Skip to contentSkip to conversation

sched/numa, mm: do not promote folios to nodes not set N_MEMORY

PATCH Superseded Byungchul Park 3 participants 11 messages Feb 14, 2024 – Feb 16, 2024

Subsystems (2)

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 sched/numa, mm: do not promote folios to nodes not set N_MEMORY

While running qemu with a configuration where some CPUs don't have their local memory and with a kernel numa balancing on, the following oops has been observed. It's because of null pointers of ->zone_pgdat of zones of those nodes that are not initialized at booting time. So should avoid nodes not set N_MEMORY from getting promoted.

Quoted 61 lines
BUG: unable to handle page fault for address: 00000000000033f3
#PF: supervisor read access in kernel mode
#PF: error_code(0x0000) - not-present page
PGD 0 P4D 0
Oops: 0000 [#1] PREEMPT SMP NOPTI
CPU: 2 PID: 895 Comm: masim Not tainted 6.6.0-dirty #255
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS
   rel-1.16.0-0-gd239552ce722-prebuilt.qemu.org 04/01/2014
RIP: 0010:wakeup_kswapd (./linux/mm/vmscan.c:7812)
Code: (omitted)
RSP: 0000:ffffc90004257d58 EFLAGS: 00010286
RAX: ffffffffffffffff RBX: ffff88883fff0480 RCX: 0000000000000003
RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88883fff0480
RBP: ffffffffffffffff R08: ff0003ffffffffff R09: ffffffffffffffff
R10: ffff888106c95540 R11: 0000000055555554 R12: 0000000000000003
R13: 0000000000000000 R14: 0000000000000000 R15: ffff88883fff0940
FS:  00007fc4b8124740(0000) GS:ffff888827c00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00000000000033f3 CR3: 000000026cc08004 CR4: 0000000000770ee0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
PKRU: 55555554
Call Trace:
 <TASK>

? __die
? page_fault_oops
? __pte_offset_map_lock
? exc_page_fault
? asm_exc_page_fault
? wakeup_kswapd
migrate_misplaced_page
__handle_mm_fault
handle_mm_fault
do_user_addr_fault
exc_page_fault
asm_exc_page_fault

RIP: 0033:0x55b897ba0808
Code: (omitted)
RSP: 002b:00007ffeefa821a0 EFLAGS: 00010287
RAX: 000055b89983acd0 RBX: 00007ffeefa823f8 RCX: 000055b89983acd0
RDX: 00007fc2f8122010 RSI: 0000000000020000 RDI: 000055b89983acd0
RBP: 00007ffeefa821a0 R08: 0000000000000037 R09: 0000000000000075
R10: 0000000000000000 R11: 0000000000000202 R12: 0000000000000000
R13: 00007ffeefa82410 R14: 000055b897ba5dd8 R15: 00007fc4b8340000
 </TASK>
Modules linked in:
CR2: 00000000000033f3
---[ end trace 0000000000000000  ]---
RIP: 0010:wakeup_kswapd (./linux/mm/vmscan.c:7812)
Code: (omitted)
RSP: 0000:ffffc90004257d58 EFLAGS: 00010286
RAX: ffffffffffffffff RBX: ffff88883fff0480 RCX: 0000000000000003
RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88883fff0480
RBP: ffffffffffffffff R08: ff0003ffffffffff R09: ffffffffffffffff
R10: ffff888106c95540 R11: 0000000055555554 R12: 0000000000000003
R13: 0000000000000000 R14: 0000000000000000 R15: ffff88883fff0940
FS:  00007fc4b8124740(0000) GS:ffff888827c00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00000000000033f3 CR3: 000000026cc08004 CR4: 0000000000770ee0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Quoted 2 lines

PKRU: 55555554
note: masim[895] exited with irqs disabled

---
 kernel/sched/fair.c | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)
kernel/sched/fair.c +17 -0
index d7a3c63a2171..6d215cc85f14 100644
@@ -1828,6 +1828,23 @@ bool should_numa_migrate_memory(struct task_struct *p, struct folio *folio,
 	int dst_nid = cpu_to_node(dst_cpu);
 	int last_cpupid, this_cpupid;
 
