mm: Initialise per_cpu_nodestats for all online pgdats at boot

Paul Mackerras and Reza Arbab reported that machines with memoryless nodes
fails when vmstats are refreshed. Paul reported an oops as follows

[    1.713998] Unable to handle kernel paging request for data at address 0xff7a10000
[    1.714164] Faulting instruction address: 0xc000000000270cd0
[    1.714304] Oops: Kernel access of bad area, sig: 11 [#1]
[    1.714414] SMP NR_CPUS=2048 NUMA PowerNV
[    1.714530] Modules linked in:
[    1.714647] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 4.7.0-kvm+ #118
[    1.714786] task: c000000ff0680010 task.stack: c000000ff0704000
[    1.714926] NIP: c000000000270cd0 LR: c000000000270ce8 CTR: 0000000000000000
[    1.715093] REGS: c000000ff0707900 TRAP: 0300   Not tainted  (4.7.0-kvm+)
[    1.715232] MSR: 9000000102009033 <SF,HV,VEC,EE,ME,IR,DR,RI,LE,TM[E]>  CR: 846b6824  XER: 20000000
[    1.715748] CFAR: c000000000008768 DAR: 0000000ff7a10000 DSISR: 42000000 SOFTE: 1
GPR00: c000000000270d08 c000000ff0707b80 c0000000011fb200 0000000000000000
GPR04: 0000000000000800 0000000000000000 0000000000000000 0000000000000000
GPR08: ffffffffffffffff 0000000000000000 0000000ff7a10000 c00000000122aae0
GPR12: c000000000a1e440 c00000000fb80000 c00000000000c188 0000000000000000
GPR16: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
GPR20: 0000000000000000 0000000000000000 0000000000000000 c000000000cecad0
GPR24: c000000000d035b8 c000000000d6cd18 c000000000d6cd18 c000001fffa86300
GPR28: 0000000000000000 c000001fffa96300 c000000001230034 c00000000122eb18
[    1.717484] NIP [c000000000270cd0] refresh_zone_stat_thresholds+0x80/0x240
[    1.717568] LR [c000000000270ce8] refresh_zone_stat_thresholds+0x98/0x240
[    1.717648] Call Trace:
[    1.717687] [c000000ff0707b80] [c000000000270d08] refresh_zone_stat_thresholds+0xb8/0x240 (unreliable)

Both supplied potential fixes but one potentially misses checks and another
had redundant initialisations. This version initialises per_cpu_nodestats
on a per-pgdat basis instead of on a per-zone basis.

Reported-by: Paul Mackerras <redacted>
Reported-by: Reza Arbab <redacted>
Signed-off-by: Mel Gorman <redacted>
---
This has been compile-tested and boot-tested on a 32-bit KVM only. A
memoryless system was not available to test the patch with. A confirmation
from Paul and Reza that it resolves their problem is welcome.

 mm/page_alloc.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 39a372a2a1d6..fb975cec3518 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c

@@ -5257,11 +5257,6 @@ static void __meminit setup_zone_pageset(struct zone *zone)
 	zone->pageset = alloc_percpu(struct per_cpu_pageset);
 	for_each_possible_cpu(cpu)
 		zone_pageset_init(zone, cpu);
-
-	if (!zone->zone_pgdat->per_cpu_nodestats) {
-		zone->zone_pgdat->per_cpu_nodestats =
-			alloc_percpu(struct per_cpu_nodestat);
-	}
 }
 
 /*

@@ -5270,10 +5265,15 @@ static void __meminit setup_zone_pageset(struct zone *zone)
  */
 void __init setup_per_cpu_pageset(void)
 {
+	struct pglist_data *pgdat;
 	struct zone *zone;
 
 	for_each_populated_zone(zone)
 		setup_zone_pageset(zone);
+
+	for_each_online_pgdat(pgdat)
+		pgdat->per_cpu_nodestats =
+			alloc_percpu(struct per_cpu_nodestat);
 }
 
 static noinline __ref

On Thu, Aug 04, 2016 at 10:24:04AM +0100, Mel Gorman wrote:

This has been compile-tested and boot-tested on a 32-bit KVM only. A
memoryless system was not available to test the patch with. A confirmation
from Paul and Reza that it resolves their problem is welcome.

Works for me. Thanks, Mel!

