arm64: do not enforce strict 16 byte alignment to stack pointer
authorColin Ian King <colin.king@canonical.com>
Wed, 11 May 2016 16:56:54 +0000 (17:56 +0100)
committerWill Deacon <will.deacon@arm.com>
Thu, 12 May 2016 13:20:49 +0000 (14:20 +0100)
copy_thread should not be enforcing 16 byte aligment and returning
-EINVAL. Other architectures trap misaligned stack access with SIGBUS
so arm64 should follow this convention, so remove the strict enforcement
check.

For example, currently clone(2) fails with -EINVAL when passing
a misaligned stack and this gives little clue to what is wrong. Instead,
it is arguable that a SIGBUS on the fist access to a misaligned stack
allows one to figure out that it is a misaligned stack issue rather
than trying to figure out why an unconventional (and undocumented)
-EINVAL is being returned.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
arch/arm64/kernel/process.c

index ad4a7e1..48eea68 100644 (file)
@@ -265,9 +265,6 @@ int copy_thread(unsigned long clone_flags, unsigned long stack_start,
                if (stack_start) {
                        if (is_compat_thread(task_thread_info(p)))
                                childregs->compat_sp = stack_start;
-                       /* 16-byte aligned stack mandatory on AArch64 */
-                       else if (stack_start & 15)
-                               return -EINVAL;
                        else
                                childregs->sp = stack_start;
                }