Re: [fake numa not working] [PATCH 1/1] mm/fake-numa: allow later numa node hotplug

From: Hyeonggon Yoo
Date: Sun Jan 19 2025 - 21:37:43 EST




On 1/20/2025 11:16 AM, Zi Yan wrote:
On Sun Jan 19, 2025 at 7:37 PM EST, Hyeonggon Yoo wrote:


On 1/18/2025 4:09 AM, Zi Yan wrote:
On 17 Jan 2025, at 0:00, Hyeonggon Yoo wrote:

On 1/6/2025 9:06 PM, Bruno Faccini wrote:
Current fake-numa implementation prevents new Numa
nodes to be later hot-plugged by drivers.
A common symptom of this limitation is the
"node <X> was absent from the node_possible_map"
message by associated warning in mm/memory_hotplug.c:
add_memory_resource().
This comes from the lack of remapping in both
pxm_to_node_map[] and node_to_pxm_map[] tables
to take fake-numa nodes into account and thus
triggers collisions with original and physical nodes
only-mapping that had been determined from BIOS tables.
This patch fixes this by doing the necessary node-ids
translation in both pxm_to_node_map[]/node_to_pxm_map[]
tables.
node_distance[] table has also been fixed accordingly.

Signed-off-by: Bruno Faccini <bfaccini@xxxxxxxxxx>

Hi Bruno,

This commit causes WARN() and disables fakenuma
when I pass "numa=fake=4U" on my QEMU setup.

Attaching WARN() log and git bisect log to help debugging.

Is your VM getting 4 NUMA nodes at the end?

No, it only gets node 0 at the end.

Got it.


Can you also share your QEMU command line and your kernel config?

The config is attached (fakenuma.config)

QEMU command line:
$ qemu-system-x86_64 \
-cpu host \
-enable-kvm \
-smp $(nproc) \
-m 32G \
-nographic \
-kernel arch/x86/boot/bzImage \
-initrd ../../rootfs.cpio.gz \
-nic user,model=virtio-net-pci \
-append "console=ttyS0 numa=fake=4U"

This can help Bruno reproduce the issue locally. At least, I
cannot reproduce it on mm-everything-2025-01-16-23-18.

Still reproduced on mm-everything-2025-01-18-04-55 with my setup.

I also
notice that my VM does not have “No NUMA configuration found”
(in your log) in its kernel log and that might explain
why I could not reproduce it.

Did you do NUMA configuration (-numa in QEMU) for your VM?
I didn't, but it's still expected to work 'cause it's fake numa.

No. It seems that the "No NUMA configuration found" is caused by your
qemu, where dummy_numa_init() is used (maybe ACPI SRAT is not
populated?)

FYI, Yes. ACPI SRAT is not populated.

$ qemu-system-x86_64 --version
QEMU emulator version 6.2.0 (Debian 1:6.2+dfsg-2ubuntu6.24)
Copyright (c) 2003-2021 Fabrice Bellard and the QEMU Project developers

I used the patch below to emulate the situation and
reproduced the issue. Since dummy_numa_init() is used when there is
no underlying NUMA architecture, NUMA initialization fails, or NUMA
is disabled on the command line (the last does not apply here),
it should be properly handled. I will let Bruno look into it.

Thank you for the info.

diff --git a/arch/x86/mm/numa.c b/arch/x86/mm/numa.c
index 64e5cdb2460a..70dfd9bfb23f 100644
--- a/arch/x86/mm/numa.c
+++ b/arch/x86/mm/numa.c
@@ -223,7 +223,7 @@ static int __init dummy_numa_init(void)
*/
void __init x86_numa_init(void)
{
- if (!numa_off) {
+ if (0 && !numa_off) {
#ifdef CONFIG_ACPI_NUMA
if (!numa_init(x86_acpi_numa_init))
return;


[WARN() splat]

