htc-flounder: Kworker hangs and makes the entire kernel panic

Created by: vartom

I see this stock kernel with all its small flaws. Delays are caused by regular attempts to find the firmware on the Wifi module and sar0_CY8C.img. I assume that all other files of the firmware are not available to the kernel either.

By * postmarketOS Bot on 2018-04-02T09:54:00

The weird thing is that I've built the kernel with a correct wifi firmware path already, the Wifi bits itself should work. No idea about the sensor firmware (plus there was something with a synaptics firmware image updater like the sar0_**img thing initially, which I disabled for now.), where to change the config to point towards the right path. By default, firmware is loaded from /vendor/firmware on Android, but I followed standard procedure and made a firmware package that installs to /lib/firmware/postmarketos and told the kernel to at least look for wifi firmware there. As I said, no idea if and where I can change the settings for the rest.

By lillywho on 2018-04-02T10:58:45

Created by: vartom

As far as I know, the kernel searches for files where they are located for the android. This is clearly spelled out. In other ways, guarantees that the firmware will not be found. At a minimum, you must completely preserve the directory structure of / vendor / firmware / files or / vendor / firmware / tegra13x / files. The conclusion of this is you can mount the partition vendor you can try to completely preserve the structure of folders of firmware. try removing postmarketos from the destination directory

By * postmarketOS Bot on 2018-04-02T13:25:02

Created by: Decatf

It's using the request_firmware API which means the firmware_class.path cmdline arg can be used to add a directory the firmware search paths.

Probably something like firmware_class.path=/lib/firmware/postmarketos

By * postmarketOS Bot on 2018-04-02T16:05:28

The firmware folder structure is preserved, that's been my aim all this time. If the firmware.class_path arg works, there'll be no need for moving one level up from /lib/firmware/postmarketos.

@vartom I already tested mounting /vendor via initfs hook and it made no difference. In fact, the kernel even hangs and panics when sitting on the initfs debug shell.

@ollieparanoid If I add the argument to the cmdline, would it work or would it fail because at this stage of booting there's only the contents of initramfs available?

By lillywho on 2018-04-03T00:32:40

Created by: ollieparanoid

It depends on what the kernel does - does it crash if the firmware files are not in the initramfs? The firmware files do not get copied to the initramfs by default. You could add a hook that either copies them, or mounts Android's firmware partition, but I think both aren't nice solutions.

Is there possibly another kernel that you could try which has less issues (@mirh: any recommendations?)

PS: Consider attaching large logs as text file instead (you can upload them to GitHub as well).

By * postmarketOS Bot on 2018-04-03T00:45:13

@ollieparanoid /vendor is the firmware partition for Nexus devices and I've tried that, to no avail. My guess is that the firmware should be available as early into booting as possible, since all the drivers are requesting the firmware and causing these hangs if it isn't found.

By lillywho on 2018-04-03T01:54:28

Created by: vartom

@Alexandroid99 Adding a firmware to initramfs is just as good a solution. The main thing is to keep the entire directory structure (/ vendor / firmware / files , / vendor / firmware / tegra13x / files).

I already tested mounting /vendor via initfs hook and it made no difference.

I guess you made a mistake somewhere doing this. In ubuntu it worked, as well as copying the firmware in / lib / firmware .

By * postmarketOS Bot on 2018-04-03T01:59:16

Hm, I had this in the hook, just like the Ubuntu port on XDA:

 #!/bin/sh

mkdir /vendor
mount -t ext4 -o ro /dev/mmcblk0p24 /vendor

This is run by the initramfs just like in the Ubuntu kernel, so I'm not getting why it's not working. Or I screwed up the rebuild and the changes just weren't implemented. I'll find out in a bit when I can reboot into my Ubuntu again.

By lillywho on 2018-04-03T02:10:57

@ollieparanoid Did I miss anything or is there no way to add files to the initfs permanently? Would be handy if the files existed in the initfs from the start. For now, I'm going to try the cmdline argument for the firmware path.

By lillywho on 2018-04-03T02:58:02

Could it also be possible that I need to enable deviceinfo_bootimg_qcdt="true" for flounder? The is option is about appended device trees and in the kernel config there's mention of appending multiple dtbs to the kernel:

CONFIG_BUILD_ARM64_APPENDED_DTB_IMAGE=y
CONFIG_BUILD_ARM64_APPENDED_DTB_IMAGE_NAMES="tegra132-flounder-xaxb tegra132-flounder-xc tegra132-flounder-xdxepvt tegra132-flounder_lte-xaxbxcxdpvt"

bootimg-analyze didn't suggest that option, so could it be that qcdt only applies to arm32 or something?

