ARM: OMAP2+: Fix suspcious RCU usage splats for omap_enter_idle_coupled
Subsystems (3)
Tony Lindgren
maintainerPATCH 1/1 ARM: OMAP2+: Fix suspcious RCU usage splats for omap_enter_idle_coupled
We get suspcious RCU usage splats with cpuidle in several places in omap_enter_idle_coupled() with the kernel debug options enabled:
RCU used illegally from extended quiescent state!
...
(_raw_spin_lock_irqsave)
(omap_enter_idle_coupled+0x17c/0x2d8)
(omap_enter_idle_coupled)
(cpuidle_enter_state)
(cpuidle_enter_state_coupled)
(cpuidle_enter)
Let's use RCU_NONIDLE to suppress these splats. Things got changed around with commit 1098582a0f6c ("sched,idle,rcu: Push rcu_idle deeper into the idle path") that started triggering these warnings.
For the tick_broadcast related calls, ideally we'd just switch over to using CPUIDLE_FLAG_TIMER_STOP for omap_enter_idle_coupled() to have the generic cpuidle code handle the tick_broadcast related calls for us and then just drop the tick_broadcast calls here.
But we're currently missing the call in the common cpuidle code for tick_broadcast_enable() that CPU1 hotplug needs as described in earlier commit 50d6b3cf9403 ("ARM: OMAP2+: fix lack of timer interrupts on CPU1 after hotplug").
--- arch/arm/mach-omap2/cpuidle44xx.c | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-)
arch/arm/mach-omap2/cpuidle44xx.c +8 -8
@@ -151,10 +151,10 @@ static int omap_enter_idle_coupled(struct cpuidle_device *dev, (cx->mpu_logic_state ==PWRDM_POWER_OFF );/* Enter broadcast mode for periodic timers */ -tick_broadcast_enable (); +RCU_NONIDLE (tick_broadcast_enable ());/* Enter broadcast mode for one-shot timers */ -tick_broadcast_enter (); +RCU_NONIDLE (tick_broadcast_enter ());/* * Call idle CPU PM enter notifier chain so that
@@ -166,7 +166,7 @@ static int omap_enter_idle_coupled(struct cpuidle_device *dev,if (dev->cpu ==0 ) {pwrdm_set_logic_retst (mpu_pd, cx->mpu_logic_state); -omap_set_pwrdm_state (mpu_pd, cx->mpu_state); +RCU_NONIDLE (omap_set_pwrdm_state (mpu_pd, cx->mpu_state));/* * Call idle CPU cluster PM enter notifier chain
@@ -179,7 +179,7 @@ static int omap_enter_idle_coupled(struct cpuidle_device *dev, index =1 ; cx = state_ptr + index;pwrdm_set_logic_retst (mpu_pd, cx->mpu_logic_state); -omap_set_pwrdm_state (mpu_pd, cx->mpu_state); +RCU_NONIDLE (omap_set_pwrdm_state (mpu_pd, cx->mpu_state)); mpuss_can_lose_context =0 ; } }
@@ -195,9 +195,9 @@ static int omap_enter_idle_coupled(struct cpuidle_device *dev, mpuss_can_lose_context)gic_dist_disable (); -clkdm_deny_idle (cpu_clkdm[1 ]); -omap_set_pwrdm_state (cpu_pd[1 ],PWRDM_POWER_ON ); -clkdm_allow_idle (cpu_clkdm[1 ]); +RCU_NONIDLE (clkdm_deny_idle (cpu_clkdm[1 ])); +RCU_NONIDLE (omap_set_pwrdm_state (cpu_pd[1 ],PWRDM_POWER_ON )); +RCU_NONIDLE (clkdm_allow_idle (cpu_clkdm[1 ]));if (IS_PM44XX_ERRATUM (PM_OMAP4_ROM_SMP_BOOT_ERRATUM_GICD ) && mpuss_can_lose_context) {
@@ -225,7 +225,7 @@ static int omap_enter_idle_coupled(struct cpuidle_device *dev,cpu_pm_exit (); cpu_pm_out: -tick_broadcast_exit (); +RCU_NONIDLE (tick_broadcast_exit ()); fail:cpuidle_coupled_parallel_barrier (dev, &abort_barrier);
Signature · 1 line
2.30.0
Quoted 26 lines · Tony Lindgren
We get suspcious RCU usage splats with cpuidle in several places in
omap_enter_idle_coupled()with the kernel debug options enabled:RCU used illegally from extended quiescent state!
...
(_raw_spin_lock_irqsave)
(omap_enter_idle_coupled+0x17c/0x2d8)
(omap_enter_idle_coupled)
(cpuidle_enter_state)
(cpuidle_enter_state_coupled)
(cpuidle_enter)Let's use
RCU_NONIDLEto suppress these splats. Things got changed around with commit1098582a0f6c("sched,idle,rcu: Pushrcu_idledeeper into the idle path") that started triggering these warnings.For the
tick_broadcastrelated calls, ideally we'd just switch over to usingCPUIDLE_FLAG_TIMER_STOPforomap_enter_idle_coupled()to have the generic cpuidle code handle thetick_broadcastrelated calls for us and then just drop thetick_broadcastcalls here.But we're currently missing the call in the common cpuidle code for
tick_broadcast_enable()that CPU1 hotplug needs as described in earlier commit50d6b3cf9403("ARM: OMAP2+: fix lack of timer interrupts on CPU1 after hotplug").
Quoted 4 lines
Cc Daniel Lezcano <[email protected]>Cc Paul E. McKenney <[email protected]>Cc Russell King <[email protected]>Signed-off-by Tony Lindgren <[email protected]>
From an RCU viewpoint:
Quoted 62 lines
--- arch/arm/mach-omap2/cpuidle44xx.c | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-)arch/arm/mach-omap2/cpuidle44xx.c +8
-8@@ -151,10 +151,10 @@ static int omap_enter_idle_coupled(struct cpuidle_device *dev, (cx->mpu_logic_state ==PWRDM_POWER_OFF );/* Enter broadcast mode for periodic timers */ -tick_broadcast_enable (); +RCU_NONIDLE (tick_broadcast_enable ());/* Enter broadcast mode for one-shot timers */ -tick_broadcast_enter (); +RCU_NONIDLE (tick_broadcast_enter ());/* * Call idle CPU PM enter notifier chain so that @@ -166,7 +166,7 @@ static int omap_enter_idle_coupled(struct cpuidle_device *dev,if (dev->cpu ==0 ) {pwrdm_set_logic_retst (mpu_pd, cx->mpu_logic_state); -omap_set_pwrdm_state (mpu_pd, cx->mpu_state); +RCU_NONIDLE (omap_set_pwrdm_state (mpu_pd, cx->mpu_state));/* * Call idle CPU cluster PM enter notifier chain @@ -179,7 +179,7 @@ static int omap_enter_idle_coupled(struct cpuidle_device *dev, index =1 ; cx = state_ptr + index;pwrdm_set_logic_retst (mpu_pd, cx->mpu_logic_state); -omap_set_pwrdm_state (mpu_pd, cx->mpu_state); +RCU_NONIDLE (omap_set_pwrdm_state (mpu_pd, cx->mpu_state)); mpuss_can_lose_context =0 ; } }@@ -195,9 +195,9 @@ static int omap_enter_idle_coupled(struct cpuidle_device *dev, mpuss_can_lose_context)gic_dist_disable (); -clkdm_deny_idle (cpu_clkdm[1 ]); -omap_set_pwrdm_state (cpu_pd[1 ],PWRDM_POWER_ON ); -clkdm_allow_idle (cpu_clkdm[1 ]); +RCU_NONIDLE (clkdm_deny_idle (cpu_clkdm[1 ])); +RCU_NONIDLE (omap_set_pwrdm_state (cpu_pd[1 ],PWRDM_POWER_ON )); +RCU_NONIDLE (clkdm_allow_idle (cpu_clkdm[1 ]));if (IS_PM44XX_ERRATUM (PM_OMAP4_ROM_SMP_BOOT_ERRATUM_GICD ) && mpuss_can_lose_context) {@@ -225,7 +225,7 @@ static int omap_enter_idle_coupled(struct cpuidle_device *dev,cpu_pm_exit (); cpu_pm_out: -tick_broadcast_exit (); +RCU_NONIDLE (tick_broadcast_exit ()); fail:cpuidle_coupled_parallel_barrier (dev, &abort_barrier);Signature · 1 line
2.30.0