[ 0.009676] No NUMA configuration found
[ 0.009677] Faking a node at [mem 0x0000000000000000-0x000000083fffffff]
[ 0.009705] Fake node size 16895MB too small, increasing to 16896MB
[ 0.009708] Faking node 0 at [mem 0x0000000000001000-0x0000000420000fff] (16896MB)
[ 0.009711] Faking node 1 at [mem 0x0000000420001000-0x000000083fffffff] (16895MB)
[ 0.009738] ------------[ cut here ]------------
[ 0.009739] -1 max nid when expected 1
[ 0.009760] WARNING: CPU: 0 PID: 0 at drivers/acpi/numa/srat.c:120 fix_pxm_node_maps+0x577/0x810
[ 0.009766] Modules linked in:
[ 0.009769] CPU: 0 UID: 0 PID: 0 Comm: swapper Not tainted 6.13.0-rc6+ #31
[ 0.009771] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.15.0-1 04/01/2014
[ 0.009772] RIP: 0010:fix_pxm_node_maps+0x577/0x810
[ 0.009774] Code: 8b 85 40 ff ff ff 44 8b 8d 44 ff ff ff 8b 85 48 ff ff ff e9 83 fb ff ff 44 89 f2 44 89 ce 48 c7 c7 aa 2f 84 8d e8 b9 71 01 fe <0f> 0b 41 b8 ff ff ff ff e9 ba fb ff ff 48 c7 c7 a0 d8 d1 8d 44 89
[ 0.009776] RSP: 0000:ffffffff8da03be0 EFLAGS: 00010046 ORIG_RAX: 0000000000000000
[ 0.009778] RAX: 0000000000000000 RBX: 0000000000000400 RCX: 0000000000000000
[ 0.009779] RDX: 0000000000000000 RSI: 0000000000000000 RDI: 0000000000000000
[ 0.009780] RBP: ffffffff8da03ca8 R08: 0000000000000000 R09: 0000000000000000
[ 0.009781] R10: 0000000000000000 R11: 0000000000000000 R12: 0000000000000001
[ 0.009782] R13: 00000000000007ff R14: 0000000000000001 R15: 0000000000000000
[ 0.009783] FS: 0000000000000000(0000) GS:ffffffff8de8c000(0000) knlGS:0000000000000000
[ 0.009784] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 0.009785] CR2: ffff90e4bde01000 CR3: 00000001bd03e000 CR4: 00000000000200b0
[ 0.009788] Call Trace:
[ 0.009790] <TASK>
[ 0.009792] ? show_regs+0x71/0x90
[ 0.009797] ? __warn+0x8d/0x150
[ 0.009799] ? fix_pxm_node_maps+0x577/0x810
[ 0.009802] ? report_bug+0x1ab/0x1c0
[ 0.009805] ? fixup_exception+0x27/0x380
[ 0.009809] ? early_fixup_exception+0xa2/0xf0
[ 0.009813] ? do_early_exception+0x28/0x90
[ 0.009816] ? early_idt_handler_common+0x2f/0x3a
[ 0.009819] ? fix_pxm_node_maps+0x577/0x810
[ 0.009822] ? numa_cleanup_meminfo+0x8c/0x5b0
[ 0.009828] numa_emulation+0x4e3/0xad0
[ 0.009830] ? __pfx_dummy_numa_init+0x10/0x10
[ 0.009833] ? __pfx_dummy_numa_init+0x10/0x10
[ 0.009836] numa_memblks_init+0x10c/0x2c0
[ 0.009838] ? __pfx_dummy_numa_init+0x10/0x10
[ 0.009840] numa_init+0x61/0x3e0
[ 0.009842] ? topology_register_boot_apic+0x2a/0x40
[ 0.009846] x86_numa_init+0x65/0x80
[ 0.009848] initmem_init+0xe/0x20
[ 0.009850] setup_arch+0x9d8/0xfa0
[ 0.009852] ? _printk+0x58/0x90
[ 0.009855] start_kernel+0x5f/0xb50
[ 0.009860] x86_64_start_reservations+0x18/0x30
[ 0.009863] x86_64_start_kernel+0xc0/0x110
[ 0.009864] ? setup_ghcb+0xe/0x140
[ 0.009867] common_startup_64+0x13e/0x141
[ 0.009871] </TASK>
[ 0.009871] ---[ end trace 0000000000000000 ]---

[Git bisect log]

