Hi Paul,
On 6/23/2021 3:30 PM, Paul E. McKenney wrote:
Hello, Aneesh!
Yesterday evening's next-20210622 testing gave me the following
kernel-build error:
ld: mm/mremap.o: in function `move_huge_pud':
/home/git/linux-next/mm/mremap.c:372: undefined reference to `__compiletime_assert_395'
Bisection landed on this commit:
257121c5aabe ("mm/mremap: convert huge PUD move to separate helper")
I have no idea how this commit relates to that error message, but
reverting this commit on top of next-20210622 really does get rid of
the problem.
This was reported by Naresh Kamboju last week:
https://lore.kernel.org/r/CA+G9fYsWHE5Vu9T3FV-vtHHbVFJWEF=bmjQxwaZs3uVYef028g@xxxxxxxxxxxxxx/
It seems to be related to the fact that older versions of clang cannot
figure out that the HPAGE_PUD case in move_pgt_entry() is never called
when CONFIG_TRANSPARENT_HUGE is unset so it is not eliminated, which is
problematic because flush_pud_tlb_range() evaluates to BUILD_BUG() in
that case, which is called within move_huge_pud().
The following reproducer provokes this error:
tools/testing/selftests/rcutorture/bin/kvm.sh --allcpus --torture lock --configs LOCK07 --build-only --kconfig "CONFIG_DEBUG_LOCK_ALLOC=y CONFIG_PROVE_LOCKING=y" --kmake-arg "CC=clang-11"
Run the above command in the top-level directory of your -next source
tree, and using this compiler:
$ clang-11 -v
Ubuntu clang version 11.1.0-++20210428103817+1fdec59bffc1-1~exp1~20210428204431.166
Target: x86_64-pc-linux-gnu
Thoughts?
Aneesh had a patch in that thread above that would resolve the issue, it
just needs to be sent to Andrew for a fixup patch.
Cheers,
Nathan
