On Thu, 13 Jan 2000, Linus Torvalds wrote:
> HOWEVER, I don't think this is going to be a huge issue in most cases. And
> if people don't need non-DMA memory, then the pages we "swapped" out are
> going to stay in RAM anyway, so it's not going to hurt us.
>
> Anyway, I obviously do agree that I may well be wrong, and that real life
> is going to come back and bite us, and we'll end up having to not do it
> this way. However, I'd prefer trying the "conceptually simple" path first,
> and only if it turns out that yes, I was completely wrong, do we try to
> fix it up with magic heuristics etc.
hm., i think we'll see this with ISA soundcards (still the majority) if
used as modules. Right now kswapd just gives up too easy and says 'no such
page', on a box with lots of RAM and all DMA allocated in process VM
space.
Anyway, the patch and suggestion of passing in a single zone is i believe
completely wrong, because it advances mm->swap_address, which unfairly
selects a given range to be checked for only one zone. So i think it's
either zone-bitmaps (or equivalent multi-zone logic) or what you
suggested, to have no zone-awareness in swap_out() for now at all.
(i believe this is also going to bite us with the IA64 port - kswapd will
have no information to free pages from the right node, we could solve this
already with a zone bitmap, or by starting per-zone kswapds. The latter
one looks like overkill to me, but it's conceptually cleaner than bitmaps
and and does not have a limitation on the number of zones. Might not be a
highprio issue though.)
-- mingo
-
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 : Sat Jan 15 2000 - 21:00:25 EST