case DRM_CAP_SYNCOBJ : req->value =drm_core_check_feature (dev,DRIVER_SYNCOBJ );return 0 ; +case DRM_CAP_DUMB_BUFFER : + req->value |= dev->driver->dumb_create ?1 :0 ; +return 0 ; }/* Other caps only work with KMS drivers */
@@ -251,10 +254,6 @@ static int drm_getcap(struct drm_device *dev, void *data, struct drm_file *file_return -ENOTSUPP ;switch (req->capability) { -case DRM_CAP_DUMB_BUFFER : -if (dev->driver->dumb_create) - req->value =1 ; -break ;case DRM_CAP_VBLANK_HIGH_CRTC : req->value =1 ;break ;
Signature · 6 lines
2.16.3 _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
Maarten Lankhorst
maintainerOp 04-05-18 om 16:25 schreef Ezequiel Garcia:
Quoted 19 lines
It's perfectly possible to get dumb buffers out of drivers that don't support modeset. This is the case of vgem, which can be used to export dmabuf to run various tests. Therefore, move the dumb buffer capability check before the mode set check.
Inspired by commit
f3f4c4d68a28("drm: AllowCAP_PRIMEon !MODESET").Fixes d5264ed3823a ("drm: Return -ENOTSUPP when called for KMS cap with a non-KMS driver")Signed-off-by Ezequiel Garcia <[email protected]>--- drivers/gpu/drm/drm_ioctl.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-)drivers/gpu/drm/drm_ioctl.c +0
-0
drivers/gpu/drm/drm_ioctl.c static int drm_getcap(struct drm_device *dev, void *data, struct drm_file *file_ · line 248
Quoted 5 lines
drivers/gpu/drm/drm_ioctl.c +2
-0@@ -244,6 +244,9 @@ static int drm_getcap(struct drm_device *dev, void *data, struct drm_file *file_case DRM_CAP_SYNCOBJ : req->value =drm_core_check_feature (dev,DRIVER_SYNCOBJ );return 0 ; +case DRM_CAP_DUMB_BUFFER : + req->value |= dev->driver->dumb_create ?1 :0 ;
Why the |= ?
Quoted 5 lines
drivers/gpu/drm/drm_ioctl.c +1
-0+return 0 ; }/* Other caps only work with KMS drivers */
drivers/gpu/drm/drm_ioctl.c static int drm_getcap(struct drm_device *dev, void *data, struct drm_file *file_ · line 259
Quoted 10 lines
drivers/gpu/drm/drm_ioctl.c +0
-4@@ -251,10 +254,6 @@ static int drm_getcap(struct drm_device *dev, void *data, struct drm_file *file_return -ENOTSUPP ;switch (req->capability) { -case DRM_CAP_DUMB_BUFFER : -if (dev->driver->dumb_create) - req->value =1 ; -break ;case DRM_CAP_VBLANK_HIGH_CRTC : req->value =1 ;break ;
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
Quoted 15 lines · Maarten Lankhorst
Op 04-05-18 om 16:25 schreef Ezequiel Garcia:
Quoted 14 lines
It's perfectly possible to get dumb buffers out of drivers that don't support modeset. This is the case of vgem, which can be used to export dmabuf to run various tests. Therefore, move the dumb buffer capability check before the mode set check.
Inspired by commit
f3f4c4d68a28("drm: AllowCAP_PRIMEon !MODESET").Fixes d5264ed3823a ("drm: Return -ENOTSUPP when called for KMS cap with a non-KMS driver")Signed-off-by Ezequiel Garcia <[email protected]>--- drivers/gpu/drm/drm_ioctl.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-)
Quoted 12 lines
drivers/gpu/drm/drm_ioctl.c static int drm_getcap(struct drm_device *dev, void *data, struct drm_file *file_ · line 248
Quoted 10 lines
drivers/gpu/drm/drm_ioctl.c +2
-0@@ -244,6 +244,9 @@ static int drm_getcap(struct drm_device *dev, void *data, struct drm_file *file_case DRM_CAP_SYNCOBJ : req->value =drm_core_check_feature (dev,DRIVER_SYNCOBJ );return 0 ; +case DRM_CAP_DUMB_BUFFER : + req->value |= dev->driver->dumb_create ?1 :0 ;Why the |= ?
I'd say that's a silly copy-paste glitch. Will send a v2.
Quoted 7 lines
Quoted 7 lines
drivers/gpu/drm/drm_ioctl.c +1
-0+return 0 ; }/* Other caps only work with KMS drivers */ @@ -251,10 +254,6 @@ static int drm_getcap(struct drm_device *dev, void *data, struct drm_file *file_return -ENOTSUPP ;
Quoted 10 lines
drivers/gpu/drm/drm_ioctl.c static int drm_getcap(struct drm_device *dev, void *data, struct drm_file *file_ · line 259
Quoted 8 lines
drivers/gpu/drm/drm_ioctl.c +0
-4switch (req->capability) { -case DRM_CAP_DUMB_BUFFER : -if (dev->driver->dumb_create) - req->value =1 ; -break ;case DRM_CAP_VBLANK_HIGH_CRTC : req->value =1 ;break ;
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
PATCH v2 1/1 drm: Allow CAP_DUMB_BUFFERS on !MODESET
In reply to Ezequiel GarciaIt's perfectly possible to get dumb buffers out of drivers that don't support modeset. This is the case of vgem, which can be used to export dmabuf to run various tests.
Inspired by commit f3f4c4d68a28 ("drm: Allow CAP_PRIME on !MODESET").
