Suspend to idle was found to not work on Goldmont CPU recently.
And the issue was triggered due to:
1. On Goldmont the CPU in idle can only be woken up via IPIs,
not POLLING mode:
Commit 08e237fa56a1 ("x86/cpu: Add workaround for MONITOR
instruction erratum on Goldmont based CPUs")
2. When the CPU is entering suspend to idle process, the
_TIF_POLLING_NRFLAG is kept on, due to cpuidle_enter_s2idle()
doesn't properly match call_cpuidle().
3. Commit b2a02fc43a1f ("smp: Optimize send_call_function_single_ipi()")
makes use of _TIF_POLLING_NRFLAG to avoid sending IPIs to
idle CPUs.
4. As a result, some IPIs related functions might not work
well during suspend to idle on Goldmont. For example, one
suspected victim:
tick_unfreeze() -> timekeeping_resume() -> hrtimers_resume()
-> clock_was_set() -> on_each_cpu() might wait forever,
because the IPIs will not be sent to the CPUs which are
sleeping with _TIF_POLLING_NRFLAG set, and Goldmont CPU
could not be woken up by only setting _TIF_NEED_RESCHED
on the monitor address.
Clear the _TIF_POLLING_NRFLAG flag before entering suspend to idle,
and let the driver's enter_s2idle() to decide whether to set
_TIF_POLLING_NRFLAG or not. So that to avoid the scenario described
above and keep the context consistent with before.
Fixes: b2a02fc43a1f ("smp: Optimize send_call_function_single_ipi()")
Reported-by: kbuild test robot <[email protected]>
Cc: "Rafael J. Wysocki" <[email protected]>
Cc: Peter Zijlstra (Intel) <[email protected]>
Cc: Len Brown <[email protected]>
Signed-off-by: Chen Yu <[email protected]>
---
v2: According to Peter's review, v1 is racy, if someone already
set TIF_NEED_RESCHED this patch just clear POLLING and go to sleep.
Check TIF_NEED_RESCHED before entering suspend to idle and
adjust the naming to be consistent with call_cpuidle().
v3: According to Rafael, it would be better to do the simplest fix
first and then do the cleanup on top of it.
---
drivers/cpuidle/cpuidle.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/drivers/cpuidle/cpuidle.c b/drivers/cpuidle/cpuidle.c
index c149d9e20dfd..e092789187c6 100644
--- a/drivers/cpuidle/cpuidle.c
+++ b/drivers/cpuidle/cpuidle.c
@@ -13,6 +13,7 @@
#include <linux/mutex.h>
#include <linux/sched.h>
#include <linux/sched/clock.h>
+#include <linux/sched/idle.h>
#include <linux/notifier.h>
#include <linux/pm_qos.h>
#include <linux/cpu.h>
@@ -186,7 +187,7 @@ int cpuidle_enter_s2idle(struct cpuidle_driver *drv, struct cpuidle_device *dev)
* be frozen safely.
*/
index = find_deepest_state(drv, dev, U64_MAX, 0, true);
- if (index > 0)
+ if (index > 0 && !current_clr_polling_and_test())
enter_s2idle_proper(drv, dev, index);
return index;
--
2.17.1
On Tue, Jun 23, 2020 at 8:30 AM Chen Yu <[email protected]> wrote:
>
> Suspend to idle was found to not work on Goldmont CPU recently.
> And the issue was triggered due to:
>
> 1. On Goldmont the CPU in idle can only be woken up via IPIs,
> not POLLING mode:
> Commit 08e237fa56a1 ("x86/cpu: Add workaround for MONITOR
> instruction erratum on Goldmont based CPUs")
> 2. When the CPU is entering suspend to idle process, the
> _TIF_POLLING_NRFLAG is kept on, due to cpuidle_enter_s2idle()
> doesn't properly match call_cpuidle().
> 3. Commit b2a02fc43a1f ("smp: Optimize send_call_function_single_ipi()")
> makes use of _TIF_POLLING_NRFLAG to avoid sending IPIs to
> idle CPUs.
> 4. As a result, some IPIs related functions might not work
> well during suspend to idle on Goldmont. For example, one
> suspected victim:
> tick_unfreeze() -> timekeeping_resume() -> hrtimers_resume()
> -> clock_was_set() -> on_each_cpu() might wait forever,
> because the IPIs will not be sent to the CPUs which are
> sleeping with _TIF_POLLING_NRFLAG set, and Goldmont CPU
> could not be woken up by only setting _TIF_NEED_RESCHED
> on the monitor address.
>
> Clear the _TIF_POLLING_NRFLAG flag before entering suspend to idle,
> and let the driver's enter_s2idle() to decide whether to set
> _TIF_POLLING_NRFLAG or not. So that to avoid the scenario described
> above and keep the context consistent with before.
>
> Fixes: b2a02fc43a1f ("smp: Optimize send_call_function_single_ipi()")
> Reported-by: kbuild test robot <[email protected]>
> Cc: "Rafael J. Wysocki" <[email protected]>
> Cc: Peter Zijlstra (Intel) <[email protected]>
> Cc: Len Brown <[email protected]>
> Signed-off-by: Chen Yu <[email protected]>
Applied (based on the previous discussion) with some subject and
changelog edits.
Thanks!
> ---
> v2: According to Peter's review, v1 is racy, if someone already
> set TIF_NEED_RESCHED this patch just clear POLLING and go to sleep.
> Check TIF_NEED_RESCHED before entering suspend to idle and
> adjust the naming to be consistent with call_cpuidle().
>
> v3: According to Rafael, it would be better to do the simplest fix
> first and then do the cleanup on top of it.
> ---
> drivers/cpuidle/cpuidle.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/cpuidle/cpuidle.c b/drivers/cpuidle/cpuidle.c
> index c149d9e20dfd..e092789187c6 100644
> --- a/drivers/cpuidle/cpuidle.c
> +++ b/drivers/cpuidle/cpuidle.c
> @@ -13,6 +13,7 @@
> #include <linux/mutex.h>
> #include <linux/sched.h>
> #include <linux/sched/clock.h>
> +#include <linux/sched/idle.h>
> #include <linux/notifier.h>
> #include <linux/pm_qos.h>
> #include <linux/cpu.h>
> @@ -186,7 +187,7 @@ int cpuidle_enter_s2idle(struct cpuidle_driver *drv, struct cpuidle_device *dev)
> * be frozen safely.
> */
> index = find_deepest_state(drv, dev, U64_MAX, 0, true);
> - if (index > 0)
> + if (index > 0 && !current_clr_polling_and_test())
> enter_s2idle_proper(drv, dev, index);
>
> return index;
> --
> 2.17.1
>