arch_numa: Restore nid checks before registering a memblock with a node

PATCH In review Marc Zyngier 2 participants 4 messages Dec 1, 2024

Subsystems (2)

Stable branch Driver core

lore

Share & tools

lore

Marc Zyngier

author

2024-12-01 09:27 UTC #0 lore raw

PATCH v2 1/1 arch_numa: Restore nid checks before registering a memblock with a node

Commit 767507654c22 ("arch_numa: switch over to numa_memblks") significantly cleaned up the NUMA registration code, but also dropped a significant check that was refusing to accept to configure a memblock with an invalid nid.

On "quality hardware" such as my ThunderX machine, this results in a kernel that dies immediately:

[    0.000000] Booting Linux on physical CPU 0x0000000000 [0x431f0a10]
[    0.000000] Linux version 6.12.0-00013-g8920d74cf8db (maz@valley-girl) (gcc (Debian 12.2.0-14) 12.2.0, GNU ld (GNU Binutils for Debian) 2.40) #3872 SMP PREEMPT Wed Nov 27 15:25:49 GMT 2024
[    0.000000] KASLR disabled due to lack of seed
[    0.000000] Machine model: Cavium ThunderX CN88XX board
[    0.000000] efi: EFI v2.4 by American Megatrends
[    0.000000] efi: ESRT=0xffce0ff18 SMBIOS 3.0=0xfffb0000 ACPI 2.0=0xffec60000 MEMRESERVE=0xffc905d98
[    0.000000] esrt: Reserving ESRT space from 0x0000000ffce0ff18 to 0x0000000ffce0ff50.
[    0.000000] earlycon: pl11 at MMIO 0x000087e024000000 (options '115200n8')
[    0.000000] printk: legacy bootconsole [pl11] enabled
[    0.000000] NODE_DATA(0) allocated [mem 0xff6754580-0xff67566bf]
[    0.000000] Unable to handle kernel paging request at virtual address 0000000000001d40
[    0.000000] Mem abort info:
[    0.000000]   ESR = 0x0000000096000004
[    0.000000]   EC = 0x25: DABT (current EL), IL = 32 bits
[    0.000000]   SET = 0, FnV = 0
[    0.000000]   EA = 0, S1PTW = 0
[    0.000000]   FSC = 0x04: level 0 translation fault
[    0.000000] Data abort info:
[    0.000000]   ISV = 0, ISS = 0x00000004, ISS2 = 0x00000000
[    0.000000]   CM = 0, WnR = 0, TnD = 0, TagAccess = 0
[    0.000000]   GCS = 0, Overlay = 0, DirtyBit = 0, Xs = 0
[    0.000000] [0000000000001d40] user address but active_mm is swapper
[    0.000000] Internal error: Oops: 0000000096000004 [#1] PREEMPT SMP
[    0.000000] Modules linked in:
[    0.000000] CPU: 0 UID: 0 PID: 0 Comm: swapper Not tainted 6.12.0-00013-g8920d74cf8db #3872
[    0.000000] Hardware name: Cavium ThunderX CN88XX board (DT)
[    0.000000] pstate: a00000c5 (NzCv daIF -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
[    0.000000] pc : sparse_init_nid+0x54/0x428
[    0.000000] lr : sparse_init+0x118/0x240
[    0.000000] sp : ffff800081da3cb0
[    0.000000] x29: ffff800081da3cb0 x28: 0000000fedbab10c x27: 0000000000000001
[    0.000000] x26: 0000000ffee250f8 x25: 0000000000000001 x24: ffff800082102cd0
[    0.000000] x23: 0000000000000001 x22: 0000000000000000 x21: 00000000001fffff
[    0.000000] x20: 0000000000000001 x19: 0000000000000000 x18: ffffffffffffffff
[    0.000000] x17: 0000000001b00000 x16: 0000000ffd130000 x15: 0000000000000000
[    0.000000] x14: 00000000003e0000 x13: 00000000000001c8 x12: 0000000000000014
[    0.000000] x11: ffff800081e82860 x10: ffff8000820fb2c8 x9 : ffff8000820fb490
[    0.000000] x8 : 0000000000ffed20 x7 : 0000000000000014 x6 : 00000000001fffff
[    0.000000] x5 : 00000000ffffffff x4 : 0000000000000000 x3 : 0000000000000000
[    0.000000] x2 : 0000000000000000 x1 : 0000000000000040 x0 : 0000000000000007
[    0.000000] Call trace:
[    0.000000]  sparse_init_nid+0x54/0x428
[    0.000000]  sparse_init+0x118/0x240
[    0.000000]  bootmem_init+0x70/0x1c8
[    0.000000]  setup_arch+0x184/0x270
[    0.000000]  start_kernel+0x74/0x670
[    0.000000]  __primary_switched+0x80/0x90
[    0.000000] Code: f865d804 d37df060 cb030000 d2800003 (b95d4084)
[    0.000000] ---[ end trace 0000000000000000 ]---
[    0.000000] Kernel panic - not syncing: Attempted to kill the idle task!
[    0.000000] ---[ end Kernel panic - not syncing: Attempted to kill the idle task! ]---

while previous kernel versions were able to recognise how brain-damaged the machine is, and only build a fake node.

