From 7c4a976cd55972b68c75a978f171b6db5df4ce66 Mon Sep 17 00:00:00 2001 From: Viresh Kumar Date: Tue, 7 Jul 2015 10:14:35 +0200 Subject: clockevents: Allow set-state callbacks to be optional Its mandatory for the drivers to provide set_state_{oneshot|periodic}() (only if related modes are supported) and set_state_shutdown() callbacks today, if they are implementing the new set-state interface. But this leads to unnecessary noop callbacks for drivers which don't want to implement them. Over that, it will lead to a full function call for nothing really useful. Lets make all set-state callbacks optional. Suggested-by: Daniel Lezcano Signed-off-by: Viresh Kumar Signed-off-by: Daniel Lezcano Link: http://lkml.kernel.org/r/1436256875-15562-1-git-send-email-daniel.lezcano@linaro.org Signed-off-by: Thomas Gleixner --- kernel/time/clockevents.c | 24 +++++++++--------------- 1 file changed, 9 insertions(+), 15 deletions(-) (limited to 'kernel') diff --git a/kernel/time/clockevents.c b/kernel/time/clockevents.c index 08ccc3da3ca0..50eb107f1198 100644 --- a/kernel/time/clockevents.c +++ b/kernel/time/clockevents.c @@ -120,19 +120,25 @@ static int __clockevents_switch_state(struct clock_event_device *dev, /* The clockevent device is getting replaced. Shut it down. */ case CLOCK_EVT_STATE_SHUTDOWN: - return dev->set_state_shutdown(dev); + if (dev->set_state_shutdown) + return dev->set_state_shutdown(dev); + return 0; case CLOCK_EVT_STATE_PERIODIC: /* Core internal bug */ if (!(dev->features & CLOCK_EVT_FEAT_PERIODIC)) return -ENOSYS; - return dev->set_state_periodic(dev); + if (dev->set_state_periodic) + return dev->set_state_periodic(dev); + return 0; case CLOCK_EVT_STATE_ONESHOT: /* Core internal bug */ if (!(dev->features & CLOCK_EVT_FEAT_ONESHOT)) return -ENOSYS; - return dev->set_state_oneshot(dev); + if (dev->set_state_oneshot) + return dev->set_state_oneshot(dev); + return 0; case CLOCK_EVT_STATE_ONESHOT_STOPPED: /* Core internal bug */ @@ -471,18 +477,6 @@ static int clockevents_sanity_check(struct clock_event_device *dev) if (dev->features & CLOCK_EVT_FEAT_DUMMY) return 0; - /* New state-specific callbacks */ - if (!dev->set_state_shutdown) - return -EINVAL; - - if ((dev->features & CLOCK_EVT_FEAT_PERIODIC) && - !dev->set_state_periodic) - return -EINVAL; - - if ((dev->features & CLOCK_EVT_FEAT_ONESHOT) && - !dev->set_state_oneshot) - return -EINVAL; - return 0; } -- cgit v1.2.3 From 8eb231261fdd20768db23863d00ef277de4b0543 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Tue, 7 Jul 2015 14:11:00 +0200 Subject: tick/broadcast: Prevent hrtimer recursion The hrtimer based broadcast vehicle can cause a hrtimer recursion which went unnoticed until we changed the hrtimer expiry code to keep track of the currently running timer. local_timer_interrupt() local_handler() hrtimer_interrupt() expire_hrtimers() broadcast_hrtimer() send_ipis() local_handler() hrtimer_interrupt() .... Solution is simple: Prevent the local handler call from the broadcast code when the broadcast 'device' is hrtimer based. [ Split out from a larger combo patch ] Tested-by: Sudeep Holla Signed-off-by: Thomas Gleixner Cc: Suzuki Poulose Cc: Lorenzo Pieralisi Cc: Catalin Marinas Cc: Rafael J. Wysocki Cc: Peter Zijlstra Cc: Preeti U Murthy Cc: Ingo Molnar Link: http://lkml.kernel.org/r/alpine.DEB.2.11.1507070929360.3916@nanos --- kernel/time/tick-broadcast.c | 16 +++++++++++++++- 1 file changed, 15 insertions(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/time/tick-broadcast.c b/kernel/time/tick-broadcast.c index d39f32cdd1b5..a76204089f78 100644 --- a/kernel/time/tick-broadcast.c +++ b/kernel/time/tick-broadcast.c @@ -265,8 +265,22 @@ static bool tick_do_broadcast(struct cpumask *mask) * Check, if the current cpu is in the mask */ if (cpumask_test_cpu(cpu, mask)) { + struct clock_event_device *bc = tick_broadcast_device.evtdev; + cpumask_clear_cpu(cpu, mask); - local = true; + /* + * We only run the local handler, if the broadcast + * device is not hrtimer based. Otherwise we run into + * a hrtimer recursion. + * + * local timer_interrupt() + * local_handler() + * expire_hrtimers() + * bc_handler() + * local_handler() + * expire_hrtimers() + */ + local = !