[PATCH] selftests/pidfd: align stack to fix SP alignment exception

[Will Deacon]

On Wed, Mar 19, 2025 at 10:59:57AM +0800, Shuai Xue wrote:
> + ARM maintainers for review.
> 
> 在 2025/3/12 14:15, Shuai Xue 写道:
> > The pidfd_test fails on the ARM64 platform with the following error:
> > 
> >      Bail out! pidfd_poll check for premature notification on child thread exec test: Failed
> > 
> > When exception-trace is enabled, the kernel logs the details:
> > 
> >      #echo 1 > /proc/sys/debug/exception-trace
> >      #dmesg | tail -n 20
> >      [48628.713023] pidfd_test[1082142]: unhandled exception: SP Alignment, ESR 0x000000009a000000, SP/PC alignment exception in pidfd_test[400000+4000]
> >      [48628.713049] CPU: 21 PID: 1082142 Comm: pidfd_test Kdump: loaded Tainted: G        W   E      6.6.71-3_rc1.al8.aarch64 #1
> >      [48628.713051] Hardware name: AlibabaCloud AliServer-Xuanwu2.0AM-1UC1P-5B/AS1111MG1, BIOS 1.2.M1.AL.P.157.00 07/29/2023
> >      [48628.713053] pstate: 60001800 (nZCv daif -PAN -UAO -TCO -DIT +SSBS BTYPE=-c)
> >      [48628.713055] pc : 0000000000402100
> >      [48628.713056] lr : 0000ffff98288f9c
> >      [48628.713056] sp : 0000ffffde49daa8
> >      [48628.713057] x29: 0000000000000000 x28: 0000000000000000 x27: 0000000000000000
> >      [48628.713060] x26: 0000000000000000 x25: 0000000000000000 x24: 0000000000000000
> >      [48628.713062] x23: 0000000000000000 x22: 0000000000000000 x21: 0000000000400e80
> >      [48628.713065] x20: 0000000000000000 x19: 0000000000402650 x18: 0000000000000000
> >      [48628.713067] x17: 00000000004200d8 x16: 0000ffff98288f40 x15: 0000ffffde49b92c
> >      [48628.713070] x14: 0000000000000000 x13: 0000000000000000 x12: 0000000000000000
> >      [48628.713072] x11: 0000000000001011 x10: 0000000000402100 x9 : 0000000000000010
> >      [48628.713074] x8 : 00000000000000dc x7 : 3861616239346564 x6 : 000000000000000a
> >      [48628.713077] x5 : 0000ffffde49daa8 x4 : 000000000000000a x3 : 0000ffffde49daa8
> >      [48628.713079] x2 : 0000ffffde49dadc x1 : 0000ffffde49daa8 x0 : 0000000000000000
> > 
> > According to ARM ARM D1.3.10.2 SP alignment checking:
> > 
> > > When the SP is used as the base address of a calculation, regardless of
> > > any offset applied by the instruction, if bits [3:0] of the SP are not
> > > 0b0000, there is a misaligned SP.
> > 
> > To fix it, align the stack with 16 bytes.
> > 
> > Signed-off-by: Shuai Xue <xueshuai at linux.alibaba.com>
> > ---
> >   tools/testing/selftests/pidfd/pidfd_test.c | 2 +-
> >   1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/tools/testing/selftests/pidfd/pidfd_test.c b/tools/testing/selftests/pidfd/pidfd_test.c
> > index c081ae91313a..ec161a7c3ff9 100644
> > --- a/tools/testing/selftests/pidfd/pidfd_test.c
> > +++ b/tools/testing/selftests/pidfd/pidfd_test.c
> > @@ -33,7 +33,7 @@ static bool have_pidfd_send_signal;
> >   static pid_t pidfd_clone(int flags, int *pidfd, int (*fn)(void *))
> >   {
> >   	size_t stack_size = 1024;
> > -	char *stack[1024] = { 0 };
> > +	char *stack[1024] __attribute__((aligned(16))) = {0};
> >   #ifdef __ia64__
> >   	return __clone2(fn, stack, stack_size, flags | SIGCHLD, NULL, pidfd);

The arm64 alignment requirement is correct, but I don't really grok what
this code is trying to do. If it's not using CLONE_VM, why bother passing
a separate stack? If it _is_ using CLONE_VM, then surely the stack needs
to be allocated somewhere other than the caller stack?

Will