This seems to have been quite an oversite in Xen's NUMA support, albeit it
probably because NUMA was arch-specific at first.

Take the opportunity to sort the includes.

Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
---
CC: Anthony PERARD <anthony.perard@vates.tech>
CC: Michal Orzel <michal.orzel@amd.com>
CC: Jan Beulich <jbeulich@suse.com>
CC: Julien Grall <julien@xen.org>
CC: Roger Pau Monné <roger.pau@citrix.com>
CC: Stefano Stabellini <sstabellini@kernel.org>

https://gitlab.com/xen-project/hardware/xen-staging/-/pipelines/1773336913

For the !NUMA case, cpu_to_node() defaults to 0 rather than NUMA_NO_NODE,
making memflags nonzero.  I'm not sure how wise this is, but ARM runs seem
happy with it.
---
 xen/common/percpu.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/xen/common/percpu.c b/xen/common/percpu.c
index c6ecd95a0874..cdd70acbeaf3 100644
--- a/xen/common/percpu.c
+++ b/xen/common/percpu.c
@@ -1,8 +1,9 @@
 /* SPDX-License-Identifier: GPL-2.0-only */
-#include <xen/percpu.h>
 #include <xen/cpu.h>
 #include <xen/init.h>
 #include <xen/mm.h>
+#include <xen/numa.h>
+#include <xen/percpu.h>
 #include <xen/rcupdate.h>
 #include <xen/sections.h>
 
@@ -27,6 +28,8 @@ void __init percpu_init_areas(void)
 
 static int init_percpu_area(unsigned int cpu)
 {
+    nodeid_t node = cpu_to_node(cpu);
+    unsigned int memflags = node != NUMA_NO_NODE ? MEMF_node(node) : 0;
     char *p;
 
     if ( __per_cpu_offset[cpu] != INVALID_PERCPU_AREA )
@@ -34,7 +37,7 @@ static int init_percpu_area(unsigned int cpu)
                ? 0
                : -EBUSY;
 
-    if ( (p = alloc_xenheap_pages(PERCPU_ORDER, 0)) == NULL )
+    if ( (p = alloc_xenheap_pages(PERCPU_ORDER, memflags)) == NULL )
         return -ENOMEM;
 
     memset(p, 0, __per_cpu_data_end - __per_cpu_start);

base-commit: 7f8c2dd829358406dc9e50d5e386e4d2b20b4303
-- 
2.39.5

On 17.04.2025 10:53, Andrew Cooper wrote:
> This seems to have been quite an oversite in Xen's NUMA support, albeit it

Nit: oversight?

> probably because NUMA was arch-specific at first.
> 
> Take the opportunity to sort the includes.
> 
> Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>

Reviewed-by: Jan Beulich <jbeulich@suse.com>

> For the !NUMA case, cpu_to_node() defaults to 0 rather than NUMA_NO_NODE,
> making memflags nonzero.  I'm not sure how wise this is, but ARM runs seem
> happy with it.

As to Arm: May be worth noting in the commit message that this is unlikely
to have any useful effect when SEPARATE_XENHEAP=y. I guess our only option
to address that (if any Arm32 NUMA systems exist) would be to use vmap().

Jan