Skip to contentSkip to conversation

KASAN: slab-use-after-free Write in flush_tlb_func

syzbot Superseded syzbot 6 participants 13 messages Jul 2, 2025 – Jul 6, 2025

lore
Share & tools
lore

syzbot

author syzbot
#0 lore raw

Hello,

syzbot found the following issue on:

HEAD commit:    1343433ed389 Add linux-next specific files for 20250630
git tree:       linux-next
console output: https://syzkaller.appspot.com/x/log.txt?x=15b3e3d4580000
kernel config:  https://syzkaller.appspot.com/x/.config?x=c1ce97baf6bd6397
dashboard link: https://syzkaller.appspot.com/bug?extid=084b6e5bc1016723a9c4
compiler:       Debian clang version 20.1.6 (++20250514063057+1e4d39e07757-1~exp1~20250514183223.118), Debian LLD 20.1.6
syz repro:      https://syzkaller.appspot.com/x/repro.syz?x=15716770580000

Downloadable assets:
disk image: https://storage.googleapis.com/syzbot-assets/c3387c64e9ec/disk-1343433e.raw.xz
vmlinux: https://storage.googleapis.com/syzbot-assets/abf15e85d8dd/vmlinux-1343433e.xz
kernel image: https://storage.googleapis.com/syzbot-assets/081c344403bc/bzImage-1343433e.xz

The issue was bisected to:

commit a12a498a9738db65152203467820bb15b6102bd2

Author: Yury Norov [NVIDIA] <[email protected]>
Date:   Mon Jun 23 00:00:08 2025 +0000

smp: Don't wait for remote work done if not needed in smp_call_function_many_cond()

bisection log:  https://syzkaller.appspot.com/x/bisect.txt?x=15ee348c580000
final oops:     https://syzkaller.appspot.com/x/report.txt?x=17ee348c580000
console output: https://syzkaller.appspot.com/x/log.txt?x=13ee348c580000

IMPORTANT: if you fix the issue, please add the following tag to the commit:

Reported-by [email protected]
Fixes a12a498a9738 ("smp: Don't wait for remote work done if not needed in smp_call_function_many_cond()")

==================================================================

BUG: KASAN: slab-use-after-free in instrument_atomic_write include/linux/instrumented.h:82 [inline]
BUG: KASAN: slab-use-after-free in clear_bit include/asm-generic/bitops/instrumented-atomic.h:41 [inline]
BUG: KASAN: slab-use-after-free in cpumask_clear_cpu include/linux/cpumask.h:628 [inline]
BUG: KASAN: slab-use-after-free in flush_tlb_func+0x23d/0x6c0 arch/x86/mm/tlb.c:1132
Write of size 8 at addr ffff88805f1dca80 by task kworker/1:1/43

