diff options
author | Ard Biesheuvel <ardb@kernel.org> | 2020-01-21 10:39:12 +0100 |
---|---|---|
committer | Ingo Molnar <mingo@kernel.org> | 2020-01-25 10:14:37 +0100 |
commit | ac6119e7f25b842fc061e8aec88c4f32d3bc28ef (patch) | |
tree | dc6102d57008c1b047c3255a0c41e41d86f47267 /security | |
parent | cada0b6dbb8c07ab353d209437243bec8cc7c64a (diff) | |
download | linux-stable-ac6119e7f25b842fc061e8aec88c4f32d3bc28ef.tar.gz linux-stable-ac6119e7f25b842fc061e8aec88c4f32d3bc28ef.tar.bz2 linux-stable-ac6119e7f25b842fc061e8aec88c4f32d3bc28ef.zip |
efi/x86: Disable instrumentation in the EFI runtime handling code
We already disable KASAN instrumentation specifically for the
EFI routines that are known to dereference memory addresses that
KASAN does not know about, avoiding false positive KASAN splats.
However, as it turns out, having GCOV or KASAN instrumentation enabled
interferes with the compiler's ability to optimize away function calls
that are guarded by IS_ENABLED() checks that should have resulted in
those references to have been const-propagated out of existence. But
with instrumenation enabled, we may get build errors like:
ld: arch/x86/platform/efi/efi_64.o: in function `efi_thunk_set_virtual_address_map':
ld: arch/x86/platform/efi/efi_64.o: in function `efi_set_virtual_address_map':
in builds where CONFIG_EFI=y but CONFIG_EFI_MIXED or CONFIG_X86_UV are not
defined, even though the invocations are conditional on IS_ENABLED() checks
against the respective Kconfig symbols.
So let's disable instrumentation entirely for this subdirectory, which
isn't that useful here to begin with.
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: linux-efi@vger.kernel.org
Diffstat (limited to 'security')
0 files changed, 0 insertions, 0 deletions