On Wed, Sep 16, 2020 at 12:40:30AM +0100, Andrew Cooper wrote: > It's worse than that. Even when stating that %rsp is modified in the > asm, the generated code sequence is still buggy, for recent Clang and GCC. > > https://godbolt.org/z/ccz9v7 > > It's clearly not safe to ever use memory operands with pushf/popf asm > fragments. So I went and singlestepped your snippet in gdb. And it all seems to work - it is simply a bit confusing: :-) eflags 0x246 [ PF ZF IF ] => 0x000055555555505d <main+13>: 9c pushfq 0x7fffffffe440: 0x00007fffffffe540 0x0000000000000000 0x7fffffffe450: 0x0000000000000000 0x00007ffff7e0ecca 0x7fffffffe460: 0x00007fffffffe548 0x00000001ffffe7c9 0x7fffffffe470: 0x0000555555555050 0x00007ffff7e0e8f8 0x7fffffffe480: 0x0000000000000000 0x0c710afd7e78681b those lines under the "=>" line are the stack contents printed with $ x/10gx $sp Then, we will pop into 0x8(%rsp): => 0x55555555505e <main+14>: popq 0x8(%rsp) 0x7fffffffe438: 0x0000000000000346 0x00007fffffffe540 0x7fffffffe448: 0x0000000000000000 0x0000000000000000 0x7fffffffe458: 0x00007ffff7e0ecca 0x00007fffffffe548 0x7fffffffe468: 0x00000001ffffe7c9 0x0000555555555050 0x7fffffffe478: 0x00007ffff7e0e8f8 0x0000000000000000 Now, POP copies the value pointed to by %rsp, *increments* the stack pointer and *then* computes the effective address of the operand. It says so in the SDM too (thanks peterz!): "If the ESP register is used as a base register for addressing a destination operand in memory, the POP instruction computes the effective address of the operand after it increments the ESP register." *That*s why, FLAGS is in 0x7fffffffe448! which is %rsp + 8. Basically flags is there *twice* on the stack: (gdb) x/10x 0x7fffffffe438 0x7fffffffe438: 0x0000000000000346 0x00007fffffffe540 ^^^^^^^^^^^^^^^^^^ 0x7fffffffe448: 0x0000000000000346 0x0000000000000000 ^^^^^^^^^^^^^^^^^^ 0x7fffffffe458: 0x00007ffff7e0ecca 0x00007fffffffe548 0x7fffffffe468: 0x00000001ffffe7c9 0x0000555555555050 0x7fffffffe478: 0x00007ffff7e0e8f8 0x0000000000000000 and now we read the second copy into %rsi. => 0x555555555062 <main+18>: mov 0x8(%rsp),%rsi 0x7fffffffe440: 0x00007fffffffe540 0x0000000000000346 0x7fffffffe450: 0x0000000000000000 0x00007ffff7e0ecca 0x7fffffffe460: 0x00007fffffffe548 0x00000001ffffe7c9 0x7fffffffe470: 0x0000555555555050 0x00007ffff7e0e8f8 0x7fffffffe480: 0x0000000000000000 0x0c710afd7e78681b Looks like it works as designed. -- Regards/Gruss, Boris. https://people.kernel.org/tglx/notes-about-netiquette