(bc->features & CLOCK_EVT_FEAT_HRTIMER); } if (!cpumask_empty(mask)) { -- cgit v1.2.3 From e0454311903d3fd0f12a86c9e65d7b271c2bb05d Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Tue, 7 Jul 2015 14:07:27 +0200 Subject: tick/broadcast: Sanity check the shutdown of the local clock_event The broadcast code shuts down the local clock event unconditionally even if no broadcast device is installed or if the broadcast device is hrtimer based. Add proper sanity checks. [ Split out from a larger combo patch ] Reported-and-tested-by: Sudeep Holla Signed-off-by: Thomas Gleixner Cc: Suzuki Poulose Cc: Lorenzo Pieralisi Cc: Catalin Marinas Cc: Rafael J. Wysocki Cc: Peter Zijlstra Cc: Preeti U Murthy Cc: Ingo Molnar Link: http://lkml.kernel.org/r/alpine.DEB.2.11.1507070929360.3916@nanos --- kernel/time/tick-broadcast.c | 23 ++++++++++++++++------- 1 file changed, 16 insertions(+), 7 deletions(-) (limited to 'kernel') diff --git a/kernel/time/tick-broadcast.c b/kernel/time/tick-broadcast.c index a76204089f78..9877d0b0aefc 100644 --- a/kernel/time/tick-broadcast.c +++ b/kernel/time/tick-broadcast.c @@ -159,7 +159,7 @@ int tick_device_uses_broadcast(struct clock_event_device *dev, int cpu) { struct clock_event_device *bc = tick_broadcast_device.evtdev; unsigned long flags; - int ret; + int ret = 0; raw_spin_lock_irqsave(&tick_broadcast_lock, flags); @@ -221,13 +221,14 @@ int tick_device_uses_broadcast(struct clock_event_device *dev, int cpu) * If we kept the cpu in the broadcast mask, * tell the caller to leave the per cpu device * in shutdown state. The periodic interrupt - * is delivered by the broadcast device. + * is delivered by the broadcast device, if + * the broadcast device exists and is not + * hrtimer based. */ - ret = cpumask_test_cpu(cpu, tick_broadcast_mask); + if (bc && !(bc->features & CLOCK_EVT_FEAT_HRTIMER)) + ret = cpumask_test_cpu(cpu, tick_broadcast_mask); break; default: - /* Nothing to do */ - ret = 0; break; } } @@ -373,8 +374,16 @@ void tick_broadcast_control(enum tick_broadcast_mode mode) case TICK_BROADCAST_ON: cpumask_set_cpu(cpu, tick_broadcast_on); if (!cpumask_test_and_set_cpu(cpu, tick_broadcast_mask)) { - if (tick_broadcast_device.mode == - TICKDEV_MODE_PERIODIC) + /* + * Only shutdown the cpu local device, if: + * + * - the broadcast device exists + * - the broadcast device is not a hrtimer based one + * - the broadcast device is in periodic mode to + * avoid a hickup during switch to oneshot mode + */ + if (bc && !(bc->features & CLOCK_EVT_FEAT_HRTIMER) && + tick_broadcast_device.mode == TICKDEV_MODE_PERIODIC) clockevents_shutdown(dev); } break; -- cgit v1.2.3 From f32dd117051185da6e923b35491a44d7debeeea5 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Tue, 7 Jul 2015 16:29:38 +0200 Subject: tick/broadcast: Make idle check independent from mode and config Currently the broadcast busy check, which prevents the idle code from going into deep idle, works only in one shot mode. If NOHZ and HIGHRES are off (config or command line) there is no sanity check at all, so under certain conditions cpus are allowed to go into deep idle, where the local timer