kcsan: Support WEAK_MEMORY with Clang where no objtool support exists
authorMarco Elver <elver@google.com>
Tue, 30 Nov 2021 11:44:33 +0000 (12:44 +0100)
committerPaul E. McKenney <paulmck@kernel.org>
Fri, 10 Dec 2021 00:42:28 +0000 (16:42 -0800)
commitbd3d5bd1a0ad386475ea7a3de8a91e7d8a600536
tree0e66c5199f20a5757e3763a92be0158cccac35a2
parenta015b7085979b12e55f67f3b86be0321fff6be3f
kcsan: Support WEAK_MEMORY with Clang where no objtool support exists

Clang and GCC behave a little differently when it comes to the
__no_sanitize_thread attribute, which has valid reasons, and depending
on context either one could be right.

Traditionally, user space ThreadSanitizer [1] still expects instrumented
builtin atomics (to avoid false positives) and __tsan_func_{entry,exit}
(to generate meaningful stack traces), even if the function has the
attribute no_sanitize("thread").

[1] https://clang.llvm.org/docs/ThreadSanitizer.html#attribute-no-sanitize-thread

GCC doesn't follow the same policy (for better or worse), and removes
all kinds of instrumentation if no_sanitize is added. Arguably, since
this may be a problem for user space ThreadSanitizer, we expect this may
change in future.

Since KCSAN != ThreadSanitizer, the likelihood of false positives even
without barrier instrumentation everywhere, is much lower by design.

At least for Clang, however, to fully remove all sanitizer
instrumentation, we must add the disable_sanitizer_instrumentation
attribute, which is available since Clang 14.0.

Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
include/linux/compiler_types.h
lib/Kconfig.kcsan