User:Knuxify/Draft:Troubleshooting/Development issues

From postmarketOS Wiki
Hint Before reading this page, check if the main Troubleshooting page does not already have tips for troubleshooting your issue.

This page contains more specific troubleshooting steps for developing new ports or performing development actions. General information about the troubleshooting process (where to find logs, etc.) can be found on the main Troubleshooting page.


  • Always look at the log file for a detailed error message: ~/.local/var/pmbootstrap/log.txt
  • Make sure, that you have the latest version. Your bug may already be fixed.
  • Please report the bugs you have found in the issues, so we can fix them.
  • Syslog is enabled, and will log to shared memory once the OS has booted. It can be read with the logread tool in BusyBox. The syslog is currently not preserved across reboots.

See also: How to report issues

Specific troubleshooting pages

Port/device troubleshooting

References

Porting process, preparation and various information

See also: Troubleshooting#pmbootstrap and build related errors

My device runs the Linux kernel (e.g. Android), but the vendor does not provide the source

Icon TODO: Move to porting guide
  1. Ask kindly via e-mail.
  2. If they don't provide the source, team up with gpl-violations.org. You have the legal right to get the source, because the Linux kernel is under the GPLv2!
  3. Consider building generic kernel for your device.
  4. Another option is using prebuilt kernels

My Android device does not have partition labels, how do I know where to put the boot.img file etc?

Icon TODO: Move to porting guide?
  • Search online for the partitions, they are probably documented somewhere.
  • Try to dump all partitions. Run `file` on them, it should roughly tell you the contents. If that does not help, run binwalk.
    • file output: check if it is a kernel (vmlinuz), boot.img file or just "data".
    • In case you have "data", don't flash anything there, we probably can't generate that format yet. Please open an issue in GitHub or ask on the channel.
  • You can also look at recovery zips from other Android ROMs, that were released for your device. Extract it and look at META-INF/com/google/android/updater-script, search for commands starting with run_program("/sbin/busybox", "dd", they are for flashing partitions.
    • In one example, there was run_program("/sbin/busybox", "dd", "if=/tmp/boot.blob", "of=/dev/block/mmcblk0p4");, which means that a boot.blob file gets flashed to /dev/block/mmcblk0p4. That boot.blob file is also in the recovery zip - run the file program on it (see above).

Proprietary blobs

Icon TODO: Move this to porting guide

Many devices require proprietary blobs for enabling certain peripherals (wifi/bluetooth/cellular/GPS). These repositories appear to contain blobs for a large percentage of devices out there.

Errors on the device

Touch screen device present but not outputting any touch events

Icon TODO: Device specific issue

For some devices (at least motorola-shamu and motorola-lux), the touch screen I2C device is present but not activated at boot, and must be activated before it produces any output.

For the motorola-lux, this is done in the following way:

echo 1 > /sys/devices/soc.0/78b8000.i2c/i2c-4/4-0020/drv_irq

The path to the correct touch device depends on the device. This can be added to an initfs hook to activate the touch screen automatically at boot.

During initramfs generation: "modprobe: WARNING: Module drm not found in directory /lib/modules/..."

This means, that a module can not be put in an initramfs. When you are not using a mainline kernel, it is likely that there are no modules configured, but the code is built directly into the kernel. This means, it is expected and can be ignored.

I have ported postmarketOS to my device and the screen doesn't get updated

Icon TODO: Isn't this msm-fb-refresher?

There is a known bug where if you execute cat /sys/class/graphics/fb0/modes > /sys/class/graphics/fb0/mode exactly one frame is drawn, and the screen isn't updated otherwise.

Device reboots

  • If your device always reboots after the same time (e.g. 2 minutes) after booting, you probably need to feed (or disable) a watchdog. Try to search your kernel config for watchdog, and also the internet for watchdog + your phone name.
  • Android: The kernel config option CONFIG_ANDROID_LOW_MEMORY_KILLER may cause reboots. Please check if you have disabled it. If this helps you, please let us know, so we can consider checking for this option.
  • Android: it is possible to get the dmesg log from just before the crash after the reboot if you enable some kernel options.

User interface is unresponsive

  • If you have a situation where an application or display manager user interface is unresponsive, you can use TTYescape to kill the offending process, power cycle the device, or otherwise access helpful command line utilities.

SSH / udev / touch screen doesn't work

  • Make sure that DEVTMPFS is enabled in your kernel config. Do not enable DEVTMPFS_MOUNT.

I can't access the framebuffer device

"if running weston from ssh, use --tty to specify a tty"

Specify a tty in weston's commandline. The correct syntax is: weston --tty=1.

