Troubleshooting:display

From postmarketOS

Generic debugging steps

Look at ~/.xsession-errors and ~/.xsession-errors.old

When lightDM starts your session, it redirects its output in that file. Including if that's a Wayland session!

If both files are empty, you can expect your graphical interface to have started without error. If they aren't created, lightDM may not have been started successfully. You can also try this: Display_manager#CanGraphical_issue .

Changing the panning

Some devices might have a framebuffer virtual size that's twice as big as the panel resolution, this is a common double buffering method. The application renders to one side of the display while the panning is set to the other side. You can try changing the panning with the `panning` file in the sysfs directory.

$ cat /sys/class/graphics/fb0/virtual_size
todo: add file contents here
$ cat /sys/class/graphics/fb0/pan
todo: add file contents here
$ echo 0 0 > /sys/class/graphics/fb0/pan

LightDM no longer starts Xorg on device without GPU driver

Set logind-check-graphical=false in /usr/share/lightdm/lightdm.conf.d/some-file.conf.

See also: pmaports!3827, https://github.com/canonical/lightdm/commit/77a7c6b7b8ca896b98ef43826641bdd520650bfb

Generic troubleshooting (any manufacturer)

Screen does not refresh

Symptoms of this problem may, for instance, be that the desktop environment starts and is displayed with initial content, but is frozen at that moment and never updates. At first it may seem to be a touchscreen problem, but the lack of automatic changes (e.g. clock showing current time) can point of frozen display instead.

Check if your screen refreshes when you do the following:

$ cat /sys/class/graphics/fb0/modes > /sys/class/graphics/fb0/mode

If it does, add msm-fb-refresher to the depends= of your device-package. It will basically do the same automatically. Despite the name, the tool is not specific to MSM at all.

(If the above command does not refresh your screen, you could try the msm-fb-refresher anyway.)

Note Make sure you zap the device rootfs after enabling the msm refresher, otherwise it will only get installed as a service and not enabled by the postmarketos-base install script (you'll only see a second of splash-screen otherwise and the refresher won't be restarted after exiting the initramfs

Colors are wrong in some way

Different framebuffer modes exist where the order of colors and transparency differ. Typically you have R (red) G (green) B (blue) and A (transparency), and they might be supplied as RGBA, BGRA, ARGB or another variant. If the order is wrong you can have different issues:

  • Everything is red-ish: example
  • Blue and red are swapped: example
  • Screen appears to show nothing (observed by switching from RGBA (which is correct) to ARGB)

To try to find a fix you need to locate the place(s) where the order is defined or set, somewhere in drivers/video. If the suggested patches in the subsections below do not work for you then have a look in the Debugging the kernel driver-section.

Qualcomm MSM devices

On Qualcomm devices the correct file should be drivers/video/msm/msm_fb.c or drivers/video/msm/mdss/mdss_fb.c. Changing the mode can (hopefully) be done in one of the following ways:

diff --git a/drivers/video/msm/mdss/mdss_fb.c b/drivers/video/msm/mdss/mdss_fb.c
index 53112ca6..69673017 100644
--- a/drivers/video/msm/mdss/mdss_fb.c
+++ b/drivers/video/msm/mdss/mdss_fb.c
@@ -569,7 +569,7 @@ static int mdss_fb_probe(struct platform_device *pdev)
 	mfd->bl_level = 0;
 	mfd->bl_scale = 1024;
	mfd->bl_min_lvl = 30;
-	mfd->fb_imgType = MDP_RGBA_8888;
+	mfd->fb_imgType = MDP_RGB_888;
 
 	mfd->pdev = pdev;
 	if (pdata->next)

imgType could also be MDP_RGBA_8888 like with motorola-condor and probably some other MSM8210 devices

Alternative patch (requires CONFIG_FB_MSM_DEFAULT_DEPTH_ARGB8888=y in the kernel config):

