| Message ID | 20231023074613.41327-1-aesteve@redhat.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | Fix cursor planes with virtualized drivers | expand |
On Monday, October 23rd, 2023 at 09:46, Albert Esteve <aesteve@redhat.com> wrote: > Link to the IGT test covering this patch (already merged): > https://lists.freedesktop.org/archives/igt-dev/2023-July/058427.html Hmm. IGT should not be merged before the kernel, because as long as the kernel is not merged there might be some uAPI changes. > Mutter patch: > https://lists.freedesktop.org/archives/igt-dev/2023-July/058427.html Seems like this link is same as IGT? Copy-pasta fail maybe?
On Mon, Oct 23, 2023 at 9:55 AM Simon Ser <contact@emersion.fr> wrote: > On Monday, October 23rd, 2023 at 09:46, Albert Esteve <aesteve@redhat.com> > wrote: > > > Link to the IGT test covering this patch (already merged): > > https://lists.freedesktop.org/archives/igt-dev/2023-July/058427.html > > Hmm. IGT should not be merged before the kernel, because as long as the > kernel is not merged there might be some uAPI changes. > Right, but uAPI header was not updated on the IGT side. As per suggestion of the maintainers, I added a static variable that matches the definition on this patch: https://lists.freedesktop.org/archives/igt-dev/2023-August/058803.html +/** + * Clients which do set cursor hotspot and treat the cursor plane + * like a mouse cursor should set this property. + */ +#define LOCAL_DRM_CLIENT_CAP_CURSOR_PLANE_HOTSPOT 6 Once this patch gets upstreamed, the localized definition will be removed, replaced by the real one. > > Mutter patch: > > https://lists.freedesktop.org/archives/igt-dev/2023-July/058427.html > > Seems like this link is same as IGT? Copy-pasta fail maybe? > > Ah yes, my bad, this is the correct link: https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3337
On Monday, October 23rd, 2023 at 10:14, Albert Esteve <aesteve@redhat.com> wrote: > On Mon, Oct 23, 2023 at 9:55 AM Simon Ser <contact@emersion.fr> wrote: > > > On Monday, October 23rd, 2023 at 09:46, Albert Esteve <aesteve@redhat.com> wrote: > > > > > Link to the IGT test covering this patch (already merged): > > > https://lists.freedesktop.org/archives/igt-dev/2023-July/058427.html > > > > Hmm. IGT should not be merged before the kernel, because as long as the > > kernel is not merged there might be some uAPI changes. > > Right, but uAPI header was not updated on the IGT side. As per suggestion of the > maintainers, I added a static variable that matches the definition on this patch: > https://lists.freedesktop.org/archives/igt-dev/2023-August/058803.html > > +/** > + * Clients which do set cursor hotspot and treat the cursor plane > + * like a mouse cursor should set this property. > + */ > +#define LOCAL_DRM_CLIENT_CAP_CURSOR_PLANE_HOTSPOT 6 > > Once this patch gets upstreamed, the localized definition will be removed, > replaced by the real one. What if this patch gets delayed and another patch using the same number is merged into the kernel first? What if someone finds a design flaw in the uAPI and it needs to be completely changed? The IGT test would then be completely broken. As a rule of thumb: never merge user-space patches before kernel. As soon as the kernel part is merged, it's fine to locally copy definitions if desirable. > > > Mutter patch: > > > https://lists.freedesktop.org/archives/igt-dev/2023-July/058427.html > > > > Seems like this link is same as IGT? Copy-pasta fail maybe? > > Ah yes, my bad, this is the correct link: > https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3337 Thanks!
Albert Esteve <aesteve@redhat.com> writes: Hello, [...] > >> > Mutter patch: >> > https://lists.freedesktop.org/archives/igt-dev/2023-July/058427.html >> >> Seems like this link is same as IGT? Copy-pasta fail maybe? >> >> > Ah yes, my bad, this is the correct link: > https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3337 The mutter chages are already in good shape and the MR has even be approved by a mutter developer. Any objections to merge the series ?
On Wednesday, November 22nd, 2023 at 13:49, Javier Martinez Canillas <javierm@redhat.com> wrote:
> Any objections to merge the series ?
No objections from me :)
Simon Ser <contact@emersion.fr> writes: Hello Simon, > On Wednesday, November 22nd, 2023 at 13:49, Javier Martinez Canillas <javierm@redhat.com> wrote: > >> Any objections to merge the series ? > > No objections from me :) > Perfect, I'll merge this series then to unblock the mutter MR. Thanks again!