---
Changes from v1:
* Drop the bitwise-OR assignment op
drivers/gpu/drm/drm_ioctl.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-)
drivers/gpu/drm/drm_ioctl.c +3 -4
@@ -244,6 +244,9 @@ static int drm_getcap(struct drm_device *dev, void *data, struct drm_file *file_case DRM_CAP_SYNCOBJ : req->value =drm_core_check_feature (dev,DRIVER_SYNCOBJ );return 0 ; +case DRM_CAP_DUMB_BUFFER : + req->value = dev->driver->dumb_create ?1 :0 ; +return 0 ; }/* Other caps only work with KMS drivers */
@@ -251,10 +254,6 @@ static int drm_getcap(struct drm_device *dev, void *data, struct drm_file *file_return -ENOTSUPP ;switch (req->capability) { -case DRM_CAP_DUMB_BUFFER : -if (dev->driver->dumb_create) - req->value =1 ; -break ;case DRM_CAP_VBLANK_HIGH_CRTC : req->value =1 ;break ;
Signature · 6 lines
2.16.3 _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
Quoted 5 lines · Ezequiel Garcia
It's perfectly possible to get dumb buffers out of drivers that don't support modeset. This is the case of vgem, which can be used to export dmabuf to run various tests.
Inspired by commit
f3f4c4d68a28("drm: AllowCAP_PRIMEon !MODESET").
Prime makes sense, because render-only drivers _really_ want to be able to share buffers.
But dumb buffers are really meant for dumb userspace running on kms drivers only, there's no need ever to tell userspace that dumb buffers are supported on non-kms drivers. It's kinda abuse of ioctls, but oh well, uapi is fixed forever. And you can still call the ioctls if you know they're there (which is always the case for the render-side of drm, there's no generic alloc ioctl for those).
-Daniel
Quoted 41 lines
Fixes d5264ed3823a ("drm: Return -ENOTSUPP when called for KMS cap with a non-KMS driver")Signed-off-by Ezequiel Garcia <[email protected]>---
Changes from v1:
* Drop the bitwise-OR assignment opdrivers/gpu/drm/drm_ioctl.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-)drivers/gpu/drm/drm_ioctl.c +3
-4@@ -244,6 +244,9 @@ static int drm_getcap(struct drm_device *dev, void *data, struct drm_file *file_case DRM_CAP_SYNCOBJ : req->value =drm_core_check_feature (dev,DRIVER_SYNCOBJ );return 0 ; +case DRM_CAP_DUMB_BUFFER : + req->value = dev->driver->dumb_create ?1 :0 ; +return 0 ; }/* Other caps only work with KMS drivers */ @@ -251,10 +254,6 @@ static int drm_getcap(struct drm_device *dev, void *data, struct drm_file *file_return -ENOTSUPP ;switch (req->capability) { -case DRM_CAP_DUMB_BUFFER : -if (dev->driver->dumb_create) - req->value =1 ; -break ;case DRM_CAP_VBLANK_HIGH_CRTC : req->value =1 ;break ;Signature · 6 lines
2.16.3 _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
Signature · 7 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
Hey Daniel,
Quoted 10 lines · Daniel Vetter
Quoted 5 lines · Ezequiel Garcia
It's perfectly possible to get dumb buffers out of drivers that don't support modeset. This is the case of vgem, which can be used to export dmabuf to run various tests.
Inspired by commit
f3f4c4d68a28("drm: AllowCAP_PRIMEon !MODESET").Prime makes sense, because render-only drivers _really_ want to be able to share buffers.
Quoted 7 lines
But dumb buffers are really meant for dumb userspace running on kms drivers only, there's no need ever to tell userspace that dumb buffers are supported on non-kms drivers. It's kinda abuse of ioctls, but oh well, uapi is fixed forever. And you can still call the ioctls if you know they're there (which is always the case for the render-side of drm, there's no generic alloc ioctl for those).
Right, I see.
Well, I was mostly interested in dry testing, and vgem seemed a good candidate for a dma-buf exporter. For instance, this would be useful to test dmabuf video4linux pipelines.
Now, is this is a nack, then the other path is to support proper prime operations on virtio-gpu (which I'm also working on).
Thanks for reviewing!
Eze
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
Quoted 21 lines · Ezequiel Garcia
Hey Daniel,
Quoted 10 lines · Daniel Vetter
Quoted 5 lines · Ezequiel Garcia
It's perfectly possible to get dumb buffers out of drivers that don't support modeset. This is the case of vgem, which can be used to export dmabuf to run various tests.
Inspired by commit
f3f4c4d68a28("drm: AllowCAP_PRIMEon !MODESET").Prime makes sense, because render-only drivers _really_ want to be able to share buffers.
Quoted 7 lines
But dumb buffers are really meant for dumb userspace running on kms drivers only, there's no need ever to tell userspace that dumb buffers are supported on non-kms drivers. It's kinda abuse of ioctls, but oh well, uapi is fixed forever. And you can still call the ioctls if you know they're there (which is always the case for the render-side of drm, there's no generic alloc ioctl for those).
Quoted 5 lines
Right, I see.
Well, I was mostly interested in dry testing, and vgem seemed a good candidate for a dma-buf exporter. For instance, this would be useful to test dmabuf video4linux pipelines.
You can still use the ioctl ... we have tons of tests using vgem for buffer import/export in our igt gpu tests. Testing prime is exactly what vgem is for!
What you don't need is confuse generic userspace by enabling the getparam for dumb buffers. But such vgem-specific tests aren't generic userspace.
https://cgit.freedesktop.org/drm/igt-gpu-tools/
https://cgit.freedesktop.org/drm/igt-gpu-tools/tree/tests/prime_vgem.c
Cheers, Daniel
Quoted 6 lines
Now, is this is a nack, then the other path is to support proper prime operations on virtio-gpu (which I'm also working on).
Thanks for reviewing!
Eze
Signature · 7 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel