Jens Axboe wrote:
> Hi,
> 
> This applies on top of the splice #3 just posted, adding support for
> moving of pages. The caller can use the SPLICE_F_MOVE flag to the splice
> syscall to ask the kernel to try and move pages, if needed.
> 
> Disclaimer: this works for me, but may have vm issues that I missed.
> CC'ing Nick :-)
> 

Like Andrew said, you can't check PageLRU without holding zone->lru_lock.
The page release code can get away with it only because the page refcount
is 0 at that point. Also, you can't reliably remove pages from the LRU
unless the refcount is 0. Ever.

The following (untested) is something like what I had in mind, and should
get stealing closer to working. I've only given it a quick review so far
(btw. why do you only unlock the page if it hasn't been stolen?)

With this patch, the ->steal will indicate if the page had been on the
LRU or not. If not, then add it; if yes, then do nothing.

There is no caller of ->steal yet that wants the page off the LRU (is
there?). That's a bit harder.

---