CPU: 1 UID: 0 PID: 43 Comm: kworker/1:1 Not tainted 6.16.0-rc4-next-20250630-syzkaller #0 PREEMPT(full) 
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 05/07/2025
Workqueue: usb_hub_wq hub_event
Call Trace:
 <IRQ>
 dump_stack_lvl+0x189/0x250 lib/dump_stack.c:120
 print_address_description mm/kasan/report.c:408 [inline]
 print_report+0xd2/0x2b0 mm/kasan/report.c:521
 kasan_report+0x118/0x150 mm/kasan/report.c:634
 check_region_inline mm/kasan/generic.c:-1 [inline]
 kasan_check_range+0x2b0/0x2c0 mm/kasan/generic.c:189
 instrument_atomic_write include/linux/instrumented.h:82 [inline]
 clear_bit include/asm-generic/bitops/instrumented-atomic.h:41 [inline]
 cpumask_clear_cpu include/linux/cpumask.h:628 [inline]
 flush_tlb_func+0x23d/0x6c0 arch/x86/mm/tlb.c:1132
 csd_do_func kernel/smp.c:134 [inline]
 __flush_smp_call_function_queue+0x370/0xaa0 kernel/smp.c:540
 __sysvec_call_function_single+0xa8/0x3d0 arch/x86/kernel/smp.c:271
 instr_sysvec_call_function_single arch/x86/kernel/smp.c:266 [inline]
 sysvec_call_function_single+0x9e/0xc0 arch/x86/kernel/smp.c:266
 </IRQ>
 <TASK>
 asm_sysvec_call_function_single+0x1a/0x20 arch/x86/include/asm/idtentry.h:709
RIP: 0010:console_flush_all+0x7f7/0xc40 kernel/printk/printk.c:3227
Code: 48 21 c3 0f 85 e9 01 00 00 e8 65 2d 1f 00 48 8b 5c 24 20 4d 85 f6 75 07 e8 56 2d 1f 00 eb 06 e8 4f 2d 1f 00 fb 48 8b 44 24 28 <42> 80 3c 20 00 74 08 48 89 df e8 3a 41 83 00 48 8b 1b 48 8b 44 24
RSP: 0018:ffffc90000b36fc0 EFLAGS: 00000293
RAX: 1ffffffff1d36a63 RBX: ffffffff8e9b5318 RCX: ffff88801faa9e00
RDX: 0000000000000000 RSI: 0000000000000000 RDI: 0000000000000000
RBP: ffffc90000b37110 R08: ffffffff8fa17437 R09: 1ffffffff1f42e86
R10: dffffc0000000000 R11: fffffbfff1f42e87 R12: dffffc0000000000
R13: 0000000000000001 R14: 0000000000000200 R15: ffffffff8e9b52c0
 __console_flush_and_unlock kernel/printk/printk.c:3285 [inline]
 console_unlock+0xc4/0x270 kernel/printk/printk.c:3325
 vprintk_emit+0x5b7/0x7a0 kernel/printk/printk.c:2450
 dev_vprintk_emit+0x337/0x3f0 drivers/base/core.c:4916
 dev_printk_emit+0xe0/0x130 drivers/base/core.c:4927
 _dev_info+0x10a/0x160 drivers/base/core.c:4985
 announce_device+0x117/0x2c0 drivers/usb/core/hub.c:2417
 usb_new_device+0x4ef/0x16f0 drivers/usb/core/hub.c:2680
 hub_port_connect drivers/usb/core/hub.c:5571 [inline]
 hub_port_connect_change drivers/usb/core/hub.c:5711 [inline]
 port_event drivers/usb/core/hub.c:5871 [inline]
 hub_event+0x2941/0x4a00 drivers/usb/core/hub.c:5953
 process_one_work kernel/workqueue.c:3239 [inline]
 process_scheduled_works+0xae1/0x17b0 kernel/workqueue.c:3322
 worker_thread+0x8a0/0xda0 kernel/workqueue.c:3403
 kthread+0x70e/0x8a0 kernel/kthread.c:463
 ret_from_fork+0x3fc/0x770 arch/x86/kernel/process.c:148
 ret_from_fork_asm+0x1a/0x30 arch/x86/entry/entry_64.S:245
 </TASK>

Allocated by task 5962:
 kasan_save_stack mm/kasan/common.c:47 [inline]
 kasan_save_track+0x3e/0x80 mm/kasan/common.c:68
 unpoison_slab_object mm/kasan/common.c:319 [inline]
 __kasan_slab_alloc+0x6c/0x80 mm/kasan/common.c:345
 kasan_slab_alloc include/linux/kasan.h:250 [inline]
 slab_post_alloc_hook mm/slub.c:4180 [inline]
 slab_alloc_node mm/slub.c:4229 [inline]
 kmem_cache_alloc_noprof+0x1c1/0x3c0 mm/slub.c:4236
 dup_mm kernel/fork.c:1466 [inline]
 copy_mm+0xdb/0x4b0 kernel/fork.c:1528
 copy_process+0x1706/0x3c00 kernel/fork.c:2168
 kernel_clone+0x21e/0x870 kernel/fork.c:2598
 __do_sys_clone kernel/fork.c:2741 [inline]
 __se_sys_clone kernel/fork.c:2725 [inline]
 __x64_sys_clone+0x18b/0x1e0 kernel/fork.c:2725
 do_syscall_x64 arch/x86/entry/syscall_64.c:63 [inline]
 do_syscall_64+0xfa/0x3b0 arch/x86/entry/syscall_64.c:94
 entry_SYSCALL_64_after_hwframe+0x77/0x7f

