On Thu, 11 May 2000, Jeff V. Merkey wrote:
> >
> > (3) should in theory work, you just need to fill in page->buffers for it,
> > and right now that may or may not do what you want. See the 99-pre7 code
> > to sync the dang thing for you - it may be more along what you want done.
> > And right now there is no way to "plug" the write - once page->buffers has
> > been filled in, the write-back can happen at any time.
>
> Who does the write back? It looks like it's the buffer cache. So Pre7
> fixes this flushing problem?
Well, pre-7 does it differently.
Earlier, what happened was that when the MM layer noticed that a page had
dirty buffers on it, it would just ask bdflush to start writing things
out. This does not work if those per-page dirty buffers aren't on the
bdflush LRU lists - which they don't actually have to be. I'm considering
getting rid of the bdflush LRU lists, and just relying on the page LRU
list instead - because the MM layer does need the latter, and the former
is really just another form of the same thing.
What pre-7 does is that when it needs more memory and the page it would
like to free has buffers attached to it, pre-7 will just cause those
buffers to be asycnhronously written out. It still cannot free the page,
but the next time around it likely _will_ be able to free it, so the
memory pressure automatically keeps the dirty buffer writeout going.
Whether this suits your needs or not is unclear. It makes more sense to
me, the old bdflush approach was really just a question of being lazy
("Oh, duh, we have dirty stuff, let's just do this one-liner that will
cause it to be flushed eventually"). And it's a lot more flexible, and
potentially makes it possible to just remove a lot of the old "global
buffer LRU" code.
(The page LRU is not as exact as the per-buffer LRU was, but it has other
advantages, the main one obviously being the natural integration with the
VM back-pressure. And "exact LRU" is probably not called for anyway - the
locality that the page cache offers might actually be advantageous).
Linus
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Mon May 15 2000 - 21:00:19 EST