Hi,

I recently attempted to install an out-of-tree module (an old version of
the octeon mmc driver) on an octeon machine running big endian 32-bit
mips. In this configuration, the kernel is 64-bit, but userspace is 32-bit.

The module build failed with this error:

of_device_id is defined like this in mod_devicetable.h:

The size of this structure is 200 bytes and not 196 byte as modpost
claims because the kernel is compiled as 64-bit (so the pointer at the
end is 8 bytes instead of 4).

I think there is a bug in the way Debian compiles modpost, because using
upstream's modpost works correctly. I see that the
real-XXX/devicetable-offsets.s file gets compiled with the target
compiler, but not with the target flags so flags such as "-mabi=64
-mnoabicalls" which are used in 64-bit MIPS kernels are not used. This
will cause the compiler to default to the userspace ABI which is 32-bit
in this case.

I'm not sure what the correct solution is here. Using the target flags
directly might make modpost specific to a kernel flavour. Hacking in
some special flags for mips* doesn't sound great either.

I have not tested any other platforms, but I am guessing this issue will
affect other platforms when the size of pointers in userspace and the
kernel differ.

James

Hi James,

Going through some older bugreports for src:linux found the above; Is
this still an issue? I have not digged deeper at all.

Regards,
Salvatore

Hi

This bug was filed for a very old kernel or the bug is old itself
without resolution.

If you can reproduce it with

- the current version in unstable/testing
- the latest kernel from backports

please reopen the bug, see https://www.debian.org/Bugs/server-control
for details.

Regards,
Salvatore

Hi

This bug was filed for a very old kernel or the bug is old itself
without resolution.

If you can reproduce it with

- the current version in unstable/testing
- the latest kernel from backports

please reopen the bug, see https://www.debian.org/Bugs/server-control
for details.

Regards,
Salvatore

Hello,

This is still reproducible in unstable (linux 7.2.12-1), with armhf
userland and the arm64 kernel.

Steps, as root on an arm64 machine:

  mmdebstrap --variant=apt --architectures=armhf,arm64 \
      --include=linux-headers-arm64:arm64 sid sid-armhf
  chroot sid-armhf apt-get update
  chroot sid-armhf apt-get install -y reform2-lpc-dkms

The DKMS build fails:

  ERROR: modpost: reform2_lpc: type mismatch between of_tis_spi_match[] and
MODULE_DEVICE_TABLE(of, ...)

apt installs linux-kbuild-7.2.12+deb14:armhf to satisfy the headers'
dependency. Replacing it with the arm64 build makes the same DKMS build
succeed:

  chroot sid-armhf apt-get install -y linux-kbuild-7.2.12+deb14:arm64

debian/rules.d/scripts/mod/Makefile.real generates devicetable-offsets.h
with the host compiler, and real-*/types.h only gives kernel_ulong_t the
size of the target. Pointer and long members keep the size of the host,
so in modpost.real-lsb-64 built on armhf, struct of_device_id is 196
bytes instead of 200. Upstream generates the offsets with the target
compiler, so it does not have this problem.

There are a few ways to fix this, but I can't think of one that's
obviously correct.

 1. Generate the offsets for the 64-bit variants with a compiler for a
    64-bit target. This gives the exact layout, but needs clang or a
    cross gcc as a build dependency, and Debian has no 64-bit cross gcc
    that runs on armhf.

 2. Rewrite the pointer and long members of mod_devicetable.h as
    kernel_ulong_t before generating the offsets. This needs no new
    dependencies, but has to match the declarations in the header.


This came up on Raspberry Pi OS, where the 32-bit userland is used with
the 64-bit kernel.

Cheers,

Serge