Re: [PATCH 14/15] x86/numa: remove redundant iteration over memblock.reserved

On 07/28/20 at 05:15pm, Mike Rapoport wrote:

On Tue, Jul 28, 2020 at 07:02:54PM +0800, Baoquan He wrote:

quoted

On 07/28/20 at 08:11am, Mike Rapoport wrote:

quoted

From: Mike Rapoport <redacted>

numa_clear_kernel_node_hotplug() function first traverses numa_meminfo
regions to set node ID in memblock.reserved and than traverses
memblock.reserved to update reserved_nodemask to include node IDs that were
set in the first loop.

Remove redundant traversal over memblock.reserved and update
reserved_nodemask while iterating over numa_meminfo.

Signed-off-by: Mike Rapoport <redacted>
---
 arch/x86/mm/numa.c | 26 ++++++++++----------------
 1 file changed, 10 insertions(+), 16 deletions(-)

diff --git a/arch/x86/mm/numa.c b/arch/x86/mm/numa.c
index 8ee952038c80..4078abd33938 100644
--- a/arch/x86/mm/numa.c
+++ b/arch/x86/mm/numa.c

@@ -498,31 +498,25 @@ static void __init numa_clear_kernel_node_hotplug(void)
 	 * and use those ranges to set the nid in memblock.reserved.
 	 * This will split up the memblock regions along node
 	 * boundaries and will set the node IDs as well.
+	 *
+	 * The nid will also be set in reserved_nodemask which is later
+	 * used to clear MEMBLOCK_HOTPLUG flag.
+	 *
+	 * [ Note, when booting with mem=nn[kMG] or in a kdump kernel,
+	 *   numa_meminfo might not include all memblock.reserved
+	 *   memory ranges, because quirks such as trim_snb_memory()
+	 *   reserve specific pages for Sandy Bridge graphics.
+	 *   These ranges will remain with nid == MAX_NUMNODES. ]
 	 */
 	for (i = 0; i < numa_meminfo.nr_blks; i++) {
 		struct numa_memblk *mb = numa_meminfo.blk + i;
 		int ret;
 
 		ret = memblock_set_node(mb->start, mb->end - mb->start, &memblock.reserved, mb->nid);
+		node_set(mb->nid, reserved_nodemask);

Really? This will set all node id into reserved_nodemask. But in the
current code, it's setting nid into memblock reserved region which
interleaves with numa_memoinfo, then get those nid and set it in
reserved_nodemask. This is so different, with my understanding. Please
correct me if I am wrong.

You are right, I've missed the intersections of numa_meminfo with
memblock.reserved.

x86 interaction with membock is so, hmm, interesting...

Yeah, numa_clear_kernel_node_hotplug() intends to find out any memory node
which has reserved memory, then make it as unmovable. Setting all node
id into reserved_nodemask will break the use case of hot removing hotpluggable
boot memory after system bootup.