| Message ID | E1gX31P-0004wA-Eq@rmk-PC.armlinux.org.uk (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [RFC] ARM: omap2: fix lack of timer interrupts on CPU1 after hotplug | expand |
* Russell King <rmk+kernel@armlinux.org.uk> [181212 11:49]: > We call tick_broadcast_enter() in omap_enter_idle_coupled(), which one > would expect would take care of the issue, but internally this only > deals with one-shot local timers - tick_broadcast_enable() on the other > hand only deals with periodic local timers. So, we need to call both. This works just fine for me for NO_HZ and cpuidle: Tested-by: Tony Lindgren <tony@atomide.com>
* Tony Lindgren <tony@atomide.com> [181212 17:53]: > * Russell King <rmk+kernel@armlinux.org.uk> [181212 11:49]: > > We call tick_broadcast_enter() in omap_enter_idle_coupled(), which one > > would expect would take care of the issue, but internally this only > > deals with one-shot local timers - tick_broadcast_enable() on the other > > hand only deals with periodic local timers. So, we need to call both. > > This works just fine for me for NO_HZ and cpuidle: > > Tested-by: Tony Lindgren <tony@atomide.com> And also works for me without NO_HZ too FYI. Tony
On Wed, Dec 12, 2018 at 10:57:41AM -0800, Tony Lindgren wrote: > * Tony Lindgren <tony@atomide.com> [181212 17:53]: > > * Russell King <rmk+kernel@armlinux.org.uk> [181212 11:49]: > > > We call tick_broadcast_enter() in omap_enter_idle_coupled(), which one > > > would expect would take care of the issue, but internally this only > > > deals with one-shot local timers - tick_broadcast_enable() on the other > > > hand only deals with periodic local timers. So, we need to call both. > > > > This works just fine for me for NO_HZ and cpuidle: > > > > Tested-by: Tony Lindgren <tony@atomide.com> > > And also works for me without NO_HZ too FYI. Hi Tony, Could you pick this patch up and submit for 5.0-rc please? Thanks.
* Russell King - ARM Linux admin <linux@armlinux.org.uk> [190207 13:46]: > On Wed, Dec 12, 2018 at 10:57:41AM -0800, Tony Lindgren wrote: > > * Tony Lindgren <tony@atomide.com> [181212 17:53]: > > > * Russell King <rmk+kernel@armlinux.org.uk> [181212 11:49]: > > > > We call tick_broadcast_enter() in omap_enter_idle_coupled(), which one > > > > would expect would take care of the issue, but internally this only > > > > deals with one-shot local timers - tick_broadcast_enable() on the other > > > > hand only deals with periodic local timers. So, we need to call both. > > > > > > This works just fine for me for NO_HZ and cpuidle: > > > > > > Tested-by: Tony Lindgren <tony@atomide.com> > > > > And also works for me without NO_HZ too FYI. > > Hi Tony, > > Could you pick this patch up and submit for 5.0-rc please? Oh sorry. I thought you had already merged it since I added my Tested-by. Hmm I guess I need to start adding also an Acked-by to signal it's OK to pick some misc patches. Anyways, applying into omap-for-v5.0/fixes-v2. Thanks, Tony
diff --git a/arch/arm/mach-omap2/cpuidle44xx.c b/arch/arm/mach-omap2/cpuidle44xx.c index a8b291f00109..dae514c8276a 100644 --- a/arch/arm/mach-omap2/cpuidle44xx.c +++ b/arch/arm/mach-omap2/cpuidle44xx.c @@ -152,6 +152,10 @@ static int omap_enter_idle_coupled(struct cpuidle_device *dev, mpuss_can_lose_context = (cx->mpu_state == PWRDM_POWER_RET) && (cx->mpu_logic_state == PWRDM_POWER_OFF); + /* Enter broadcast mode for periodic timers */ + tick_broadcast_enable(); + + /* Enter broadcast mode for one-shot timers */ tick_broadcast_enter(); /* @@ -218,15 +222,6 @@ static int omap_enter_idle_coupled(struct cpuidle_device *dev, return index; } -/* - * For each cpu, setup the broadcast timer because local timers - * stops for the states above C1. - */ -static void omap_setup_broadcast_timer(void *arg) -{ - tick_broadcast_enable(); -} - static struct cpuidle_driver omap4_idle_driver = { .name = "omap4_idle", .owner = THIS_MODULE, @@ -319,8 +314,5 @@ int __init omap4_idle_init(void) if (!cpu_clkdm[0] || !cpu_clkdm[1]) return -ENODEV; - /* Configure the broadcast timer on each cpu */ - on_each_cpu(omap_setup_broadcast_timer, NULL, 1); - return cpuidle_register(idle_driver, cpu_online_mask); }
If we have a kernel configured for periodic timer interrupts, and we have cpuidle enabled, then we end up with CPU1 losing timer interupts after a hotplug. This can manifest itself in RCU stall warnings, or userspace becoming unresponsive. The problem is that the kernel initially wants to use the TWD timer for interrupts, but the TWD loses context when we enter the C3 cpuidle state. Nothing reprograms the TWD after idle. We have solved this in the past by switching to broadcast timer ticks, and cpuidle44xx switches to that mode at boot time. However, there is nothing to switch from periodic mode local timers after a hotplug operation. We call tick_broadcast_enter() in omap_enter_idle_coupled(), which one would expect would take care of the issue, but internally this only deals with one-shot local timers - tick_broadcast_enable() on the other hand only deals with periodic local timers. So, we need to call both. Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk> --- arch/arm/mach-omap2/cpuidle44xx.c | 16 ++++------------ 1 file changed, 4 insertions(+), 12 deletions(-)