From [https://cascardo.eti.br/patches/0001-fix-video-argb-setting.patch 0001-fix-video-argb-setting.patch]
diff --git a/drivers/video/msm/msm_fb.c b/drivers/video/msm/msm_fb.c
index 251a5cbb753d..dcd9262a88f4 100644
--- a/drivers/video/msm/msm_fb.c
+++ b/drivers/video/msm/msm_fb.c
@@ -1342,16 +1342,16 @@ static int msm_fb_register(struct msm_fb_data_type *mfd)
 		fix->xpanstep = 1;
 		fix->ypanstep = 1;
 		var->vmode = FB_VMODE_NONINTERLACED;
-		var->blue.offset = 0;
-		var->green.offset = 8;
-		var->red.offset = 16;
+		var->blue.offset = 24;
+		var->green.offset = 16;
+		var->red.offset = 8;
 		var->blue.length = 8;
 		var->green.length = 8;
 		var->red.length = 8;
 		var->blue.msb_right = 0;
 		var->green.msb_right = 0;
 		var->red.msb_right = 0;
-		var->transp.offset = 24;
+		var->transp.offset = 0;
 		var->transp.length = 8;
 		bpp = 4;
 		break;
@@ -1361,16 +1361,16 @@ static int msm_fb_register(struct msm_fb_data_type *mfd)
 		fix->xpanstep = 1;
 		fix->ypanstep = 1;
 		var->vmode = FB_VMODE_NONINTERLACED;
-		var->blue.offset = 8;
-		var->green.offset = 16;
-		var->red.offset = 24;
+		var->blue.offset = 16;
+		var->green.offset = 8;
+		var->red.offset = 0;
 		var->blue.length = 8;
 		var->green.length = 8;
 		var->red.length = 8;
 		var->blue.msb_right = 0;
 		var->green.msb_right = 0;
 		var->red.msb_right = 0;
-		var->transp.offset = 0;
+		var->transp.offset = 24;
 		var->transp.length = 8;
 		bpp = 4;
 		break;
@@ -2260,15 +2260,15 @@ static int msm_fb_check_var(struct fb_var_screeninfo *var, struct fb_info *info)
 		/* Figure out if the user meant RGBA or ARGB
 		   and verify the position of the RGB components */
 
-		if (var->transp.offset == 24) {
-			if ((var->blue.offset != 0) ||
-			    (var->green.offset != 8) ||
-			    (var->red.offset != 16))
-				return -EINVAL;
-		} else if (var->transp.offset == 0) {
-			if ((var->blue.offset != 8) ||
+		if (var->transp.offset == 0) {
+			if ((var->blue.offset != 24) ||
 			    (var->green.offset != 16) ||
-			    (var->red.offset != 24))
+			    (var->red.offset != 8))
+				return -EINVAL;
+		} else if (var->transp.offset == 24) {
+			if ((var->blue.offset != 16) ||
+			    (var->green.offset != 8) ||
+			    (var->red.offset != 0))
 				return -EINVAL;
 		} else
 			return -EINVAL;
@@ -2365,7 +2365,7 @@ static int msm_fb_set_par(struct fb_info *info)
 		break;
 
 	case 32:
-		if (var->transp.offset == 24)
+		if (var->transp.offset == 0)
 			mfd->fb_imgType = MDP_ARGB_8888;
 		else
 			mfd->fb_imgType = MDP_RGBA_8888;

You might also need to add the mode, as done for linux-xiaomi-santoni:

diff --git a/drivers/video/msm/mdss/mdss_fb.c b/drivers/video/msm/mdss/mdss_fb.c
index 075ee8a3880..b4531f66a06 100644
--- a/drivers/video/msm/mdss/mdss_fb.c
+++ b/drivers/video/msm/mdss/mdss_fb.c
@@ -868,7 +868,8 @@ static int mdss_fb_probe(struct platform_device *pdev)
 	mfd->bl_scale = 1024;
 	mfd->bl_min_lvl = 30;
 	mfd->ad_bl_level = 0;
-	mfd->fb_imgType = MDP_RGBA_8888;
+	// Default framebuffer format.
+	mfd->fb_imgType = MDP_BGRA_8888;
 	mfd->calib_mode_bl = 0;
 
 	if (mfd->panel.type == MIPI_VIDEO_PANEL ||
@@ -2143,6 +2144,25 @@ static int mdss_fb_register(struct msm_fb_data_type *mfd)
 		bpp = 4;
 		break;
 
+	case MDP_BGRA_8888:
+		fix->type = FB_TYPE_PACKED_PIXELS;
+		fix->xpanstep = 1;
+		fix->ypanstep = 1;
+		var->vmode = FB_VMODE_NONINTERLACED;
+		var->blue.offset = 0;
+		var->green.offset = 8;
+		var->red.offset = 16;
+		var->blue.length = 8;
+		var->green.length = 8;
+		var->red.length = 8;
+		var->blue.msb_right = 0;
+		var->green.msb_right = 0;
+		var->red.msb_right = 0;
+		var->transp.offset = 24;
+		var->transp.length = 8;
+		bpp = 4;
+		break;
+
 	case MDP_YCRYCB_H2V1:
 		fix->type = FB_TYPE_INTERLEAVED_PLANES;
 		fix->xpanstep = 2;

Another possible workaround might be setting the default color depth to 16 bits per pixel (see this comment).

