diff mbox series

[RFC,v2,03/21] x86/numa_emulation: fix fake NUMA in uniform case

Message ID 20181226133351.229014333@intel.com (mailing list archive)
State New, archived
Headers show
Series PMEM NUMA node and hotness accounting/migration | expand

Commit Message

Fengguang Wu Dec. 26, 2018, 1:14 p.m. UTC 
From: Fan Du <fan.du@intel.com>

The index of numa_meminfo is expected to the same as of numa_meminfo.blk[].
and numa_remove_memblk_from break the expectation.

2S system does not break, because

before numa_remove_memblk_from
index  nid
0	0
1	1

after numa_remove_memblk_from

index  nid
0	1
1	1

If you try to configure uniform fake node in 4S system.
index  nid
0	0
1	1
2       2
3	3

node 3 will be removed by numa_remove_memblk_from when iterate index 2.
so we only create fake node for 3 physcial node, and a portion of memroy
wasted as much as it hit lost pages checking in numa_meminfo_cover_memory.

Signed-off-by: Fan Du <fan.du@intel.com>

---
 arch/x86/mm/numa_emulation.c |   16 +++++++++++++++-
 1 file changed, 15 insertions(+), 1 deletion(-)
diff mbox series

Patch

--- linux.orig/arch/x86/mm/numa_emulation.c	2018-12-23 19:20:51.570664269 +0800
+++ linux/arch/x86/mm/numa_emulation.c	2018-12-23 19:20:51.566664364 +0800
@@ -381,7 +381,21 @@  void __init numa_emulation(struct numa_m
 		goto no_emu;
 
 	memset(&ei, 0, sizeof(ei));
-	pi = *numa_meminfo;
+
+	{
+		/* Make sure the index is identical with nid */
+		struct numa_meminfo *mi = numa_meminfo;
+		int nid;
+
+		for (i = 0; i < mi->nr_blks; i++) {
+			nid = mi->blk[i].nid;
+			pi.blk[nid].nid = nid;
+			pi.blk[nid].start = mi->blk[i].start;
+			pi.blk[nid].end = mi->blk[i].end;
+		}
+		pi.nr_blks = mi->nr_blks;
+
+	}
 
 	for (i = 0; i < MAX_NUMNODES; i++)
 		emu_nid_to_phys[i] = NUMA_NO_NODE;