irq: Reorder time handling against HARDIRQ_OFFSET on IRQ entry
cover irq: Reorder time handling against HARDIRQ_OFFSET on IRQ entry
There are two issues with the current layout of tick_irq_enter() as it's called before HARDIRQ_OFFSET is incremented:
1) It's not correctly handled by lockdep which doesn't consider it as hardirq context. And jiffies/timekeeping update take a few interesting locks.
2) Softirqs need to be explicitly disabled around it to prevent ksoftirqd from being spuriously woken up.
The current call dependency prevents tick_irq_enter() from being moved after HARDIRQ_OFFSET: account_irq_enter_time() needs to be called before HARDIRQ_OFFSET incrementation due to cputime index dispatch and it must be called after tick_irq_enter() which updates the clocks that may be necessary for cputime accounting.
Here is a proposal to fix this layout.
(The EXPORT_SYMBOL_GPL() in vtime will likely disappear in the next take as they don't seem to be necessary anymore, but I'll need to check that thoroughly).
git://git.kernel.org/pub/scm/linux/kernel/git/frederic/linux-dynticks.git
irq/core
HEAD: 9502ee20aed8bb847176e1d7d83ccd0625430744
Frederic Weisbecker (4):
sched/vtime: Consolidate IRQ time accounting
s390/vtime: Convert to consolidated IRQ time accounting
irqtime: Move irqtime entry accounting after irq offset incrementation
irq: Call tick_irq_enter() inside HARDIRQ_OFFSET
arch/ia64/kernel/time.c | 22 +++++++--- arch/powerpc/kernel/time.c | 60 ++++++++++++++++++++-------- arch/s390/include/asm/vtime.h | 1 - arch/s390/kernel/vtime.c | 60 ++++++++++++++++++---------- include/linux/hardirq.h | 4 +- include/linux/vtime.h | 18 ++++----- kernel/sched/cputime.c | 75 +++++++++++++++++++++++++++-------- kernel/softirq.c | 16 +++----- 8 files changed, 176 insertions(+), 80 deletions(-)
Signature · 1 line
2.25.1
RFC PATCH 1/4 sched/vtime: Consolidate IRQ time accounting
The 3 architectures implementing CONFIG_VIRT_CPU_ACCOUNTING_NATIVE all have their own version of irq time accounting that dispatch the cputime to the appropriate index: hardirq, softirq, system, idle, guest... from an all-in-one function.
Instead of having these ad-hoc versions, move the cputime destination dispatch decision to the core code and leave only the actual per-index cputime accounting to the architecture.
For now only ia64 and powerpc are handled. s390 will need a slightly different treatment.
--- arch/ia64/kernel/time.c | 22 ++++++++++---- arch/powerpc/kernel/time.c | 60 +++++++++++++++++++++++++++----------- arch/s390/kernel/vtime.c | 15 ++++++---- include/linux/vtime.h | 8 ++--- kernel/sched/cputime.c | 21 +++++++++++-- 5 files changed, 92 insertions(+), 34 deletions(-)
arch/ia64/kernel/time.c +17 -5
@@ -138,12 +138,8 @@ void vtime_account_kernel(struct task_struct *tsk)struct thread_info *ti =task_thread_info (tsk);__u64 stime =vtime_delta (tsk); -if ((tsk->flags &PF_VCPU ) && !irq_count ()) +if (tsk->flags &PF_VCPU ) ti->gtime += stime; -else if (hardirq_count ()) - ti->hardirq_time += stime; -else if (in_serving_softirq ()) - ti->softirq_time += stime;else ti->stime += stime; }
@@ -156,6 +152,22 @@ void vtime_account_idle(struct task_struct *tsk) ti->idle_time +=vtime_delta (tsk); } +void vtime_account_softirq (struct task_struct *tsk) +{ +struct thread_info *ti =task_thread_info (tsk); + + ti->softirq_time +=vtime_delta (tsk); +} +EXPORT_SYMBOL_GPL (vtime_account_softirq); + +void vtime_account_hardirq (struct task_struct *tsk) +{ +struct thread_info *ti =task_thread_info (tsk); + + ti->hardirq_time +=vtime_delta (tsk); +} +EXPORT_SYMBOL_GPL (vtime_account_hardirq); +#endif /* CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */ static irqreturn_t
arch/powerpc/kernel/time.c +43 -17
94 lines not on this page · raw message
arch/s390/kernel/vtime.c +10 -5
@@ -226,7 +226,7 @@ void vtime_flush(struct task_struct *tsk)* Update process times based on virtual cpu times stored by entry.S * to the lowcore fields user_timer, system_timer & steal_clock. */ -void vtime_account_irq_enter (struct task_struct *tsk) +void vtime_account_kernel (struct task_struct *tsk) {u64 timer;
@@ -245,12 +245,17 @@ void vtime_account_irq_enter(struct task_struct *tsk)virt_timer_forward (timer); } -EXPORT_SYMBOL_GPL (vtime_account_irq_enter); - -void vtime_account_kernel (struct task_struct *tsk) -__attribute__((alias ("vtime_account_irq_enter" )));EXPORT_SYMBOL_GPL (vtime_account_kernel); +void vtime_account_irq_enter (struct task_struct *tsk) +__attribute__((alias ("vtime_account_kernel" ))); +EXPORT_SYMBOL_GPL (vtime_account_irq_enter); + +void vtime_account_irq_exit (struct task_struct *tsk) +__attribute__((alias ("vtime_account_kernel" ))); +EXPORT_SYMBOL_GPL (vtime_account_irq_exit); + +/* * Sorted add to a list. List is linear searched until first bigger * element is found.
