On Thu, May 18, 2023 at 12:17:18PM +0800, Haibo Xu wrote:
...
> > The idea of these *to_str functions is to dump output that can be
> > copy+pasted into a reg array (hence the trailing commas in print_reg
> > lines). So we can't just print random lines here or return '##UNKOWN##',
> > as that won't compile. Instead, the default should return
> >
> > str_with_index("KVM_REG_RISCV_CONFIG_REG(##)", reg_off)
> >
>
> Thanks for sharing the detailed idea, will fix it in next version!
I guess we could also return a string like,
"KVM_REG_RISCV_CONFIG_REG(##) /* UNKNOWN */"
as that would still compile and also convey the message that this
register doesn't have a name because the test doesn't know it yet.
...
> > We should share all the code above, except print_reg(), with aarch64.
> > I'll send a patch series that splits the arch-neutral code out of
> > the aarch64 test that you can base this test on.
> >
>
> Good idea! I will rebase the patch based on your work.
>
Ok, I've pushed patches to [1]. This series introduces two things to KVM
selftests. Primarily it splits the aarch64/get-reg-list test into a
cross-arch get-reg-list test and an $ARCH_DIR/get-reg-list.o object file,
which the cross-arch test depends on. To do that, it also introduces the
concept of a "split test", a test that has a cross-arch part which depends
on an arch-specific part. Using a split test is cleaner than the
#ifdeffery we usually do for cross-arch tests.
I've added kvmarm@xxxxxxxxxxxxxxx, Marc, Oliver, and Sean to the CC of
this message. You'll want to add them when you post v2 as well.
[1] https://github.com/jones-drew/linux/commits/arm64/kself/get-reg-list
Thanks,
drew
