6.5-stable review patch. If anyone has any objections, please let me know.
------------------
From: Mårten Lindahl <marten.lindahl@xxxxxxxx>
[ Upstream commit 8922ba71c969d2a0c01a94372a71477d879470de ]
If a panic is triggered by a hrtimer interrupt all online cpus will be
notified and set offline. But as highlighted by commit 19dbdcb8039c
("smp: Warn on function calls from softirq context") this call should
not be made synchronous with disabled interrupts:
softdog: Initiating panic
Kernel panic - not syncing: Software Watchdog Timer expired
WARNING: CPU: 1 PID: 0 at kernel/smp.c:753 smp_call_function_many_cond
unwind_backtrace:
show_stack
dump_stack_lvl
__warn
warn_slowpath_fmt
smp_call_function_many_cond
smp_call_function
crash_smp_send_stop.part.0
machine_crash_shutdown
__crash_kexec
panic
softdog_fire
__hrtimer_run_queues
hrtimer_interrupt
Make the smp call for machine_crash_nonpanic_core() asynchronous.
Signed-off-by: Mårten Lindahl <marten.lindahl@xxxxxxxx>
Signed-off-by: Russell King (Oracle) <rmk+kernel@xxxxxxxxxxxxxxx>
Signed-off-by: Sasha Levin <sashal@xxxxxxxxxx>
---
arch/arm/kernel/machine_kexec.c | 14 +++++++++++++-
1 file changed, 13 insertions(+), 1 deletion(-)
diff --git a/arch/arm/kernel/machine_kexec.c b/arch/arm/kernel/machine_kexec.c
index 46364b699cc30..5d07cf9e0044d 100644
--- a/arch/arm/kernel/machine_kexec.c
+++ b/arch/arm/kernel/machine_kexec.c
@@ -94,16 +94,28 @@ static void machine_crash_nonpanic_core(void *unused)
}
}
+static DEFINE_PER_CPU(call_single_data_t, cpu_stop_csd) =
+ CSD_INIT(machine_crash_nonpanic_core, NULL);
+
void crash_smp_send_stop(void)
{
static int cpus_stopped;
unsigned long msecs;
+ call_single_data_t *csd;
+ int cpu, this_cpu = raw_smp_processor_id();
if (cpus_stopped)
return;
atomic_set(&waiting_for_crash_ipi, num_online_cpus() - 1);
- smp_call_function(machine_crash_nonpanic_core, NULL, false);
+ for_each_online_cpu(cpu) {
+ if (cpu == this_cpu)
+ continue;
+
+ csd = &per_cpu(cpu_stop_csd, cpu);
+ smp_call_function_single_async(cpu, csd);
+ }
+
msecs = 1000; /* Wait at most a second for the other cpus to stop */
while ((atomic_read(&waiting_for_crash_ipi) > 0) && msecs) {
mdelay(1);
--
2.40.1