# bad: [f378252a2168c2fbf8fc08b635061e5f6748c1f2] kasan: sw_tags: use str_on_off() helper in kasan_init_sw_tags()
git bisect bad f378252a2168c2fbf8fc08b635061e5f6748c1f2
# good: [cbc5dde0a461240046e8a41c43d7c3b76d5db952] fs/proc: fix softlockup in __read_vmcore (part 2)
git bisect good cbc5dde0a461240046e8a41c43d7c3b76d5db952
# good: [fb73203263021f2ee9dd54d280b1c543d10acd76] mm: move common part of pagetable_*_ctor to helper
git bisect good fb73203263021f2ee9dd54d280b1c543d10acd76
# bad: [8fd966c9310f14d5bbe653cf087853582170aad6] mm, swap: remove old allocation path for HDD
git bisect bad 8fd966c9310f14d5bbe653cf087853582170aad6
# bad: [8646971f02651146554233d63fdd721f5f060973] mm: shmem: skip swapcache for swapin of synchronous swap device
git bisect bad 8646971f02651146554233d63fdd721f5f060973
# good: [b8ba614eaa8dd31702c61e7df7231b1b18b99259] mm/damon/paddr: report filter-passed bytes back for DAMOS_STAT action
git bisect good b8ba614eaa8dd31702c61e7df7231b1b18b99259
# good: [fd0935b8e9e8c2edf05a3d0ffa09d0aa3e9cf2dd] Docs/ABI/damon: document per-region DAMOS filter-passed bytes stat file
git bisect good fd0935b8e9e8c2edf05a3d0ffa09d0aa3e9cf2dd
# good: [3ffd3cd7e2e8793d3b5fbb83c03668c47ab6d599] selftests/damon: remove tests for DAMON debugfs interface
git bisect good 3ffd3cd7e2e8793d3b5fbb83c03668c47ab6d599
# good: [bf9f93f0b7bff8af780330f5094775a39caf3612] mm/damon: remove DAMON debugfs interface
git bisect good bf9f93f0b7bff8af780330f5094775a39caf3612
# bad: [1746744be6ff271e29345a5be4cc144aa33b10ab] mm/memmap: prevent double scanning of memmap by kmemleak
git bisect bad 1746744be6ff271e29345a5be4cc144aa33b10ab
# bad: [ca19a5ab8756ef78c34d02d2b8c266a9cc7bf57d] mm/fake-numa: allow later numa node hotplug
git bisect bad ca19a5ab8756ef78c34d02d2b8c266a9cc7bf57d
# first bad commit: [ca19a5ab8756ef78c34d02d2b8c266a9cc7bf57d] mm/fake-numa: allow later numa node hotplug

---
drivers/acpi/numa/srat.c | 86 ++++++++++++++++++++++++++++++++++++
include/acpi/acpi_numa.h | 5 +++
include/linux/numa_memblks.h | 3 ++
mm/numa_emulation.c | 45 ++++++++++++++++---
mm/numa_memblks.c | 2 +-
5 files changed, 133 insertions(+), 8 deletions(-)

