diff options
author | Michael Ellerman <mpe@ellerman.id.au> | 2023-02-21 14:03:31 +0100 |
---|---|---|
committer | Michael Ellerman <mpe@ellerman.id.au> | 2023-02-28 04:32:34 +0100 |
commit | f8b2336f15f3bc30e37ce5c052cde5b6319bb6df (patch) | |
tree | 542e4a4262207e1d770062f064ba745a8b3f86c2 /.mailmap | |
parent | powerpc/vmlinux.lds: Add .text.asan/tsan sections (diff) | |
download | linux-f8b2336f15f3bc30e37ce5c052cde5b6319bb6df.tar.xz linux-f8b2336f15f3bc30e37ce5c052cde5b6319bb6df.zip |
powerpc: Avoid dead code/data elimination when using recordmcount
Although powerpc now has objtool mcount support, it's not enabled in all
configurations due to dependencies.
On those configurations, with some linkers (binutils 2.37 at least),
it's still possible to hit the dreaded "recordmcount bug", eg. errors
such as:
CC kernel/kexec_file.o
Cannot find symbol for section 10: .text.unlikely.
kernel/kexec_file.o: failed
make[1]: *** [scripts/Makefile.build:287 : kernel/kexec_file.o] Error 1
Those errors are much more prevalent when building with
CONFIG_LD_DEAD_CODE_DATA_ELIMINATION, because it places every function
in a separate section.
CONFIG_LD_DEAD_CODE_DATA_ELIMINATION is marked experimental and is not
enabled in any powerpc defconfigs or by major distros. Although it does
have at least some users on 32-bit where kernel size tends to be more
important.
Avoid the build errors by blocking CONFIG_LD_DEAD_CODE_DATA_ELIMINATION
when the build is using recordmcount, rather than objtool. In practice
that means for 64-bit big endian builds, or 64-bit clang builds - both
because they lack CONFIG_MPROFILE_KERNEL.
On 32-bit objtool is always used, so
CONFIG_LD_DEAD_CODE_DATA_ELIMINATION is still available there.
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20230221130331.2714199-1-mpe@ellerman.id.au
Diffstat (limited to '.mailmap')
0 files changed, 0 insertions, 0 deletions