Good patch set. will apply them after review and test. Thanks, Inki Dae
2012/12/26 Prathyush K <prathyus...@samsung.com> > This patchset has more fixes for fixing more issues related to wait for > vblank. > > Patch 1: add mixer apply function > The mixer window enabled flag is being done in both > exynos_drm_hdmi as > well as exynos_mixer. This patch moves the flag to one place and adds the > mixer apply function similar to fimd apply function. > > Patch 2: fimd: call win_disable only when window is enabled > Adds a condition to call win_disable only if the window is enabled. > > Patch 3: enable vblank in fimd wait for vblank > If vblank is disabled, then we turn enable vblank temporarily > before > wait for vblank and disable it afterwards. This ensures that the wait will > not timeout. > > Patch 4: fimd: clear channel before enabling iommu > This patch is required if the fimd window channel is already active > before fimd probe. This could happen if we turn on fimd during uboot. This > patch ensures that we do not get a page fault upon iommu initialization. If > any channel is active, we disable the channel and wait for vblank. > > Patch 5: mixer: do not finish a pageflip if layer update in progress > Exynos5 supports only 2 layer updates for mixer. This patch > ensures each layer is updated only once per vsync. Also, in case a layer > update is pending, the irq handler exits early. This ensures there is > no corruption on screen as well as there is no page fault during buffer > freeing. > > Patch 6: add complete_scanout interface > Before freeing a framebuffer, we call complete_scanout encoder > function which just calls wait for vblank of all the encoders. > This is not a very optimized method since a framebuffer might not > be in use by a crtc and we end up waiting for vblank unnecessarily. > Instead, this patch modifies the wait_for_vblank interface to > complete_scanout interface. In this function, each crtc must > check if the fb is currently being used. If it is being used, > it must wait for vsync (or even disable a window if necessary) > and ensure that the buffer is no longer used by crtc before returning. > So if a crtc is not actually reading from the buffer, the complete > scanout function will just return and not wait for vsync. > > Patch 7: hdmi: add complete_scanout function > Patch 8: fimd: add complete_scanout function > These two patches add the respective complete_scanout functions > for fimd and mixer. These functions check the actual base and shadow > registers of fimd/mixer to see if a particular dma-address is in use > and will wait_for_vblank only if it is in use. Also for mixer, if the > base address matches the fb's dma address, then the window is disabled. > This is because a layer update is already set when the base register > was updated. So unless we disable the window, we cannot free the buffer. > > Akshu Agrawal (1): > drm/exynos: fimd: clear channel before enabling iommu > > Prathyush K (6): > drm/exynos: add mixer apply function > drm/exynos: fimd: call win_disable only when window is enabled > drm/exynos: enable vblank in fimd wait for vblank > drm/exynos: add complete_scanout interface > drm/exynos: hdmi: add complete_scanout function > drm/exynos: fimd: add complete_scanout function > > Sean Paul (1): > drm/exynos: mixer: do not finish a pageflip if layer update in > progress > > drivers/gpu/drm/exynos/exynos_drm_drv.h | 7 ++- > drivers/gpu/drm/exynos/exynos_drm_encoder.c | 23 ++++---- > drivers/gpu/drm/exynos/exynos_drm_encoder.h | 4 +- > drivers/gpu/drm/exynos/exynos_drm_fb.c | 13 ++++- > drivers/gpu/drm/exynos/exynos_drm_fimd.c | 81 > ++++++++++++++++++++++++-- > drivers/gpu/drm/exynos/exynos_drm_hdmi.c | 26 +++------ > drivers/gpu/drm/exynos/exynos_drm_hdmi.h | 4 +- > drivers/gpu/drm/exynos/exynos_mixer.c | 90 > ++++++++++++++++++++++++++--- > drivers/gpu/drm/exynos/regs-mixer.h | 1 + > include/video/samsung_fimd.h | 1 + > 10 files changed, 202 insertions(+), 48 deletions(-) > > -- > 1.8.0 > > _______________________________________________ > dri-devel mailing list > dri-devel@lists.freedesktop.org > http://lists.freedesktop.org/mailman/listinfo/dri-devel >
_______________________________________________ dri-devel mailing list dri-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/dri-devel