Exynos devices

On newer exynos devices you typically need to edit a file in drivers/video/fbdev/exynos/SOMETHING/decon_dsi.c.

To fix the problem where red and blue were swapped for dreamlte (exynos8895) the following patch worked:

diff --git a/drivers/video/fbdev/exynos/dpu/decon_dsi.c b/drivers/video/fbdev/exynos/dpu/decon_dsi.c
index eeaa2ba19778..3e597a145432 100644
--- a/drivers/video/fbdev/exynos/dpu/decon_dsi.c
+++ b/drivers/video/fbdev/exynos/dpu/decon_dsi.c
@@ -862,11 +862,11 @@ int decon_check_var(struct fb_var_screeninfo *var, struct fb_info *info)
 	case 24:
 		/* our 24bpp is unpacked, so 32bpp */
 		var->bits_per_pixel	= 32;
-		var->red.offset		= 16;
+		var->red.offset		= 0;
 		var->red.length		= 8;
 		var->green.offset	= 8;
 		var->green.length	= 8;
-		var->blue.offset	= 0;
+		var->blue.offset	= 16;
 		var->blue.length	= 8;
 		break;

How this patch was found is described in the Debugging the kernel driver-section.

Qualcomm framebuffer (mdss, mdp3)

MSM framebuffer doesn't work

We found out that removing the 3D driver support makes the MSM framebuffer work. See #66.

CONFIG_MSM_KGSL=n

-> Device Drivers
    -> Graphics support
        -> MSM 3D Graphics driver

On devices using the mdss_fb driver, adding mdss-fb-init-hack to the dependencies of the device package (example) might help.

Screen output flicker and inactivity

If you notice a very brief flicker upon boot, but immediately disappears and screen turns black, you might have a buggy framebuffer Qualcomm MDSS MDP3 driver.

You can try patching the source code by removing the function assigned to the mdp3_interface->off_fnc in the mdp3_ctrl.c file.

See example patch: [1]

Screen output messed up

If the output on the screen is not aligned correctly (see picture) it's probably because you have a wrong value in /sys/class/graphics/fb0/bits_per_pixel.

You can try patching the source code by setting a different video mode (RGB/ARGB/RGBA/YCRYCB/...).

This may also help, when your X11 is segfaulting (see #779).

See example patch: 06_fix_mdss_fb_rgb_mode.patch

Screen is blank outside of Weston

Something like that should help, see #603:

$ echo 0 > /sys/class/graphics/fb0/blank
$ echo 1 > /sys/devices/fd900000.qcom,mdss_mdp/qcom,mdss_fb_primary.134/leds/lcd-backlight/brightness

Debugging the kernel driver

Generally to understand what happens in the kernel you want to print messages to dmesg with information like error codes or the value of some variable.

The quickest way to try different kernel patches is to obtain the kernel source (git clone or download a source tar archive), add debug messages, build the kernel with envkernel.sh and then flash it. The whole build and flash procedure is described in Compiling kernels with envkernel.sh. Compiling and flashing only the kernel is faster than compiling and re-installing the kernel and rootfs everytime, which is done when you run pmbootstrap install [--android-recovery-zip].

To add debug messages you need to find interesting places where to add those though. If you have a display issue then drivers/video/ should be where to look.

Example: error in Xorg.0.log

In pmbootstrap#1618 the screen just showed noise, and /var/log/Xorg.0.log contained lots of these errors: (EE) FBDEV(0): FBIOPUTCMAP: Invalid argument. It was desired to get more information about the error (what exactly is the invalid argument?).

greping for FBIOPUTCMAP in the kernel source showed that the error comes from drivers/video/fbmem.c, and so some suitable debug messages could be:

diff --git a/drivers/video/fbmem.c b/drivers/video/fbmem.c
index 3e6371dc9038..a3ccb86baec6 100755
--- a/drivers/video/fbmem.c
+++ b/drivers/video/fbmem.c
@@ -1096,9 +1096,15 @@ static long do_fb_ioctl(struct fb_info *info, unsigned int cmd,
 		ret = copy_to_user(argp, &fix, sizeof(fix)) ? -EFAULT : 0;
 		break;
 	case FBIOPUTCMAP:
-		if (copy_from_user(&cmap, argp, sizeof(cmap)))
+			printk(KERN_ERR "PMOS DEBUG: calling copy_from_user...\n");
+		if (copy_from_user(&cmap, argp, sizeof(cmap))) {
+			printk(KERN_ERR "PMOS DEBUG: copy_from_user failed!\n");
 			return -EFAULT;
+		}
+		printk(KERN_ERR "PMOS DEBUG: copy_from_user ok!\n");
+		printk(KERN_ERR "PMOS DEBUG: calling fb_set_user_cmap...\n");
 		ret = fb_set_user_cmap(&cmap, info);
+		printk(KERN_ERR "PMOS DEBUG: fb_set_user_cmap returned %d\n", ret);
 		break;
 	case FBIOGETCMAP:
 		if (copy_from_user(&cmap, argp, sizeof(cmap)))