diff --git a/drivers/acpi/numa/srat.c b/drivers/acpi/numa/srat.c
index bec0dcd1f9c3..59fffe34c9d0 100644
--- a/drivers/acpi/numa/srat.c
+++ b/drivers/acpi/numa/srat.c
@@ -81,6 +81,92 @@ int acpi_map_pxm_to_node(int pxm)
}
EXPORT_SYMBOL(acpi_map_pxm_to_node);
+#ifdef CONFIG_NUMA_EMU
+/*
+ * Take max_nid - 1 fake-numa nodes into account in both
+ * pxm_to_node_map()/node_to_pxm_map[] tables.
+ */
+int __init fix_pxm_node_maps(int max_nid)
+{
+ static int pxm_to_node_map_copy[MAX_PXM_DOMAINS] __initdata
+ = { [0 ... MAX_PXM_DOMAINS - 1] = NUMA_NO_NODE };
+ static int node_to_pxm_map_copy[MAX_NUMNODES] __initdata
+ = { [0 ... MAX_NUMNODES - 1] = PXM_INVAL };
+ int i, j, index = -1, count = 0;
+ nodemask_t nodes_to_enable;
+
+ if (numa_off || srat_disabled())
+ return -1;
+
+ /* find fake nodes PXM mapping */
+ for (i = 0; i < MAX_NUMNODES; i++) {
+ if (node_to_pxm_map[i] != PXM_INVAL) {
+ for (j = 0; j <= max_nid; j++) {
+ if ((emu_nid_to_phys[j] == i) &&
+ WARN(node_to_pxm_map_copy[j] != PXM_INVAL,
+ "Node %d is already binded to PXM %d\n",
+ j, node_to_pxm_map_copy[j]))
+ return -1;
+ if (emu_nid_to_phys[j] == i) {
+ node_to_pxm_map_copy[j] =
+ node_to_pxm_map[i];
+ if (j > index)
+ index = j;
+ count++;
+ }
+ }
+ }
+ }
+ if (WARN(index != max_nid, "%d max nid when expected %d\n",
+ index, max_nid))
+ return -1;
+
+ nodes_clear(nodes_to_enable);
+
+ /* map phys nodes not used for fake nodes */
+ for (i = 0; i < MAX_NUMNODES; i++) {
+ if (node_to_pxm_map[i] != PXM_INVAL) {
+ for (j = 0; j <= max_nid; j++)
+ if (emu_nid_to_phys[j] == i)
+ break;
+ /* fake nodes PXM mapping has been done */
+ if (j <= max_nid)
+ continue;
+ /* find first hole */
+ for (j = 0;
+ j < MAX_NUMNODES &&
+ node_to_pxm_map_copy[j] != PXM_INVAL;
+ j++)
+ ;
+ if (WARN(j == MAX_NUMNODES,
+ "Number of nodes exceeds MAX_NUMNODES\n"))
+ return -1;
+ node_to_pxm_map_copy[j] = node_to_pxm_map[i];
+ node_set(j, nodes_to_enable);
+ count++;
+ }
+ }
+
+ /* creating reverse mapping in pxm_to_node_map[] */
+ for (i = 0; i < MAX_NUMNODES; i++)
+ if (node_to_pxm_map_copy[i] != PXM_INVAL &&
+ pxm_to_node_map_copy[node_to_pxm_map_copy[i]] == NUMA_NO_NODE)
+ pxm_to_node_map_copy[node_to_pxm_map_copy[i]] = i;
+
+ /* overwrite with new mapping */
+ for (i = 0; i < MAX_NUMNODES; i++) {
+ node_to_pxm_map[i] = node_to_pxm_map_copy[i];
+ pxm_to_node_map[i] = pxm_to_node_map_copy[i];
+ }
+
+ /* enable other nodes found in PXM for hotplug */
+ nodes_or(numa_nodes_parsed, nodes_to_enable, numa_nodes_parsed);
+
+ pr_debug("found %d total number of nodes\n", count);
+ return 0;
+}
+#endif
+
static void __init
acpi_table_print_srat_entry(struct acpi_subtable_header *header)
{
diff --git a/include/acpi/acpi_numa.h b/include/acpi/acpi_numa.h
index b5f594754a9e..99b960bd473c 100644
--- a/include/acpi/acpi_numa.h
+++ b/include/acpi/acpi_numa.h
@@ -17,11 +17,16 @@ extern int node_to_pxm(int);
extern int acpi_map_pxm_to_node(int);
extern unsigned char acpi_srat_revision;
extern void disable_srat(void);
+extern int fix_pxm_node_maps(int max_nid);
extern void bad_srat(void);
extern int srat_disabled(void);
#else /* CONFIG_ACPI_NUMA */
+static inline int fix_pxm_node_maps(int max_nid)
+{
+ return 0;
+}
static inline void disable_srat(void)
{
}
diff --git a/include/linux/numa_memblks.h b/include/linux/numa_memblks.h
index cfad6ce7e1bd..dd85613cdd86 100644
--- a/include/linux/numa_memblks.h
+++ b/include/linux/numa_memblks.h
@@ -29,7 +29,10 @@ int __init numa_cleanup_meminfo(struct numa_meminfo *mi);
int __init numa_memblks_init(int (*init_func)(void),
bool memblock_force_top_down);
+extern int numa_distance_cnt;
+
#ifdef CONFIG_NUMA_EMU
+extern int emu_nid_to_phys[MAX_NUMNODES];
int numa_emu_cmdline(char *str);