Albert Esteve <aesteve@redhat.com> writes: > v6: Shift DRIVER_CURSOR_HOTSPOT flag bit to BIT(9), since BIT(8) > was already taken by DRIVER_GEM_GPUVA. > > v5: Add a change with documentation from Michael, based on his discussion > with Pekka and bump the kernel version DRM_CLIENT_CAP_CURSOR_PLANE_HOTSPOT > might be introduced with to 6.6. > > v4: Make drm_plane_create_hotspot_properties static, rename > DRM_CLIENT_CAP_VIRTUALIZED_CURSOR_PLANE to DRM_CLIENT_CAP_CURSOR_PLANE_HOTSPOT > and some minor stylistic fixes for things found by Javier and Pekka > in v3. > > v3: Renames, fixes and cleanups suggested by Daniel, Simon and Pekka > after v2. There's no major changes in functionality. Please let me know > if I missed anything, it's been a while since v2. > > Virtualized drivers have had a lot of issues with cursor support on top > of atomic modesetting. This set both fixes the long standing problems > with atomic kms and virtualized drivers and adds code to let userspace > use atomic kms on virtualized drivers while preserving functioning > seamless cursors between the host and guest. > > The first change in the set is one that should be backported as far as > possible, likely 5.4 stable, because earlier stable kernels do not have > virtualbox driver. The change makes virtualized drivers stop exposing > a cursor plane for atomic clients, this fixes mouse cursor on all well > formed compositors which will automatically fallback to software cursor. > > The rest of the changes until the last one ports the legacy hotspot code > to atomic plane properties. > > Finally the last change introduces userspace API to let userspace > clients advertise the fact that they are aware of additional restrictions > placed upon the cursor plane by virtualized drivers and lets them use > atomic kms with virtualized drivers (the clients are expected to set > hotspots correctly when advertising support for virtual cursor plane). > > Link to the IGT test covering this patch (already merged): > https://lists.freedesktop.org/archives/igt-dev/2023-July/058427.html > > Mutter patch: > https://lists.freedesktop.org/archives/igt-dev/2023-July/058427.html > > Michael Banack (1): > drm: Introduce documentation for hotspot properties > > Zack Rusin (8): > drm: Disable the cursor plane on atomic contexts with virtualized > drivers > drm/atomic: Add support for mouse hotspots > drm/vmwgfx: Use the hotspot properties from cursor planes > drm/qxl: Use the hotspot properties from cursor planes > drm/vboxvideo: Use the hotspot properties from cursor planes > drm/virtio: Use the hotspot properties from cursor planes > drm: Remove legacy cursor hotspot code > drm: Introduce DRM_CLIENT_CAP_CURSOR_PLANE_HOTSPOT > Pushed to drm-misc (drm-misc-next). Thanks!
v6: Shift DRIVER_CURSOR_HOTSPOT flag bit to BIT(9), since BIT(8) was already taken by DRIVER_GEM_GPUVA. v5: Add a change with documentation from Michael, based on his discussion with Pekka and bump the kernel version DRM_CLIENT_CAP_CURSOR_PLANE_HOTSPOT might be introduced with to 6.6. v4: Make drm_plane_create_hotspot_properties static, rename DRM_CLIENT_CAP_VIRTUALIZED_CURSOR_PLANE to DRM_CLIENT_CAP_CURSOR_PLANE_HOTSPOT and some minor stylistic fixes for things found by Javier and Pekka in v3. v3: Renames, fixes and cleanups suggested by Daniel, Simon and Pekka after v2. There's no major changes in functionality. Please let me know if I missed anything, it's been a while since v2. Virtualized drivers have had a lot of issues with cursor support on top of atomic modesetting. This set both fixes the long standing problems with atomic kms and virtualized drivers and adds code to let userspace use atomic kms on virtualized drivers while preserving functioning seamless cursors between the host and guest. The first change in the set is one that should be backported as far as possible, likely 5.4 stable, because earlier stable kernels do not have virtualbox driver. The change makes virtualized drivers stop exposing a cursor plane for atomic clients, this fixes mouse cursor on all well formed compositors which will automatically fallback to software cursor. The rest of the changes until the last one ports the legacy hotspot code to atomic plane properties. Finally the last change introduces userspace API to let userspace clients advertise the fact that they are aware of additional restrictions placed upon the cursor plane by virtualized drivers and lets them use atomic kms with virtualized drivers (the clients are expected to set hotspots correctly when advertising support for virtual cursor plane). Link to the IGT test covering this patch (already merged): https://lists.freedesktop.org/archives/igt-dev/2023-July/058427.html Mutter patch: https://lists.freedesktop.org/archives/igt-dev/2023-July/058427.html Michael Banack (1): drm: Introduce documentation for hotspot properties Zack Rusin (8): drm: Disable the cursor plane on atomic contexts with virtualized drivers drm/atomic: Add support for mouse hotspots drm/vmwgfx: Use the hotspot properties from cursor planes drm/qxl: Use the hotspot properties from cursor planes drm/vboxvideo: Use the hotspot properties from cursor planes drm/virtio: Use the hotspot properties from cursor planes drm: Remove legacy cursor hotspot code drm: Introduce DRM_CLIENT_CAP_CURSOR_PLANE_HOTSPOT Documentation/gpu/drm-kms.rst | 6 ++ drivers/gpu/drm/drm_atomic_state_helper.c | 14 +++ drivers/gpu/drm/drm_atomic_uapi.c | 20 ++++ drivers/gpu/drm/drm_ioctl.c | 9 ++ drivers/gpu/drm/drm_plane.c | 120 +++++++++++++++++++++- drivers/gpu/drm/qxl/qxl_display.c | 14 ++- drivers/gpu/drm/qxl/qxl_drv.c | 2 +- drivers/gpu/drm/vboxvideo/vbox_drv.c | 2 +- drivers/gpu/drm/vboxvideo/vbox_mode.c | 4 +- drivers/gpu/drm/virtio/virtgpu_drv.c | 2 +- drivers/gpu/drm/virtio/virtgpu_plane.c | 8 +- drivers/gpu/drm/vmwgfx/vmwgfx_drv.c | 2 +- drivers/gpu/drm/vmwgfx/vmwgfx_kms.c | 9 +- include/drm/drm_drv.h | 9 ++ include/drm/drm_file.h | 12 +++ include/drm/drm_framebuffer.h | 12 --- include/drm/drm_plane.h | 14 +++ include/uapi/drm/drm.h | 25 +++++ 18 files changed, 245 insertions(+), 39 deletions(-)