+	/*
+	 * A node of dst_nid might not have its local memory. Promoting
+	 * a folio to the node is meaningless. What's even worse, oops
+	 * can be observed by the null pointer of ->zone_pgdat in
+	 * various points of the code during migration.
+	 *
+	 * For instance, oops has been observed at CPU2 while qemu'ing:
+	 *
+	 * {qemu} \
+	 *    -numa node,nodeid=0,mem=1G,cpus=0-1 \
+	 *    -numa node,nodeid=1,cpus=2-3 \
+	 *    -numa node,nodeid=2,mem=8G \
+	 *    ...
+	 */
+	if (!node_state(dst_nid, N_MEMORY))
+		return false;
+
 	/*
 	 * The pages in slow memory node should be migrated according
 	 * to hot/cold instead of private/shared.
Signature · 1 line
2.17.1
#1 lore raw

Hi,

Quoted 69 lines · On Wed, Feb 14, 2024 at 12:53:55PM +0900 Byungchul Park wrote:

While running qemu with a configuration where some CPUs don't have their local memory and with a kernel numa balancing on, the following oops has been observed. It's because of null pointers of ->zone_pgdat of zones of those nodes that are not initialized at booting time. So should avoid nodes not set N_MEMORY from getting promoted.

Quoted 63 lines
BUG: unable to handle page fault for address: 00000000000033f3
#PF: supervisor read access in kernel mode
#PF: error_code(0x0000) - not-present page
PGD 0 P4D 0
Oops: 0000 [#1] PREEMPT SMP NOPTI
CPU: 2 PID: 895 Comm: masim Not tainted 6.6.0-dirty #255
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS
   rel-1.16.0-0-gd239552ce722-prebuilt.qemu.org 04/01/2014
RIP: 0010:wakeup_kswapd (./linux/mm/vmscan.c:7812)
Code: (omitted)
RSP: 0000:ffffc90004257d58 EFLAGS: 00010286
RAX: ffffffffffffffff RBX: ffff88883fff0480 RCX: 0000000000000003
RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88883fff0480
RBP: ffffffffffffffff R08: ff0003ffffffffff R09: ffffffffffffffff
R10: ffff888106c95540 R11: 0000000055555554 R12: 0000000000000003
R13: 0000000000000000 R14: 0000000000000000 R15: ffff88883fff0940
FS:  00007fc4b8124740(0000) GS:ffff888827c00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00000000000033f3 CR3: 000000026cc08004 CR4: 0000000000770ee0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
PKRU: 55555554
Call Trace:
 <TASK>

? __die
? page_fault_oops
? __pte_offset_map_lock
? exc_page_fault
? asm_exc_page_fault
? wakeup_kswapd
migrate_misplaced_page
__handle_mm_fault
handle_mm_fault
do_user_addr_fault
exc_page_fault
asm_exc_page_fault

RIP: 0033:0x55b897ba0808
Code: (omitted)
RSP: 002b:00007ffeefa821a0 EFLAGS: 00010287
RAX: 000055b89983acd0 RBX: 00007ffeefa823f8 RCX: 000055b89983acd0
RDX: 00007fc2f8122010 RSI: 0000000000020000 RDI: 000055b89983acd0
RBP: 00007ffeefa821a0 R08: 0000000000000037 R09: 0000000000000075
R10: 0000000000000000 R11: 0000000000000202 R12: 0000000000000000
R13: 00007ffeefa82410 R14: 000055b897ba5dd8 R15: 00007fc4b8340000
 </TASK>
Modules linked in:
CR2: 00000000000033f3
---[ end trace 0000000000000000  ]---
RIP: 0010:wakeup_kswapd (./linux/mm/vmscan.c:7812)
Code: (omitted)
RSP: 0000:ffffc90004257d58 EFLAGS: 00010286
RAX: ffffffffffffffff RBX: ffff88883fff0480 RCX: 0000000000000003
RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88883fff0480
RBP: ffffffffffffffff R08: ff0003ffffffffff R09: ffffffffffffffff
R10: ffff888106c95540 R11: 0000000055555554 R12: 0000000000000003
R13: 0000000000000000 R14: 0000000000000000 R15: ffff88883fff0940
FS:  00007fc4b8124740(0000) GS:ffff888827c00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00000000000033f3 CR3: 000000026cc08004 CR4: 0000000000770ee0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400

PKRU: 55555554
note: masim[895] exited with irqs disabled

I think you could trim the down a little bit.