Weston and fbsplash -s <image> do not work

If Weston returns an error because it could not access /dev/tty1, even when you specify it on the command line as above, you may need to enable Virtual Terminals in the kernel config.
You can enable CONFIG_VT using pmbootstrap menuconfig and checking under Device Drivers -> Character devices -> Virtual terminal

ERROR: unsupported 'crypto_LUKS' filesystem (/dev/mmcblk2p2)

See postmarketos-release-upgrade#15.

Flashing problems

Invalid sparse file format at header magic

In some devices (e.g.: bullhead, titan), when flashing the system (pmbootstrap flasher flash_rootfs), it works but fastboot complains of Invalid sparse file format at header magic.

$ ./pmbootstrap.py flasher flash_rootfs
[23:02:58] (native) install android-tools
[23:02:58] (native) flash rootfs image
target reported max download size of 536870912 bytes
Invalid sparse file format at header magic
sending sparse 'system' 1/1 (274101 KB)...
OKAY [  9.452s]
writing 'system' 1/1...
OKAY [ 12.584s]
finished. total time: 22.036s
[23:03:20] Done

It happens when your device expects a sparse system image. The workaround is as follows:

  • Set deviceinfo_flash_sparse=true in deviceinfo

To do this, edit the file in ./local/var/pmbootstrap/cache_git/pmaports/device/your_device_name/deviceinfo and add deviceinfo_flash_sparse="true". When finished, run "pmbootstrap install" to rebuild the file system

For more information, see issue #299

Login screen hangs

When switching from an installation on the system-partition to an installation on the userdata-partition, or when switching from a downstream-kernel to a mainline-kernel, it can happen, that the system boots up but hangs before login.

Logread via SSH may print a loop of

Jan 1 02:45:10 device-xy daemon.info init: can't open /dev/ttyHSL0: No such file or directory
Jan 1 02:45:10 device-xy daemon.info init: process '/sbin/getty -L ttyHSL0 115200 vt100' (pid 9119) exited. Scheduling for restart.

Try to erase the partitions before flashing: fastboot erase system && fastboot erase userdata && fastboot erase cache

Heimdall: Flashing fails at 7%

You might also be getting the error "ERROR: Failed to confirm end of file transfer sequence"

It might be, that your device only accepts "sparse images" (#1643). Try setting deviceinfo_flash_sparse="true" in the deviceinfo.

FAILED (remote: Unknown chunk type)

This is a bootloader limitation. Flash with netcat as a workaround, or use this wiki's search feature to come up with more related information. Feel free to organize this in a better way, but this should be a good starting point. See also #1735.

flasher (fastboot) is stuck in < waiting for any device >

Maybe you need to install udev rules on your host Linux distribution. Arch has the android-udev package for example, which uses the rules from here.

If that does not help, it might be necessary to set the flash_fastboot_vendor_id in your deviceinfo (this was necessary for the Amazon Kindle Fire HDX). Try to search the web for your device name and "fastboot vendor id".

If you were previously able to use fastboot and it stopped working, sometimes pmbootstrap shutdown to restart the chroot helps (#1633). Also try to reboot the device into fastboot again.

ERROR: Failed to confirm end of file transfer sequence!

This message can mean anything. To get more information you can instead create an android-recovery-zip and install that from recovery. The installation will probably fail now as well, but at least this time you will have logs in /tmp/postmarketos/.

One possible reason for the error is simply that the image is too large for the partition, in which case you need to decrease the size, or maybe re-partition your device.

The whole error message might look something like

[...]
Uploading BOOT
100%
ERROR: Failed to confirm end of file transfer sequence!
ERROR: BOOT upload failed!

Ending session...
ERROR: Failed to send end session packet!
Releasing device interface...

pmbootstrap and build related errors

Alpine, chroots and pmbootstrap

pmbootstrap is broken! But it worked yesterday!

Sorry! Revert to the last commit, that was known to work (git checkout $commit) and report an issue, if there is not already one. Consult with others in the IRC/matrix channel to find out, which commit was the last one that worked.

ERROR: This device does not exist anymore, check <https://postmarketos.org/renamed> to see if it was renamed

One case where this might happen is if you remove pmbootstrap's work path, and then run a pmbootstrap init. As you might notice it is still remembering values and devices from previous executions.

To solve this you need to remove (or rename) $XDG_CONFIG/pmbootstrap.cfg (i.e ~/.config/pmbootstrap.cfg). After doing so pmbootstrap init should work fine.

BusyBox does something different than the standard tool for the job, and this breaks building my package!

Example: unzip in BusyBox doesn't extract symlinks like the normal unzip implementation (#82).

Solution: Simply add unzip to the makedepends in the APKBUILD, and the "real" unzip will be used (or in that case, try to get the sources in another archive format, such as .tar.bz2, .tar.xz, .tar.gz).

WARNING: No provider for dependency...

