vfscanf is miscompiled by gcc 4.4, here is a shorter testcase that is
miscompiled also by 4.6:
extern void *alloca (__SIZE_TYPE__);
extern void bar (long *, char *);
long
foo (long x)
{
long buf[400];
char *p = alloca (x);
bar (buf, p);
return buf[0];
}
at -O2 -m64 results in:
...
bl .bar
nop
addi 1,31,3328
ld 3,112(31)
ld 0,16(1)
ld 31,-8(1)
mtlr 0
blr
Note that sched2 swapped incorrectly the stack frame release and ld 3,112(31),
so the ld insn reads from memory location 3216 below the stack (which is much
lower than 288 bytes of ppc64 red zone).
If a signal comes in in between addi and ld and something clobbers the stack,
this function (or, with 4.4 vfscanf) will return incorrect value.
There is nothing in the epilogue stack release insn that would make it a memory
barrier.
--
Summary: ppc64 glibc miscompilation
Product: gcc
Version: 4.6.0
Status: UNCONFIRMED
Keywords: wrong-code
Severity: normal
Priority: P3
Component: target
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: jakub at gcc dot gnu dot org
GCC target triplet: powerpc64-linux
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44199
