Re: [PATCH 1/2] drm/mxsfb: Fix runtime PM for unpowering lcdif block
On Fri, 2018-07-13 at 12:57 +0200, Stefan Agner wrote: > On 05.06.2018 19:11, Leonard Crestez wrote: > > Adding imx6sl/sx lcdif nodes in a power domain currently does work, it > > results in black/corrupted screens or hangs. While the driver does > > enable runtime pm it does not deal correctly with the block being > > unpowered. > > > > Fix by adding pm_runtime_get/put_sync to mxsfb_pipe_enable/disable. > > > > The mxsfb_plane_atomic_update function can get called before > > mxsfb_pipe_enable while the block is not powered. When this happens the > > write to LCDIF_NEXT_BUF is lost causing random corrupted pixels on > > unblank. > > > > Fix this by splitting the writing of gem->paddr to nextbuf into a > > mxsfb_update_hw_next_buf functiona and also calling it from > > mxsfb_crtc_mode_set_nofb. > > > > Also add fields to mxsfb_drv to keep track of enabled/suspended states. > > +static void mxsfb_update_hw_next_buf(struct mxsfb_drm_private *mxsfb) > > +{ > > + struct drm_framebuffer *fb = mxsfb->pipe.plane.state->fb; > > + struct drm_gem_cma_object *gem; > > + > > + if (!fb) > > + return; > > + > > + gem = drm_fb_cma_get_gem_obj(fb, 0); > > + if (!gem) > > + return; > > + > > + writel(gem->paddr, mxsfb->base + mxsfb->devdata->next_buf); > > +} > > + > > static void mxsfb_crtc_mode_set_nofb(struct mxsfb_drm_private *mxsfb) > > { > > struct drm_display_mode *m = >pipe.crtc.state->adjusted_mode; > > const u32 bus_flags = mxsfb->connector.display_info.bus_flags; > > u32 vdctrl0, vsync_pulse_len, hsync_pulse_len; > > @@ -268,35 +283,41 @@ static void mxsfb_crtc_mode_set_nofb(struct > > mxsfb_drm_private *mxsfb) > >mxsfb->base + LCDC_VDCTRL3); > > > > writel(SET_DOTCLK_H_VALID_DATA_CNT(m->hdisplay), > >mxsfb->base + LCDC_VDCTRL4); > > > > + mxsfb_update_hw_next_buf(mxsfb); > > mxsfb_disable_axi_clk(mxsfb); > > } > > > > void mxsfb_crtc_enable(struct mxsfb_drm_private *mxsfb) > > { > > + if (mxsfb->enabled) > > + return; > > + > > mxsfb_crtc_mode_set_nofb(mxsfb); > > mxsfb_enable_controller(mxsfb); > > + > > + mxsfb->enabled = true; > > Is this state keeping really necessary? Sort of. The code is indeed a bit complicated, there are 3 possible states: "disabled", "enabled", "enabled but suspended" When the device resumes it should enable the controller only if it was enabled before (not if screen is blanked). After some digging I found that suspend/resume could be better implemented using drm_mode_config_helper_suspend/resume. This should automatically remember the crtc state "properly" without keeping state in the driver. One notable issue is that the lcdif block is only powered while the crtc is enabled but mxsfb_plane_atomic_update can be called outside that. So we need to check if the CRTC is enabled inside mxsfb_plane_atomic_update and avoid writing to hardware. Instead the HW write for next is delayed for the next mxsfb_crtc_mode_set_nofb. Does this make sense? Being unable to update the plane if the crtc is not enabled seems like it would be a common limitation of simple display controllers. Writing the FB paddr from mxsfb_crtc_mode_set_nofb feels wrong and I'm not sure it behaves correctly: when the display is resumed it seems to output an initial frame of incorrect data. It's possible the enable sequence needs to be adjusted. ___ dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel
Re: [PATCH 1/2] drm/mxsfb: Fix runtime PM for unpowering lcdif block
On 05.06.2018 19:11, Leonard Crestez wrote: > Adding imx6sl/sx lcdif nodes in a power domain currently does work, it > results in black/corrupted screens or hangs. While the driver does > enable runtime pm it does not deal correctly with the block being > unpowered. > > Fix by adding pm_runtime_get/put_sync to mxsfb_pipe_enable/disable. > > The mxsfb_plane_atomic_update function can get called before > mxsfb_pipe_enable while the block is not powered. When this happens the > write to LCDIF_NEXT_BUF is lost causing random corrupted pixels on > unblank. > > Fix this by splitting the writing of gem->paddr to nextbuf into a > mxsfb_update_hw_next_buf functiona and also calling it from > mxsfb_crtc_mode_set_nofb. > > Also add fields to mxsfb_drv to keep track of enabled/suspended states. > > Signed-off-by: Robert Chiras > Signed-off-by: Leonard Crestez That patch seems to go in the right direction. One comment below. > > --- > drivers/gpu/drm/mxsfb/mxsfb_crtc.c | 38 +++- > drivers/gpu/drm/mxsfb/mxsfb_drv.c | 72 +- > drivers/gpu/drm/mxsfb/mxsfb_drv.h | 3 ++ > 3 files changed, 100 insertions(+), 13 deletions(-) > > This was initially written by Robert for imx8m but I tested it also > works on imx6sx/imx6sl to DISPMIX power domain. > > Tested on imx6sl-evk and imx6sx-sdb with SEIKO 43WVF1G panel by > blanking and unblanking via sysfs and suspend/resume > > Testing requires a modified config (to enable MXFSB_DRM): > CONFIG_DRM_MXSFB=y > CONFIG_DRM_PANEL_SEIKO_43WVF1G=y > CONFIG_FB_MXS=n > > It also requires dts changes to enable the DISPMIX power domain. The > dts changes might break drivers so this patch attempts to fix the lcdif > driver first. > > Patch 2 is a RFC of imx6sl changes, imx6sx is a bit more complicated > and it also interacts with PCI. > > diff --git a/drivers/gpu/drm/mxsfb/mxsfb_crtc.c > b/drivers/gpu/drm/mxsfb/mxsfb_crtc.c > index 0abe77675b76..cce2ec1c80ae 100644 > --- a/drivers/gpu/drm/mxsfb/mxsfb_crtc.c > +++ b/drivers/gpu/drm/mxsfb/mxsfb_crtc.c > @@ -194,10 +194,25 @@ static int mxsfb_reset_block(void __iomem *reset_addr) > return ret; > > return clear_poll_bit(reset_addr, MODULE_CLKGATE); > } > > +static void mxsfb_update_hw_next_buf(struct mxsfb_drm_private *mxsfb) > +{ > + struct drm_framebuffer *fb = mxsfb->pipe.plane.state->fb; > + struct drm_gem_cma_object *gem; > + > + if (!fb) > + return; > + > + gem = drm_fb_cma_get_gem_obj(fb, 0); > + if (!gem) > + return; > + > + writel(gem->paddr, mxsfb->base + mxsfb->devdata->next_buf); > +} > + > static void mxsfb_crtc_mode_set_nofb(struct mxsfb_drm_private *mxsfb) > { > struct drm_display_mode *m = >pipe.crtc.state->adjusted_mode; > const u32 bus_flags = mxsfb->connector.display_info.bus_flags; > u32 