Normal behavior. it means, that your "user" repository (the repository with all custom compiled packaged from pmOS, such as your device package and kernel) does not have the dependencies from the official Alpine Linux repositories (such as cryptsetup, iw, htop, ...) when indexing it. But these dependencies are in the official repositories, so they will be found when you try to install something. Usually the package indexing is done on all repositories at once in Alpine (then you wouldn't get the error), but pmOS does it only on the "user" repository.

ERROR: busybox-1.27.0-r1.post-install: script exited with error 127

Programs running in (foreign arch) chroots cannot be executed. See #1972.

ERROR: /mnt/pmbootstrap-packages/x86_64/qemu-user-static-repack-2.8-r6.apk: UNTRUSTED signature

You don't have your own package signing key imported, which gets generated when you compile a package for the first time. pmbootstrap should automatically import it, but when you hit ^C during that, it may go wrong.

Your generated key is in ~/.local/var/pmbootstrap/config_abuild, all imported keys are in ~/.local/var/pmbootstrap/config_apk_keys.

A quick fix is deleting all packages:

./pmbootstrap zap -p

See issue #270 for more information and don't hesitate to reopen if this didn't work for you.

Installed version newer than the version in the repositories

You have a program installed in one of your chroots with a higher version, than what is available in the binary package repositories (either from upstream or the packages you have built yourself).

Example how this can happen:

  • Git branch switch to one which increased the version of postmarketos-mkinitfs
  • Build postmarketos-mkinitfs
  • Install it in a chroot (e.g. by running pmbootstrap install)
  • Branch switched to master
  • All packages got zapped, but not the chroots
  • Build postmarketos-mkinitfs again (this time it has a lower version count!)

Solution: Simply zap your chroot! When it gets set up again, it will use the current package version.

It's not installing my changed package

You have changed a package but it's using the one from the repository and not your changed build.

You probably didn't build it or not for the correct architecture. pmbootstrap build defaults to the host architecture (usually x86_64), if you need that package on your device, use the --arch flag.

Newer version in binary package repositories than in aports folder

You have built a package earlier, and the branch you are currently working in has a lower version in the aports folder.

Example how this can happen:

  • Git branch switch to a feature branch which increased the version of postmarketos-mkinitfs
  • Build postmarketos-mkinitfs
  • Switch to master branch
  • Run just about any pmbootstrap command, it will print this warning:
 Package 'postmarketos-mkinitfs' in your aports folder has version X, but the binary package repositories already have version Y

Solution: Run pmbootstrap zap -m to delete the mismatched binary packages.

Note: If that does not help it, you can manually delete that specific packages from your aports folder: sudo rm ~/.local/var/pmbootstrap/packages/*/postmarketos-mkinitfs*.apk and run pmbootstrap index to re-index all packages. See also: #466, !474


If you want to use older package built locally from pmaports:

  1. Build the package using pmbootstrap build <pkg> --force to ignore newer version present in binary repos
  2. Enter chroot, e.g. device's rootfs: pmbootstrap chroot -r
  3. Install package in specific version, e.g. apk add linux-postmarketos-qcom-sdm845==6.2.0-r3

The warning (package <pkg>: aport version <v1> is lower than <v2> from the binary repository. <v2> will be used when installing <pkg>.) will still appear during e.g. pmbootstrap install, but your package will stay at specified version until chroot is wiped.

"1 error; XXXX MiB in XXX packages" but no visible error, "Command failed (exit code 1): apk add ..."

Sometimes if package installation fails on a previous command, the error carries over to all subsequent apk operations - that's the "1 error" mentioned in the line. Check the log further back to see if you can't find the exact error there.

To fix this, try running pmbootstrap chroot apk fix (you may need to pass e.g. the --rootfs flag or the --arch=... flag depending on which chroot the error happened in). This will attempt to fix the encountered errors, and if it fails, it will print the error message again.

Alternatively, you could try zapping the relevant chroot with pmbootstrap zap and re-doing your command.

mkbootimg: error: argument --dt: can't open '/boot/dt.img'

This happens when you have deviceinfo_bootimg_qcdt enabled (which is correct when pmbootstrap generated it that way), but your kernel does not generate the /boot/dt.img file. Please adjust your kernel APKBUILD as follows:

  • add dtbtool to makedepends
  • extend your build() function like this
  • and your package() function like that

See #688 for a plan to automatically generate the linux APKBUILD right.

Permission denied: .../pmbootstrap/work/chroot_native/proc/1/cwd'

It looks like you have created your work folder as: pmbootstrap/work, where pmbootstrap is the folder with the checked out source code. That is not supported. Please run pmbootstrap shutdown, then delete that folder, then run pmbootstrap init again and choose another work folder (e.g. ~/.local/var/pmbootstrap).

