powerpc: Fix smp_send_stop NMI IPI handling
authorNicholas Piggin <npiggin@gmail.com>
Wed, 25 Apr 2018 02:17:53 +0000 (12:17 +1000)
committerMichael Ellerman <mpe@ellerman.id.au>
Wed, 25 Apr 2018 10:38:08 +0000 (20:38 +1000)
commitac61c1156623455c46701654abd8c99720bceea1
treec7f94e67f672b815fc5815d3f4ce7af623359f3e
parent682e6b4da5cbe8e9a53f979a58c2a9d7dc997175
powerpc: Fix smp_send_stop NMI IPI handling

The NMI IPI handler for a receiving CPU increments nmi_ipi_busy_count
over the handler function call, which causes later smp_send_nmi_ipi()
callers to spin until the call is finished.

The stop_this_cpu() function never returns, so the busy count is never
decremeted, which can cause the system to hang in some cases. For
example panic() will call smp_send_stop() early on which calls
stop_this_cpu() on other CPUs, then later in the reboot path,
pnv_restart() will call smp_send_stop() again, which hangs.

Fix this by adding a special case to the stop_this_cpu() handler to
decrement the busy count, because it will never return.

Now that the NMI/non-NMI versions of stop_this_cpu() are different,
split them out into separate functions rather than doing #ifdef tricks
to share the body between the two functions.

Fixes: 6bed3237624e3 ("powerpc: use NMI IPI for smp_send_stop")
Reported-by: Abdul Haleem <abdhalee@linux.vnet.ibm.com>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
[mpe: Split out the functions, tweak change log a bit]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
arch/powerpc/kernel/smp.c