VM idle page zeroing

Matthew Dillon dillon at apollo.backplane.com
Sun May 16 17:24:14 PDT 2010

:Any particular reason for moving the thread to the last cpu?

    No, It is just an experiment. There is no need to put it on cpu 0
    on a SMP system since numerous dedicated threads already run on cpu 0.
    It can't hurt, anyhow.

:Something that occured to me - right now, we have a thread that wakes up
:every 1/10th of a second, checks if there are pages to zero. At long
:stretches of idle time, we just end up waking and sleeping a whole lot.
:Instead, perhaps, when vm_page_zero_check() sees that we hit ZIDLE_HIGH we
:should perhaps sleep for a lot longer and drop down to the usual rate when
:we hit ZIDLE_LOW?
:-- vs

    Yes and no.  Yes it is probably worth doing as sleep states become
    more advanced.  But on the flip side the actual cpu use is miniscule,
    less than 0.01%.  Essentially 10uS once every 1/10 second and with
    no locking requirements in the no-work case.


More information about the Kernel mailing list