| Message ID | 20200615173611.15349-1-yu.c.chen@intel.com (mailing list archive) |
|---|---|
| State | Superseded, archived |
| Headers | show |
| Series | [RFC] PM / s2idle: Clear _TIF_POLLING_NRFLAG before suspend to idle | expand |
On Tue, Jun 16, 2020 at 01:36:11AM +0800, Chen Yu 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 POLL 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. > 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. *sigh*... just what we need. > @@ -186,8 +187,10 @@ 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(); > enter_s2idle_proper(drv, dev, index); > + } > > return index; > } So how is that commit 08e237fa56a1 not suffient? That makes mwait_idle_with_hints() DTRT for this 'functionally challenged' piece of hardware. AFAICT intel_enter_s2idle() uses mwait_idle_with_hints(). What am I missing?
On Mon, Jun 15, 2020 at 08:40:41PM +0200, Peter Zijlstra wrote: > > @@ -186,8 +187,10 @@ 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(); > > enter_s2idle_proper(drv, dev, index); > > + } > > > > return index; > > } > > So how is that commit 08e237fa56a1 not suffient? That makes > mwait_idle_with_hints() DTRT for this 'functionally challenged' piece of > hardware. > > AFAICT intel_enter_s2idle() uses mwait_idle_with_hints(). > > What am I missing? What's missing is that cpuidle_enter_s2idle() doesn't properly match call_cpuidle(). Something like so then. Your version is racy, if someone already set TIF_NEED_RESCHED you just clear POLLING and go to sleep. --- diff --git a/drivers/cpuidle/cpuidle.c b/drivers/cpuidle/cpuidle.c index c149d9e20dfd..81bee8d03c6d 100644 --- a/drivers/cpuidle/cpuidle.c +++ b/drivers/cpuidle/cpuidle.c @@ -133,8 +133,8 @@ int cpuidle_find_deepest_state(struct cpuidle_driver *drv, } #ifdef CONFIG_SUSPEND -static void enter_s2idle_proper(struct cpuidle_driver *drv, - struct cpuidle_device *dev, int index) +static void s2idle_enter(struct cpuidle_driver *drv, + struct cpuidle_device *dev, int index) { ktime_t time_start, time_end; @@ -168,6 +168,15 @@ static void enter_s2idle_proper(struct cpuidle_driver *drv, dev->states_usage[index].s2idle_usage++; } +static int call_s2idle(struct cpuidle_driver *drv, struct cpuidle_device *dev, + int index) +{ + if (!current_clr_polling_and_test()) + s2idle_enter(drv, dev, index); + + return index; +} + /** * cpuidle_enter_s2idle - Enter an idle state suitable for suspend-to-idle. * @drv: cpuidle driver for the given CPU. @@ -187,7 +196,7 @@ int cpuidle_enter_s2idle(struct cpuidle_driver *drv, struct cpuidle_device *dev) */ index = find_deepest_state(drv, dev, U64_MAX, 0, true); if (index > 0) - enter_s2idle_proper(drv, dev, index); + call_s2idle(drv, dev, index); return index; }
On Mon, Jun 15, 2020 at 09:31:54PM +0200, Peter Zijlstra wrote: > On Mon, Jun 15, 2020 at 08:40:41PM +0200, Peter Zijlstra wrote: > > > > @@ -186,8 +187,10 @@ 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(); > > > enter_s2idle_proper(drv, dev, index); > > > + } > > > > > > return index; > > > } > > > > So how is that commit 08e237fa56a1 not suffient? That makes > > mwait_idle_with_hints() DTRT for this 'functionally challenged' piece of > > hardware. > > > > AFAICT intel_enter_s2idle() uses mwait_idle_with_hints(). > > > > What am I missing? > > What's missing is that cpuidle_enter_s2idle() doesn't properly match > call_cpuidle(). > Right. > Something like so then. Your version is racy, if someone already set > TIF_NEED_RESCHED you just clear POLLING and go to sleep. > Got it, I'll test the patch below. Thanks, Chenyu > --- > > diff --git a/drivers/cpuidle/cpuidle.c b/drivers/cpuidle/cpuidle.c > index c149d9e20dfd..81bee8d03c6d 100644 > --- a/drivers/cpuidle/cpuidle.c > +++ b/drivers/cpuidle/cpuidle.c > @@ -133,8 +133,8 @@ int cpuidle_find_deepest_state(struct cpuidle_driver *drv, > } > > #ifdef CONFIG_SUSPEND > -static void enter_s2idle_proper(struct cpuidle_driver *drv, > - struct cpuidle_device *dev, int index) > +static void s2idle_enter(struct cpuidle_driver *drv, > + struct cpuidle_device *dev, int index) > { > ktime_t time_start, time_end; > > @@ -168,6 +168,15 @@ static void enter_s2idle_proper(struct cpuidle_driver *drv, > dev->states_usage[index].s2idle_usage++; > } > > +static int call_s2idle(struct cpuidle_driver *drv, struct cpuidle_device *dev, > + int index) > +{ > + if (!current_clr_polling_and_test()) > + s2idle_enter(drv, dev, index); > + > + return index; > +} > + > /** > * cpuidle_enter_s2idle - Enter an idle state suitable for suspend-to-idle. > * @drv: cpuidle driver for the given CPU. > @@ -187,7 +196,7 @@ int cpuidle_enter_s2idle(struct cpuidle_driver *drv, struct cpuidle_device *dev) > */ > index = find_deepest_state(drv, dev, U64_MAX, 0, true); > if (index > 0) > - enter_s2idle_proper(drv, dev, index); > + call_s2idle(drv, dev, index); > > return index; > }
diff --git a/drivers/cpuidle/cpuidle.c b/drivers/cpuidle/cpuidle.c index c149d9e20dfd..d17dad362d34 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> @@ -186,8 +187,10 @@ 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(); enter_s2idle_proper(drv, dev, index); + } return index; }
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 POLL 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. 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. I don't find a way in Ubuntu to update the firmware of Goldmont and check if the issue was gone, a fix patch would do no harm. 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 <lkp@intel.com> Cc: Peter Zijlstra (Intel) <peterz@infradead.org> Signed-off-by: Chen Yu <yu.c.chen@intel.com> --- drivers/cpuidle/cpuidle.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)