Description:
The latest version of the patch introduces a __cacheline_aligned_in_smp
macro. It
- applies on lock by lock base
- leaves the handling of the lock unchanged
- leaves the UP code unchanged
The __cacheline_aligned_in_smp macro expands only in the SMP
case. It puts the cacheline aligned data in a special section which
is only for cacheline aligned data. This guarantees that the cacheline
is exclusive for this lock.
(There exists also a version of the macro with 4 underscores
____cacheline_aligned_in_smp in cache.h which does not use a separate
section, this is not the same, because in this case the cacheline
is not exclusive for the lock)
|