Re: "FATAL: kernel too old" when running mmdebootstrap'd riscv64 chroot
On 23 Oct 2018, at 17:13, Luke Kenneth Casson Leighton <lkcl@lkcl.net> wrote:
>
> On Tue, Oct 23, 2018 at 12:00 PM James Clarke <jrtc27@debian.org> wrote:
>>
>> On 23 Oct 2018, at 11:56, John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de> wrote:
>>>
>>> On 10/23/18 12:31 PM, James Clarke wrote:
>>>> Unless I'm missing something, this argument is false? Sure you need a 4.15
>>>> kernel if you want to run riscv64 natively, but this is using qemu-user on
>>>> amd64; all the system calls will be translated to the amd64 versions that have
>>>> existed for years. If not, qemu-user itself would be imposing a minimum kernel
>>>> version.
>>>
>>> Yes, I wanted to outline that as well.
>>>
>>>> NB: This is coming from glibc as it sets arch_minimum_kernel=4.15.0 in
>>>> sysdeps/unix/sysv/linux/riscv/configure.ac.
>>>
>>> I usually bootstrap with:
>>>
>>> $ debootstrap --foreign --no-check-gpg --variant=buildd --include=debian-ports-archive-keyring --arch=riscv64 unstable sid-riscv64-sbuild http://ftp.ports.debian.org/debian-ports
>>>
>>> Then build a fresh qemu-user from git, copy it into the chroot to usr/bin/qemu-riscv64-static,
>>> then change into the chroot and run ./debbootstrap/debbootstrap --second-stage.
>>>
>>> Never had any issues.
>
> yep i'm getting the exact same "kernel too old" with this as well:
> # strace -o log.txt -ff /usr/sbin/chroot sid-riscv64-sbuild
> ./debootstrap/debootstrap --second-stage
Does the chroot have the updated qemu? Is it called the right thing?
> question. why, if i am using the latest libc6 (2.27-6) from sid, and
> it works perfectly with linux 4.13, is it specifically the case that
> the riscv configuration sets a minimum of 4.15.0? it does not make
> any sense.
Because it can't possibly work on a pre-4.15 riscv64 kernel, and it's better to
detect the old kernel rather than failing at some later point in an
unintelligible way because a system call was missing.
> https://packages.debian.org/sid/libc6
>
> i mean, if i installed an updated version of libc6 with an "apt-get
> install libc6" which had the same hard mandatory requirement, on the
> next boot i'd be absolutely f****d!
Yes, and in fact, on most architectures, 3.8 is set as the minimum, so if
you're running an old kernel you can have precisely this issue. I believe the
libc6 maintainer scripts detect this and bail out so as not to render your
system broken. This minimum will probably be bumped at some point in the future
when glibc wants to use a newer system call, but is generally pretty
conservative for precisely this reason.
James
> the correlation "4.15 was the first riscv64 kernel therefore it shall
> be set as the absolute minimum requirement" does not logically follow.
>
> l.
Reply to: