x86/retpoline: Fix retpoline unwind
authorPeter Zijlstra <peterz@infradead.org>
Thu, 16 Apr 2020 12:34:26 +0000 (14:34 +0200)
committerPeter Zijlstra <peterz@infradead.org>
Thu, 30 Apr 2020 18:14:34 +0000 (20:14 +0200)
commitcc1ac9c792810b93783a7de344f428922af8d98c
tree33da496a06b118211f1178a55f3d30d51f96abec
parent34fdce6981b96920ced4e0ee56e9db3fb03a33f0
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
arch/x86/include/asm/asm-prototypes.h
arch/x86/include/asm/nospec-branch.h
arch/x86/lib/retpoline.S