exynos4 kernel is too big for p4note

Presumably following !3560 (merged), the post-upgrade script for postmarketos-mkinitfs fails because dd: error writing '/dev/mmcblk2p5': No space left on device. This affects p4note, and likely also other Exynos4 devices such as samsung-m0. The kernel currently present on v22.06 is fine.

added category::device device-category::community release-edge labels

changed the description

By Newbyte on 2022-11-14T09:28:35

changed milestone to %v22.12 Release

Yup facing the same issue on samsung-espresso3g, the support of more features in kernel makes the size too big for our devices (with 8mb limit). In the future this will no doubtably increase, so why not use a minimal config for devices with small boot partitions?

By Mighty on 2022-11-15T15:05:52

Some ideas:

• Can we reduce the initramfs size, either by removing things we don't need or by compressing it better?
• Flash a bootloader in the partition that loads the kernel + initramfs from another partition?
• As proposed by @sicelo, add a mechanism to disable certain kernel options for some devices. Not my favorite solution, I'd rather have them consistent across all devices in main and community. But if we don't have a choice... maybe the easiest version of that would be changing the CI test so it doesn't fail anymore for some devices.

By Oliver Smith on 2022-11-17T09:04:45

Flash a bootloader in the partition that loads the kernel + initramfs from another partition?

It's something I'd like to do, but the U-Boot port for samsung-m0 only works if you use it to replace S-Boot. If you put it on the kernel partition, it doesn't work. I'm not willing to require users to do something that risky.

By Newbyte on 2022-11-17T09:04:45

Can we reduce the initramfs size, either by removing things we don't need or by compressing it better?

Most of the size of the modules dir, at least for the amd64 initramfs built for the qemu image, is in these modules.* files. How do these work, are they all required at boot? Why is there a modules.dep that's an ascii text file, and a modules.dep.bin that's a binary file (and thus probably not likely to compress very well...) ?

/tmp/initramfs/lib/modules/5.15.79-0-lts # ls -lah
total 4.7M
drwxr-xr-x 3 root root 4.0K Nov 18 21:33 .
drwxr-xr-x 3 root root 4.0K Nov 18 21:33 ..
drwxr-xr-x 7 root root 4.0K Nov 18 21:33 kernel
-rw-r--r-- 1 root root 1.1M Nov 18 21:33 modules.alias
-rw-r--r-- 1 root root 1.1M Nov 18 21:33 modules.alias.bin
-rw-r--r-- 1 root root 4.7K Nov 18 21:33 modules.builtin
-rw-r--r-- 1 root root 3.7K Nov 18 21:33 modules.builtin.alias.bin
-rw-r--r-- 1 root root 7.4K Nov 18 21:33 modules.builtin.bin
-rw-r--r-- 1 root root  37K Nov 18 21:33 modules.builtin.modinfo
-rw-r--r-- 1 root root 492K Nov 18 21:33 modules.dep
-rw-r--r-- 1 root root 642K Nov 18 21:33 modules.dep.bin
-rw-r--r-- 1 root root  458 Nov 18 21:33 modules.devname
-rw-r--r-- 1 root root 134K Nov 18 21:33 modules.order
-rw-r--r-- 1 root root 1.4K Nov 18 21:33 modules.softdep
-rw-r--r-- 1 root root 555K Nov 18 21:33 modules.symbols
-rw-r--r-- 1 root root 671K Nov 18 21:33 modules.symbols.bin

I'm not sure if these exist in other initramfs images created for other devices though... so maybe this is a red herring for helping this particular situation.

By clayton craft on 2022-11-19T05:40:47

ok seems like some of them are used by busybox modprobe, at least that's what I managed to infer after staring at this patch in aports: https://gitlab.alpinelinux.org/alpine/aports/-/blob/master/main/busybox/0009-depmod-support-generating-kmod-binary-index-files.patch

maybe it means we can then drop the text versions of the files? (e.g. drop modules.dep but keep modules.dep.bin?)

anyone have any idea about that?

By clayton craft on 2022-11-19T05:55:23

Brought this up in the team meeting, some notes:

• @craftyguy has WIP patches to use different compression methods for the initramfs, this could probably be used soon -> could use zstd?
• @z3ntu figured that we don't compress the kernel modules for all kernels, we might be able to save a good amount of space there (still gets compressed with gzip with the rest of the initramfs currently)

@MightyM17: do you want to give compressing the kernel modules a try?

By Oliver Smith on 2022-11-18T14:07:24

That would reduce the kernel size? I cant understand how. Also @dartvader316 has managed to reduce the kernel size so there should be MR coming soon

By Mighty on 2022-11-18T14:07:24

no, it wouldn't directly reduce the size of the kernel, but it would possibly free up some space in the boot partition that could be occupied by the kernel.

By clayton craft on 2022-11-19T11:17:35

mentioned in merge request !3653 (merged)

By Mighty on 2022-11-20T06:37:26

mentioned in issue boot-deploy#10

By Antoni Aloy Torrens on 2022-11-21T14:39:07

mentioned in commit 421bf41e

By Newbyte on 2022-12-01T21:38:56

mentioned in merge request !3656 (merged)

By Oliver Smith on 2022-12-02T08:01:02

mentioned in commit 1611fb50

By Newbyte on 2022-12-02T09:47:18

I have the same problem on my n8000 p4note. #1812 (closed)

By DvaMishkiLapa on 2022-12-04T12:15:50

Thanks for checking in. This issue will be fixed by !3656 (merged) once it is merged.

By Newbyte on 2022-12-04T12:15:50

mentioned in issue #1812 (closed)

By Newbyte on 2022-12-04T12:15:08

added status::confirmed label

mentioned in commit ab590f21

By Newbyte on 2022-12-04T17:15:08

closed with commit ab590f21

closed

mentioned in commit e9b3f075

By Newbyte on 2022-12-04T17:17:20