New Raspberry Pi Kernels & Related Packages

ARM
745

This patch from yesterday looks significant.

746

I thought so also yesterday when I saw it but after compiling it and testing not much changed. It still will not boot with fkms using V3D rendering. It boots with kms using V3D rendering and seems to work but I get a kernel backtrace that bothers me but it could be because they disabled CONFIG_ RPIVID_VIDEO because it will not compile:

backtrace
backtrace1664Ɨ954 191 KB

With their latest upgrade to kernel 5.10.52 the issues with 5.13 and 5.14 has crept in to it. I do not want to release any kernel that does not perform with all of the assorted things here that has been working in the past our members use. The RPi releases do not have all of the use cases we have here. With all of the mega patching they have done with the last few kernels to port their work to match upstream they appear to be busy fixing one issue at a time and it looks like they are starting with issues with in their own images.

I am thinking very seriously about going back to the kernel 5.10.46 before all of the mega patch work began where everything seems to work for everyone and freezing it until they get things sorted out. Lately every time I see a commit each day in one of the 3 kernels I re-compile and test. This is starting to get old; some days it is 3 kernels to re-compile.

747

Should one of these be kms?

I guess you could have the rc kernel ā€œbrokenā€, but yeah, a fully working kernel is likely preferred for testing and stable.

748

I corrected it. Thanksā€¦

I double checked by using their own config with make bcm2711_defconfig.
.

749

Have you tried disable_fw_kms_setup=1?

No clue if it helpsā€¦ just tossing it out there.

750

Did not change anything.

751

I built the current rpi 5.14-rc2 with bcm2711_defconfig and booting with UEFI:

With kms, no backtrace, clean boots except for:
hdmi-audio-codec hdmi-audio-codec.3.auto: ASoC: error at snd_soc_dai_startup on i2s-hifi: -19

and it works with v3d enabled or disabled.

With fkms, a black screen and unresponsive keyboard.

752

Very similar to what I see also on all 3 of the latest kernels. I never tested the audio because I just did not care with all of the other things going on and not to mention them disabling a feature because it would not compile. I was also getting the hdmi-audio-codec errors.

753

If I switch between v3d and llvmpipe by the use of the disable-v3d suse overlay, I have working sound. vc4 with digital hdmi and bcm2835 for the headphone jack. In both cases, I load the vc4-kms-v3d-pi4 overlay.

754

Or maybe it is time to move to kms as the default? The re-base of the code to upstream makes it the ā€œdefaultā€ now, at least as far as the code is concerned. Evidenced by the RPI folks having to patch, to get fkms working.

755

If anyone is interested in the use of the disable-v3d overlay:

$ nano disable-v3d.dts

/dts-v1/;

/ {
        compatible = "brcm,bcm2711";

        fragment@0 {
                target = <0xffffffff>;

                __overlay__ {
                        status = "disabled";
                };
        };

        __fixups__ {
                v3d = "/fragment@0:target:0";
        };
};

Next, compile it into the overlay:

sudo dtc -@ -Hepapr -I dts -O dtb -o /boot/overlays/disable-v3d.dtbo disable-v3d.dts

Finally, add it to config.txt (order matters):

dtparam=audio=on
dtoverlay=vc4-kms-v3d-pi4
dtoverlay=disable-v3d

You will have vc4 for hdmi sound and 2d video and llvmpipe rendering. Comment out dtoverlay=disable-v3d and you will have vc4 for hdmi sound and 2d video and v3d rendering. Nice and simple, and benefits from stable device detection and configuration. bcm2835 provides the headphone jack in both instances.

For Plasma, you can not use xf86-video-fbturbo-git. But for other desktops it may work, but you will need to manage the 99-fbturbo.conf file.

Edit: You can incorporate u-boot and have a boot menu which selectively loads the disable-v3d overlay.

Edit 2: And you want these as kernel arguments: snd_bcm2835.enable_compat_alsa=0 and snd_bcm2835.enable_hdmi=1

756

So far, I have not seen a backtrace with 5.14-rc2 + kms + v3d.

757

Yep kms, + v3d seems to be working ok. The problem is not everyone can use kms + v3d. I can not with my Vizio tv (never have been able to) which I suspect will be true for most using tvā€™s for monitors; fkms + v3d does fine on my Vizio. But my old Gateway monitor has no issues with either mode.

Are you still using bcm2711_defconfig?

758

Yes 5.14.0-rc2-v8, not even a local name change.

759

Iā€™m uploading the packages with all of our modules plugged in for you to test.

760

Of course test this with kms + v3d only. It should have all modules enabled as in 5.12.

md5sum:
03cb63592b0b6e853a53aba1d2e1c95a 5.14-rc2.tar

https://drive.google.com/file/d/1rmvly427-Kr2OcePZ99SLgkXxVmDPcaM/view?usp=sharing

761

I tested the latest upstream linux-rc3 @strit built and uploaded to unstable this mourning and it seems to be ok with the pi4. Of course no V3D renderingā€¦

rc3
rc3824Ɨ987 51.4 KB

762

Have you tried editing the Kconfig file so you can add CONFIG_DRM_V3D=m?

This patch is not applied to upstream. Donā€™t do this.

763

The test rc kernel, seems to be good so far. Pretty clean dmesg and journalctl, no backtrace.

764

Do not know what you are getting at. Present config in what you are testing:

config