Atomic() vs k_sched_lock() vs irq_lock() vs k_mutex_lock()
|
Kim Bøndergaard
We have an internal discussion about the recommended usage of the 4 methods mentioned in subjects, when the goal is to protect an assignment of a variable (u32 - any platform) within one task so it is not resulting in an invalid reading from a higher priority task.
Mutex'es are usually good since they only disturb the threads being involved in accessing the shared variable - but are they overkill in this case? irq_lock() sounds like something fast but effects the entire system k_sched_lock() is fast (according to the manual) but effects all threads as I see it.
atomic is probably a good option, but what's the efficiency on e.g. an ARM platform not natively supporting atomic's.
Any ideas or recommendation about when to use what solution?
Kim Bøndergaard
Prevas A/S Team Manager / Systems Architect Hedeager 3, DK-8200 Aarhus N Phone +45 3315 9090
|
|
|
|
Boie, Andrew P
You should be using the APIs in include/atomic.h
On arches that don't natively support atomic operations, these get implemented in terms of irq_lock()
Andrew
From: devel@... [mailto:devel@...]
On Behalf Of Kim Bøndergaard
Sent: Monday, April 1, 2019 6:53 AM To: Zephyr-devel@... Subject: [Zephyr-devel] Atomic() vs k_sched_lock() vs irq_lock() vs k_mutex_lock()
We have an internal discussion about the recommended usage of the 4 methods mentioned in subjects, when the goal is to protect an assignment of a variable (u32 - any platform) within one task so it is not resulting in an invalid reading from a higher priority task.
Mutex'es are usually good since they only disturb the threads being involved in accessing the shared variable - but are they overkill in this case? irq_lock() sounds like something fast but effects the entire system k_sched_lock() is fast (according to the manual) but effects all threads as I see it.
atomic is probably a good option, but what's the efficiency on e.g. an ARM platform not natively supporting atomic's.
Any ideas or recommendation about when to use what solution?
Kim Bøndergaard Phone +45 3315 9090
|
|
|