kernel compiling in woof-ce

[musher0]

Hello all.

I agree with pelo on this one.

We developers can do research about a new kernel, but middle-of-the-road
users will appreciate finding a brief explanation of the features of any new
kernel and why it is better than the one published before it.

BFN.

[0xdawg]

.... a brief explanation ....

there are brief explanations all over the web.

feel free to educate yourself for in-depth informations

[stemsee]

I ususally look here at the changelog

https://cdn.kernel.org/pub/linux/kernel ... og-4.10.11

http://news.softpedia.com/news/linux-ke ... 4952.shtml

[musher0]

That's quite technical.
Isn't there an explanation in "layman's terms" somewhere?

[Sailor Enceladus]

That's quite technical.
Isn't there an explanation in "layman's terms" somewhere?

The problem is that everything in the kernel is technical (and greek to the layman) I think.

[stemsee]

So I have recently re-arranged ubuild.sh and forked it to nubuild.sh

The script is much shorter with only one build function which builds x86_64 32-pae and 32-nopae kernels and modules, switching each build arch according to arc variable.

The directory structure is different now with sources and vanilla and patches all in the parent directory. The only remaining sub directories are in dist where the packages are finalised. This makes it easier to navigate and view, and also for locating files etc.

There are no more build.conf files! Also only 3 DOTconfig files for x64 pae nopae, which can be reconfigure internally with 'nubuild.sh' '888' option.

I have not yet added recent improvements from woof-ce>kernel-kit , because sukk-2 seems fine without them for the moment.

Still testing.

EDIT: built 4.10.12 and 4.10.13 successfully.

[stemsee]

Oops! Nubuild couldn't build a 3.x kernel. Now it can.

[stemsee]

I was just re-configureing the DOTconfig for x64 and I noticed the option to specify path to files or dirs for the kernel compile to build (in?) the cpio (initrd) with the kernel build.

Code: Select all

CONFIG_INITRAMFS_SOURCE="/initrd/mnt/dev_save/FatHouse/initrd-tree"

[stemsee]

I just built a x64 4.10.13 kernel with initrd built in @ 40mb. Works fine.

In my initrd tree there were modules from previous 4.10.13 build which had not been configured fully for my hardware.

I also booted same kernel with external initrd with modules.sfs inside and everything worked fine too.

So it is possible to build the kernel with kernel modules inside the initrd, inside the kernel. As it is possible to specify several directories and/or files for the cpio archive at kernel compile time, it seems possible to also build in the main sfs, probably just need to code initrd init to look for it.

So now we are realistically and easily able to compile a puppy linux distribution as one vmlinuz file.

grub cfg/menu/lst

Code: Select all

kernel vmlinuz

[stemsee]

Re-compiled with full compliment of modules and kernel-modules.sfs put in initrd tree; removed initrd from grub.cfg boot entry. Installed new kernel (56MB) and re-booted flawlessly to desktop.
The kernel booted more verbosely.

eth0 is now sit0.


I did tried compiling the main sfs in, but @ 2GB compilation stopped at cpio phase and failed.
should work with a smaller main sfs.

[LateAdopter]

The kernel booted more verbosely.

With kernel 4.10 the loglevel is set in DOTconfig and it defaults to 7 when you first save a 4.10 DOTconfig. Previously the kernel kit patched it to 3. This may be the cause of the extra rhubarb when booting.

[stemsee]

Ahh! Could be

Thanks LatAdopter!

[stemsee]

After some research seems cpio archive is limited to 2GB size, in general.

re-compiled 4.7.1 with fatdog initrd tree, including modules and main sfs, builds successfully but at boot kernel panic!
Any ideas?

[Dry Falls]

After some research seems cpio archive is limited to 2GB size, in general.

re-compiled 4.7.1 with fatdog initrd tree, including modules and main sfs, builds successfully but at boot kernel panic!
Any ideas?

Not many comers to this question. Aside from the verbosity issue, the 64k dotconfig in the new suuk does not support xz compression. Aside from that, I've been playing with a humungous initrd in JL64 and it requires

Code: Select all

rootfstype=ramfs

as in fatdog or

Code: Select all

rootwait rw

as in quirky. A number of smallish sfs files can be in the initrd, but large ones may present a problem. In lighthouse, even if the base file is inside the initrd, a pre-existing base anywhere in the system (top of partition) will be mounted, so pmedia and pdev1 boot parameters are also needed.

Without the rootfs parameter, memory allocation error and kernel panic everytime. >400M initrd and I get "init too big" error or "no working init". Go figure!

df

update: Further testing with humongous initrd in JL64 -- no need for above boot parameters. rootfs in pupstate is automatic (initrd=377M) so I assume that is taken care of by Tazoc's init script. adding only 10M to that gives a "no suitable memory target" error. I think this is a new kernel issue. I notice the fatdog initrd is 344M.

[stemsee]

Kernel-4.12.12 x64 kernel and modules

https://drive.google.com/file/d/0B1gMz4 ... FVfRlhSbmc

[stemsee]

Today I configured and compiled 4.13.3 x64 for Atom cherry trail cpu. It boots but xserver refuses connections. I searched the net for a solution but nothing worked. Does anyone know what is the problem. How to change group at cli?

The config is attached with fake gz extension.

[stemsee]

recomplied 4.13.3 using fd 4.7.1 config with all intel and atom cherry trail options enabled.
running fathouse on x8300 atom 2-in-1 detachable touch tablet.

multitouch screen working.
bt working
wifi working
touchpad and buttons one finger
brightness working
cpu temp works
hdmi sound working

soundcards found speakers not yet working....using stereo bt headset
cam not found
sd-card reader not working
vol buttons not working

https://drive.google.com/file/d/0B1gMz4 ... RzT2hCcTA/

[stemsee]

SUKK-2_12-2017

Some really silly bugs fixed!

Stems Uattended Kernel Kit
v-12-2017

Requires minimal setting up to build and package
kernel and modules and sources.

You must load devx.sfs on puppy or have installed
development libraries such as git gcc make
for your distro.

This kit can also build official .deb packages if your OS
allows the installation of the necessary packages.
fakeroot build-essential kernel-package libssl-dev git
ncurses-dev bc linux-headers-amd64 or linux-headers-x86

Usage is simple
in term type
./nubuild.sh

For very first use I recommend using build option
'888'
and config option
'2) make menuconfig'

Then configure kernel features for your system
This will ensure your configs are up to date
and appropriately configured. Presently the configs
will be for a specfic hardware I was testing.

Follow the prompts.

Packages will be in 'dist/packages64' etc

Use at your own risk, no warranty or liability accepted.

[musher0]

Many thanks, stemsee.

Happy Holidays!

[stemsee]

De rien!