Quoted 19 lines
---
 kernel/sched/fair.c | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)
kernel/sched/fair.c +4 -0
index d7a3c63a2171..6d215cc85f14 100644
@@ -1828,6 +1828,23 @@ bool should_numa_migrate_memory(struct task_struct *p, struct folio *folio,
 	int dst_nid = cpu_to_node(dst_cpu);
 	int last_cpupid, this_cpupid;
 
+	/*
+	 * A node of dst_nid might not have its local memory. Promoting
+	 * a folio to the node is meaningless. What's even worse, oops
+	 * can be observed by the null pointer of ->zone_pgdat in

kernel/sched/fair.c bool should_numa_migrate_memory(struct task_struct *p, struct folio *folio, · line 1844

Quoted 10 lines
kernel/sched/fair.c +9 -0
+	 * various points of the code during migration.
+	 *
 
+	 * For instance, oops has been observed at CPU2 while qemu'ing:
+	 *
+	 * {qemu} \
+	 *    -numa node,nodeid=0,mem=1G,cpus=0-1 \
+	 *    -numa node,nodeid=1,cpus=2-3 \
+	 *    -numa node,nodeid=2,mem=8G \
+	 *    ...

This part above should probably be in the commit message not in the code. The first paragraph of comment is plenty.

Otherwise, I think the check probably makes sense.

Cheers,
Phil

Quoted 11 lines
kernel/sched/fair.c +4 -0
+	 */
+	if (!node_state(dst_nid, N_MEMORY))
+		return false;
+
 	/*
 	 * The pages in slow memory node should be migrated according

* to hot/cold instead of private/shared.

Signature · 1 line
2.17.1
#2 lore raw
Quoted 109 lines · On Wed, Feb 14, 2024 at 07:31:37AM -0500 Phil Auld wrote:

Hi,

Quoted 69 lines · On Wed, Feb 14, 2024 at 12:53:55PM +0900 Byungchul Park wrote:

While running qemu with a configuration where some CPUs don't have their local memory and with a kernel numa balancing on, the following oops has been observed. It's because of null pointers of ->zone_pgdat of zones of those nodes that are not initialized at booting time. So should avoid nodes not set N_MEMORY from getting promoted.

Quoted 63 lines
BUG: unable to handle page fault for address: 00000000000033f3
#PF: supervisor read access in kernel mode
#PF: error_code(0x0000) - not-present page
PGD 0 P4D 0
Oops: 0000 [#1] PREEMPT SMP NOPTI
CPU: 2 PID: 895 Comm: masim Not tainted 6.6.0-dirty #255
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS
   rel-1.16.0-0-gd239552ce722-prebuilt.qemu.org 04/01/2014
RIP: 0010:wakeup_kswapd (./linux/mm/vmscan.c:7812)
Code: (omitted)
RSP: 0000:ffffc90004257d58 EFLAGS: 00010286
RAX: ffffffffffffffff RBX: ffff88883fff0480 RCX: 0000000000000003
RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88883fff0480
RBP: ffffffffffffffff R08: ff0003ffffffffff R09: ffffffffffffffff
R10: ffff888106c95540 R11: 0000000055555554 R12: 0000000000000003
R13: 0000000000000000 R14: 0000000000000000 R15: ffff88883fff0940
FS:  00007fc4b8124740(0000) GS:ffff888827c00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00000000000033f3 CR3: 000000026cc08004 CR4: 0000000000770ee0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
PKRU: 55555554
Call Trace:
 <TASK>

? __die
? page_fault_oops
? __pte_offset_map_lock
? exc_page_fault
? asm_exc_page_fault
? wakeup_kswapd
migrate_misplaced_page
__handle_mm_fault
handle_mm_fault
do_user_addr_fault
exc_page_fault
asm_exc_page_fault

