rms: add a comment explaining performance deficiencies of write locking

This commit is contained in:
mjg 2020-08-04 19:52:16 +00:00
parent 817d89563e
commit 07249f6844

View File

@ -872,6 +872,10 @@ db_show_rm(const struct lock_object *lock)
* Neither rms_rlock nor rms_runlock use fences. Instead compiler barriers are
* inserted to prevert reordering of generated code. Execution ordering is
* provided with the use of an IPI handler.
*
* No attempt is made to track which CPUs read locked at least once,
* consequently write locking sends IPIs to all of them. This will become a
* problem at some point. The easiest way to lessen it is to provide a bitmamp.
*/
void