stops, and are not woken up again because there is no broadcast timer installed or a hrtimer based broadcast device is not evaluated. Move tick_broadcast_oneshot_control() into the common code and provide proper subfunctions for the various config combinations. The common check in tick_broadcast_oneshot_control() is for the C3STOP misfeature flag of the local clock event device. If its not set, idle can proceed. If set, further checks are necessary. Provide checks for the trivial cases: - If broadcast is disabled in the config, then return busy - If oneshot mode (NOHZ/HIGHES) is disabled in the config, return busy if the broadcast device is hrtimer based. - If oneshot mode is enabled in the config call the original tick_broadcast_oneshot_control() function. That function needs extra checks which will be implemented in seperate patches. [ Split out from a larger combo patch ] Reported-and-tested-by: Sudeep Holla Signed-off-by: Thomas Gleixner Cc: Suzuki Poulose Cc: Lorenzo Pieralisi Cc: Catalin Marinas Cc: Rafael J. Wysocki Cc: Peter Zijlstra Cc: Preeti U Murthy Cc: Ingo Molnar Link: http://lkml.kernel.org/r/alpine.DEB.2.11.1507070929360.3916@nanos --- include/linux/tick.h | 4 ---- kernel/time/tick-broadcast.c | 26 +++++++++++--------------- kernel/time/tick-common.c | 21 +++++++++++++++++++++ kernel/time/tick-sched.h | 10 ++++++++++ 4 files changed, 42 insertions(+), 19 deletions(-) (limited to 'kernel') diff --git a/include/linux/tick.h b/include/linux/tick.h index 3741ba1a652c..6916dcb61857 100644 --- a/include/linux/tick.h +++ b/include/linux/tick.h @@ -67,11 +67,7 @@ extern void tick_broadcast_control(enum tick_broadcast_mode mode); static inline void tick_broadcast_control(enum tick_broadcast_mode mode) { } #endif /* BROADCAST */ -#if defined(CONFIG_GENERIC_CLOCKEVENTS_BROADCAST) && defined(CONFIG_TICK_ONESHOT) extern int tick_broadcast_oneshot_control(enum tick_broadcast_state state); -#else -static inline int tick_broadcast_oneshot_control(enum tick_broadcast_state state) { return 0; } -#endif static inline void tick_broadcast_enable(void) { diff --git a/kernel/time/tick-broadcast.c b/kernel/time/tick-broadcast.c index 9877d0b0aefc..ef77b16ad5df 100644 --- a/kernel/time/tick-broadcast.c +++ b/kernel/time/tick-broadcast.c @@ -685,18 +685,7 @@ static void broadcast_shutdown_local(struct clock_event_device *bc, clockevents_switch_state(dev, CLOCK_EVT_STATE_SHUTDOWN); } -/** - * tick_broadcast_oneshot_control - Enter/exit broadcast oneshot mode - * @state: The target state (enter/exit) - * - * The system enters/leaves a state, where affected devices might stop - * Returns 0 on success, -EBUSY if the cpu is used to broadcast wakeups. - * - * Called with interrupts disabled, so clockevents_lock is not - * required here because the local clock event device cannot go away - * under us. - */ -int tick_broadcast_oneshot_control(enum tick_broadcast_state state) +int __tick_broadcast_oneshot_control(enum tick_broadcast_state state) { struct clock_event_device *bc, *dev; struct tick_device *td; @@ -717,9 +706,6 @@ int tick_broadcast_oneshot_control(enum tick_broadcast_state state) td = this_cpu_ptr(&tick_cpu_device); dev = td->evtdev; - if (!