[SOLVED] Even lower screen brightness?

[shazow]

@Thomas_Weissschuh

This should be a limitation of your desktop environment/window manager.

I can certainly disable the screen altogether, regardless of desktop environment, but that’s not the same thing as turning the backlight to zero.

Admittedly I am on x11 and not wayland, not sure if there’s a workaround in wayland that works better (perhaps Plasma is doing some magic there?).

In either case, relying on dpms and temperature shifting to approximate functionality is still just workarounds for what appears to be a bug/limitation of the AMD frameworks’ firmware (a limitation that is not present on the intel version, or any other laptop I’ve used).

Perhaps it’s worth opening a fresh thread?

P.S. Hey fellow Python http alum!

[Thomas_Weissschuh]

@shazow

I can certainly disable the screen altogether, regardless of desktop environment, but that’s not the same thing as turning the backlight to zero.

This was specifically for this sentence:

@shazow

Using dpms to turn off the screen is not equivalent because then we can’t use input devices without activating the screen again.

On my sway (wayland) setup if I do the following:

swaymsg 'output eDP-1 dpms off'; sleep 10; swaymsg 'output eDP-1 dpms on'

Then I can wildly use input devices during those 10 seconds and the screen stays black.

From a kernel perspective the semantics of brightness = 0 are not well-defined.
Different drivers implement different behaviors.

There is ongoing work to get a new backlight API, with a well-defined 0 value.
But that definition is “just enough to be readable”, for “off” use dpms.

FWIW I forced the kernel to override the limits received from ACPI and set the PWM to a hard 0.
But that still didn’t turn off the screen completely.
So it may be in the GPU firmware.

[shazow]

@Thomas_Weissschuh

From a kernel perspective the semantics of brightness = 0 are not well-defined.
Different drivers implement different behaviors.

That’s fair. I won’t make an argument that the driver does not technically comply within the boundaries of the the kernel specifications.

I am making the argument that from a Framework customer perspective, particularly people in this thread (and from a consistency perspective of the Intel version of the device, and the vast majority of other consumer laptops like Thinkpads and Macbooks): Framework should consider changing this behaviour on their AMD devices.

[jwp]

The backlight turning off works both on fedora39 and 40 using those respective DE brightness sliders when set to 0 on the AMD FW13. I don’t think this has anything to do with Framework it’s whatever userspace you’re using and/or configuration that’s been changed from the initial state. I don’t know I haven’t used Xorg in years, but I would suggest testing on a fedora40 live distro in and seeing if you have the same issue. You’re the first user i’ve seen who hasn’t been able to get the backlight turned off i’ve come across on the forums.

[lectrode]

Regarding the lowest brightness (while still being on), I agree this shouldn’t be marked as “resolved”. There is a workaround that helps (redshift or other tools to darken the displayed image), but not enough IMHO.

It really does not go low enough for use in low-light environments. It is still uncomfortable/harsh to look at in the darkest of environments even when using redshift. Other laptop screens can dim the backlight considerably lower than the AMD-powered framework 16.

Case in point:

Left: framework 16
Right: thinkpad e595

Both screens are showing the same completely white image at their lowest brightness (brightnessctl s 0%) (without the use of redshift or similar software).

I should mention: in the picture, the thinkpad has no screen protector, while the framework has this “blue light blocking” one

[Leonard]

I too agree that the lowest brightness is much too high. Using night light makes it somewhat bearable, though it remains a workaround.

[RESPONDED] AMD Framework minimum brightness too high (now with measurements)

Rtings has not tested the latest Intel Framework laptop. The latest Framework 13 laptops (ie. AMD and 13th Gen Intel) use the BOE NE135GXM-N61 display panel. The older Intel Framework laptops use the BOE NE135FBM-N41 display panel. The main difference is that the NE135GXM-N61 has a matte finish, however when I looked through the datasheets I noticed a that the BOE NE135FBM-N41 can supposedly go down to 1% of max brightness whereas the NE135GXM-N61 can only go down to 5% of max brightness. P…

According to this post, the matte panel simply has a much higher rated minimum brightness than the glossy one (20 nits vs 4 nits, respectively). Matt promised to provide some more info on the subject in that same thread, but hasn’t come back to it yet.

[Lucio_Torelli]

Even if it is a hardware problem it would be great to have a native software solution from Framework. The lowest brightness on the laptop is higher than my monitor at 50% brightness. I avoid using the laptop past 19:00 because of that.

Marking something as solved without a solution is quite upsetting.

[Thomas_Weissschuh]

@Thomas_Weissschuh

FWIW I forced the kernel to override the limits received from ACPI and set the PWM to a hard 0.
But that still didn’t turn off the screen completely.
So it may be in the GPU firmware.

I tested this again, and while sending a PWM signal of 0 to the GPU does not completely turn off the display, it is still considerably less bright than by default.

