On 11/1/24 8:45 AM, Uros Bizjak wrote:
On Fri, Nov 1, 2024 at 2:18 PM Jeff Law <jeffreya...@gmail.com> wrote:
On 11/1/24 5:34 AM, Uros Bizjak wrote:
Stack pointer modifications in asm are currently not flagged in
crtl->sp_is_unchanging due to RTX pointer comparison in
notice_stack_pointer_modification_1. Pointer comparison does not detect
that "(reg/v:DI 7 sp [ current_stack_pointer ])" and "(reg/f:DI 7 sp)"
RTXes actually all correspond to the same stack pointer register.
Due to the above omission, the compiler does not detect that asm RTX
manipulates stack pointer in the following construct:
But how did you get two distinct RTXs for the stack pointer? That's not
supposed to happen IIRC.
Please see the testcase in the patch:
+register unsigned long current_stack_pointer asm ("%rsp");
+#define ASM_CALL_CONSTRAINT "+r" (current_stack_pointer)
[...]
+ asm volatile ("pushfq; push %1; pop %0; popfq"
+ : "=r" (y), ASM_CALL_CONSTRAINT
+ : "e" (-1));
When compiled for x86_64, the stack pointer RTX that is different from
the generic stack pointer RTX is created.
If we're allowing that, then I suspect there's a good number of places
that will ultimately need to be fixed. It's baked in pretty deep that
there is only one stack pointer.
So I think we need to conclude whether or not the testcase is valid or
not first.
jeff