(dev->features & CLOCK_EVT_FEAT_C3STOP)) - return 0; - raw_spin_lock(&tick_broadcast_lock); bc = tick_broadcast_device.evtdev; cpu = smp_processor_id(); @@ -961,6 +947,16 @@ bool tick_broadcast_oneshot_available(void) return bc ? bc->features & CLOCK_EVT_FEAT_ONESHOT : false; } +#else +int __tick_broadcast_oneshot_control(enum tick_broadcast_state state) +{ + struct clock_event_device *bc = tick_broadcast_device.evtdev; + + if (!bc || (bc->features & CLOCK_EVT_FEAT_HRTIMER)) + return -EBUSY; + + return 0; +} #endif void __init tick_broadcast_init(void) diff --git a/kernel/time/tick-common.c b/kernel/time/tick-common.c index 76446cb5dfe1..55e13efff1ab 100644 --- a/kernel/time/tick-common.c +++ b/kernel/time/tick-common.c @@ -343,6 +343,27 @@ out_bc: tick_install_broadcast_device(newdev); } +/** + * tick_broadcast_oneshot_control - Enter/exit broadcast oneshot mode + * @state: The target state (enter/exit) + * + * The system enters/leaves a state, where affected devices might stop + * Returns 0 on success, -EBUSY if the cpu is used to broadcast wakeups. + * + * Called with interrupts disabled, so clockevents_lock is not + * required here because the local clock event device cannot go away + * under us. + */ +int tick_broadcast_oneshot_control(enum tick_broadcast_state state) +{ + struct tick_device *td = this_cpu_ptr(&tick_cpu_device); + + if (!(td->evtdev->features & CLOCK_EVT_FEAT_C3STOP)) + return 0; + + return __tick_broadcast_oneshot_control(state); +} + #ifdef CONFIG_HOTPLUG_CPU /* * Transfer the do_timer job away from a dying cpu. diff --git a/kernel/time/tick-sched.h b/kernel/time/tick-sched.h index 42fdf4958bcc..a4a8d4e9baa1 100644 --- a/kernel/time/tick-sched.h +++ b/kernel/time/tick-sched.h @@ -71,4 +71,14 @@ extern void tick_cancel_sched_timer(int cpu); static inline void tick_cancel_sched_timer(int cpu) { } #endif +#ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST +extern int __tick_broadcast_oneshot_control(enum tick_broadcast_state state); +#else +static inline int +__tick_broadcast_oneshot_control(enum tick_broadcast_state state) +{ + return -EBUSY; +} +#endif + #endif -- cgit v1.2.3 From b78f3f3c898c824bf56ab55cfa59fc72be49c349 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Tue, 7 Jul 2015 16:34:32 +0200 Subject: tick/broadcast: Prevent deep idle if no broadcast device available Add a check for a installed broadcast device to the oneshot control function and return busy if not. [ Split out from a larger combo patch ] Reported-and-tested-by: Sudeep Holla Signed-off-by: Thomas Gleixner Cc: Suzuki Poulose Cc: Lorenzo Pieralisi Cc: Catalin Marinas Cc: Rafael J. Wysocki Cc: Peter Zijlstra Cc: Preeti U Murthy Cc: Ingo Molnar Link: http://lkml.kernel.org/r/alpine.DEB.2.11.1507070929360.3916@nanos --- kernel/time/tick-broadcast.c | 7 +++++++ 1 file changed, 7 insertions(+) (limited to 'kernel') diff --git a/kernel/time/tick-broadcast.c b/kernel/time/tick-broadcast.c index ef77b16ad5df..fad3f789beec 100644 --- a/kernel/time/tick-broadcast.c +++ b/kernel/time/tick-broadcast.c @@ -692,6 +692,13 @@ int __tick_broadcast_oneshot_control(enum tick_broadcast_state state) int cpu, ret = 0; ktime_t now; + /* + * If there is no broadcast device, tell the caller not to go + * into deep idle. + */ + if (!tick_broadcast_device.evtdev) + return -EBUSY; + /* * Periodic mode does not care about the enter/exit of power * states -- cgit v1.2.3 From e3ac79e087ffe8a1f953ed44a74acf7616cb0b25 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Tue, 7 Jul 2015 16:38:11 +0200 Subject: tick/broadcast: Move the check for periodic mode inside state handling We need to check more than the periodic mode for proper operation in all runtime combinations. To avoid code duplication move the check into the enter state handling. No functional change. [ Split out from a larger combo patch ] Reported-and-tested-by: Sudeep Holla Signed-off-by: Thomas Gleixner Cc: Suzuki Poulose Cc: Lorenzo Pieralisi Cc: Catalin Marinas Cc: Rafael J. Wysocki Cc: Peter Zijlstra Cc: Preeti U Murthy Cc: Ingo Molnar Link: http://lkml.kernel.org/r/alpine.DEB.2.11.1507070929360.3916@nanos --- kernel/time/tick-broadcast.c | 22 ++++++++-------------- 1 file changed, 8 insertions(+), 14 deletions(-) (limited to 'kernel') diff --git a/kernel/time/tick-broadcast.c b/kernel/time/tick-broadcast.c index fad3f789beec..83aa92e87a85 100644 --- a/kernel/time/tick-broadcast.c +++ b/kernel/time/tick-broadcast.c @@ -688,7 +688,6 @@ static void broadcast_shutdown_local(struct clock_event_device *bc, int __tick_broadcast_oneshot_control(enum tick_broadcast_state state) { struct clock_event_device *bc, *dev; - struct tick_device *td; int cpu, ret = 0; ktime_t now; @@ -699,25 +698,20 @@ int __tick_broadcast_oneshot_control(enum tick_broadcast_state state) if (!tick_broadcast_device.evtdev) return -EBUSY; - /* - * Periodic mode does not care about the enter/exit of power - * states - */ - if (tick_broadcast_device.mode == TICKDEV_MODE_PERIODIC) - return 0; - - /* - * We are called with preemtion disabled from the depth of the - * idle code, so we can't be moved away. - */ - td = this_cpu_ptr(&tick_cpu_device); - dev = td->evtdev; + dev = this_cpu_ptr(&tick_cpu_device)->evtdev; raw_spin_lock(&tick_broadcast_lock); bc = tick_broadcast_device.evtdev; cpu = smp_processor_id(); if (state == TICK_BROADCAST_ENTER) { + /* + * If the broadcast device is in periodic mode, we + * return. + */ + if (tick_broadcast_device.mode == TICKDEV_MODE_PERIODIC) + goto out; + if (!cpumask_test_and_set_cpu(cpu, tick_broadcast_oneshot_mask)) { WARN_ON_ONCE(cpumask_test_cpu(cpu, tick_broadcast_pending_mask)); broadcast_shutdown_local(bc, dev); -- cgit v1.2.3 From d33257264b0267a8fd20f6717abbb484c9e21130 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Tue, 7 Jul 2015 17:45:22 +0200 Subject: tick/broadcast: Return busy if periodic mode and hrtimer broadcast If the system is in periodic mode and the broadcast device is hrtimer based, return busy as we have no proper handling for this. [ Split out from a larger combo patch ] Tested-by: Sudeep Holla Signed-off-by: Thomas Gleixner Cc: Suzuki Poulose Cc: Lorenzo Pieralisi Cc: Catalin Marinas Cc: Rafael J. Wysocki Cc: Peter Zijlstra Cc: Preeti U Murthy Cc: Ingo Molnar Link: http://lkml.kernel.org/r/alpine.DEB.2.11.1507070929360.3916@nanos --- kernel/time/tick-broadcast.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/time/tick-broadcast.c b/kernel/time/tick-broadcast.c index 83aa92e87a85..da7b40fdf6d0 100644 --- a/kernel/time/tick-broadcast.c +++ b/kernel/time/tick-broadcast.c @@ -709,8 +709,12 @@ int __tick_broadcast_oneshot_control(enum tick_broadcast_state state) * If the broadcast device is in periodic mode, we * return. */ - if (tick_broadcast_device.mode == TICKDEV_MODE_PERIODIC) + if (tick_broadcast_device.mode == TICKDEV_MODE_PERIODIC) { + /* If it is a hrtimer based broadcast, return busy */ + if (bc->features & CLOCK_EVT_FEAT_HRTIMER) + ret = -EBUSY; goto out; + } if (!cpumask_test_and_set_cpu(cpu, tick_broadcast_oneshot_mask)) { WARN_ON_ONCE(cpumask_test_cpu(cpu, tick_broadcast_pending_mask)); -- cgit v1.2.3 From 0cc5281aa592d0020868f6ccaed359b4ad7b2684 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Tue, 7 Jul 2015 16:45:15 +0200 Subject: tick/broadcast: Return busy when IPI is pending Tell the idle code not to go deep if the broadcast IPI is about to arrive. [ Split out from a larger combo patch ] Tested-by: Sudeep Holla Signed-off-by: Thomas Gleixner Cc: Suzuki Poulose Cc: Lorenzo Pieralisi Cc: Catalin Marinas Cc: Rafael J. Wysocki Cc: Peter Zijlstra Cc: Preeti U Murthy Cc: Ingo Molnar Link: http://lkml.kernel.org/r/alpine.DEB.2.11.1507070929360.3916@nanos --- kernel/time/tick-broadcast.c | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-) (limited to 'kernel') diff --git