Use the memblock_validate_numa_coverage() helper to restore some sanity and a "working" system.

Fixes 767507654c22 ("arch_numa: switch over to numa_memblks")

Signed-off-by Marc Zyngier <[email protected]>

Cc Catalin Marinas <[email protected]>

Cc Will Deacon <[email protected]>

Cc Zi Yan <[email protected]>

Cc Dan Williams <[email protected]>

Cc David Hildenbrand <[email protected]>

Cc Andrew Morton <[email protected]>

Cc [email protected]

---
 drivers/base/arch_numa.c | 4 ++++
 1 file changed, 4 insertions(+)

drivers/base/arch_numa.c +4 -0

index e187016764265..c63a72a1fed64 100644
@@ -208,6 +208,10 @@ static int __init numa_register_nodes(void)
 {
 	int nid;
 
+	/* Check the validity of the memblock/node mapping */
+	if (!memblock_validate_numa_coverage(1))
+		return -EINVAL;
+
 	/* Finally register nodes. */
 	for_each_node_mask(nid, numa_nodes_parsed) {
 		unsigned long start_pfn, end_pfn;

Signature · 1 line

2.39.2

Mike Rapoport

2024-12-01 19:32 UTC #1 lore raw

Hi Marc,

Quoted 4 lines · Marc Zyngier

Commit 767507654c22 ("arch_numa: switch over to numa_memblks") significantly cleaned up the NUMA registration code, but also dropped a significant check that was refusing to accept to configure a memblock with an invalid nid.

...

Quoted 25 lines

while previous kernel versions were able to recognise how brain-damaged the machine is, and only build a fake node.
Use the memblock_validate_numa_coverage() helper to restore some sanity and a "working" system.
Fixes 767507654c22 ("arch_numa: switch over to numa_memblks")
Suggested-by Mike Rapoport <[email protected]>
Signed-off-by Marc Zyngier <[email protected]>
Cc Catalin Marinas <[email protected]>
Cc Will Deacon <[email protected]>
Cc Zi Yan <[email protected]>
Cc Dan Williams <[email protected]>
Cc David Hildenbrand <[email protected]>
Cc Andrew Morton <[email protected]>
Cc [email protected]
---
 drivers/base/arch_numa.c | 4 ++++
 1 file changed, 4 insertions(+)
drivers/base/arch_numa.c +0 -0
index e187016764265..c63a72a1fed64 100644

drivers/base/arch_numa.c static int __init numa_register_nodes(void) · line 212

Quoted 5 lines

drivers/base/arch_numa.c +2 -0

@@ -208,6 +208,10 @@ static int __init numa_register_nodes(void)
 {
 	int nid;
 
+	/* Check the validity of the memblock/node mapping */
+	if (!memblock_validate_numa_coverage(1))

I've changed this to memblock_validate_numa_coverage(0) and applied along with my patch that changed memblock_validate_numa_coverage() to work with 0:

https://git.kernel.org/pub/scm/linux/kernel/git/rppt/memblock.git/log/?h=thunderx-fix

Can you please verify that it works on your "quality hardware"?

Quoted 7 lines

drivers/base/arch_numa.c +2 -0

+		return -EINVAL;
+
 	/* Finally register nodes. */
 	for_each_node_mask(nid, numa_nodes_parsed) {
 		unsigned long start_pfn, end_pfn;

Signature · 1 line

2.39.2

Signature · 2 lines

Sincerely yours,
Mike.

