s390/idle: Provide arch specific kcpustat_field_idle()/kcpustat_field_iowait()

The former s390 specific arch_cpu_idle_time() implementation was
removed, since its implementation was racy and reported idle time
could go backwards [1].

However this removal was not necessary, since independently of the s390
architecture specific races there exists the iowait counter update race,
which can also lead to reported idle time going backwards [2].

With Frederic Weisbecker's recent cpu idle time accounting refactoring
kernel_cpustat got a sequence counter. Use this to implement s390 specific
variants of kcpustat_field_idle() and kcpustat_field_iowait(). This is
logically a revert of [1] and moves cpu idle time accounting back into s390
architecture code, which is also more precise than the dyntick idle time
accounting by nohz/scheduler.

For comparing cross cpu time stamps it is necessary to use the stcke
instead of the stckf instruction in irq entry path. Furthermore this
open-codes a sequence lock in assembler and C code, which is required to
update the irq entry time stamp to the per cpu idle_data structure in a
race free manner.

[1] commit be76ea614460 ("s390/idle: remove arch_cpu_idle_time() and corresponding code")
[2] commit ead70b752373 ("timers/nohz: Add a comment about broken iowait counter update race")

Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Acked-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>

2 files changed, 33 insertions, 0 deletions

diff --git a/include/linux/kernel_stat.h b/include/linux/kernel_stat.h
index fce1392e2140..9ca6c2259dfe 100644
--- a/include/linux/kernel_stat.h
+++ b/include/linux/kernel_stat.h
@@ -107,6 +107,30 @@ static inline unsigned long kstat_cpu_irqs_sum(unsigned int cpu)
 }
 
 #ifdef CONFIG_NO_HZ_COMMON
+
+#ifdef CONFIG_HAVE_VIRT_CPU_ACCOUNTING_IDLE
+
+static inline void kcpustat_dyntick_start(u64 now) { }
+static inline void kcpustat_dyntick_stop(u64 now) { }
+static inline void kcpustat_irq_enter(u64 now) { }
+static inline void kcpustat_irq_exit(u64 now) { }
+static inline bool kcpustat_idle_dyntick(void) { return false; }
+
+extern u64 arch_kcpustat_field_idle(int cpu);
+extern u64 arch_kcpustat_field_iowait(int cpu);
+
+static inline u64 kcpustat_field_idle(int cpu)
+{
+	return arch_kcpustat_field_idle(cpu);
+}
+
+static inline u64 kcpustat_field_iowait(int cpu)
+{
+	return arch_kcpustat_field_iowait(cpu);
+}
+
+#else /* !CONFIG_HAVE_VIRT_CPU_ACCOUNTING_IDLE */
+
 extern void kcpustat_dyntick_start(u64 now);
 extern void kcpustat_dyntick_stop(u64 now);
 extern void kcpustat_irq_enter(u64 now);
@@ -118,6 +142,9 @@ static inline bool kcpustat_idle_dyntick(void)
 {
 	return __this_cpu_read(kernel_cpustat.idle_dyntick);
 }
+
+#endif /* !CONFIG_HAVE_VIRT_CPU_ACCOUNTING_IDLE */
+
 #else
 static inline u64 kcpustat_field_idle(int cpu)
 {
diff --git a/include/linux/vtime.h b/include/linux/vtime.h
index 9dc25b04a119..82825e775499 100644
--- a/include/linux/vtime.h
+++ b/include/linux/vtime.h
@@ -42,9 +42,15 @@ extern void vtime_account_irq(struct task_struct *tsk, unsigned int offset);
 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);
+#ifdef CONFIG_HAVE_VIRT_CPU_ACCOUNTING_IDLE
+static inline void vtime_reset(void) { }
+static inline void vtime_dyntick_start(void) { }
+static inline void vtime_dyntick_stop(void) { }
+#else
 extern void vtime_reset(void);
 extern void vtime_dyntick_start(void);
 extern void vtime_dyntick_stop(void);
+#endif
 #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
 static inline void vtime_account_irq(struct task_struct *tsk, unsigned int offset) { }
 static inline void vtime_account_softirq(struct task_struct *tsk) { }