Re: [RFC] Provide in-kernel headers for making it easy to extend the kernel

[Daniel Colascione]

On Fri, Jan 25, 2019 at 11:01 AM <hpa@xxxxxxxxx> wrote:
>
> On January 24, 2019 12:59:29 PM PST, Joel Fernandes <joel@xxxxxxxxxxxxxxxxx> wrote:
> >On Thu, Jan 24, 2019 at 07:57:26PM +0100, Karim Yaghmour wrote:
> >>
> >> On 1/23/19 11:37 PM, Daniel Colascione wrote:
> >[..]
> >> > > Personally I advocated a more aggressive approach with Joel in
> >private:
> >> > > just put the darn headers straight into the kernel image, it's
> >the
> >> > > *only* artifact we're sure will follow the Android device
> >whatever
> >> > > happens to it (like built-in ftrace).
> >> >
> >> > I was thinking along similar lines. Ordinarily, we make loadable
> >> > kernel modules. What we kind of want here is a non-loadable kernel
> >> > module --- or a non-loadable section in the kernel image proper.
> >I'm
> >> > not familiar with early-stage kernel loader operation: I know it's
> >> > possible to crease discardable sections in the kernel image, but
> >can
> >> > we create sections that are never slurped into memory in the first
> >> > place? If not, maybe loading and immediately discarding the header
> >> > section is good enough.
> >>
> >> Interesting. Maybe just append it to the image but have it not loaded
> >and
> >> have a kernel parameter than enables a "/proc/kheaders" driver to
> >know where
> >> the fetch the appended headers from storage at runtime. There would
> >be no
> >> RAM loading whatsoever of the headers, just some sort of
> >> "kheaders=/dev/foobar:offset:size" parameter. If you turn the option
> >on, you
> >> get a fatter kernel image size to store on permanent storage, but no
> >impact
> >> on what's loaded at boot time.
> >
> >Embedding anything into the kernel image does impact boot time though
> >because
> >it increase the time spent by bootloader. A module OTOH would not have
> >such
> >overhead.
> >
> >Also a kernel can be booted in any number of ways other than mass
> >storage so
> >it is not a generic Linux-wide solution to have a kheaders= option like
> >that.
> >If the option is forgotten, then the running system can't use the
> >feature.
> >The other issue is it requires a kernel command line option /
> >bootloader
> >changes for that which adds more configuration burden, which not be
> >needed
> >with a module.
> >
> >> > Would such a thing really do better than LZMA? LZMA already has
> >very
> >> > clever techniques for eliminating long-range redundancies in
> >> > compressible text, including redundancies at the sub-byte level. I
> >can
> >> > certainly understand the benefit of stripping comments, since
> >removing
> >> > comments really does decrease the total amount of information the
> >> > compressor has to preserve, but I'm not sure how much the encoding
> >> > scheme you propose below would help, since it reminds me of the
> >> > encoding scheme that LZMA would discover automatically.
> >>
> >> I'm no compression algorithm expert. If you say LZMA would do the
> >> same/better than what I suggested then I have no reason to contest
> >that. My
> >> goal is to see the headers as part of the kernel image that's
> >distributed on
> >> devices so that they don't have to be chased around. I'm just trying
> >to make
> >> it as palatable as possible.
> >
> >I believe LZMA is really good at that sort of thing too.
> >
> >Also at 3.3MB of module size, I think we are really good size-wise. But
> >Dan
> >is helping look at possibly reducing further if he gets time. Many
> >modules in
> >my experience are much bigger. amdgpu.ko on my Linux machine is 6.1MB.
> >
> >I really think making it a module is the best way to make sure this is
> >bundled with the kernel on the widest number of Android and other Linux
> >systems, without incurring boot time overhead, or any other command
> >line
> >configuration burden.
> >
> >I spoke to so many people at LPC personally with other kernel
> >contributors,
> >and many folks told me one word - MODULE :D.  Even though I hesitated
> >at
> >first, now it seems the right solution.
> >
> >If no one seriously objects, I'll clean this up and post a v2 and with
> >the
> >RFC tag taken off. Thank you!
> >
> > - Joel
>
> So let me throw in a different notion.
>
> A kernel module really is nothing other than a kernel build system artifact stored in the filesystem.
>
> I really don't at any reason whatsoever why this is direct from just producing an archive and putting it in the module directory, except that the latter is far simpler.
>
> I see literally *no* problem, social or technical, you are solvin by actually making it a kernel ELF object.

Joel does have a point. Suppose we're on Android and we're testing a
random local kernel we've built. We can command the device to boot
into the bootloader, then send our locally-built kernel to the device
with "fastboot boot mykernelz". Having booted the device this way,
there's no on-disk artifact corresponding to mykernelz: we just sent
the boot kernel directly to the device's memory. Now, suppose I want
to attach DCTV or some other fancy ftrace-based analysis tool to the
device in order to study how mykernelz acts in some scenario I care
about. With Joel's approach, DCTV would be able to grab the kernel
headers from the running kernel, compile whatever kprobe or BPF
incantations needed, and have everything Just Work. If we put the
headers only on disk without any way to retrieve them at runtime, we'd
need a different path for kernel self-description in the case where a
running kernel doesn't have an on-disk representation, and that adds a
lot of complexity for everyone everywhere. By providing
guaranteed-correct kernel headers via some runtime interface, we make
a lot of things Just Work, and that has value.