vdctrl0, vsync_pulse_len, hsync_pulse_len; > @@ -268,35 +283,41 @@ static void mxsfb_crtc_mode_set_nofb(struct > mxsfb_drm_private *mxsfb) > mxsfb->base + LCDC_VDCTRL3); > > writel(SET_DOTCLK_H_VALID_DATA_CNT(m->hdisplay), > mxsfb->base + LCDC_VDCTRL4); > > + mxsfb_update_hw_next_buf(mxsfb); > mxsfb_disable_axi_clk(mxsfb); > } > > void mxsfb_crtc_enable(struct mxsfb_drm_private *mxsfb) > { > + if (mxsfb->enabled) > + return; > + > mxsfb_crtc_mode_set_nofb(mxsfb); > mxsfb_enable_controller(mxsfb); > + > + mxsfb->enabled = true; Is this state keeping really necessary? -- Stefan > } > > void mxsfb_crtc_disable(struct mxsfb_drm_private *mxsfb) > { > + if (!mxsfb->enabled) > + return; > + > mxsfb_disable_controller(mxsfb); > + > + mxsfb->enabled = false; > } > > void mxsfb_plane_atomic_update(struct mxsfb_drm_private *mxsfb, > struct drm_plane_state *state) > { > struct drm_simple_display_pipe *pipe = >pipe; > struct drm_crtc *crtc = >crtc; > - struct drm_framebuffer *fb = pipe->plane.state->fb; > struct drm_pending_vblank_event *event; > - struct drm_gem_cma_object *gem; > - > - if (!crtc) > - return; > > spin_lock_irq(>dev->event_lock); > event = crtc->state->event; > if (event) { > crtc->state->event = NULL; > @@ -307,14 +328,9 @@ void mxsfb_plane_atomic_update(struct > mxsfb_drm_private *mxsfb, > drm_crtc_send_vblank_event(crtc, event); > } > } > spin_unlock_irq(>dev->event_lock); > > - if (!fb) > - return; > - > - gem = drm_fb_cma_get_gem_obj(fb, 0); > - > mxsfb_enable_axi_clk(mxsfb); > - writel(gem->paddr, mxsfb->base + mxsfb->devdata->next_buf); > + mxsfb_update_hw_next_buf(mxsfb); > mxsfb_disable_axi_clk(mxsfb); > } > diff --git a/drivers/gpu/drm/mxsfb/mxsfb_drv.c > b/drivers/gpu/drm/mxsfb/mxsfb_drv.c > index 5cae8db9dcd4..c889cac2e275 100644 > ---
[PATCH 1/2] drm/mxsfb: Fix runtime PM for unpowering lcdif block
Adding imx6sl/sx lcdif nodes in a power domain currently does work, it results in black/corrupted screens or hangs. While the driver does enable runtime pm it does not deal correctly with the block being unpowered. Fix by adding pm_runtime_get/put_sync to mxsfb_pipe_enable/disable. The mxsfb_plane_atomic_update function can get called before mxsfb_pipe_enable while the block is not powered. When this happens the write to LCDIF_NEXT_BUF is lost causing random corrupted pixels on unblank. Fix this by splitting the writing of gem->paddr to nextbuf into a mxsfb_update_hw_next_buf functiona and also calling it from mxsfb_crtc_mode_set_nofb. Also add fields to mxsfb_drv to keep track of enabled/suspended states. Signed-off-by: Robert Chiras Signed-off-by: Leonard Crestez --- drivers/gpu/drm/mxsfb/mxsfb_crtc.c | 38 +++- drivers/gpu/drm/mxsfb/mxsfb_drv.c | 72 +- drivers/gpu/drm/mxsfb/mxsfb_drv.h | 3 ++ 3 files changed, 100 insertions(+), 13 deletions(-) This was initially written by Robert for imx8m but I tested it also works on imx6sx/imx6sl to DISPMIX power domain. Tested on imx6sl-evk and imx6sx-sdb with SEIKO 43WVF1G panel by blanking and unblanking via sysfs and suspend/resume Testing requires a modified config (to enable