https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87220
Bug ID: 87220
Summary: -fstack-check produces inefficient and wrong tests
Product: gcc
Version: unknown
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: other
Assignee: unassigned at gcc dot gnu.org
Reporter: bugdal at aerifal dot cx
Target Milestone: ---
Given the test case:
void bar(void *);
int foo()
{
char a[10];
bar(&a);
}
-fstack-check produces the code (x86_64; similar for 32-bit):
foo:
subq $4152, %rsp
orq $0, (%rsp)
addq $4128, %rsp
leaq 6(%rsp), %rdi
call bar
addq $24, %rsp
ret
This test:
1. Unnecessarily touches an extra page of stack that's not used, wasting
memory.
2. Skips over a whole page, potentially clobbering (data race) data on the
other side of a guard page.
3. Is completely unnecessary; since less than a page of stack is needed, only a
probe of the final desired stack pointer (after the third instruction) is
needed, and it's performed implicitly by the call.
I would expect at least:
foo:
subq $24, %rsp
orq $0, (%rsp)
leaq 6(%rsp), %rdi
call bar
addq $24, %rsp
ret
or ideally:
foo:
subq $24, %rsp
leaq 6(%rsp), %rdi
call bar
addq $24, %rsp
ret
The excessive stack usage, unsafety (jumping the guard page), and size and
potential performance hit from unnecessary checks seem to make -fstack-check
unsuitable for its intended purpose at this time.