RIP: 0033:0x55b897ba0808
Code: (omitted)
RSP: 002b:00007ffeefa821a0 EFLAGS: 00010287
RAX: 000055b89983acd0 RBX: 00007ffeefa823f8 RCX: 000055b89983acd0
RDX: 00007fc2f8122010 RSI: 0000000000020000 RDI: 000055b89983acd0
RBP: 00007ffeefa821a0 R08: 0000000000000037 R09: 0000000000000075
R10: 0000000000000000 R11: 0000000000000202 R12: 0000000000000000
R13: 00007ffeefa82410 R14: 000055b897ba5dd8 R15: 00007fc4b8340000
 </TASK>
Modules linked in:
CR2: 00000000000033f3
---[ end trace 0000000000000000  ]---
RIP: 0010:wakeup_kswapd (./linux/mm/vmscan.c:7812)
Code: (omitted)
RSP: 0000:ffffc90004257d58 EFLAGS: 00010286
RAX: ffffffffffffffff RBX: ffff88883fff0480 RCX: 0000000000000003
RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88883fff0480
RBP: ffffffffffffffff R08: ff0003ffffffffff R09: ffffffffffffffff
R10: ffff888106c95540 R11: 0000000055555554 R12: 0000000000000003
R13: 0000000000000000 R14: 0000000000000000 R15: ffff88883fff0940
FS:  00007fc4b8124740(0000) GS:ffff888827c00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00000000000033f3 CR3: 000000026cc08004 CR4: 0000000000770ee0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400

PKRU: 55555554
note: masim[895] exited with irqs disabled

I think you could trim the down a little bit.

Quoted 19 lines
---
 kernel/sched/fair.c | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)
kernel/sched/fair.c +4 -0
index d7a3c63a2171..6d215cc85f14 100644
@@ -1828,6 +1828,23 @@ bool should_numa_migrate_memory(struct task_struct *p, struct folio *folio,
 	int dst_nid = cpu_to_node(dst_cpu);
 	int last_cpupid, this_cpupid;
 
+	/*
+	 * A node of dst_nid might not have its local memory. Promoting
+	 * a folio to the node is meaningless. What's even worse, oops
+	 * can be observed by the null pointer of ->zone_pgdat in

kernel/sched/fair.c bool should_numa_migrate_memory(struct task_struct *p, struct folio *folio, · line 1844

Quoted 10 lines
kernel/sched/fair.c +9 -0
+	 * various points of the code during migration.
+	 *
 
+	 * For instance, oops has been observed at CPU2 while qemu'ing:
+	 *
+	 * {qemu} \
+	 *    -numa node,nodeid=0,mem=1G,cpus=0-1 \
+	 *    -numa node,nodeid=1,cpus=2-3 \
+	 *    -numa node,nodeid=2,mem=8G \
+	 *    ...

This part above should probably be in the commit message not in the code. The first paragraph of comment is plenty.

Quoted 2 lines

Otherwise, I think the check probably makes sense.

Actually, after looking at the memory.c code I wonder if this check should not be made farther up in the numa migrate machinery.

Cheers,
Phil

Quoted 19 lines

Cheers,
Phil

Quoted 11 lines
kernel/sched/fair.c +4 -0
+	 */
+	if (!node_state(dst_nid, N_MEMORY))
+		return false;
+
 	/*
 	 * The pages in slow memory node should be migrated according

* to hot/cold instead of private/shared.

Signature · 1 line
2.17.1
#3 lore raw
Quoted 110 lines · Phil Auld
Quoted 109 lines · On Wed, Feb 14, 2024 at 07:31:37AM -0500 Phil Auld wrote:

Hi,

Quoted 69 lines · On Wed, Feb 14, 2024 at 12:53:55PM +0900 Byungchul Park wrote:

While running qemu with a configuration where some CPUs don't have their local memory and with a kernel numa balancing on, the following oops has been observed. It's because of null pointers of ->zone_pgdat of zones of those nodes that are not initialized at booting time. So should avoid nodes not set N_MEMORY from getting promoted.

Quoted 63 lines
BUG: unable to handle page fault for address: 00000000000033f3
#PF: supervisor read access in kernel mode
#PF: error_code(0x0000) - not-present page
PGD 0 P4D 0
Oops: 0000 [#1] PREEMPT SMP NOPTI
CPU: 2 PID: 895 Comm: masim Not tainted 6.6.0-dirty #255
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS
   rel-1.16.0-0-gd239552ce722-prebuilt.qemu.org 04/01/2014
RIP: 0010:wakeup_kswapd (./linux/mm/vmscan.c:7812)
Code: (omitted)
RSP: 0000:ffffc90004257d58 EFLAGS: 00010286
RAX: ffffffffffffffff RBX: ffff88883fff0480 RCX: 0000000000000003
RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88883fff0480
RBP: ffffffffffffffff R08: ff0003ffffffffff R09: ffffffffffffffff
R10: ffff888106c95540 R11: 0000000055555554 R12: 0000000000000003
R13: 0000000000000000 R14: 0000000000000000 R15: ffff88883fff0940
FS:  00007fc4b8124740(0000) GS:ffff888827c00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00000000000033f3 CR3: 000000026cc08004 CR4: 0000000000770ee0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
PKRU: 55555554
Call Trace:
 <TASK>

