I've reworked the patch for trimming the page table cache to
address the objections raised to the previous version. The patch
makes no change to the existing idle-loop water-mark code; instead
it defines a trim_pgt_cache() function to free one page at a time
from the cache. It cycles through all CPUs to distribute the
trimming evenly. Currntly it's called once per
do_try_to_free_page(), but could be made part of the swap state
machine by adding a success return code.
This patch will allow machines of even modest memory size to obtain
the advantages of the page table cache when they aren't swapping,
but have full access to the memory when the system needs it.
I'm going to stop arguing about these patch sets.
But I will mention that let's just happen to remember to "fix" this
new dependency on the master kernel lock, which is created by this
patch, when we fully fine grain the fault processing path. It is as
if all my work to make it lockless now goes to shit.
I still really don't like these changes.
Later,
David S. Miller
davem@dm.cobaltmicro.com
-
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.altern.org/andrebalsa/doc/lkml-faq.html