You can test this with the following patch

diff --git a/drivers/gpu/drm/amd/display/dc/link/protocols/link_edp_panel_control.c b/drivers/gpu/drm/amd/display/dc/link/protocols/link_edp_panel_control.c
index ad9aca790dd7..42f1f476ec67 100644
--- a/drivers/gpu/drm/amd/display/dc/link/protocols/link_edp_panel_control.c
+++ b/drivers/gpu/drm/amd/display/dc/link/protocols/link_edp_panel_control.c
@@ -521,12 +521,18 @@ bool edp_set_backlight_level(const struct dc_link *link,
                uint32_t backlight_pwm_u16_16,
                uint32_t frame_ramp)
 {
+       static bool force_real_zero;
        struct dc  *dc = link->ctx->dc;
 
        DC_LOGGER_INIT(link->ctx->logger);
        DC_LOG_BACKLIGHT("New Backlight level: %d (0x%X)\n",
                        backlight_pwm_u16_16, backlight_pwm_u16_16);
 
+       if (force_real_zero && backlight_pwm_u16_16 < 5000)
+               backlight_pwm_u16_16 = 0;
+       force_real_zero = !force_real_zero;
+
+
        if (dc_is_embedded_signal(link->connector_signal)) {
                struct pipe_ctx *pipe_ctx = get_pipe_from_link(link);

And then repeatedly do echo 0 > /sys/class/backlight/amdgpu_bl1/brightness and observe the different brightness levels.

For a “permanent” “fix” use this:

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c
index 7099ff9cf8c5..6e463e5860e5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c
@@ -383,6 +383,9 @@ static int amdgpu_atif_query_backlight_caps(struct amdgpu_atif *atif)
                        characteristics.min_input_signal;
        atif->backlight_caps.max_input_signal =
                        characteristics.max_input_signal;
+
+       /* Framework Firmware returns "12" by default. */
+       atif->backlight_caps.min_input_signal = 0;
 out:
        kfree(info);
        return err;

This is just some constant in the Firmware which should be easily changable for Framework.
(I don’t see a way to easily override this as a user)

Only tested on a matte panel.

[lectrode]

Can confirm that using that kernel driver patch to force override the minimum value provided by the firmware makes it much better. With that, it’s now much more inline with expectations.

Left: framework 16
Right: thinkpad e595

with patch:

without patch:

[gaben]

I would like to experiment with this patch. Could you or @Thomas_Weissschuh point me in the direction of how to compile (?) the modified amdgpu driver and configure my system to use it? What are the keywords I should search for?

I am running Fedora 39.

[Thomas_Weissschuh]

@gaben

This should work: Building a Custom Kernel :: Fedora Docs

But further than that I have no clue about Fedora.

[Thomas_Weissschuh]

I sent a quirk to the kernel: [PATCH] drm/amd: force min_input_signal to 0 on Framework AMD 13/16 - Thomas Weißschuh
Let’s see if it can be made to work out of the box.

[Leonard]

Shouldn’t this just be enabled for the default matte panel? And not for the glossy panel and the new, recently launched panel?

Aside from that, I suppose a Framework firmware patch would be the best long-term solution?

[Thomas_Weissschuh]

@Leonard

And not for the glossy panel and the new, recently launched panel?

Depends on what it does on those panels.
Somebody would need to test it.

@Leonard

Aside from that, I suppose a Framework firmware patch would be the best long-term solution?

Yes.

[Alex_Morozov]

I did measurements for fw13 old glossy panel and fw16 panel. On each graph there are 2 curves – unpatched (higher) and patched (lower) version. Axis X – value set in /sys/.../amdgpu_bl0/brightness, axis Y – brightness measured by a sensor in some relative units.

Fw13 original glossy (~2022)

Fw16

With the patch above, for both panels min brightness goes 5.5x down.

As I understand it, framework folks aren’t going to patch firmware for 16" model, since current value is already minimal per panel’s datasheet. It works nicely outside of spec, though.

[JanW]

Just writing to add a +1 on the request to lower the minimal brightness for the ADM FW13 matte panel (not the new high-resolution one). I use my laptop for work, and the screen brightness at minimal setting is too bright to be used in a dark conference room - it’s distracting to everyone around me. Confirmed with brightnessctrl that it is actually at the minimum:
Current brightness: 1 (0%)

Does anyone know if the new 2.8K panel goes lower on a 7840U FW13 ? Might be a reason for me to upgrade, even if I don’t need the higher resolution or higher refresh rates at all. Or is minimum brightness even higher, given the higher max brightness of that screen ?

[adioe3]

Has there been any news on this? Compiling the kernel on Fedora for each update is a pain…

[Thomas_Weissschuh]

@adioe3

Has there been any news on this?

I just sent out a new patchset, any testing is welcome.
https://lore.kernel.org/lkml/20240731-amdgpu-min-backlight-quirk-v3-0-46d40bb21a62@weissschuh.net/T/#t

[GreyXor]

Thanks very much, any answers from framework team to integrate this into firmware ?

[Nelo390]

This worked for me! Does this patch reduce the maximum brightness? I don’t think it does, but I might as well make sure.