? __die
? page_fault_oops
? __pte_offset_map_lock
? exc_page_fault
? asm_exc_page_fault
? wakeup_kswapd
migrate_misplaced_page
__handle_mm_fault
handle_mm_fault
do_user_addr_fault
exc_page_fault
asm_exc_page_fault

RIP: 0033:0x55b897ba0808
Code: (omitted)
RSP: 002b:00007ffeefa821a0 EFLAGS: 00010287
RAX: 000055b89983acd0 RBX: 00007ffeefa823f8 RCX: 000055b89983acd0
RDX: 00007fc2f8122010 RSI: 0000000000020000 RDI: 000055b89983acd0
RBP: 00007ffeefa821a0 R08: 0000000000000037 R09: 0000000000000075
R10: 0000000000000000 R11: 0000000000000202 R12: 0000000000000000
R13: 00007ffeefa82410 R14: 000055b897ba5dd8 R15: 00007fc4b8340000
 </TASK>
Modules linked in:
CR2: 00000000000033f3
---[ end trace 0000000000000000  ]---
RIP: 0010:wakeup_kswapd (./linux/mm/vmscan.c:7812)
Code: (omitted)
RSP: 0000:ffffc90004257d58 EFLAGS: 00010286
RAX: ffffffffffffffff RBX: ffff88883fff0480 RCX: 0000000000000003
RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88883fff0480
RBP: ffffffffffffffff R08: ff0003ffffffffff R09: ffffffffffffffff
R10: ffff888106c95540 R11: 0000000055555554 R12: 0000000000000003
R13: 0000000000000000 R14: 0000000000000000 R15: ffff88883fff0940
FS:  00007fc4b8124740(0000) GS:ffff888827c00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00000000000033f3 CR3: 000000026cc08004 CR4: 0000000000770ee0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400

PKRU: 55555554
note: masim[895] exited with irqs disabled

I think you could trim the down a little bit.

Quoted 19 lines
---
 kernel/sched/fair.c | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)
kernel/sched/fair.c +4 -0
index d7a3c63a2171..6d215cc85f14 100644
@@ -1828,6 +1828,23 @@ bool should_numa_migrate_memory(struct task_struct *p, struct folio *folio,
 	int dst_nid = cpu_to_node(dst_cpu);
 	int last_cpupid, this_cpupid;
 
+	/*
+	 * A node of dst_nid might not have its local memory. Promoting
+	 * a folio to the node is meaningless. What's even worse, oops
+	 * can be observed by the null pointer of ->zone_pgdat in

kernel/sched/fair.c bool should_numa_migrate_memory(struct task_struct *p, struct folio *folio, · line 1844

Quoted 10 lines
kernel/sched/fair.c +9 -0
+	 * various points of the code during migration.
+	 *
 
+	 * For instance, oops has been observed at CPU2 while qemu'ing:
+	 *
+	 * {qemu} \
+	 *    -numa node,nodeid=0,mem=1G,cpus=0-1 \
+	 *    -numa node,nodeid=1,cpus=2-3 \
+	 *    -numa node,nodeid=2,mem=8G \
+	 *    ...

This part above should probably be in the commit message not in the code. The first paragraph of comment is plenty.

Quoted 5 lines
Quoted 2 lines

Otherwise, I think the check probably makes sense.

Actually, after looking at the memory.c code I wonder if this check should not be made farther up in the numa migrate machinery.

First of all, we cannot avoid hinting fault. It's because no one knows which node a task eventually runs on until a hinting fault occurs. So should let it go get hinting fault *and then* we can make the decision if we can migrate the folio or not. Assuming that, IMHO, should_numa_migrate_memory() is a good place to make it.

Thoughts? Am I missing something?

