https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78158
--- Comment #14 from Dmitry Vyukov <dvyukov at google dot com> ---
Right, we need to catch completely bogus values.
But the question is: are these values extending up or down? :)
Because special flag was added after the HLE flags...
So how about 12 bits? :)
--- lib/tsan/rtl/tsan_interface_atomic.cc (revision 298354)
+++ lib/tsan/rtl/tsan_interface_atomic.cc (working copy)
@@ -450,10 +450,30 @@
// C/C++
+static morder covert_morder(morder mo) {
+ if (flags()->force_seq_cst_atomics)
+ return (morder)mo_seq_cst;
+
+ // Filter out additional memory order flags:
+ // MEMMODEL_SYNC = 1 << 15
+ // __ATOMIC_HLE_ACQUIRE = 1 << 16
+ // __ATOMIC_HLE_RELEASE = 1 << 17
+ //
+ // HLE is an optimization, and we pretend that elision always fails.
+ // MEMMODEL_SYNC is used when lowering __sync_ atomics,
+ // since we use __sync_ atomics for actual atomic operations,
+ // we can safely ignore it as well.
+ // It's unclear if these special flags extend up or down, because
+ // MEMMODEL_SYNC was added after HLE flags. So we mask lower 12 bits to
+ // allow some future special flags and catch completely bogus memory order
+ // values at the same time.
+ return mo & 0xfff;
+}
+
#define SCOPED_ATOMIC(func, ...) \
const uptr callpc = (uptr)__builtin_return_address(0); \
uptr pc = StackTrace::GetCurrentPc(); \
- mo = flags()->force_seq_cst_atomics ? (morder)mo_seq_cst : mo; \
+ mo = covert_morder(mo); \
ThreadState *const thr = cur_thread(); \
if (thr->ignore_interceptors) \
return NoTsanAtomic##func(__VA_ARGS__); \
