]> git.ipfire.org Git - thirdparty/kernel/stable.git/commitdiff
powerpc: smp_send_stop do not offline stopped CPUs
authorNicholas Piggin <npiggin@gmail.com>
Sat, 19 May 2018 04:35:53 +0000 (14:35 +1000)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Fri, 24 Aug 2018 11:06:43 +0000 (13:06 +0200)
[ Upstream commit de6e5d38417e6cdb005843db420a2974993d36ff ]

Marking CPUs stopped by smp_send_stop as offline can cause warnings
due to cross-CPU wakeups. This trace was noticed on a busy system
running a sysrq+c crash test, after the injected crash:

WARNING: CPU: 51 PID: 1546 at kernel/sched/core.c:1179 set_task_cpu+0x22c/0x240
CPU: 51 PID: 1546 Comm: kworker/u352:1 Tainted: G      D
Workqueue: mlx5e mlx5e_update_stats_work [mlx5_core]
[...]
NIP [c00000000017c21c] set_task_cpu+0x22c/0x240
LR [c00000000017d580] try_to_wake_up+0x230/0x720
Call Trace:
[c000000001017700] runqueues+0x0/0xb00 (unreliable)
[c00000000017d580] try_to_wake_up+0x230/0x720
[c00000000015a214] insert_work+0x104/0x140
[c00000000015adb0] __queue_work+0x230/0x690
[c000003fc5007910] [c00000000015b26c] queue_work_on+0x5c/0x90
[c0080000135fc8f8] mlx5_cmd_exec+0x538/0xcb0 [mlx5_core]
[c008000013608fd0] mlx5_core_access_reg+0x140/0x1d0 [mlx5_core]
[c00800001362777c] mlx5e_update_pport_counters.constprop.59+0x6c/0x90 [mlx5_core]
[c008000013628868] mlx5e_update_ndo_stats+0x28/0x90 [mlx5_core]
[c008000013625558] mlx5e_update_stats_work+0x68/0xb0 [mlx5_core]
[c00000000015bcec] process_one_work+0x1bc/0x5f0
[c00000000015ecac] worker_thread+0xac/0x6b0
[c000000000168338] kthread+0x168/0x1b0
[c00000000000b628] ret_from_kernel_thread+0x5c/0xb4

This happens because firstly the CPU is not really offline in the
usual sense, processes and interrupts have not been migrated away.
Secondly smp_send_stop does not happen atomically on all CPUs, so
one CPU can have marked itself offline, while another CPU is still
running processes or interrupts which can affect the first CPU.

Fix this by just not marking the CPU as offline. It's more like
frozen in time, so offline does not really reflect its state properly
anyway. There should be nothing in the crash/panic path that walks
online CPUs and synchronously waits for them, so this change should
not introduce new hangs.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Sasha Levin <alexander.levin@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
arch/powerpc/kernel/smp.c

index 9ca7148b5881a098abf7bf0a3eff99536f99a7ec..6d6cf14009cfaffa37e57f6a62491c2ef76ef5ec 100644 (file)
@@ -579,9 +579,6 @@ static void nmi_stop_this_cpu(struct pt_regs *regs)
        nmi_ipi_busy_count--;
        nmi_ipi_unlock();
 
-       /* Remove this CPU */
-       set_cpu_online(smp_processor_id(), false);
-
        spin_begin();
        while (1)
                spin_cpu_relax();
@@ -596,9 +593,6 @@ void smp_send_stop(void)
 
 static void stop_this_cpu(void *dummy)
 {
-       /* Remove this CPU */
-       set_cpu_online(smp_processor_id(), false);
-
        hard_irq_disable();
        spin_begin();
        while (1)