On Tue, Dec 22, 2020 at 05:19:51PM +0100, Rafael J. Wysocki wrote:
On 12/22/2020 2:37 AM, Frederic Weisbecker wrote:
With Paul, we've been thinking that the idle loop wasn't twisted enough yet to deserve 2020.
rcutorture, after some recent parameter changes, has been complaining about a hung task.
It appears that rcu_idle_enter() may wake up a NOCB kthread but this happens after the last generic need_resched() check. Some cpuidle drivers fix it by chance but many others don't.
Here is a proposed bunch of fixes. I will need to also fix the rcu_user_enter() case, likely using irq_work, since nohz_full requires irq work to support self IPI.
Also more generally, this raise the question of local task wake_up() under disabled interrupts. When a wake up occurs in a preempt disabled section, it gets handled by the outer preempt_enable() call. There is no similar mechanism when a wake up occurs with interrupts disabled. I guess it is assumed to be handled, at worst, in the next tick. But a local irq work would provide instant preemption once IRQs are re-enabled. Of course this would only make sense in CONFIG_PREEMPTION, and when the tick is disabled...
git://git.kernel.org/pub/scm/linux/kernel/git/frederic/linux-dynticks.git sched/idle
HEAD: f2fa6e4a070c1535b9edc9ee097167fd2b15d235
Thanks, Frederic
Frederic Weisbecker (4): sched/idle: Fix missing need_resched() check after rcu_idle_enter() cpuidle: Fix missing need_resched() check after rcu_idle_enter() ARM: imx6q: Fix missing need_resched() check after rcu_idle_enter() ACPI: processor: Fix missing need_resched() check after rcu_idle_enter()
arch/arm/mach-imx/cpuidle-imx6q.c | 7 ++++++- drivers/acpi/processor_idle.c | 10 ++++++++-- drivers/cpuidle/cpuidle.c | 33 +++++++++++++++++++++++++-------- kernel/sched/idle.c | 18 ++++++++++++------ 4 files changed, 51 insertions(+), 17 deletions(-)
Please feel free to add
Reviewed-by: Rafael J. Wysocki rafael.j.wysocki@intel.com
to all patches in the series.
I would guess that they will take some other path to mainline, but I have queued these to cut down on rcutorture's whining. ;-)
Thanx, Paul