Marc Zyngier

author

2024-12-01 19:49 UTC #2 lore raw

Hi Mike,

Quoted 46 lines · Mike Rapoport

Hi Marc,
Quoted 4 lines · Marc Zyngier
Commit 767507654c22 ("arch_numa: switch over to numa_memblks") significantly cleaned up the NUMA registration code, but also dropped a significant check that was refusing to accept to configure a memblock with an invalid nid.
...
Quoted 25 lines
while previous kernel versions were able to recognise how brain-damaged the machine is, and only build a fake node.
Use the memblock_validate_numa_coverage() helper to restore some sanity and a "working" system.
Fixes 767507654c22 ("arch_numa: switch over to numa_memblks")
Suggested-by Mike Rapoport <[email protected]>
Signed-off-by Marc Zyngier <[email protected]>
Cc Catalin Marinas <[email protected]>
Cc Will Deacon <[email protected]>
Cc Zi Yan <[email protected]>
Cc Dan Williams <[email protected]>
Cc David Hildenbrand <[email protected]>
Cc Andrew Morton <[email protected]>
Cc [email protected]
---
 drivers/base/arch_numa.c | 4 ++++
 1 file changed, 4 insertions(+)
drivers/base/arch_numa.c +0 -0
index e187016764265..c63a72a1fed64 100644
drivers/base/arch_numa.c static int __init numa_register_nodes(void) · line 212
Quoted 5 lines
drivers/base/arch_numa.c +2 -0
@@ -208,6 +208,10 @@ static int __init numa_register_nodes(void)
 {
 	int nid;
 
+	/* Check the validity of the memblock/node mapping */
+	if (!memblock_validate_numa_coverage(1))
I've changed this to memblock_validate_numa_coverage(0) and applied along with my patch that changed memblock_validate_numa_coverage() to work with 0:

Quoted 3 lines

https://git.kernel.org/pub/scm/linux/kernel/git/rppt/memblock.git/log/?h=thunderx-fix
Can you please verify that it works on your "quality hardware"?

Commit 427c6179e159b in your tree still has memblock_validate_numa_coverage(1). Forgot to push out the updated version?

Flipping this to 0 locally, I have verified that this still allows the old thing to trudge along:

root@duodenum:~# uname -a
Linux duodenum 6.12.0-12115-g427c6179e159-dirty #3896 SMP PREEMPT Sun Dec  1 19:43:13 GMT 2024 aarch64

Thanks again,

M.

Signature · 1 line

Without deviation from the norm, progress is not possible.

Mike Rapoport

2024-12-01 20:06 UTC #3 lore raw

Quoted 50 lines · Marc Zyngier

Hi Mike,
Quoted 46 lines · Mike Rapoport
Hi Marc,
Quoted 4 lines · Marc Zyngier
Commit 767507654c22 ("arch_numa: switch over to numa_memblks") significantly cleaned up the NUMA registration code, but also dropped a significant check that was refusing to accept to configure a memblock with an invalid nid.
...
Quoted 25 lines
while previous kernel versions were able to recognise how brain-damaged the machine is, and only build a fake node.
Use the memblock_validate_numa_coverage() helper to restore some sanity and a "working" system.
Fixes 767507654c22 ("arch_numa: switch over to numa_memblks")
Suggested-by Mike Rapoport <[email protected]>
Signed-off-by Marc Zyngier <[email protected]>
Cc Catalin Marinas <[email protected]>
Cc Will Deacon <[email protected]>
Cc Zi Yan <[email protected]>
Cc Dan Williams <[email protected]>
Cc David Hildenbrand <[email protected]>
Cc Andrew Morton <[email protected]>
Cc [email protected]
---
 drivers/base/arch_numa.c | 4 ++++
 1 file changed, 4 insertions(+)
drivers/base/arch_numa.c +0 -0
index e187016764265..c63a72a1fed64 100644
drivers/base/arch_numa.c static int __init numa_register_nodes(void) · line 212
Quoted 5 lines
drivers/base/arch_numa.c +2 -0
@@ -208,6 +208,10 @@ static int __init numa_register_nodes(void)
 {
 	int nid;
 
+	/* Check the validity of the memblock/node mapping */
+	if (!memblock_validate_numa_coverage(1))
I've changed this to memblock_validate_numa_coverage(0) and applied along with my patch that changed memblock_validate_numa_coverage() to work with 0:

Quoted 6 lines

Quoted 3 lines
https://git.kernel.org/pub/scm/linux/kernel/git/rppt/memblock.git/log/?h=thunderx-fix
Can you please verify that it works on your "quality hardware"?
Commit 427c6179e159b in your tree still has memblock_validate_numa_coverage(1). Forgot to push out the updated version?

Argh, indeed.

Quoted 5 lines

Flipping this to 0 locally, I have verified that this still allows the old thing to trudge along:
root@duodenum:~# uname -a
Linux duodenum 6.12.0-12115-g427c6179e159-dirty #3896 SMP PREEMPT Sun Dec  1 19:43:13 GMT 2024 aarch64

Thanks for testing!

Quoted 3 lines

Thanks again,
	M.

Signature · 2 lines

Sincerely yours,
Mike.