| Message ID | 20200930112023.121821-1-ulf.hansson@linaro.org |
|---|---|
| State | New |
| Headers | show |
| Series | ARM: imx6q: Fixup RCU usage for cpuidle | expand |
On Wed, Sep 30, 2020 at 01:20:23PM +0200, Ulf Hansson wrote: > The commit eb1f00237aca ("lockdep,trace: Expose tracepoints"), started to > expose us for tracepoints. For imx6q cpuidle, this leads to an RCU splat > according to below. > > [6.870684] [<c0db7690>] (_raw_spin_lock) from [<c011f6a4>] (imx6q_enter_wait+0x18/0x9c) > [6.878846] [<c011f6a4>] (imx6q_enter_wait) from [<c09abfb0>] (cpuidle_enter_state+0x168/0x5e4) > > To fix the problem, let's assign the corresponding idlestate->flags the > CPUIDLE_FLAG_RCU_IDLE bit, which enables us to call rcu_idle_enter|exit() > at the proper point. > > Reported-by: Dong Aisheng <aisheng.dong@nxp.com> > Suggested-by: Peter Zijlstra <peterz@infradead.org> > Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org> Yep, that looks about right. Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> > --- > arch/arm/mach-imx/cpuidle-imx6q.c | 4 +++- > 1 file changed, 3 insertions(+), 1 deletion(-) > > diff --git a/arch/arm/mach-imx/cpuidle-imx6q.c b/arch/arm/mach-imx/cpuidle-imx6q.c > index 24dd5bbe60e4..094337dc1bc7 100644 > --- a/arch/arm/mach-imx/cpuidle-imx6q.c > +++ b/arch/arm/mach-imx/cpuidle-imx6q.c > @@ -24,7 +24,9 @@ static int imx6q_enter_wait(struct cpuidle_device *dev, > imx6_set_lpm(WAIT_UNCLOCKED); > raw_spin_unlock(&cpuidle_lock); > > + rcu_idle_enter(); > cpu_do_idle(); > + rcu_idle_exit(); > > raw_spin_lock(&cpuidle_lock); > if (num_idle_cpus-- == num_online_cpus()) > @@ -44,7 +46,7 @@ static struct cpuidle_driver imx6q_cpuidle_driver = { > { > .exit_latency = 50, > .target_residency = 75, > - .flags = CPUIDLE_FLAG_TIMER_STOP, > + .flags = CPUIDLE_FLAG_TIMER_STOP | CPUIDLE_FLAG_RCU_IDLE, > .enter = imx6q_enter_wait, > .name = "WAIT", > .desc = "Clock off", > -- > 2.25.1 >
On Wed, Sep 30, 2020 at 1:33 PM Peter Zijlstra <peterz@infradead.org> wrote: > > On Wed, Sep 30, 2020 at 01:20:23PM +0200, Ulf Hansson wrote: > > The commit eb1f00237aca ("lockdep,trace: Expose tracepoints"), started to > > expose us for tracepoints. For imx6q cpuidle, this leads to an RCU splat > > according to below. > > > > [6.870684] [<c0db7690>] (_raw_spin_lock) from [<c011f6a4>] (imx6q_enter_wait+0x18/0x9c) > > [6.878846] [<c011f6a4>] (imx6q_enter_wait) from [<c09abfb0>] (cpuidle_enter_state+0x168/0x5e4) > > > > To fix the problem, let's assign the corresponding idlestate->flags the > > CPUIDLE_FLAG_RCU_IDLE bit, which enables us to call rcu_idle_enter|exit() > > at the proper point. > > > > Reported-by: Dong Aisheng <aisheng.dong@nxp.com> > > Suggested-by: Peter Zijlstra <peterz@infradead.org> > > Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org> > > Yep, that looks about right. > > Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> > > > --- > > arch/arm/mach-imx/cpuidle-imx6q.c | 4 +++- > > 1 file changed, 3 insertions(+), 1 deletion(-) > > > > diff --git a/arch/arm/mach-imx/cpuidle-imx6q.c b/arch/arm/mach-imx/cpuidle-imx6q.c > > index 24dd5bbe60e4..094337dc1bc7 100644 > > --- a/arch/arm/mach-imx/cpuidle-imx6q.c > > +++ b/arch/arm/mach-imx/cpuidle-imx6q.c > > @@ -24,7 +24,9 @@ static int imx6q_enter_wait(struct cpuidle_device *dev, > > imx6_set_lpm(WAIT_UNCLOCKED); > > raw_spin_unlock(&cpuidle_lock); > > > > + rcu_idle_enter(); > > cpu_do_idle(); > > + rcu_idle_exit(); > > > > raw_spin_lock(&cpuidle_lock); > > if (num_idle_cpus-- == num_online_cpus()) > > @@ -44,7 +46,7 @@ static struct cpuidle_driver imx6q_cpuidle_driver = { > > { > > .exit_latency = 50, > > .target_residency = 75, > > - .flags = CPUIDLE_FLAG_TIMER_STOP, > > + .flags = CPUIDLE_FLAG_TIMER_STOP | CPUIDLE_FLAG_RCU_IDLE, > > .enter = imx6q_enter_wait, > > .name = "WAIT", > > .desc = "Clock off", > > -- Applied as 5.9-rc8 material, thanks!
diff --git a/arch/arm/mach-imx/cpuidle-imx6q.c b/arch/arm/mach-imx/cpuidle-imx6q.c index 24dd5bbe60e4..094337dc1bc7 100644 --- a/arch/arm/mach-imx/cpuidle-imx6q.c +++ b/arch/arm/mach-imx/cpuidle-imx6q.c @@ -24,7 +24,9 @@ static int imx6q_enter_wait(struct cpuidle_device *dev, imx6_set_lpm(WAIT_UNCLOCKED); raw_spin_unlock(&cpuidle_lock); + rcu_idle_enter(); cpu_do_idle(); + rcu_idle_exit(); raw_spin_lock(&cpuidle_lock); if (num_idle_cpus-- == num_online_cpus()) @@ -44,7 +46,7 @@ static struct cpuidle_driver imx6q_cpuidle_driver = { { .exit_latency = 50, .target_residency = 75, - .flags = CPUIDLE_FLAG_TIMER_STOP, + .flags = CPUIDLE_FLAG_TIMER_STOP | CPUIDLE_FLAG_RCU_IDLE, .enter = imx6q_enter_wait, .name = "WAIT", .desc = "Clock off",
The commit eb1f00237aca ("lockdep,trace: Expose tracepoints"), started to expose us for tracepoints. For imx6q cpuidle, this leads to an RCU splat according to below. [6.870684] [<c0db7690>] (_raw_spin_lock) from [<c011f6a4>] (imx6q_enter_wait+0x18/0x9c) [6.878846] [<c011f6a4>] (imx6q_enter_wait) from [<c09abfb0>] (cpuidle_enter_state+0x168/0x5e4) To fix the problem, let's assign the corresponding idlestate->flags the CPUIDLE_FLAG_RCU_IDLE bit, which enables us to call rcu_idle_enter|exit() at the proper point. Reported-by: Dong Aisheng <aisheng.dong@nxp.com> Suggested-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org> --- arch/arm/mach-imx/cpuidle-imx6q.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)