include/linux/vtime.h +3 -5
@@ -84,11 +84,9 @@ static inline void vtime_init_idle(struct task_struct *tsk, int cpu) { }#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE extern void vtime_account_irq_enter (struct task_struct *tsk); -static inline void vtime_account_irq_exit (struct task_struct *tsk) -{ -/* On hard|softirq exit we always account to hard|softirq cputime */ -vtime_account_kernel (tsk); -} +extern void vtime_account_irq_exit (struct task_struct *tsk); +extern void vtime_account_softirq (struct task_struct *tsk); +extern void vtime_account_hardirq (struct task_struct *tsk);extern void vtime_flush (struct task_struct *tsk);#else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */ static inline void vtime_account_irq_enter (struct task_struct *tsk) { }
kernel/sched/cputime.c +19 -2
@@ -429,12 +429,29 @@ void vtime_task_switch(struct task_struct *prev)#ifndef __ARCH_HAS_VTIME_ACCOUNTvoid vtime_account_irq_enter (struct task_struct *tsk) { -if (!in_interrupt () &&is_idle_task (tsk)) +if (hardirq_count ()) { +vtime_account_hardirq (tsk); + }else if (in_serving_softirq ()) { +vtime_account_softirq (tsk); + }else if (is_idle_task (tsk)) {vtime_account_idle (tsk); -else + }else {vtime_account_kernel (tsk); + } }EXPORT_SYMBOL_GPL (vtime_account_irq_enter); + +void vtime_account_irq_exit (struct task_struct *tsk) +{ +WARN_ON_ONCE (in_task ()); + +if (hardirq_count ()) { +vtime_account_hardirq (tsk); + }else if (in_serving_softirq ()) { +vtime_account_softirq (tsk); + } +} +EXPORT_SYMBOL_GPL (vtime_account_irq_exit);#endif /* __ARCH_HAS_VTIME_ACCOUNT */ void cputime_adjust(struct task_cputime *curr, struct prev_cputime *prev,
Signature · 1 line
2.25.1
RFC PATCH 2/4 s390/vtime: Convert to consolidated IRQ time accounting
In reply to Frederic Weisbeckers390 has its own version of IRQ time accounting because it doesn't account the idle time the same way the other architectures do. Only the actual idle sleep time is accounted as idle time, the rest of the idle task execution is accounted as system time.
However converting it to the consolidated IRQ time accounting is easy:
just keep the current behaviour and redirect generic idle time accounting to system time accounting.
This removes the need to maintain an ad-hoc implementation of cputime dispatch decision.
--- arch/s390/include/asm/vtime.h | 1 - arch/s390/kernel/vtime.c | 57 ++++++++++++++++++++++------------- kernel/sched/cputime.c | 2 -- 3 files changed, 36 insertions(+), 24 deletions(-)
arch/s390/include/asm/vtime.h +0 -1
@@ -2,7 +2,6 @@#ifndef _S390_VTIME_H#define _S390_VTIME_H -#define __ARCH_HAS_VTIME_ACCOUNT#define __ARCH_HAS_VTIME_TASK_SWITCH#endif /* _S390_VTIME_H */
arch/s390/kernel/vtime.c +36 -21
75 lines not on this page · raw message
kernel/sched/cputime.c +0 -2
@@ -426,7 +426,6 @@ void vtime_task_switch(struct task_struct *prev)* time spent by the CPU when it's in low power mode) must override * vtime_account(). */ -#ifndef __ARCH_HAS_VTIME_ACCOUNTvoid vtime_account_irq_enter (struct task_struct *tsk) {if (hardirq_count ()) {
@@ -452,7 +451,6 @@ void vtime_account_irq_exit(struct task_struct *tsk) } }EXPORT_SYMBOL_GPL (vtime_account_irq_exit); -#endif /* __ARCH_HAS_VTIME_ACCOUNT */ void cputime_adjust (struct task_cputime *curr,struct prev_cputime *prev,u64 *ut,u64 *st)
Signature · 1 line
2.25.1
RFC PATCH 3/4 sched/irqtime: Move irqtime entry accounting after irq offset incrementation
In reply to Frederic WeisbeckerIRQ time entry is currently accounted before HARDIRQ_OFFSET or SOFTIRQ_OFFSET are incremented. This is convenient to decide to which index the cputime to account is dispatched.
