cvs commit: src/sys/kern vfs_mount.c

Matthew Dillon dillon at apollo.backplane.com
Mon Apr 7 11:31:42 PDT 2008


:[I guess this got lost, reposting]
:
:...
:> I have the feeling that just calling lwkt_yield() doesn't necessarily=20
:> help.  I still get the stutters.  I'd need something to yield to=20
:> runnable userland processes, not just processes with the same prio (I=20
:> suspect we're running in a critical section here, rendering the yield=20
:> futile)
:>=20
:> cheers
:>   simon
:>=20

    Yah, it won't help for kernel vs userland thread.

    An explicit tsleep would be reasonable.  E.g. tsleep for one tick
    every 1000 vnodes or something like that.  The code is primarily run
    from the syncer thread to flush all vnodes once every 30 seconds.
    It is not time critical.

					-Matt
					Matthew Dillon 
					<dillon at backplane.com>





More information about the Commits mailing list