mmap_sem contention issues

mmap_sem contention issues

[Andy Lutomirski]

I'm having serious latency problems due to mmap_sem contention.  I
have a real-time thread that has (soft) page faults on locked pages,
and it blocks for multiple milliseconds on
(call_rwsem_down_read_failed do_page_fault page_fault).  Can this be
fixed?

Some ideas:

1. Drop mmap_sem during the filesystem part of mmap and munmap.
(MAP_POPULATE in particular is a disaster -- using it will easily
increase latency from a few milliseconds to a respectable fraction of
a second.)

2. Come up with some way to lock specific vm_area_structs for read
access without taking mmap_sem at all.  This looks unpleasant with the
current rbtree structure -- something like a radix tree might work
much better if the nodes were to contain their own locks.

--Andy

-- 
Andy Lutomirski
AMA Capital Management, LLC

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>