By lillywho on 2018-04-03T04:41:58

Created by: mirh

The Lineage tree he's using should already be the most updated one.

But is it normal to have tegra124 messages, even though N9 is tegra132?

By * postmarketOS Bot on 2018-04-03T10:39:28

I think I've read somewhere that the two are similar and/or share drivers here and there. Can't remember.

By lillywho on 2018-04-03T12:42:13

Created by: mirh

One has a Cortex-A15 32 bit CPU, while the other is nvdia's own 64 bit architecture. It's kinda of basic to know, tbf.

Aside of that they are kinda the same. Some topics for you, if you were wondering. https://archlinuxarm.org/forum/viewtopic.php?f=49&t=9688 https://github.com/linux-shield

By * postmarketOS Bot on 2018-04-03T20:46:43

Created by: ollieparanoid

is there no way to add files to the initfs permanently?

This is possible now, see aports/main/postmarketos-mkinitfs-hook-debug-shell/20-debug-shell.files. So far this has only been tested with adding binaries though (with resolving the libraries, and adding them as well), so it might not work for regular files. That would be a simple change though.

Could it also be possible that I need to enable deviceinfo_bootimg_qcdt="true" for flounder?

If it was missing, then it would not boot. See QCDT.

@Alexandroid99: I think the easiest workaround is trying to get the initramfs hook working that mounts /vendor, as you've mentioned in https://github.com/postmarketOS/pmbootstrap/issues/1384#issuecomment-378105299. After installing the hook, you can double-check if it was added properly with pmbootstrap initfs ls. The -s parameter is handy to skip rebuilds (they get triggered pretty often right now, we could improve that). Finally, you could use the debug-shell hook to debug whether your hook is working or not. Simply execute the commands manually via telnet, and check the output and what dmesg says.

If all that does not help, maybe your problem is not the missing firmware, but a watchdog?

By * postmarketOS Bot on 2018-04-04T00:48:38

@ollieparanoid What do I do if the hook isn't copied? So far it appears that it really isn't being copied despite my usage of the standard approach present in other device configs:

install -Dm644 "$srcdir"/initfs-hook.sh "$pkgdir"/etc/postmarketos-mkinitfs/hooks/00-${pkgname}.sh

Also, how would I go about adding an entry to /etc/fstab ? There isn't a specific wiki page about that point, only some user comment that mentions a device config for reference that doesn't exist anymore https://wiki.postmarketos.org/wiki/User_talk:Babalooey

By lillywho on 2018-04-06T09:37:32

Alrighty, I bypassed the issue of the hook not being copied by modding the debug shell hook and pasting my commands into it. The kernel still freezes even though /vendor is present. Dammit.

By lillywho on 2018-04-06T10:25:07

Created by: ollieparanoid

@ollieparanoid What do I do if the hook isn't copied? So far it appears that it really isn't being copied despite my usage of the standard approach present in other device configs

Just to make sure:

Have your rebuilt your device package with force for the target arch (e.g. armhf)? (pmbootstrap build --force --arch=armhf device-...)
Have you made sure that the updated package was installated in the chroot? (e.g. by running pmbootstrap zap and pmbootstrap install again)

If you do it that way, it should get updated for sure.

Also, how would I go about adding an entry to /etc/fstab ?

Note that the /etc/fstab is only present after the init script inside the initramfs is through and the real root partition is mounted. So in case you want to mount something in the initramfs, I would recommend to hardcode mount /dev/path/to/device /path/to/mountpoint instead.

You can find a few examples for fstab changes with grep -r fstab in the aports folder, e.g. device-samsung-i9305/device-samsung-i9305.post-install:

#!/bin/sh

# Create mount point and /efs mount point in fstab
if [ -z "$(grep /efs /etc/fstab)" ]; then
    mkdir -p /efs
    echo "/dev/mmcblk0p1    /efs    auto    ro  0   0" >> /etc/fstab
fi

Create such a post-install file and add/change the install= line in your device APKBUILD to:

install="$pkgname.post-install"

Just to rule out that the /vendor partition was really mounted, make sure your custom hook starts with 10- (so it gets executed before the debug-shell hook, which starts with 20-) and install both of them. Then use the telnet shell to check if /vendor/ is mounted (and if it is not, check if you can mount it via telnet and adjust your hook).

By * postmarketOS Bot on 2018-04-06T19:36:51

moved to pmaports#53 (closed)

By Bart Ribbers on 2018-09-17T21:56:21

htc-flounder: Kworker hangs and makes the entire kernel panic

Child items 0

Activity