[PATCH 3/5] xen/mm: Optimise getting free page counts per NUMA node

[Bernhard Kaindl <bernhard.kaindl@xxxxxxxxxx>]

From: Alejandro Vallejo <alejandro.vallejo@xxxxxxxxx>

Introduce per-node free page counters which enhance the
efficiency of avail_node_heap_pages():

- It no longer needs to iterate over all zones of a node.

- It is utilised by the numainfo hypercall and the debug-key 'u'
  to display NUMA information in the printk buffer

- This aggregate will be needed for the node-specific claims
  feature to determine the number of free pages in a node in
  the hot path of get_free_buddy() without looping over all
  zones of a node.

Signed-off-by: Alejandro Vallejo <alejandro.vallejo@xxxxxxxxx>
Signed-off-by: Bernhard Kaindl <bernhard.kaindl@xxxxxxxxxx>

---

Applied Jan's review:
https://lists.xenproject.org/archives/html/xen-devel/2026-03/msg00144.html

Changes:
- Removed accessor macro
- Abandoned conversion to unsigned long
---
 xen/common/page_alloc.c | 19 ++++++++++++++++++-
 1 file changed, 18 insertions(+), 1 deletion(-)

diff --git a/xen/common/page_alloc.c b/xen/common/page_alloc.c
index 0fab1630e318..95bae26d1c1f 100644
--- a/xen/common/page_alloc.c
+++ b/xen/common/page_alloc.c
@@ -485,7 +485,20 @@ static unsigned long node_need_scrub[MAX_NUMNODES];
 
 static unsigned long *avail[MAX_NUMNODES];
 static long total_avail_pages;
+/*
+ * Sum of the free pages in all zones of that node.
+ * Provided via sysctl by NUMA node placement decisions of domain builders and
+ * for monitoring. It is also logged with debug-key 'u' for NUMA debugging.
+ */
+static long node_avail_pages[MAX_NUMNODES];
 
+/*
+ * The global heap lock, protecting access to the heap and related structures.
+ * It protects the heap and claims of the buddy allocator and d->*claims.
+ * The locking order is: d->page_alloc_lock (optional) -> heap_lock
+ * - Numerous external callers holding d->page_alloc_lock call functions
+ *   taking the heap_lock. Note: Violating it would cause an ABBA deadlock.
+ */
 static DEFINE_SPINLOCK(heap_lock);
 static long outstanding_claims; /* total outstanding claims by all domains */
 
@@ -1097,6 +1110,8 @@ static struct page_info *alloc_heap_pages(
         }
     }
 
+    ASSERT(node_avail_pages[node] >= request);
+    node_avail_pages[node] -= request;
     ASSERT(avail[node][zone] >= request);
     avail[node][zone] -= request;
     total_avail_pages -= request;
@@ -1287,6 +1302,7 @@ static int reserve_offlined_page(struct page_info *head)
         if ( !page_state_is(cur_head, offlined) )
             continue;
 
+        node_avail_pages[node]--;
         avail[node][zone]--;
         total_avail_pages--;
         ASSERT(total_avail_pages >= 0);
@@ -1611,6 +1627,7 @@ static void free_heap_pages(
         }
     }
 
+    node_avail_pages[node] += 1 << order;
     avail[node][zone] += 1 << order;
     total_avail_pages += 1 << order;
     if ( need_scrub )
@@ -2878,7 +2895,7 @@ unsigned long avail_domheap_pages_region(
 
 unsigned long avail_node_heap_pages(unsigned int nodeid)
 {
-    return avail_heap_pages(MEMZONE_XEN, NR_ZONES -1, nodeid);
+    return node_avail_pages[nodeid];
 }
 
 
-- 
2.39.5