diff options
author | Peter Zijlstra <peterz@infradead.org> | 2020-04-16 14:34:26 +0200 |
---|---|---|
committer | Peter Zijlstra <peterz@infradead.org> | 2020-04-30 20:14:34 +0200 |
commit | cc1ac9c792810b93783a7de344f428922af8d98c (patch) | |
tree | 33da496a06b118211f1178a55f3d30d51f96abec /tools/objtool/check.c | |
parent | 34fdce6981b96920ced4e0ee56e9db3fb03a33f0 (diff) | |
download | linux-stable-cc1ac9c792810b93783a7de344f428922af8d98c.tar.gz linux-stable-cc1ac9c792810b93783a7de344f428922af8d98c.tar.bz2 linux-stable-cc1ac9c792810b93783a7de344f428922af8d98c.zip |
x86/retpoline: Fix retpoline unwind
Currently objtool cannot understand retpolines, and thus cannot
generate ORC unwind information for them. This means that we cannot
unwind from the middle of a retpoline.
The recent ANNOTATE_INTRA_FUNCTION_CALL and UNWIND_HINT_RET_OFFSET
support in objtool enables it to understand the basic retpoline
construct. A further problem is that the ORC unwind information is
alternative invariant; IOW. every alternative should have the same
ORC, retpolines obviously violate this. This means we need to
out-of-line them.
Since all GCC generated code already uses out-of-line retpolines, this
should not affect performance much, if anything.
This will enable objtool to generate valid ORC data for the
out-of-line copies, which means we can correctly and reliably unwind
through a retpoline.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Josh Poimboeuf <jpoimboe@redhat.com>
Link: https://lkml.kernel.org/r/20200428191700.210835357@infradead.org
Diffstat (limited to 'tools/objtool/check.c')
0 files changed, 0 insertions, 0 deletions