Unfortunately it prevents tick_irq_enter() from being called under HARDIRQ_OFFSET because tick_irq_enter() has to be called before the IRQ entry accounting due to the necessary clock catch up. As a result we don't benefit from appropriate lockdep coverage on tick_irq_enter().
To prepare for fixing this, move the IRQ entry cputime accounting after the preempt offset is incremented. This requires the cputime dispatch code to handle the extra offset.
--- include/linux/hardirq.h | 4 +-- include/linux/vtime.h | 10 +++++--- kernel/sched/cputime.c | 56 ++++++++++++++++++++++++++++++----------- kernel/softirq.c | 2 +- 4 files changed, 51 insertions(+), 21 deletions(-)
include/linux/hardirq.h +2 -2
@@ -32,9 +32,9 @@ static __always_inline void rcu_irq_enter_check_tick(void)*/ #define __irq_enter () \ do { \ + preempt_count_add(HARDIRQ_OFFSET); \ + lockdep_hardirq_enter(); \ account_irq_enter_time(current); \ - preempt_count_add(HARDIRQ_OFFSET); \ - lockdep_hardirq_enter(); \ } while (0)/*
include/linux/vtime.h +6 -4
@@ -96,21 +96,23 @@ static inline void vtime_flush(struct task_struct *tsk) { }#ifdef CONFIG_IRQ_TIME_ACCOUNTING -extern void irqtime_account_irq (struct task_struct *tsk); +extern void irqtime_account_enter (struct task_struct *tsk); +extern void irqtime_account_exit (struct task_struct *tsk);#else -static inline void irqtime_account_irq (struct task_struct *tsk) { } +static inline void irqtime_account_enter (struct task_struct *tsk) { } +static inline void irqtime_account_exit (struct task_struct *tsk) { }#endif static inline void account_irq_enter_time (struct task_struct *tsk) {vtime_account_irq_enter (tsk); -irqtime_account_irq (tsk); +irqtime_account_enter (tsk); }static inline void account_irq_exit_time (struct task_struct *tsk) {vtime_account_irq_exit (tsk); -irqtime_account_irq (tsk); +irqtime_account_exit (tsk); }#endif /* _LINUX_KERNEL_VTIME_H */
kernel/sched/cputime.c +42 -14
81 lines not on this page · raw message
kernel/softirq.c +1 -1
@@ -315,9 +315,9 @@ asmlinkage __visible void __softirq_entry __do_softirq(void) current->flags &= ~PF_MEMALLOC ; pending =local_softirq_pending (); -account_irq_enter_time (current);__local_bh_disable_ip (_RET_IP_,SOFTIRQ_OFFSET ); +account_irq_enter_time (current); in_hardirq =lockdep_softirq_start (); restart:
Signature · 1 line
2.25.1
RFC PATCH 4/4 irq: Call tick_irq_enter() inside HARDIRQ_OFFSET
In reply to Frederic WeisbeckerNow that account_irq_enter_time() is called after HARDIRQ_OFFSET has been incremented, there is nothing left that prevents us from also moving tick_irq_enter() after HARDIRQ_OFFSET is incremented.
The desired outcome is to remove the nasty hack that prevents softirqs from being raised through ksoftirqd instead of the hardirq bottom half. Also tick_irq_enter() then becomes appropriately covered by lockdep.
--- kernel/softirq.c | 14 +++++--------- 1 file changed, 5 insertions(+), 9 deletions(-)
kernel/softirq.c +5 -9
@@ -377,16 +377,12 @@ asmlinkage __visible void __softirq_entry __do_softirq(void)*/ void irq_enter_rcu (void ) { -if (is_idle_task (current) && !in_interrupt ()) { -/* -* Prevent raise_softirq from needlessly waking up ksoftirqd -* here, as softirq will be serviced on return from interrupt. -*/ -local_bh_disable (); +__irq_enter_raw (); + +if (is_idle_task (current) && (irq_count () ==HARDIRQ_OFFSET ))tick_irq_enter (); -_local_bh_enable (); - } -__irq_enter (); + +account_irq_enter_time (current); }/**
Signature · 1 line
2.25.1