Re: [PATCH v3 1/2] Provide in-kernel headers for making it easy to extend the kernel

[Qais Yousef]

On 02/28/19 22:26, Joel Fernandes wrote:
> On Fri, Mar 01, 2019 at 11:28:26AM +0900, Masami Hiramatsu wrote:
> > Hi Joel,
> 
> Hi Masami,
> 
> > On Thu, 28 Feb 2019 10:00:54 -0500
> > Joel Fernandes <joel@xxxxxxxxxxxxxxxxx> wrote:
> > 
> > > > Hmm, isn't it easier to add kernel-headers package on Android?
> > > 
> > > I have already been down that road. In the Android ecosystem, the Android
> > > teams only provide a "userspace system image" which goes on the system
> > > partition of the flash (and a couple other images are also provided but
> > > system is the main one). The system image cannot contain GPL source code. It
> > > is also not possible to put kernel headers for every kernel version on the
> > > system images that ship and is not practical. Android boots on 1000s of forked
> > > kernels. It does not make sense to provide headers on the system image for
> > > every kernel version and I already had many discussions on the subject with
> > > the teams, it is something that is just not done. Now for kernel modules,
> > > there's another image called the "vendor image" which is flashed onto the
> > > vendor parition, this is where kernel modules go.  This vendor image is not
> > > provided by Google for non-Pixel devices. So we have no control over what
> > > goes there BUT we do know that kernel modules that are enabled will go there,
> > > and we do have control over enforcing that certain kernel modules should be
> > > built and available as they are mandatory for Android to function properly.
> > > We would also possibly make it a built-in option as well. Anyway my point is
> > > keeping it in the kernel is really the easiest and the smartest choice IMO.
> > 
> > Sorry, I'm not convinced yet. This sounds like "because Android decided not
> > to put the header files on vendor partition, but kernel module is OK"
> > Why don't google ask vendors to put their kernel headers (or header tarball)
> > on vendor partition instead?
> 
> May be Google can do that, but I think you missed the point of the patches.
> Making it a module is not mandatory, people can build it into the kernel as
> well (CONFIG_IKHEADERS_PROC=y). In this case, the proc entry will be
> available on boot without any dependency on the filesystem. If you go through
> the other threads such as folks from ARM who replied, they just boot a kernel
> image for debug purpose and want headers on device available without any
> additional step of copying headers to the filesystem. And folks from Google
> also said that they wanted a built-in option.

Yes I do see this patch a useful addition. When you need the header and you're
dealing with multiple devices and kernel trees and versions - having the
headers part of the kernel just makes it easier to use them when you need them.
Especially sometimes when you change a config option (like enabling a new
syscall e.g: bpf) it's very easy to forget that the headers has changed as well
and you need to push an updated copy.

FWIW most of the time I run on non-android systems for development/debugging
purposes. I use the built-in version although I can see a module version
helpful too. You can save some space if your kernel image partition is small
and it's easy to install the module along with all other modules when
rebuilding the kernel than remembering to update the headers. It's a (very)
nice convenience.

--
Qais Yousef

> 
> There are many usecases for this, I have often run into issues with Linux
> over the years not only with Android, but other distros, where I boot custom
> kernels with no linux-headers package. This is quite painful. It is
> convenient to have it as /proc file since the file is dependent on kernel
> being booted up and this will work across all Linux distros and systems. I
> feel that if you can keep an open mind about it, you will see that a lot of
> people will use this feature if it is accepted and there is a lot of positive
> feedback in earlier posts of this set.
> 
> > > > > The code to read the headers is based on /proc/config.gz code and uses
> > > > > the same technique to embed the headers.
> > > > > 
> > > > > To build a module, the below steps have been tested on an x86 machine:
> > > > > modprobe kheaders
> > > > > rm -rf $HOME/headers
> > > > > mkdir -p $HOME/headers
> > > > > tar -xvf /proc/kheaders.tar.xz -C $HOME/headers >/dev/null
> > > > > cd my-kernel-module
> > > > > make -C $HOME/headers M=$(pwd) modules
> > > > > rmmod kheaders
> > > > 
> > > > It seems a bit complex, but no difference from compared with carrying
> > > > kheaders.tar.gz. I think we would better have a psudo filesystem
> > > > which can mount this compressed header file directly :) Then it becomes
> > > > simpler, like
> > > > 
> > > > modprobe headerfs
> > > > mkdir $HOME/headers
> > > > mount -t headerfs $HOME/headers
> > > > 
> > > > And this doesn't consume any disk-space.
> > > 
> > > I felt using a compressed tar is really the easiest way because of all the
> > > tools are already available.
> > 
> > As I asked above, if the pure tarball is useful, you can simply ask vendors
> > to put the header tarball on their vendor directory. I feel making it as
> > a module is not a right way.
> 
> I don't see what is the drawback of making it a module, it makes it well
> integrated into kernel build and ecosystem. I also didn't see any
> justification you're providing about why it cannot be a module. If you go
> through this and earlier threads, a lot of people are Ok with having a module
> option. And I asked several top kernel maintainers at LPC and many people
> suggested having it as a module.
> 
> > > There isn't a compressed in-ram filesystem right
> > > now that I'm aware off that can achieve the kind of high compression ratio
> > > this patchset does.
> > 
> > I think if linux can support something like tarfs(or compressed initramfs)
> > in kernel, it gives linux an improvement not only a hack. :-)
> 
> Agreed, that sounds like a good idea. I will consider doing it once the
> series in its current form can be accepted. I am saying so since this series
> is simple, and I can do that as a next step since that idea will take a lot
> of time to implement. But I am keen on doing it.
> 
> thanks,
> 
>  - Joel
>