Hi Sebastian,
On Tue, Mar 18, 2014 at 09:40:57PM +0000, Sebastian Capella wrote:
> Use of tracers in local_irq_disable is causes abort loops when called
> with irqs disabled using a temporary stack. Replace local_irq_disable
> with raw_local_irq_disable instead to avoid tracers.
Do you have any more information about these aborts? At the time we call
local_irq_disable, the stack is still intact, so if the issue is simply
related to having any tracers active at the call_with_stack invocation, we'd
be better off disabling tracing here altogether.
Will
> Signed-off-by: Sebastian Capella <[email protected]>
> Cc: Russell King <[email protected]>
> Cc: Andrew Morton <[email protected]>
> Cc: Will Deacon <[email protected]>
> Cc: Robin Holt <[email protected]>
> Cc: Thomas Gleixner <[email protected]>
> Cc: Konstantin Khlebnikov <[email protected]>
> Cc: Steven Capper <[email protected]>
> Cc: Stephen Warren <[email protected]>
> Cc: Tejun Heo <[email protected]>
> ---
> arch/arm/kernel/process.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/arch/arm/kernel/process.c b/arch/arm/kernel/process.c
> index 92f7b15..f58b723 100644
> --- a/arch/arm/kernel/process.c
> +++ b/arch/arm/kernel/process.c
> @@ -100,7 +100,7 @@ void soft_restart(unsigned long addr)
> u64 *stack = soft_restart_stack + ARRAY_SIZE(soft_restart_stack);
>
> /* Disable interrupts first */
> - local_irq_disable();
> + raw_local_irq_disable();
> local_fiq_disable();
>
> /* Disable the L2 if we're the last man standing. */
> --
> 1.7.9.5
>
>
Hi Will,
On 14 April 2014 03:53, Will Deacon <[email protected]> wrote:
> On Tue, Mar 18, 2014 at 09:40:57PM +0000, Sebastian Capella wrote:
>> Use of tracers in local_irq_disable is causes abort loops when called
>> with irqs disabled using a temporary stack. Replace local_irq_disable
>> with raw_local_irq_disable instead to avoid tracers.
>
> Do you have any more information about these aborts? At the time we call
> local_irq_disable, the stack is still intact, so if the issue is simply
> related to having any tracers active at the call_with_stack invocation, we'd
> be better off disabling tracing here altogether.
This is specifically for when soft_restart is called in the
hibernation path with tracers enabled. At that point, we've already
switched to a temporary stack, and when we call local_irq_disable,
we'll see this:
In the local_irq_disable, it ends up calling trace_hardirqs_off
(CONFIG_TRACE_IRQFLAGS_SUPPORT is enabled), which calls
trace_hardirqs_off_caller which checks lockdep_recursion in the
current task, but we've switched to a temporary stack with the
call_with_stack, and get_current is returning NULL. This
triggers a data abort, which calls trace_hardirqs_off
again and so on.
We originally had a patch which added a soft restart called
soft_restart_noirq which avoided the irq disable.
You can see the discussion here:
https://patchwork.kernel.org/patch/3677591/
Thanks!
Sebastian