diff options
author | Nathan Chancellor <nathan@kernel.org> | 2024-11-24 05:30:19 +0300 |
---|---|---|
committer | Yury Norov <yury.norov@gmail.com> | 2024-12-30 21:29:25 +0300 |
commit | f9d2ee3f51d60100c10d4db64da4413f69e81993 (patch) | |
tree | e09e9605aff3506fa859100c222834eb66aa9c3f /tools/perf/scripts/python/exported-sql-viewer.py | |
parent | 4463a445a64b719e6f501d80dcc5872dde42eb73 (diff) | |
download | linux-f9d2ee3f51d60100c10d4db64da4413f69e81993.tar.xz |
riscv: Always inline bitops
When building allmodconfig + ThinLTO with certain versions of clang,
arch_set_bit() may not be inlined, resulting in a modpost warning:
WARNING: modpost: vmlinux: section mismatch in reference: arch_set_bit+0x58 (section: .text.arch_set_bit) -> numa_nodes_parsed (section: .init.data)
acpi_numa_rintc_affinity_init() calls arch_set_bit() via __node_set()
with numa_nodes_parsed, which is marked as __initdata. If arch_set_bit()
is not inlined, modpost will flag that it is being called with data that
will be freed after init.
As acpi_numa_rintc_affinity_init() is marked as __init, there is not
actually a functional issue here. However, the bitop functions should be
marked as __always_inline, so that they work consistently for init and
non-init code, which the comment in include/linux/nodemask.h alludes to.
This matches s390 and x86's implementations.
Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Signed-off-by: Yury Norov <yury.norov@gmail.com>
Diffstat (limited to 'tools/perf/scripts/python/exported-sql-viewer.py')
0 files changed, 0 insertions, 0 deletions