MXFSB_DRM): CONFIG_DRM_MXSFB=y CONFIG_DRM_PANEL_SEIKO_43WVF1G=y CONFIG_FB_MXS=n It also requires dts changes to enable the DISPMIX power domain. The dts changes might break drivers so this patch attempts to fix the lcdif driver first. Patch 2 is a RFC of imx6sl changes, imx6sx is a bit more complicated and it also interacts with PCI. diff --git a/drivers/gpu/drm/mxsfb/mxsfb_crtc.c b/drivers/gpu/drm/mxsfb/mxsfb_crtc.c index 0abe77675b76..cce2ec1c80ae 100644 --- a/drivers/gpu/drm/mxsfb/mxsfb_crtc.c +++ b/drivers/gpu/drm/mxsfb/mxsfb_crtc.c @@ -194,10 +194,25 @@ static int mxsfb_reset_block(void __iomem *reset_addr) return ret; return clear_poll_bit(reset_addr, MODULE_CLKGATE); } +static void mxsfb_update_hw_next_buf(struct mxsfb_drm_private *mxsfb) +{ + struct drm_framebuffer *fb = mxsfb->pipe.plane.state->fb; + struct drm_gem_cma_object *gem; + + if (!fb) + return; + + gem = drm_fb_cma_get_gem_obj(fb, 0); + if (!gem) + return; + + writel(gem->paddr, mxsfb->base + mxsfb->devdata->next_buf); +} + static void mxsfb_crtc_mode_set_nofb(struct mxsfb_drm_private *mxsfb) { struct drm_display_mode *m = >pipe.crtc.state->adjusted_mode; const u32 bus_flags = mxsfb->connector.display_info.bus_flags; u32 vdctrl0, vsync_pulse_len, hsync_pulse_len; @@ -268,35 +283,41 @@ static void mxsfb_crtc_mode_set_nofb(struct mxsfb_drm_private *mxsfb) mxsfb->base + LCDC_VDCTRL3); writel(SET_DOTCLK_H_VALID_DATA_CNT(m->hdisplay), mxsfb->base + LCDC_VDCTRL4); + mxsfb_update_hw_next_buf(mxsfb); mxsfb_disable_axi_clk(mxsfb); } void mxsfb_crtc_enable(struct mxsfb_drm_private *mxsfb) { + if (mxsfb->enabled) + return; + mxsfb_crtc_mode_set_nofb(mxsfb); mxsfb_enable_controller(mxsfb); + + mxsfb->enabled = true; } void mxsfb_crtc_disable(struct mxsfb_drm_private *mxsfb) { + if (!mxsfb->enabled) + return; + mxsfb_disable_controller(mxsfb); + + mxsfb->enabled = false; } void mxsfb_plane_atomic_update(struct mxsfb_drm_private *mxsfb, struct drm_plane_state *state) { struct drm_simple_display_pipe *pipe = >pipe; struct drm_crtc *crtc = >crtc; - struct drm_framebuffer *fb = pipe->plane.state->fb; struct drm_pending_vblank_event *event; - struct drm_gem_cma_object *gem; - - if (!crtc) - return; spin_lock_irq(>dev->event_lock); event = crtc->state->event; if (event) { crtc->state->event = NULL; @@ -307,14 +328,9 @@ void mxsfb_plane_atomic_update(struct mxsfb_drm_private *mxsfb, drm_crtc_send_vblank_event(crtc, event); } } spin_unlock_irq(>dev->event_lock); - if (!fb) - return; - - gem = drm_fb_cma_get_gem_obj(fb, 0); - mxsfb_enable_axi_clk(mxsfb); - writel(gem->paddr, mxsfb->base + mxsfb->devdata->next_buf); + mxsfb_update_hw_next_buf(mxsfb); mxsfb_disable_axi_clk(mxsfb); } diff --git a/drivers/gpu/drm/mxsfb/mxsfb_drv.c b/drivers/gpu/drm/mxsfb/mxsfb_drv.c index 5cae8db9dcd4..c889cac2e275 100644 --- a/drivers/gpu/drm/mxsfb/mxsfb_drv.c +++ b/drivers/gpu/drm/mxsfb/mxsfb_drv.c @@ -100,23 +100,27 @@ static const struct drm_mode_config_funcs mxsfb_mode_config_funcs = { static void mxsfb_pipe_enable(struct drm_simple_display_pipe *pipe, struct drm_crtc_state *crtc_state) { struct mxsfb_drm_private *mxsfb =