Tested-by: Reza Arbab <redacted>

-- 
Reza Arbab

On Thu, Aug 04, 2016 at 10:24:04AM +0100, Mel Gorman wrote:

Paul Mackerras and Reza Arbab reported that machines with memoryless nodes
fails when vmstats are refreshed. Paul reported an oops as follows

[    1.713998] Unable to handle kernel paging request for data at address 0xff7a10000
[    1.714164] Faulting instruction address: 0xc000000000270cd0
[    1.714304] Oops: Kernel access of bad area, sig: 11 [#1]
[    1.714414] SMP NR_CPUS=2048 NUMA PowerNV
[    1.714530] Modules linked in:
[    1.714647] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 4.7.0-kvm+ #118
[    1.714786] task: c000000ff0680010 task.stack: c000000ff0704000
[    1.714926] NIP: c000000000270cd0 LR: c000000000270ce8 CTR: 0000000000000000
[    1.715093] REGS: c000000ff0707900 TRAP: 0300   Not tainted  (4.7.0-kvm+)
[    1.715232] MSR: 9000000102009033 <SF,HV,VEC,EE,ME,IR,DR,RI,LE,TM[E]>  CR: 846b6824  XER: 20000000
[    1.715748] CFAR: c000000000008768 DAR: 0000000ff7a10000 DSISR: 42000000 SOFTE: 1
GPR00: c000000000270d08 c000000ff0707b80 c0000000011fb200 0000000000000000
GPR04: 0000000000000800 0000000000000000 0000000000000000 0000000000000000
GPR08: ffffffffffffffff 0000000000000000 0000000ff7a10000 c00000000122aae0
GPR12: c000000000a1e440 c00000000fb80000 c00000000000c188 0000000000000000
GPR16: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
GPR20: 0000000000000000 0000000000000000 0000000000000000 c000000000cecad0
GPR24: c000000000d035b8 c000000000d6cd18 c000000000d6cd18 c000001fffa86300
GPR28: 0000000000000000 c000001fffa96300 c000000001230034 c00000000122eb18
[    1.717484] NIP [c000000000270cd0] refresh_zone_stat_thresholds+0x80/0x240
[    1.717568] LR [c000000000270ce8] refresh_zone_stat_thresholds+0x98/0x240
[    1.717648] Call Trace:
[    1.717687] [c000000ff0707b80] [c000000000270d08] refresh_zone_stat_thresholds+0xb8/0x240 (unreliable)

Both supplied potential fixes but one potentially misses checks and another
had redundant initialisations. This version initialises per_cpu_nodestats
on a per-pgdat basis instead of on a per-zone basis.

Reported-by: Paul Mackerras <redacted>
Reported-by: Reza Arbab <redacted>
Signed-off-by: Mel Gorman <redacted>

That works, thanks.

Tested-by: Paul Mackerras <redacted>

On Thu, Aug 04, 2016 at 10:24:04AM +0100, Mel Gorman wrote:

[    1.713998] Unable to handle kernel paging request for data at address 0xff7a10000
[    1.714164] Faulting instruction address: 0xc000000000270cd0
[    1.714304] Oops: Kernel access of bad area, sig: 11 [#1]
[    1.714414] SMP NR_CPUS=2048 NUMA PowerNV
[    1.714530] Modules linked in:
[    1.714647] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 4.7.0-kvm+ #118
[    1.714786] task: c000000ff0680010 task.stack: c000000ff0704000
[    1.714926] NIP: c000000000270cd0 LR: c000000000270ce8 CTR: 0000000000000000
[    1.715093] REGS: c000000ff0707900 TRAP: 0300   Not tainted  (4.7.0-kvm+)
[    1.715232] MSR: 9000000102009033 <SF,HV,VEC,EE,ME,IR,DR,RI,LE,TM[E]>  CR: 846b6824  XER: 20000000
[    1.715748] CFAR: c000000000008768 DAR: 0000000ff7a10000 DSISR: 42000000 SOFTE: 1
GPR00: c000000000270d08 c000000ff0707b80 c0000000011fb200 0000000000000000
GPR04: 0000000000000800 0000000000000000 0000000000000000 0000000000000000
GPR08: ffffffffffffffff 0000000000000000 0000000ff7a10000 c00000000122aae0
GPR12: c000000000a1e440 c00000000fb80000 c00000000000c188 0000000000000000
GPR16: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
GPR20: 0000000000000000 0000000000000000 0000000000000000 c000000000cecad0
GPR24: c000000000d035b8 c000000000d6cd18 c000000000d6cd18 c000001fffa86300
GPR28: 0000000000000000 c000001fffa96300 c000000001230034 c00000000122eb18
[    1.717484] NIP [c000000000270cd0] refresh_zone_stat_thresholds+0x80/0x240
[    1.717568] LR [c000000000270ce8] refresh_zone_stat_thresholds+0x98/0x240
[    1.717648] Call Trace:
[    1.717687] [c000000ff0707b80] [c000000000270d08] refresh_zone_stat_thresholds+0xb8/0x240 (unreliable)

