On 02/09/2017 06:05 AM, Anshuman Khandual wrote:

On 02/08/2017 10:48 PM, Vlastimil Babka wrote:

quoted

On 02/08/2017 03:01 PM, Anshuman Khandual wrote:

quoted

This implements allocation isolation for CDM nodes in buddy allocator by
discarding CDM memory zones all the time except in the cases where the
gfp
flag has got __GFP_THISNODE or the nodemask contains CDM nodes in cases
where it is non NULL (explicit allocation request in the kernel or user
process MPOL_BIND policy based requests).

Signed-off-by: Anshuman Khandual <redacted>
---
 mm/page_alloc.c | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)

diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 40908de..7d8c82a 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c

@@ -64,6 +64,7 @@
 #include <linux/page_owner.h>
 #include <linux/kthread.h>
 #include <linux/memcontrol.h>
+#include <linux/node.h>

 #include <asm/sections.h>
 #include <asm/tlbflush.h>

@@ -2908,6 +2909,24 @@ get_page_from_freelist(gfp_t gfp_mask, unsigned

int order, int alloc_flags,
         struct page *page;
         unsigned long mark;

+        /*
+         * CDM nodes get skipped if the requested gfp flag
+         * does not have __GFP_THISNODE set or the nodemask
+         * does not have any CDM nodes in case the nodemask
+         * is non NULL (explicit allocation requests from
+         * kernel or user process MPOL_BIND policy which has
+         * CDM nodes).
+         */
+        if (is_cdm_node(zone->zone_pgdat->node_id)) {
+            if (!(gfp_mask & __GFP_THISNODE)) {
+                if (!ac->nodemask)
+                    continue;
+
+                if (!nodemask_has_cdm(*ac->nodemask))
+                    continue;

nodemask_has_cdm() looks quite expensive, combined with the loop here
that's O(n^2). But I don't understand why you need it. If there is no
cdm node in the nodemask, then we never reach this code with a cdm node,
because the zonelist iterator already checks the nodemask? Am I missing
something?

A CDM zone can be selected during zonelist iteration if

	(1) If nodemask is NULL (where all zones are eligible)

		(1) Skip it if __GFP_THISNODE is not mentioned
		(2) Pick it if __GFP_THISNODE is mentioned

	(2) If nodemask has CDM (where CDM zones are eligible)

		(1) Pick it if nodemask has CDM
		(2) Pick it if __GFP_THISNODE is mentioned

(1) (1) Enforces the primary isolation
(2) (1) Is the only option which could be O(n^2) as the worst case

Checking for both the zone being a CDM zone and the nodemask containing
CDM node has to happen together for (2) (1). But we dont run into this
option unless we have first checked if request contains __GFP_THISNODE
and that nodemask is really a non NULL value. Hence the number cases
getting into (2) (1) should be less. IIUC only the user space MPOL_BIND
ones will come here.

Maybe I'm still missing something, but when you do nodemask_has_cdm() above then 
we already passed "if (!ac->nodemask) continue" which means ac->nodemask is not 
null, which means the zonelist iterator already did the filtering on 
ac->nodemask, and if this zone passed the filter and it's a cdm zone, then it 
has to be set in the nodemask?

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>