InterlockedBitTestAndSet() bug

Look at InterlockedBitTestAndSet and Co. from include/psdk/winnt.h:

_asm_ _volatile_("lock "
"btsl %2,%1\n\t"
"sbbl %0,%0\n\t"
:"=r" (OldBit),"=m" (*Base)
:"Ir" (Bit)
: "memory");

There should be "+m" instead of "=m", I believe.

Modifier "=" means that this operand is write-only for this instruction: the previous value is discarded and replaced by output data. GCC will assume that the value in this operand before the instruction is dead and need not be generated. "+" means that this operand is both read and written by the instruction.

For example:

ExInitializePushLock(Lock);
ExAcquirePushLockExclusive(Lock);

GCC may think that ExInitializePushLock() is not needed here and remove it, which is wrong.

P.S. See also intrin_x86.h