On 09/11/18 17:28, Sudeep Holla wrote:
> On Fri, Nov 9, 2018 at 4:10 PM Marc Zyngier <[email protected]> wrote:
>>
> [...]
>
>>
>> See bb42ca474010 and d003d029cea8 for details.
>>
>> Now, activating this workaround leads to lockdep being really angry,
>> most likely because the cpus_read_lock is not taken, which is a change
>> in behaviour...
>>
>> I'm trying to dig into this now.
>>
>
> Yes we found similar issue in kernel/sched/core.c sched_init_smp
> There's a fix with detailed description in -next
> (Commit 40fa3780bac2 ("sched/core: Take the hotplug lock in sched_init_smp()")
>
> The behaviour changed since commit cb538267ea1e ("jump_label/lockdep:
> Assert we hold the hotplug lock for _cpuslocked() operations")
I indeed came to the same conclusion, but the fix is slightly less than
obvious. I have the following arm64-specific crap, but it is pretty
terrible:
diff --git a/arch/arm64/kernel/time.c b/arch/arm64/kernel/time.c
index f258636273c9..9e96e9eaca9b 100644
--- a/arch/arm64/kernel/time.c
+++ b/arch/arm64/kernel/time.c
@@ -36,6 +36,7 @@
#include <linux/clocksource.h>
#include <linux/clk-provider.h>
#include <linux/acpi.h>
+#include <linux/cpu.h>
#include <clocksource/arm_arch_timer.h>
@@ -69,7 +70,9 @@ void __init time_init(void)
u32 arch_timer_rate;
of_clk_init(NULL);
+ cpus_read_lock();
timer_probe();
+ cpus_read_unlock();
tick_setup_hrtimer_broadcast();
Qian, can you please let me know if this helps? If it does, we'll have
to think of something a bit better...
Thanks,
M.
--
Jazz is not dead. It just smells funny...