Example: red and blue colors are swapped

In pmaports#916 red and blue were switched on the display. As described in Colors are wrong in some way this most likely meant that that the framebuffer mode had to be changed somewhere. greping for BGRA, RGBA and red\.offset in drivers/video/ showed quite a lot of places to investigate. Since this was on an exynos device files in drivers/video/fbdev/exynos seemed especially interesting.

Since for many Qualcomm devices it was possible to simply change the mode from BGRA to RGBA or vice-versa, the first thing tried was this patch:

diff --git a/drivers/video/fbdev/exynos/decon_8890/decon_core.c b/drivers/video/fbdev/exynos/decon_8890/decon_core.c
index 2b65c68d8c74..388e94077847 100644
--- a/drivers/video/fbdev/exynos/decon_8890/decon_core.c
+++ b/drivers/video/fbdev/exynos/decon_8890/decon_core.c
@@ -3924,7 +3924,8 @@ static int decon_probe(struct platform_device *pdev)
                decon->vpp_used[decon->default_idma] = true;
                memset(&config, 0, sizeof(struct decon_win_config));
                config.vpp_parm.addr[0] = fbinfo->fix.smem_start;
-               config.format = DECON_PIXEL_FORMAT_RGBA_8888;
+               pr_err("PMOS DEBUG: decon_8890/decon_core: setting format to BGRA\n");
+               config.format = DECON_PIXEL_FORMAT_BGRA_8888;
                config.src.w = fbinfo->var.xres;
                config.src.h = fbinfo->var.yres;
                config.src.f_w = fbinfo->var.xres;
diff --git a/drivers/video/fbdev/exynos/dpu/decon_core.c b/drivers/video/fbdev/exynos/dpu/decon_core.c
index 152c86e06d74..662c0dc1a74f 100644
--- a/drivers/video/fbdev/exynos/dpu/decon_core.c
+++ b/drivers/video/fbdev/exynos/dpu/decon_core.c
@@ -4076,7 +4076,8 @@ static int decon_initial_display(struct decon_device *decon, bool is_colormap)
        set_bit(decon->dt.dft_idma, &decon->prev_used_dpp);
        memset(&config, 0, sizeof(struct decon_win_config));
        config.dpp_parm.addr[0] = fbinfo->fix.smem_start;
-       config.format = DECON_PIXEL_FORMAT_BGRA_8888;
+       pr_err("PMOS DEBUG: decon_8890/decon_dsi: setting format to RGBA\n");
+       config.format = DECON_PIXEL_FORMAT_RGBA_8888;
        config.src.w = fbinfo->var.xres;
        config.src.h = fbinfo->var.yres;
        config.src.f_w = fbinfo->var.xres;
diff --git a/drivers/video/fbdev/exynos/dual_dpu/decon_core.c b/drivers/video/fbdev/exynos/dual_dpu/decon_core.c
index d02af9c7e30a..1b9aa3ee57b1 100644
--- a/drivers/video/fbdev/exynos/dual_dpu/decon_core.c
+++ b/drivers/video/fbdev/exynos/dual_dpu/decon_core.c
@@ -3735,7 +3735,8 @@ static int decon_initial_display(struct decon_device *decon, bool is_colormap)
        set_bit(decon->dt.dft_idma, &decon->prev_used_dpp);
        memset(&config, 0, sizeof(struct decon_win_config));
        config.dpp_parm.addr[0] = fbinfo->fix.smem_start;
-       config.format = DECON_PIXEL_FORMAT_BGRA_8888;
+       pr_err("PMOS DEBUG: dual_dpu/decon_core: setting format to RGBA\n");
+       config.format = DECON_PIXEL_FORMAT_RGBA_8888;
        config.src.w = fbinfo->var.xres;
        config.src.h = fbinfo->var.yres;
        config.src.f_w = fbinfo->var.xres;