a/kernel/time/tick-broadcast.c b/kernel/time/tick-broadcast.c index da7b40fdf6d0..70b47bc928b9 100644 --- a/kernel/time/tick-broadcast.c +++ b/kernel/time/tick-broadcast.c @@ -725,11 +725,15 @@ int __tick_broadcast_oneshot_control(enum tick_broadcast_state state) * if the cpu local event is earlier than the * broadcast event. If the current CPU is in * the force mask, then we are going to be - * woken by the IPI right away. + * woken by the IPI right away; we return + * busy, so the CPU does not try to go deep + * idle. */ - if (!cpumask_test_cpu(cpu, tick_broadcast_force_mask) && - dev->next_event.tv64 < bc->next_event.tv64) + if (cpumask_test_cpu(cpu, tick_broadcast_force_mask)) { + ret = -EBUSY; + } else if (dev->next_event.tv64 < bc->next_event.tv64) { tick_broadcast_set_event(bc, cpu, dev->next_event); + } } /* * If the current CPU owns the hrtimer broadcast -- cgit v1.2.3 From d5113e13a550bc9c2b53cc9944b8a06453c4a0a1 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Tue, 7 Jul 2015 16:43:04 +0200 Subject: tick/broadcast: Check for hrtimer broadcast active early If the current cpu is the one which has the hrtimer based broadcast queued then we better return busy immediately instead of going through loops and hoops to figure that out. [ Split out from a larger combo patch ] Tested-by: Sudeep Holla Signed-off-by: Thomas Gleixner Cc: Suzuki Poulose Cc: Lorenzo Pieralisi Cc: Catalin Marinas Cc: Rafael J. Wysocki Cc: Peter Zijlstra Cc: Preeti U Murthy Cc: Ingo Molnar Link: http://lkml.kernel.org/r/alpine.DEB.2.11.1507070929360.3916@nanos --- kernel/time/tick-broadcast.c | 36 ++++++++++++++++++++++++++---------- 1 file changed, 26 insertions(+), 10 deletions(-) (limited to 'kernel') diff --git a/kernel/time/tick-broadcast.c b/kernel/time/tick-broadcast.c index 70b47bc928b9..c8d731ac9563 100644 --- a/kernel/time/tick-broadcast.c +++ b/kernel/time/tick-broadcast.c @@ -705,6 +705,17 @@ int __tick_broadcast_oneshot_control(enum tick_broadcast_state state) cpu = smp_processor_id(); if (state == TICK_BROADCAST_ENTER) { + /* + * If the current CPU owns the hrtimer broadcast + * mechanism, it cannot go deep idle and we do not add + * the CPU to the broadcast mask. We don't have to go + * through the EXIT path as the local timer is not + * shutdown. + */ + ret = broadcast_needs_cpu(bc, cpu); + if (ret) + goto out; + /* * If the broadcast device is in periodic mode, we * return. @@ -718,7 +729,10 @@ int __tick_broadcast_oneshot_control(enum tick_broadcast_state state) if (!cpumask_test_and_set_cpu(cpu, tick_broadcast_oneshot_mask)) { WARN_ON_ONCE(cpumask_test_cpu(cpu, tick_broadcast_pending_mask)); + + /* Conditionally shut down the local timer. */ broadcast_shutdown_local(bc, dev); + /* * We only reprogram the broadcast timer if we * did not mark ourself in the force mask and @@ -733,18 +747,20 @@ int __tick_broadcast_oneshot_control(enum tick_broadcast_state state) ret = -EBUSY; } else if (dev->next_event.tv64 < bc->next_event.tv64) { tick_broadcast_set_event(bc, cpu, dev->next_event); + /* + * In case of hrtimer broadcasts the + * programming might have moved the + * timer to this cpu. If yes, remove + * us from the broadcast mask and + * return busy. + */ + ret = broadcast_needs_cpu(bc, cpu); + if (ret) { + cpumask_clear_cpu(cpu, + tick_broadcast_oneshot_mask); + } } } - /* - * If the current CPU owns the hrtimer broadcast - * mechanism, it cannot go deep idle and we remove the - * CPU from the broadcast mask. We don't have to go - * through the EXIT path as the local timer is not - * shutdown. - */ - ret = broadcast_needs_cpu(bc, cpu); - if (ret) - cpumask_clear_cpu(cpu, tick_broadcast_oneshot_mask); } else { if (cpumask_test_and_clear_cpu(cpu, tick_broadcast_oneshot_mask)) { clockevents_switch_state(dev, CLOCK_EVT_STATE_ONESHOT); -- cgit v1.2.3 From c4288334818c81c946acb23d2319881f58c3d497 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Sun, 5 Jul 2015 20:53:17 +0000 Subject: tick/broadcast: Handle spurious interrupts gracefully Andriy reported that on a virtual machine the warning about negative expiry time in the clock events programming code triggered: hpet: hpet0 irq 40 for MSI hpet: hpet1 irq 41 for MSI Switching to clocksource hpet WARNING: at kernel/time/clockevents.c:239 [] clockevents_program_event+0xdb/0xf0 [] tick_handle_periodic_broadcast+0x41/0x50 [] timer_interrupt+0x15/0x20 When the second hpet is installed as a per cpu timer the broadcast event is not longer required and stopped, which sets the next_evt of the broadcast device to KTIME_MAX. If after that a spurious interrupt happens on the broadcast device, then the current code blindly handles it and tries to reprogram the broadcast device afterwards, which adds the period to next_evt. KTIME_MAX + period results in a negative expiry value causing the WARN_ON in the clockevents code to trigger. Add a proper check for the state of the broadcast device into the interrupt handler and return if the interrupt is spurious. [ Folded in pointer fix from Sudeep ] Reported-by: Andriy Gapon Signed-off-by: Thomas Gleixner Cc: Sudeep Holla Cc: Peter Zijlstra Cc: Preeti U Murthy Link: http://lkml.kernel.org/r/20150705205221.802094647@linutronix.de --- kernel/time/tick-broadcast.c | 7 +++++++ 1 file changed, 7 insertions(+) (limited to 'kernel') diff --git a/kernel/time/tick-broadcast.c b/kernel/time/tick-broadcast.c index c8d731ac9563..ee3cf942d6eb 100644 --- a/kernel/time/tick-broadcast.c +++ b/kernel/time/tick-broadcast.c @@ -316,6 +316,13 @@ static void tick_handle_periodic_broadcast(struct clock_event_device *dev) bool bc_local; raw_spin_lock(&tick_broadcast_lock); + + /* Handle spurious interrupts gracefully */ + if (clockevent_state_shutdown(tick_broadcast_device.evtdev)) { + raw_spin_unlock(&tick_broadcast_lock); + return; + } + bc_local = tick_do_periodic_broadcast(); if (clockevent_state_oneshot(dev)) { -- cgit v1.2.3 From c4d029f2d43b39de7b9299e8b58102a442ba86f8 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Sat, 11 Jul 2015 14:26:34 +0200 Subject: tick/broadcast: Prevent NULL pointer dereference Dan reported that the recent changes to the broadcast code introduced a potential NULL dereference. Add the proper check. Fixes: e0454311903d "tick/broadcast: Sanity check the shutdown of the local clock_event" Reported-by: Dan Carpenter Signed-off-by: Thomas Gleixner --- kernel/time/tick-broadcast.c | 18 ++++++++++-------- 1 file changed, 10 insertions(+), 8 deletions(-) (limited to 'kernel') diff --git a/kernel/time/tick-broadcast.c b/kernel/time/tick-broadcast.c index ee3cf942d6eb..52b9e199b5ac 100644 --- a/kernel/time/tick-broadcast.c +++ b/kernel/time/tick-broadcast.c @@ -409,14 +409,16 @@ void tick_broadcast_control(enum tick_broadcast_mode mode) break; } - if (cpumask_empty(tick_broadcast_mask)) { - if (!bc_stopped) - clockevents_shutdown(bc); - } else if (bc_stopped) { - if (tick_broadcast_device.mode == TICKDEV_MODE_PERIODIC) - tick_broadcast_start_periodic(bc); - else - tick_broadcast_setup_oneshot(bc); + if (bc) { + if (cpumask_empty(tick_broadcast_mask)) { + if (!bc_stopped) + clockevents_shutdown(bc); + } else if (bc_stopped) { + if (tick_broadcast_device.mode == TICKDEV_MODE_PERIODIC) + tick_broadcast_start_periodic(bc); + else + tick_broadcast_setup_oneshot(bc); + } } raw_spin_unlock(&tick_broadcast_lock); } -- cgit v1.2.3