On Tue, Mar 03, 2020 at 08:04:05AM -0500, Jonathan Marek wrote: > What Xorg prints doesn't mean anything. I don't think there will be errors > in dmesg, you need to run something that does pageflips as fast as possible > and see that the refresh rate is still 60. (modetest with -v, glmark-drm are > examples) I assume that you mean modetest from https://gitlab.freedesktop.org/mesa/drm/tree/master/tests/modetest ? Here's the modeset connector information: id encoder status name size (mm) modes encoders 32 31 connected DSI-1 62x110 1 31 modes: index name refresh (Hz) hdisp hss hse htot vdisp vss vse vtot) #0 1080x1920 71.71 1080 1082 1084 1086 1920 1922 1924 1926 150000 flags: ; type: preferred, driver And the page flip results... $ modetest -v -s 32:1080x1920 trying to open device 'msm'...done setting mode 1080x1920-71.71Hz@XR24 on connectors 32, crtc 50 failed to set gamma: Function not implemented freq: 13.50Hz freq: 13.51Hz freq: 13.51Hz It's the same results with and without Ville's patch. Here's the beginning of the glmark2 results with the x11-gl flavor: ======================================================= glmark2 2017.07 ======================================================= OpenGL Information GL_VENDOR: freedreno GL_RENDERER: FD330 GL_VERSION: 3.1 Mesa 20.0.0-devel ======================================================= [build] use-vbo=false: FPS: 26 FrameTime: 38.462 ms [build] use-vbo=true: FPS: 26 FrameTime: 38.462 ms [texture] texture-filter=nearest: FPS: 26 FrameTime: 38.462 ms [texture] texture-filter=linear: FPS: 26 FrameTime: 38.462 ms [texture] texture-filter=mipmap: FPS: 27 FrameTime: 37.037 ms [shading] shading=gouraud: FPS: 27 FrameTime: 37.037 ms [shading] shading=blinn-phong-inf: FPS: 27 FrameTime: 37.037 ms [shading] shading=phong: FPS: 27 FrameTime: 37.037 ms [shading] shading=cel: FPS: 26 FrameTime: 38.462 ms [bump] bump-render=high-poly: FPS: 27 FrameTime: 37.037 ms [bump] bump-render=normals: FPS: 27 FrameTime: 37.037 ms [bump] bump-render=height: FPS: 27 FrameTime: 37.037 ms [effect2d] kernel=0,1,0;1,-4,1;0,1,0;: FPS: 25 FrameTime: 40.000 ms [effect2d] kernel=1,1,1,1,1;1,1,1,1,1;1,1,1,1,1;: FPS: 26 FrameTime: 38.462 ms [pulsar] light=false:quads=5:texture=false: FPS: 26 FrameTime: 38.462 ms [desktop] blur-radius=5:effect=blur:passes=1:separable=true:windows=4: FPS: 26 FrameTime: 38.462 ms [desktop] effect=shadow:windows=4: FPS: 27 FrameTime: 37.037 ms ... Brian > > On 3/3/20 7:26 AM, Brian Masney wrote: > > On Mon, Mar 02, 2020 at 10:36:54PM -0500, Jonathan Marek wrote: > > > Another thing: did you verify that the panel still runs at 60hz (and not > > > dropping frames to 30hz)? IIRC that was the behavior with lower clock. > > > > Yes, the panel is running at 60 HZ according to the Xorg log with > > Ville's patch applied: > > > > modeset(0): Modeline "1080x1920"x60.0 125.50 1080 1082 1084 1086 > > 1920 1922 1924 1926 (115.6 kHz eP) > > > > I verified there's no underflow errors in dmesg. > > > > If I recall correctly, the clock speeds that was in your tree was set > > too low for the gpu_opp_table (that wouldn't cause this issue), but I > > seem to recall there were some other clock speed mismatches. The > > bandwidth requests weren't set on the RPM as well, so maybe that > > contributed to the problem. That's done upstream with the msm8974 > > interconnect driver: > > > > https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/interconnect/qcom/msm8974.c > > > > There's a separate known issue with 'pp done time out' errors that > > occur on the framebuffer that started upstream several months ago with > > the introduction of async commit support in the MSM driver. I tried > > working around this by enabling the autorefresh feature but it's not > > fully working yet and I hit a dead end since there's no docs available > > publicly for this. The grim details are at: > > > > https://lore.kernel.org/lkml/20191230020053.26016-2-masneyb@xxxxxxxxxxxxx/ > > > > So I'm still OK with Ville's patch going in. > > > > Brian > > > > > > > > > > On 3/2/20 10:28 PM, Jonathan Marek wrote: > > > > > > > > On 3/2/20 10:13 PM, Brian Masney wrote: > > > > > On Mon, Mar 02, 2020 at 03:48:22PM -0500, Jonathan Marek wrote: > > > > > > Hi, > > > > > > > > > > > > This is a command mode panel and the the msm/mdp5 driver uses > > > > > > the vrefresh > > > > > > field for the actual refresh rate, while the dotclock field is > > > > > > used for the > > > > > > DSI clocks. The dotclock needed to be a bit higher than > > > > > > necessary otherwise > > > > > > the panel would not work. > > > > > > > > > > > > If you want to get rid of the separate clock/vrefresh fields there would > > > > > > need to be some changes to msm driver. > > > > > > > > > > > > (note I hadn't made the patch with upstreaming in mind, the > > > > > > 150000 value is > > > > > > likely not optimal, just something that worked, this is something that > > > > > > should have been checked with the downstream driver) > > > > > > > > > > Is this the right clock frequency in the downstream MSM 3.4 kernel that > > > > > you're talking about? > > > > > > > > > > https://github.com/AICP/kernel_lge_hammerhead/blob/n7.1/arch/arm/mach-msm/clock-8974.c#L3326 > > > > > > > > > > > > > > > > > > No, I'm talking about the DSI clock (the driver for it is in > > > > drm/msm/dsi/). For a command mode panel the front/back porches aren't > > > > relevant, but the dsi pixel/byte clock need to be a bit higher than > > > > 1920x1080x60. Since 125498 is a little higher than 124416 that might be > > > > enough (there is also rounding of the clock values to consider). > > > > > > > > > I don't see any obvious clock values in the downstream command mode > > > > > panel configuration: > > > > > > > > > > https://github.com/AICP/kernel_lge_hammerhead/blob/n7.1/arch/arm/boot/dts/msm8974-hammerhead/msm8974-hammerhead-panel.dtsi#L591 > > > > > > > > > > > > > > > Anyways, I tried Ville's patch with the framebuffer, kmscube, and X11 > > > > > and everything appears to be working fine. You can add my Tested-by if > > > > > you end up applying this. > > > > > > > > > > Tested-by: Brian Masney <masneyb@xxxxxxxxxxxxx> > > > > > > > > > > Brian > > > > > > > > > > > > > > > > On 3/2/20 3:34 PM, Ville Syrjala wrote: > > > > > > > From: Ville Syrjälä <ville.syrjala@xxxxxxxxxxxxxxx> > > > > > > > > > > > > > > The currently listed dotclock disagrees with the currently > > > > > > > listed vrefresh rate. Change the dotclock to match the vrefresh. > > > > > > > > > > > > > > Someone tell me which (if either) of the dotclock or vreresh is > > > > > > > correct? > > > > > > > > > > > > > > Cc: Jonathan Marek <jonathan@xxxxxxxx> > > > > > > > Cc: Brian Masney <masneyb@xxxxxxxxxxxxx> > > > > > > > Cc: Linus Walleij <linus.walleij@xxxxxxxxxx> > > > > > > > Signed-off-by: Ville Syrjälä <ville.syrjala@xxxxxxxxxxxxxxx> > > > > > > > --- > > > > > > > drivers/gpu/drm/panel/panel-simple.c | 2 +- > > > > > > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > > > > > > > > > > > diff --git a/drivers/gpu/drm/panel/panel-simple.c > > > > > > > b/drivers/gpu/drm/panel/panel-simple.c > > > > > > > index b24fdf239440..f958d8dfd760 100644 > > > > > > > --- a/drivers/gpu/drm/panel/panel-simple.c > > > > > > > +++ b/drivers/gpu/drm/panel/panel-simple.c > > > > > > > @@ -3996,7 +3996,7 @@ static const struct panel_desc_dsi > > > > > > > panasonic_vvx10f004b00 = { > > > > > > > }; > > > > > > > static const struct drm_display_mode lg_acx467akm_7_mode = { > > > > > > > - .clock = 150000, > > > > > > > + .clock = 125498, > > > > > > > .hdisplay = 1080, > > > > > > > .hsync_start = 1080 + 2, > > > > > > > .hsync_end = 1080 + 2 + 2, > > > > > > > _______________________________________________ dri-devel mailing list dri-devel@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/dri-devel