Byungchul

Quoted 24 lines

Cheers,
Phil

Quoted 19 lines

Cheers,
Phil

Quoted 11 lines
kernel/sched/fair.c +4 -0
+	 */
+	if (!node_state(dst_nid, N_MEMORY))
+		return false;
+
 	/*
 	 * The pages in slow memory node should be migrated according

* to hot/cold instead of private/shared.

Signature · 1 line
2.17.1
#4 lore raw
In reply to Phil Auld
Quoted 73 lines · Phil Auld

Hi,

Quoted 69 lines · On Wed, Feb 14, 2024 at 12:53:55PM +0900 Byungchul Park wrote:

While running qemu with a configuration where some CPUs don't have their local memory and with a kernel numa balancing on, the following oops has been observed. It's because of null pointers of ->zone_pgdat of zones of those nodes that are not initialized at booting time. So should avoid nodes not set N_MEMORY from getting promoted.

Quoted 63 lines
BUG: unable to handle page fault for address: 00000000000033f3
#PF: supervisor read access in kernel mode
#PF: error_code(0x0000) - not-present page
PGD 0 P4D 0
Oops: 0000 [#1] PREEMPT SMP NOPTI
CPU: 2 PID: 895 Comm: masim Not tainted 6.6.0-dirty #255
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS
   rel-1.16.0-0-gd239552ce722-prebuilt.qemu.org 04/01/2014
RIP: 0010:wakeup_kswapd (./linux/mm/vmscan.c:7812)
Code: (omitted)
RSP: 0000:ffffc90004257d58 EFLAGS: 00010286
RAX: ffffffffffffffff RBX: ffff88883fff0480 RCX: 0000000000000003
RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88883fff0480
RBP: ffffffffffffffff R08: ff0003ffffffffff R09: ffffffffffffffff
R10: ffff888106c95540 R11: 0000000055555554 R12: 0000000000000003
R13: 0000000000000000 R14: 0000000000000000 R15: ffff88883fff0940
FS:  00007fc4b8124740(0000) GS:ffff888827c00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00000000000033f3 CR3: 000000026cc08004 CR4: 0000000000770ee0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
PKRU: 55555554
Call Trace:
 <TASK>

? __die
? page_fault_oops
? __pte_offset_map_lock
? exc_page_fault
? asm_exc_page_fault
? wakeup_kswapd
migrate_misplaced_page
__handle_mm_fault
handle_mm_fault
do_user_addr_fault
exc_page_fault
asm_exc_page_fault

RIP: 0033:0x55b897ba0808
Code: (omitted)
RSP: 002b:00007ffeefa821a0 EFLAGS: 00010287
RAX: 000055b89983acd0 RBX: 00007ffeefa823f8 RCX: 000055b89983acd0
RDX: 00007fc2f8122010 RSI: 0000000000020000 RDI: 000055b89983acd0
RBP: 00007ffeefa821a0 R08: 0000000000000037 R09: 0000000000000075
R10: 0000000000000000 R11: 0000000000000202 R12: 0000000000000000
R13: 00007ffeefa82410 R14: 000055b897ba5dd8 R15: 00007fc4b8340000
 </TASK>
Modules linked in:
CR2: 00000000000033f3
---[ end trace 0000000000000000  ]---
RIP: 0010:wakeup_kswapd (./linux/mm/vmscan.c:7812)
Code: (omitted)
RSP: 0000:ffffc90004257d58 EFLAGS: 00010286
RAX: ffffffffffffffff RBX: ffff88883fff0480 RCX: 0000000000000003
RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88883fff0480
RBP: ffffffffffffffff R08: ff0003ffffffffff R09: ffffffffffffffff
R10: ffff888106c95540 R11: 0000000055555554 R12: 0000000000000003
R13: 0000000000000000 R14: 0000000000000000 R15: ffff88883fff0940
FS:  00007fc4b8124740(0000) GS:ffff888827c00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00000000000033f3 CR3: 000000026cc08004 CR4: 0000000000770ee0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400

PKRU: 55555554
note: masim[895] exited with irqs disabled

Quoted 1 line

I think you could trim the down a little bit.

Thank you for the feedback. I will.