I've been investigating node hotplug. That path is also going to require 
initialization of per_cpu_nodestats. This worked for me:

diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
index 3894b65..41266dc 100644
--- a/mm/memory_hotplug.c
+++ b/mm/memory_hotplug.c

@@ -1219,6 +1219,7 @@ static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
  
  	/* init node's zones as empty zones, we don't have any present pages.*/
  	free_area_init_node(nid, zones_size, start_pfn, zholes_size);
+	pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
  
  	/*
  	 * The node we allocated has no zone fallback lists. For avoiding

@@ -1249,6 +1250,7 @@ static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
  static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
  {
  	arch_refresh_nodedata(nid, NULL);
+	free_percpu(pgdat->per_cpu_nodestats);
  	arch_free_nodedata(pgdat);
  	return;
  }

-- 
Reza Arbab

On Wed, Aug 10, 2016 at 12:59:40PM -0500, Reza Arbab wrote:

On Thu, Aug 04, 2016 at 10:24:04AM +0100, Mel Gorman wrote:

quoted

[    1.713998] Unable to handle kernel paging request for data at address 0xff7a10000
[    1.714164] Faulting instruction address: 0xc000000000270cd0
[    1.714304] Oops: Kernel access of bad area, sig: 11 [#1]
[    1.714414] SMP NR_CPUS=2048 NUMA PowerNV
[    1.714530] Modules linked in:
[    1.714647] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 4.7.0-kvm+ #118
[    1.714786] task: c000000ff0680010 task.stack: c000000ff0704000
[    1.714926] NIP: c000000000270cd0 LR: c000000000270ce8 CTR: 0000000000000000
[    1.715093] REGS: c000000ff0707900 TRAP: 0300   Not tainted  (4.7.0-kvm+)
[    1.715232] MSR: 9000000102009033 <SF,HV,VEC,EE,ME,IR,DR,RI,LE,TM[E]>  CR: 846b6824  XER: 20000000
[    1.715748] CFAR: c000000000008768 DAR: 0000000ff7a10000 DSISR: 42000000 SOFTE: 1
GPR00: c000000000270d08 c000000ff0707b80 c0000000011fb200 0000000000000000
GPR04: 0000000000000800 0000000000000000 0000000000000000 0000000000000000
GPR08: ffffffffffffffff 0000000000000000 0000000ff7a10000 c00000000122aae0
GPR12: c000000000a1e440 c00000000fb80000 c00000000000c188 0000000000000000
GPR16: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
GPR20: 0000000000000000 0000000000000000 0000000000000000 c000000000cecad0
GPR24: c000000000d035b8 c000000000d6cd18 c000000000d6cd18 c000001fffa86300
GPR28: 0000000000000000 c000001fffa96300 c000000001230034 c00000000122eb18
[    1.717484] NIP [c000000000270cd0] refresh_zone_stat_thresholds+0x80/0x240
[    1.717568] LR [c000000000270ce8] refresh_zone_stat_thresholds+0x98/0x240
[    1.717648] Call Trace:
[    1.717687] [c000000ff0707b80] [c000000000270d08] refresh_zone_stat_thresholds+0xb8/0x240 (unreliable)

I've been investigating node hotplug. That path is also going to require
initialization of per_cpu_nodestats. This worked for me:

Fix looks ok. Can you add a proper changelog to it including an example
oops or do you need me to do it?

-- 
Mel Gorman
SUSE Labs

On Thu, Aug 11, 2016 at 10:28:08AM +0100, Mel Gorman wrote:

Fix looks ok. Can you add a proper changelog to it including an example
oops or do you need me to do it?

Sure, no problem. Patch to follow.

