MIPS: OCTEON: Improve _machine_halt implementation.

As noted by Wladislav Wiebe:
   $ halt
   ..
   Sent SIGKILL to all processes
   Requesting system halt
   [66.729373] System halted.
   [66.733244]
   [66.734761] =====================================
   [66.739473] [ BUG: lock held at task exit time! ]
   [66.744188] 3.8.7-0-sampleversion-fct #49 Tainted: G           O
   [66.750202] -------------------------------------
   [66.754913] init/21479 is exiting with locks still held!
   [66.760234] 1 lock held by init/21479:
   [66.763990]  #0:  (reboot_mutex){+.+...}, at: [<ffffffff801776c8>] SyS_reboot+0xe0/0x218
   [66.772165]
   [66.772165] stack backtrace:
   [66.776532] Call Trace:
   [66.778992] [<ffffffff805780a8>] dump_stack+0x8/0x34
   [66.783972] [<ffffffff801618b0>] do_exit+0x610/0xa70
   [66.788948] [<ffffffff801777a8>] SyS_reboot+0x1c0/0x218
   [66.794186] [<ffffffff8013d6a4>] handle_sys64+0x44/0x64

This is an alternative fix to the one sent by Wladislav.  We kill the
watchdog for each CPU and then spin in WAIT with interrupts disabled.
This is the lowest power mode for the OCTEON.  If we were to spin with
interrupts enabled, we would get a continual stream of warning messages
and backtraces from the lockup detector, so I chose to disable
interrupts.

Signed-off-by: David Daney <david.daney@cavium.com>
Cc: Maxim Uvarov <muvarov@gmail.com>
Cc: Wladislav Wiebe <wladislav.kw@gmail.com>
Cc: linux-mips@linux-mips.org
Cc: David Daney <david.daney@cavium.com>
Patchwork: https://patchwork.linux-mips.org/patch/5324/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>