void __init numa_emu_update_cpu_to_node(int *emu_nid_to_phys,
unsigned int nr_emu_nids);
diff --git a/mm/numa_emulation.c b/mm/numa_emulation.c
index 031fb9961bf7..9d55679d99ce 100644
--- a/mm/numa_emulation.c
+++ b/mm/numa_emulation.c
@@ -8,11 +8,12 @@
#include <linux/memblock.h>
#include <linux/numa_memblks.h>
#include <asm/numa.h>
+#include <acpi/acpi_numa.h>
#define FAKE_NODE_MIN_SIZE ((u64)32 << 20)
#define FAKE_NODE_MIN_HASH_MASK (~(FAKE_NODE_MIN_SIZE - 1UL))
-static int emu_nid_to_phys[MAX_NUMNODES];
+int emu_nid_to_phys[MAX_NUMNODES];
static char *emu_cmdline __initdata;
int __init numa_emu_cmdline(char *str)
@@ -379,6 +380,7 @@ void __init numa_emulation(struct numa_meminfo *numa_meminfo, int numa_dist_cnt)
size_t phys_size = numa_dist_cnt * numa_dist_cnt * sizeof(phys_dist[0]);
int max_emu_nid, dfl_phys_nid;
int i, j, ret;
+ nodemask_t physnode_mask = numa_nodes_parsed;
if (!emu_cmdline)
goto no_emu;
@@ -395,7 +397,6 @@ void __init numa_emulation(struct numa_meminfo *numa_meminfo, int numa_dist_cnt)
* split the system RAM into N fake nodes.
*/
if (strchr(emu_cmdline, 'U')) {
- nodemask_t physnode_mask = numa_nodes_parsed;
unsigned long n;
int nid = 0;
@@ -465,9 +466,6 @@ void __init numa_emulation(struct numa_meminfo *numa_meminfo, int numa_dist_cnt)
*/
max_emu_nid = setup_emu2phys_nid(&dfl_phys_nid);
- /* commit */
- *numa_meminfo = ei;
-
/* Make sure numa_nodes_parsed only contains emulated nodes */
nodes_clear(numa_nodes_parsed);
for (i = 0; i < ARRAY_SIZE(ei.blk); i++)
@@ -475,10 +473,21 @@ void __init numa_emulation(struct numa_meminfo *numa_meminfo, int numa_dist_cnt)
ei.blk[i].nid != NUMA_NO_NODE)
node_set(ei.blk[i].nid, numa_nodes_parsed);
- numa_emu_update_cpu_to_node(emu_nid_to_phys, ARRAY_SIZE(emu_nid_to_phys));
+ /* fix pxm_to_node_map[] and node_to_pxm_map[] to avoid collision
+ * with faked numa nodes, particularly during later memory hotplug
+ * handling, and also update numa_nodes_parsed accordingly.
+ */
+ ret = fix_pxm_node_maps(max_emu_nid);
+ if (ret < 0)
+ goto no_emu;
+
+ /* commit */
+ *numa_meminfo = ei;
+
+ numa_emu_update_cpu_to_node(emu_nid_to_phys, max_emu_nid + 1);
/* make sure all emulated nodes are mapped to a physical node */
- for (i = 0; i < ARRAY_SIZE(emu_nid_to_phys); i++)
+ for (i = 0; i < max_emu_nid + 1; i++)
if (emu_nid_to_phys[i] == NUMA_NO_NODE)
emu_nid_to_phys[i] = dfl_phys_nid;
@@ -501,12 +510,34 @@ void __init numa_emulation(struct numa_meminfo *numa_meminfo, int numa_dist_cnt)
numa_set_distance(i, j, dist);
}
}
+ for (i = 0; i < numa_distance_cnt; i++) {
+ for (j = 0; j < numa_distance_cnt; j++) {
+ int physi, physj;
+ u8 dist;
+
+ /* distance between fake nodes is already ok */
+ if (emu_nid_to_phys[i] != NUMA_NO_NODE &&
+ emu_nid_to_phys[j] != NUMA_NO_NODE)
+ continue;
+ if (emu_nid_to_phys[i] != NUMA_NO_NODE)
+ physi = emu_nid_to_phys[i];
+ else
+ physi = i - max_emu_nid;
+ if (emu_nid_to_phys[j] != NUMA_NO_NODE)
+ physj = emu_nid_to_phys[j];
+ else
+ physj = j - max_emu_nid;
+ dist = phys_dist[physi * numa_dist_cnt + physj];
+ numa_set_distance(i, j, dist);
+ }
+ }
/* free the copied physical distance table */
memblock_free(phys_dist, phys_size);
return;
no_emu:
+ numa_nodes_parsed = physnode_mask;
/* No emulation. Build identity emu_nid_to_phys[] for numa_add_cpu() */
for (i = 0; i < ARRAY_SIZE(emu_nid_to_phys); i++)
emu_nid_to_phys[i] = i;
diff --git a/mm/numa_memblks.c b/mm/numa_memblks.c
index a3877e9bc878..ff4054f4334d 100644
--- a/mm/numa_memblks.c
+++ b/mm/numa_memblks.c
@@ -7,7 +7,7 @@
#include <linux/numa.h>
#include <linux/numa_memblks.h>
-static int numa_distance_cnt;
+int numa_distance_cnt;
static u8 *numa_distance;
nodemask_t numa_nodes_parsed __initdata;


Best Regards,
Yan, Zi