On Mon, Jul 4, 2011 at 12:50 PM, Owain Ainsworth <[email protected]> wrote: > So, the msync implementation in uvm says this: > > /* > * XXXCDC: do we really need this semantic? > * > * XXX Gak! If size is zero we are supposed to sync "all modified > * pages with the region containing addr". Unfortunately, we > * don't really keep track of individual mmaps so we approximate > * by flushing the range of the map entry containing addr. > * This can be incorrect if the region splits or is coalesced > * with a neighbor. > */ > > I checked in posix and can find no reference to this at all. So I would > like to suggest this diff. This of course needs quite a bit of testing > (a ports bulk build would be nice), but removing unnecessary crap makes > people happy. Otoh, if something actually depends on this then it may > need to stay.
I agree that posix doesn't require this. msync() with len=0 is supposed to be a noop AFAICT. I like the diff: shoot the code. Philip
