From b277fc793daf258877b4c0744b52f69d6e6ba22e Mon Sep 17 00:00:00 2001
From: "Aneesh Kumar K.V" <aneesh.kumar@linux.ibm.com>
Date: Tue, 4 Apr 2023 09:44:33 +0530
Subject: [PATCH] powerpc/papr_scm: Update the NUMA distance table for the
target node
References: bsc#1209999 ltc#202140 bsc#1142685 ltc#179509 FATE#327775 git-fixes
Patch-mainline: v6.3-rc7
Git-commit: b277fc793daf258877b4c0744b52f69d6e6ba22e
Platform device helper routines won't update the NUMA distance table
while creating a platform device, even if the device is present on a
NUMA node that doesn't have memory or CPU. This is especially true for
pmem devices. If the target node of the pmem device is not online, we
find the nearest online node to the device and associate the pmem device
with that online node. To find the nearest online node, we should have
the numa distance table updated correctly. Update the distance
information during the device probe.
For a papr scm device on NUMA node 3 distance_lookup_table value for
distance_ref_points_depth = 2 before and after fix is below:
Before fix:
node 3 distance depth 0 - 0
node 3 distance depth 1 - 0
node 4 distance depth 0 - 4
node 4 distance depth 1 - 2
node 5 distance depth 0 - 5
node 5 distance depth 1 - 1
After fix
node 3 distance depth 0 - 3
node 3 distance depth 1 - 1
node 4 distance depth 0 - 4
node 4 distance depth 1 - 2
node 5 distance depth 0 - 5
node 5 distance depth 1 - 1
Without the fix, the nearest numa node to the pmem device (NUMA node 3)
will be picked as 4. After the fix, we get the correct numa node which
is 5.
Fixes: da1115fdbd6e ("powerpc/nvdimm: Pick nearby online node if the device node is not online")
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://msgid.link/20230404041433.1781804-1-aneesh.kumar@linux.ibm.com
Acked-by: Michal Suchanek <msuchanek@suse.de>
---
arch/powerpc/mm/numa.c | 1 +
arch/powerpc/platforms/pseries/papr_scm.c | 7 +++++++
2 files changed, 8 insertions(+)
diff --git a/arch/powerpc/mm/numa.c b/arch/powerpc/mm/numa.c
--- a/arch/powerpc/mm/numa.c
+++ b/arch/powerpc/mm/numa.c
@@ -312,6 +312,7 @@ void update_numa_distance(struct device_
return;
}
}
+EXPORT_SYMBOL_GPL(update_numa_distance);
static int __init find_primary_domain_index(void)
{
diff --git a/arch/powerpc/platforms/pseries/papr_scm.c b/arch/powerpc/platforms/pseries/papr_scm.c
index 2f8385523a13..1a53e048ceb7 100644
--- a/arch/powerpc/platforms/pseries/papr_scm.c
+++ b/arch/powerpc/platforms/pseries/papr_scm.c
@@ -1428,6 +1428,13 @@ static int papr_scm_probe(struct platform_device *pdev)
return -ENODEV;
}
+ /*
+ * open firmware platform device create won't update the NUMA
+ * distance table. For PAPR SCM devices we use numa_map_to_online_node()
+ * to find the nearest online NUMA node and that requires correct
+ * distance table information.
+ */
+ update_numa_distance(dn);
p = kzalloc(sizeof(*p), GFP_KERNEL);
if (!p)
--
2.40.0