osq_lock: add define to use smp_cond_load_relaxed. #70
Conversation
Adds a define USE_SMP_COND_LOAD_RELAXED that enables use of smp_cond_load_relaxed in osq_lock spin (as per v5.6 kernel)
| @@ -358,6 +358,20 @@ do { \ | |||
| } \ | |||
| VAL; \ | |||
| }) | |||
|
|
|||
| #define smp_cond_load_relaxed(ptr, cond_expr) \ | |||
There was a problem hiding this comment.
| @@ -384,6 +398,19 @@ do { \ | |||
| barrier(); \ | |||
| VAL; \ | |||
| }) | |||
|
|
|||
| #define smp_cond_load_relaxed(ptr, cond_expr) ({ \ | |||
There was a problem hiding this comment.
| @@ -393,7 +393,10 @@ static bool osq_lock(uint64_t *osq, unsigned long cpu_number) | |||
| * guaranteed their existence -- this allows us to apply | |||
| * cmpxchg in an attempt to undo our queueing. | |||
| */ | |||
|
|
|||
| #if defined(USE_SMP_COND_LOAD_RELAXED) | |||
There was a problem hiding this comment.
Should this be defined and set as default since it appears to be in the latest linux kernel?
There was a problem hiding this comment.
My concern is that someone may be relying on the behaviour of the older kernel code for regression/testing.
If we agree that this isn't a concern and that we want to reflect the latest stable kernel osq_lock then we can change to make this behaviour the default.
Note that we'll also need to update the commentary at the top of osq_lock.h to reflect this.
There was a problem hiding this comment.
I think it is ok to change the code to reflect what is stable in the kernel. We should leave an option to run the old implementation that can be switched on via a define in the Makefile
Adds a define USE_SMP_COND_LOAD_RELAXED that enables use of
smp_cond_load_relaxed in osq_lock spin (as per v5.6 kernel).
To address issue #69.