After that it should work - if this still does not resolve it, please open up an issue in the GitLab issue tracker.

can't create /dev/null: Permission denied

Make sure that your working directory is mounted correctly. This can happen when working over NFS, or when you mounted the drive your working directory is on trough the file explorer (if that's the case, umount it, and mount it manually). This can probably happen on every drive mounted with nosuid or similar.

cp: write error: No space left on device (during file transfer from chrooted environment to target image)

This issue is triggered when pmbootstrap determines the wrong size for the target image, as it is unfortunately very hard to calculate the correct size independent of the filesystem on which the chroot resides. Some examples for this: #1916; #928 where more elaborate information can be found.

While fixing the problem seems impossible for every configuration, there is a workaround available (#1904 and its associated merge request !1989): Use the -E/--extra-space option of pmbootstrap to specify extra space on the resulting image file in MB. Just play around with the value until it works for you. You can specify the value also through pmbootstrap init in its Additional options.

"ERROR: Is $builddir set correctly?" for linux-* package

Check where the kernel sources archive has the kernel-related code stored and change the builddir value accordingly.

Cannot download packages from mirrors: "temporary error (try again later)"

Icon TODO: we have this in the rewrite but there's some stuff here that hasn't been carried over

Another symptom of this error is the message: "alpine-base (no such package)."

This is because of a network error - APK cannot connect to any of the package mirrors. pmbootstrap will copy your /etc/resolv.conf into the chroot. If your resolv.conf is empty or misconfigured, APK will not be able to connect to a DNS server. See: resolv.conf(5)

Build errors

Errors with grsec-based kernel on the host system

See: Troubleshooting:grsec

Errors with ecryptfs

When your "work" folder is inside an ecryptfs mounted folder, /dev/random may not work inside the chroot and you may have other issues. Try to move your work folder to another location, you can use /tmp to test if this resolves the issue. You can change the location in pmbootstrap init. See also: #61. (This doesn't seem to be an issue with newer versions of ecryptfs anymore.)

musl-*/busybox-static-* fails to build with a 404 error

Run pmbootstrap aportgen musl-armhf musl-aarch64 musl-x86_64. The reason why this fails from time to time is, that we use Alpine's binary musl lib c package for armhf/aarch64 in the native chroot, to get the cross-compiler working. So the download URL inside the aports/cross/musl-*/APKBUILD file must be in sync with the upstream version. This pmbootstrap command safely gets the new version, uses apk to verify the signature of the package, generates the hash, and updates the APKBUILD file. To make sure, that pmbootstrap aportgen receives the latest musl version, delete your HTTP cache right before by executing this command: pmbootstrap zap -hc. Don't bother making a pull-request for this, as the fix is really trivial (see this commit). Just report it as issue, if it happens.

du: unrecognized option: summarize

Update pmbootstrap.

Packaging for aarch64: ./config.guess: unable to guess system type

This happens in packages built with autoconf (./configure; make; make install), when it ships an outdated config.guess file in the source code. Adjust your APKBUILD's prepare() function like done in !698.

ERROR: ... BAD archive

If you pmbootstrap log shows something similar to the following:

(005853) [22:25:47] (native) % apk --no-progress add -u qemu-user-static-repack qemu-user-static-repack-binfmt
WARNING: Ignoring /mnt/pmbootstrap-packages/x86_64/APKINDEX.tar.gz: No such file or directory
(1/2) Installing qemu-user-static-repack (2.8-r9)
ERROR: qemu-user-static-repack-2.8-r9: BAD archive
(2/2) Installing qemu-user-static-repack-binfmt (2.8-r9)
1 error; 20 MiB in 36 packages
(005853) [22:25:53] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

...then downloading a package (in this case qemu-user-static-repack) failed. Please delete the broken package and try it again:

$ cd ~/.local/var/pmbootstrap/ # go into your work folder, as specified in 'pmbootstrap init'
$ sudo rm -v cache_apk_*/qemu-user-static-repack-* # adjust the failing package name accordingly

CROSS_COMPILE_ARM32 not defined or empty, the compat vDSO will not be built

Disable CONFIG_COMPAT_VDSO in kernel config.

boot-deploy exits with ERROR: "deviceinfo_header_version is 2, but 'deviceinfo_dtb' is missing. Set 'deviceinfo_dtb'" when compiling downstream kernel

Deviceinfo file is generated with deviceinfo_header_version set to 2 and boot-deploy expects a dtb file to be to be specified in deviceinfo. Verify that the deviceinfo_dtb entry in your deviceinfo is not empty or points to a non-existent dtb/dts in your kernel source.

Special section for stuff that needs to be moved to Troubleshooting:kernel

Icon TODO: Big TODO: move it all out!!!

../tools/dct/DrvGen: No such file or directory

This error is specific to Mediatek kernels and it occures when DrvGen tries to load libc6.so library it has been linked against but Alpine chroot doesn't support libc. One of the solutions is to copy python3dct.patch to your linux-... directory and activate it in the APKBUILD. This patch replaces precompiled DrvGen with a python3 version.