On 7/5/24 17:28, Thorsten Glaser wrote:
Helge Deller dixit:
I just did a give-back, and this time the build succeeded (on the
"c8000" buildd server). The last build failed on the "pasta" buildd
server, which is a qemu-user based build machine, while "c8000" is
a physical server which runs the *very latest* stable native
Linux kernel (v6.9.7).
Aaah, that could of course be a thing.
Of course I will try to reproduce it on "pasta" again just to
If this can be reproduced under qemu-user but not on native hardware
I can mark that test as expected to fail under qemu-user. There’s a
few of them already.
Good to know.
Perhaps qemu-user does not yet support the new constants from those
patches?
It's about how some syscalls are called, so old kernel/glibc versions
used different registers or register orderings.
This makes those calls highly incompatible, but Arnds patches cleaned
it up, so in the long run it should become better now.
Beside the kernel & glibc patches some qemu-user patches will probably
be necessary as well.
I think disabling such tests is probably best right now.
Helge