Quoted 22 lines
Quoted 21 lines
---
 kernel/sched/fair.c | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)
kernel/sched/fair.c +6 -0
index d7a3c63a2171..6d215cc85f14 100644
@@ -1828,6 +1828,23 @@ bool should_numa_migrate_memory(struct task_struct *p, struct folio *folio,
 	int dst_nid = cpu_to_node(dst_cpu);
 	int last_cpupid, this_cpupid;
 
+	/*
+	 * A node of dst_nid might not have its local memory. Promoting
+	 * a folio to the node is meaningless. What's even worse, oops
+	 * can be observed by the null pointer of ->zone_pgdat in
+	 * various points of the code during migration.
+	 *
Quoted 10 lines

kernel/sched/fair.c bool should_numa_migrate_memory(struct task_struct *p, struct folio *folio, · line 1843

Quoted 7 lines
kernel/sched/fair.c +7 -0
+	 * For instance, oops has been observed at CPU2 while qemu'ing:
+	 *
+	 * {qemu} \
+	 *    -numa node,nodeid=0,mem=1G,cpus=0-1 \
+	 *    -numa node,nodeid=1,cpus=2-3 \
+	 *    -numa node,nodeid=2,mem=8G \
+	 *    ...

This part above should probably be in the commit message not in the code. The first paragraph of comment is plenty.

I will.

Thanks. I will respin it.

Byungchul

Quoted 19 lines

Otherwise, I think the check probably makes sense.

Cheers,
Phil

Quoted 11 lines
kernel/sched/fair.c +4 -0
+	 */
+	if (!node_state(dst_nid, N_MEMORY))
+		return false;
+
 	/*
 	 * The pages in slow memory node should be migrated according
 	 * to hot/cold instead of private/shared.
Signature · 1 line
2.17.1
#5 lore raw
In reply to Byungchul Park
Quoted 5 lines · Byungchul Park

While running qemu with a configuration where some CPUs don't have their local memory and with a kernel numa balancing on, the following oops has been observed. It's because of null pointers of ->zone_pgdat of zones of those nodes that are not initialized at booting time. So should avoid nodes not set N_MEMORY from getting promoted.

Looking at free_area_init(), we call free_area_init_node() for each node found on the system.
And free_area_init_node()->free_area_init_core() inits all zones belonging to the system via zone_init_internals().
Now, I am not saying the check is wrong because we obviously do not want migrate memory to a memoryless node, but I am confused as to where we are crashing.

Signature · 2 lines
Oscar Salvador
SUSE Labs
#6 lore raw
Quoted 11 lines · Oscar Salvador
Quoted 5 lines · Byungchul Park

While running qemu with a configuration where some CPUs don't have their local memory and with a kernel numa balancing on, the following oops has been observed. It's because of null pointers of ->zone_pgdat of zones of those nodes that are not initialized at booting time. So should avoid nodes not set N_MEMORY from getting promoted.

Looking at free_area_init(), we call free_area_init_node() for each node found on the system.
And free_area_init_node()->free_area_init_core() inits all zones belonging to the system via zone_init_internals().

For normal numa nodes, node_data[] is initialized at alloc_node_data(), but it's not for memoryless node. However, the node *gets onlined* at init_cpu_to_node().

Let's look at back free_area_init(). free_area_init_node() will be called with node_data[] not set yet, because it's already *onlined*. So ->zone_pgdat cannot be initialized properly in the path you mentioned.

Byungchul

Quoted 8 lines

Now, I am not saying the check is wrong because we obviously do not want migrate memory to a memoryless node, but I am confused as to where we are crashing.

Signature · 2 lines
Oscar Salvador
SUSE Labs
#7 lore raw
Quoted 7 lines · Byungchul Park

For normal numa nodes, node_data[] is initialized at alloc_node_data(), but it's not for memoryless node. However, the node *gets onlined* at init_cpu_to_node().

Let's look at back free_area_init(). free_area_init_node() will be called with node_data[] not set yet, because it's already *onlined*. So ->zone_pgdat cannot be initialized properly in the path you mentioned.

I am might be missing something., so bear with me.

free_area_init() gets called before init_cpu_to_node() does. free_area_init_node() gets called on every possible node.

free_area_init_node then() does

pg_data_t *pgdat = NODE_DATA(nid);,

and then we call free_area_init_core().

free_area_init_core() does

free_area_init_core() does
zone_init_internals()

which ends up doing zone->zone_pgdat = NODE_DATA(nid);

If node_data[] was not set at all, we would already blow up when doing the first

  for_each_node()
    pgdat = NODE_DATA(nid);
    free_area_init_node(nid);

back in free_area_init().

Signature · 2 lines
Oscar Salvador
SUSE Labs
#8 lore raw
Quoted 27 lines · Oscar Salvador
Quoted 7 lines · Byungchul Park

For normal numa nodes, node_data[] is initialized at alloc_node_data(), but it's not for memoryless node. However, the node *gets onlined* at init_cpu_to_node().

Let's look at back free_area_init(). free_area_init_node() will be called with node_data[] not set yet, because it's already *onlined*. So ->zone_pgdat cannot be initialized properly in the path you mentioned.

I am might be missing something., so bear with me.

free_area_init() gets called before init_cpu_to_node() does. free_area_init_node() gets called on every possible node.

free_area_init_node then() does

pg_data_t *pgdat = NODE_DATA(nid);,

and then we call free_area_init_core().

free_area_init_core() does

free_area_init_core() does
zone_init_internals()

which ends up doing zone->zone_pgdat = NODE_DATA(nid);

Quoted 8 lines

If node_data[] was not set at all, we would already blow up when doing the first

  for_each_node()
    pgdat = NODE_DATA(nid);
    free_area_init_node(nid);

back in free_area_init().

It seems that I got it wrong about the reason. Let me check it again and share the reason.

Just in case, this patch is still definitely necessary tho.

Byungchul

#9 lore raw
Quoted 35 lines · Byungchul Park
Quoted 34 lines · Oscar Salvador
Quoted 7 lines · Byungchul Park

For normal numa nodes, node_data[] is initialized at alloc_node_data(), but it's not for memoryless node. However, the node *gets onlined* at init_cpu_to_node().

Let's look at back free_area_init(). free_area_init_node() will be called with node_data[] not set yet, because it's already *onlined*. So ->zone_pgdat cannot be initialized properly in the path you mentioned.

I am might be missing something., so bear with me.

free_area_init() gets called before init_cpu_to_node() does. free_area_init_node() gets called on every possible node.

free_area_init_node then() does

pg_data_t *pgdat = NODE_DATA(nid);,

and then we call free_area_init_core().

free_area_init_core() does

free_area_init_core() does
zone_init_internals()

which ends up doing zone->zone_pgdat = NODE_DATA(nid);

If node_data[] was not set at all, we would already blow up when doing the first

  for_each_node()
    pgdat = NODE_DATA(nid);
    free_area_init_node(nid);
Quoted 6 lines
Quoted 1 line

back in free_area_init().

It seems that I got it wrong about the reason. Let me check it again and share the reason.

Just in case, this patch is still definitely necessary tho.

Sorry for the confusing expression. Please don't misunderstand it. The oops has been always observed in the configuration that I descriped. I meant:

Just in case, I need to say the fix is still necessary.

Byungchul

#10 lore raw
Quoted 32 lines · Byungchul Park
Quoted 31 lines · Byungchul Park
Quoted 30 lines · Oscar Salvador
Quoted 7 lines · Byungchul Park

For normal numa nodes, node_data[] is initialized at alloc_node_data(), but it's not for memoryless node. However, the node *gets onlined* at init_cpu_to_node().

Let's look at back free_area_init(). free_area_init_node() will be called with node_data[] not set yet, because it's already *onlined*. So ->zone_pgdat cannot be initialized properly in the path you mentioned.

I am might be missing something., so bear with me.

free_area_init() gets called before init_cpu_to_node() does. free_area_init_node() gets called on every possible node.

free_area_init_node then() does

pg_data_t *pgdat = NODE_DATA(nid);,

and then we call free_area_init_core().

free_area_init_core() does

free_area_init_core() does
zone_init_internals()

which ends up doing zone->zone_pgdat = NODE_DATA(nid);

If node_data[] was not set at all, we would already blow up when doing the first

Quoted 8 lines
Quoted 8 lines
Quoted 5 lines
  for_each_node()
    pgdat = NODE_DATA(nid);
    free_area_init_node(nid);

back in free_area_init().

It seems that I got it wrong about the reason. Let me check it again and share the reason.

I analyzed it wrong. Even though the issue was gone with the patch but it's not the fix. Sorry for making you confused. I submitted the fix with another patch:

https://lore.kernel.org/lkml/[email protected]/

Byungchul