preempt: Provide preempt_[dis|en]able_nested()

On PREEMPT_RT enabled kernels, spinlocks and rwlocks are neither disabling
preemption nor interrupts. Though there are a few places which depend on
the implicit preemption/interrupt disable of those locks, e.g. seqcount
write sections, per CPU statistics updates etc.

To avoid sprinkling CONFIG_PREEMPT_RT conditionals all over the place, add
preempt_disable_nested() and preempt_enable_nested() which should be
descriptive enough.

Add a lockdep assertion for the !PREEMPT_RT case to catch callers which
do not have preemption disabled.

Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/r/20220825164131.402717-2-bigeasy@linutronix.de
This commit is contained in:
Thomas Gleixner 2022-08-25 18:41:24 +02:00
parent 521a547ced
commit 555bb4ccd1

View file

@ -421,4 +421,46 @@ static inline void migrate_enable(void) { }
#endif /* CONFIG_SMP */
/**
* preempt_disable_nested - Disable preemption inside a normally preempt disabled section
*
* Use for code which requires preemption protection inside a critical
* section which has preemption disabled implicitly on non-PREEMPT_RT
* enabled kernels, by e.g.:
* - holding a spinlock/rwlock
* - soft interrupt context
* - regular interrupt handlers
*
* On PREEMPT_RT enabled kernels spinlock/rwlock held sections, soft
* interrupt context and regular interrupt handlers are preemptible and
* only prevent migration. preempt_disable_nested() ensures that preemption
* is disabled for cases which require CPU local serialization even on
* PREEMPT_RT. For non-PREEMPT_RT kernels this is a NOP.
*
* The use cases are code sequences which are not serialized by a
* particular lock instance, e.g.:
* - seqcount write side critical sections where the seqcount is not
* associated to a particular lock and therefore the automatic
* protection mechanism does not work. This prevents a live lock
* against a preempting high priority reader.
* - RMW per CPU variable updates like vmstat.
*/
/* Macro to avoid header recursion hell vs. lockdep */
#define preempt_disable_nested() \
do { \
if (IS_ENABLED(CONFIG_PREEMPT_RT)) \
preempt_disable(); \
else \
lockdep_assert_preemption_disabled(); \
} while (0)
/**
* preempt_enable_nested - Undo the effect of preempt_disable_nested()
*/
static __always_inline void preempt_enable_nested(void)
{
if (IS_ENABLED(CONFIG_PREEMPT_RT))
preempt_enable();
}
#endif /* __LINUX_PREEMPT_H */