Commit 
377747c40657 ("ARM: entry: allow ARM-private syscalls to be
restarted") reworked the low-level syscall dispatcher to allow
restarting of ARM-private syscalls. Unfortunately, this relocated the
label used to dispatch a private syscall from the trace path, so that
the invocation would be bypassed altogether!
This causes applications to fail under strace as soon as they rely on
a private syscall (e.g. set_tls):
  set_tls(0xb6fad4c0, 0xb6fadb98, 0xb6fb1050, 0xb6fad4c0, 0xb6fb1050)
      = -1 ENOSYS (Function not implemented)
This patch fixes the label so that we correctly dispatch private
syscalls from the trace path.
Reported-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com>
Tested-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
 
        ldrcc   pc, [tbl, scno, lsl #2]         @ call sys_* routine
 
        add     r1, sp, #S_OFF
-       cmp     scno, #(__ARM_NR_BASE - __NR_SYSCALL_BASE)
+2:     cmp     scno, #(__ARM_NR_BASE - __NR_SYSCALL_BASE)
        eor     r0, scno, #__NR_SYSCALL_BASE    @ put OS number back
        bcs     arm_syscall
-2:     mov     why, #0                         @ no longer a real syscall
+       mov     why, #0                         @ no longer a real syscall
        b       sys_ni_syscall                  @ not private func
 
 #if defined(CONFIG_OABI_COMPAT) || !defined(CONFIG_AEABI)