-- 
Reza Arbab

The following oops occurs after a pgdat is hotadded:

[   86.839956] Unable to handle kernel paging request for data at address 0x00c30001
[   86.840132] Faulting instruction address: 0xc00000000022f8f4
[   86.840328] Oops: Kernel access of bad area, sig: 11 [#1]
[   86.840468] SMP NR_CPUS=2048 NUMA pSeries
[   86.840612] Modules linked in: ip6t_rpfilter ip6t_REJECT nf_reject_ipv6 ipt_REJECT nf_reject_ipv4 xt_conntrack ebtable_nat ebtable_broute bridge stp llc ebtable_filter ebtables ip6table_nat nf_conntrack_ipv6 nf_defrag_ipv6 nf_nat_ipv6 ip6table_mangle ip6table_security ip6table_raw ip6table_filter ip6_tables iptable_nat nf_conntrack_ipv4 nf_defrag_ipv4 nf_nat_ipv4 nf_nat nf_conntrack iptable_mangle iptable_security iptable_raw iptable_filter nls_utf8 isofs sg virtio_balloon uio_pdrv_genirq uio ip_tables xfs libcrc32c sr_mod cdrom sd_mod virtio_net ibmvscsi scsi_transport_srp virtio_pci virtio_ring virtio dm_mirror dm_region_hash dm_log dm_mod
[   86.842955] CPU: 0 PID: 0 Comm: swapper/0 Tainted: G        W 4.8.0-rc1-device #110
[   86.843140] task: c000000000ef3080 task.stack: c000000000f6c000
[   86.843323] NIP: c00000000022f8f4 LR: c00000000022f948 CTR: 0000000000000000
[   86.843595] REGS: c000000000f6fa50 TRAP: 0300   Tainted: G        W (4.8.0-rc1-device)
[   86.843889] MSR: 800000010280b033 <SF,VEC,VSX,EE,FP,ME,IR,DR,RI,LE,TM[E]>  CR: 84002028  XER: 20000000
[   86.844624] CFAR: d000000001d2013c DAR: 0000000000c30001 DSISR: 40000000 SOFTE: 0
GPR00: c00000000022f948 c000000000f6fcd0 c000000000f71400 0000000000000001
GPR04: 0000000000000100 0000000000000000 0000000000000000 0000000000c30000
GPR08: ffffffffffffffff 0000000000000001 0000000000c30000 00000000ffffffff
GPR12: 0000000000002200 c000000001300000 c000000000faefb4 c000000000faefa8
GPR16: c000000000f6c000 c000000000f6c080 c000000000bf15b0 c000000000f6c080
GPR20: c000000000bf4928 0000000000000000 0000000000000003 c000000000bf4968
GPR24: c0000000ffed0000 0000000000000000 0000000000000000 c000000000f6fd58
GPR28: 0000000000000001 0000000000000001 c000000000f6fcf0 c0000000ffed9c08
[   86.847747] NIP [c00000000022f8f4] refresh_cpu_vm_stats+0x1a4/0x2f0
[   86.847897] LR [c00000000022f948] refresh_cpu_vm_stats+0x1f8/0x2f0
[   86.848060] Call Trace:
[   86.848183] [c000000000f6fcd0] [c00000000022f948] refresh_cpu_vm_stats+0x1f8/0x2f0 (unreliable)

Add per_cpu_nodestats initialization to the hotplug codepath.

Signed-off-by: Reza Arbab <redacted>
---
 mm/memory_hotplug.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
index 3894b65..41266dc 100644
--- a/mm/memory_hotplug.c
+++ b/mm/memory_hotplug.c

@@ -1219,6 +1219,7 @@ static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
 
 	/* init node's zones as empty zones, we don't have any present pages.*/
 	free_area_init_node(nid, zones_size, start_pfn, zholes_size);
+	pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
 
 	/*
 	 * The node we allocated has no zone fallback lists. For avoiding

@@ -1249,6 +1250,7 @@ static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
 {
 	arch_refresh_nodedata(nid, NULL);
+	free_percpu(pgdat->per_cpu_nodestats);
 	arch_free_nodedata(pgdat);
 	return;
 }

-- 
1.8.3.1

On 12/08/16 02:04, Reza Arbab wrote:

I wonder if node_set_online() should do the allocation and offline should free.
But that would be a larger change

Balbir