After compiling, flashing and booting the kernel, the colors were still wrong, and none of the messages showed up in dmesg, so this not the right places to set the mode. Next focus was therefore on the grep results of red\.offset in drivers/video/fbdev/exynos/. red.offset was being set in drivers/video/fbdev/exynos/decon_8890/decon_dsi.c, drivers/video/fbdev/exynos/dpu/decon_dsi.c and drivers/video/fbdev/exynos/dual_dpu/decon_dsi.c so this patch was added:

diff --git a/drivers/video/fbdev/exynos/decon_8890/decon_dsi.c b/drivers/video/fbdev/exynos/decon_8890/decon_dsi.c
index 8703b56108ac..101d1460ee98 100644
--- a/drivers/video/fbdev/exynos/decon_8890/decon_dsi.c
+++ b/drivers/video/fbdev/exynos/decon_8890/decon_dsi.c
@@ -187,6 +187,8 @@ int decon_check_var(struct fb_var_screeninfo *var, struct fb_info *info)
        var->transp.offset = 0;
        var->transp.length = 0;
 
+       pr_err("PMOS_DEBUG: decon_8890/decon_dsi: bits_per_pixel %d\n", var->bits_per_pixel);
+
        switch (var->bits_per_pixel) {
        case 1:
        case 2:
diff --git a/drivers/video/fbdev/exynos/dpu/decon_dsi.c b/drivers/video/fbdev/exynos/dpu/decon_dsi.c
index eeaa2ba19778..70a645fd8f4b 100644
--- a/drivers/video/fbdev/exynos/dpu/decon_dsi.c
+++ b/drivers/video/fbdev/exynos/dpu/decon_dsi.c
@@ -812,6 +812,8 @@ int decon_check_var(struct fb_var_screeninfo *var, struct fb_info *info)
        var->transp.offset = 0;
        var->transp.length = 0;
 
+       pr_err("PMOS_DEBUG: dpu/decon_dsi: bits_per_pixel %d\n", var->bits_per_pixel);
+
        switch (var->bits_per_pixel) {
        case 1:
        case 2:
diff --git a/drivers/video/fbdev/exynos/dual_dpu/decon_dsi.c b/drivers/video/fbdev/exynos/dual_dpu/decon_dsi.c
index d01b913bc8e0..30d7391b1bfa 100644
--- a/drivers/video/fbdev/exynos/dual_dpu/decon_dsi.c
+++ b/drivers/video/fbdev/exynos/dual_dpu/decon_dsi.c
@@ -815,6 +815,8 @@ int decon_check_var(struct fb_var_screeninfo *var, struct fb_info *info)
        var->transp.offset = 0;
        var->transp.length = 0;
 
+       pr_err("PMOS_DEBUG: dual_dpu/decon_dsi: bits_per_pixel %d\n", var->bits_per_pixel);
+
        switch (var->bits_per_pixel) {
        case 1:
        case 2:

After compiling, flashing and booting the kernel only the message from dpu/decon_dsi.c appeared in dmesg, and bits_per_pixel was 32.

In the same file we have

	switch (var->bits_per_pixel) {

    [...]

	case 32:
	case 28:
	case 25:
		var->transp.length	= var->bits_per_pixel - 24;
		var->transp.offset	= 24;
		/* drop through */
	case 24:
		/* our 24bpp is unpacked, so 32bpp */
		var->bits_per_pixel	= 32;
		var->red.offset		= 16;
		var->red.length		= 8;
		var->green.offset	= 8;
		var->green.length	= 8;
		var->blue.offset	= 0;
		var->blue.length	= 8;
		break;

Each pixel is 32 bit, and they are unpacked in the order BGRA, since blue.offset=0, green.offset=8, red.offset=16 and transp.offset=24. Since we want to swap blue<->red we change those offsets:

diff --git a/drivers/video/fbdev/exynos/dpu/decon_dsi.c b/drivers/video/fbdev/exynos/dpu/decon_dsi.c
index eeaa2ba19778..3e597a145432 100644
--- a/drivers/video/fbdev/exynos/dpu/decon_dsi.c
+++ b/drivers/video/fbdev/exynos/dpu/decon_dsi.c
@@ -862,11 +862,11 @@ int decon_check_var(struct fb_var_screeninfo *var, struct fb_info *info)
 	case 24:
 		/* our 24bpp is unpacked, so 32bpp */
 		var->bits_per_pixel	= 32;
-		var->red.offset		= 16;
+		var->red.offset		= 0;
 		var->red.length		= 8;
 		var->green.offset	= 8;
 		var->green.length	= 8;
-		var->blue.offset	= 0;
+		var->blue.offset	= 16;
 		var->blue.length	= 8;
 		break;

And after compiling, flashing and booting the kernel the problem is solved!

See also