Freed by task 8084:
 kasan_save_stack mm/kasan/common.c:47 [inline]
 kasan_save_track+0x3e/0x80 mm/kasan/common.c:68
 kasan_save_free_info+0x46/0x50 mm/kasan/generic.c:576
 poison_slab_object mm/kasan/common.c:247 [inline]
 __kasan_slab_free+0x62/0x70 mm/kasan/common.c:264
 kasan_slab_free include/linux/kasan.h:233 [inline]
 slab_free_hook mm/slub.c:2417 [inline]
 slab_free mm/slub.c:4680 [inline]
 kmem_cache_free+0x18f/0x400 mm/slub.c:4782
 exit_mm+0x1da/0x2c0 kernel/exit.c:581
 do_exit+0x648/0x2300 kernel/exit.c:947
 do_group_exit+0x21c/0x2d0 kernel/exit.c:1100
 __do_sys_exit_group kernel/exit.c:1111 [inline]
 __se_sys_exit_group kernel/exit.c:1109 [inline]
 __x64_sys_exit_group+0x3f/0x40 kernel/exit.c:1109
 x64_sys_call+0x21ba/0x21c0 arch/x86/include/generated/asm/syscalls_64.h:232
 do_syscall_x64 arch/x86/entry/syscall_64.c:63 [inline]
 do_syscall_64+0xfa/0x3b0 arch/x86/entry/syscall_64.c:94
 entry_SYSCALL_64_after_hwframe+0x77/0x7f

The buggy address belongs to the object at ffff88805f1dc080
 which belongs to the cache mm_struct of size 2584

The buggy address is located 2560 bytes inside of
freed 2584-byte region [ffff88805f1dc080, ffff88805f1dca98)

The buggy address belongs to the physical page:
page: refcount:0 mapcount:0 mapping:0000000000000000 index:0xffff88805f1dcb40 pfn:0x5f1d8

head: order:3 mapcount:0 entire_mapcount:0 nr_pages_mapped:0 pincount:0 memcg:ffff88805f612801
anon flags: 0xfff00000000040(head|node=0|zone=1|lastcpupid=0x7ff)
page_type: f5(slab)

raw: 00fff00000000040 ffff88801a44bb40 0000000000000000 dead000000000001
raw: ffff88805f1dcb40 00000000000b0009 00000000f5000000 ffff88805f612801

head: 00fff00000000040 ffff88801a44bb40 0000000000000000 dead000000000001
head: ffff88805f1dcb40 00000000000b0009 00000000f5000000 ffff88805f612801
head: 00fff00000000003 ffffea00017c7601 00000000ffffffff 00000000ffffffff
head: ffffffffffffffff 0000000000000000 00000000ffffffff 0000000000000008

page dumped because: kasan: bad access detected
page_owner tracks the page as allocated
page last allocated via order 3, migratetype Unmovable, gfp_mask 0xd20c0(__GFP_IO|__GFP_FS|__GFP_NOWARN|__GFP_NORETRY|__GFP_COMP|__GFP_NOMEMALLOC), pid 5719, tgid 5719 (dhcpcd-run-hook), ts 61767719991, free_ts 61767008311
 set_page_owner include/linux/page_owner.h:32 [inline]
 post_alloc_hook+0x240/0x2a0 mm/page_alloc.c:1848
 prep_new_page mm/page_alloc.c:1856 [inline]
 get_page_from_freelist+0x21e4/0x22c0 mm/page_alloc.c:3855
 __alloc_frozen_pages_noprof+0x181/0x370 mm/page_alloc.c:5145
 alloc_pages_mpol+0x232/0x4a0 mm/mempolicy.c:2419
 alloc_slab_page mm/slub.c:2487 [inline]
 allocate_slab+0x8a/0x370 mm/slub.c:2655
 new_slab mm/slub.c:2709 [inline]
 ___slab_alloc+0xbeb/0x1410 mm/slub.c:3891
 __slab_alloc mm/slub.c:3981 [inline]
 __slab_alloc_node mm/slub.c:4056 [inline]
 slab_alloc_node mm/slub.c:4217 [inline]
 kmem_cache_alloc_noprof+0x283/0x3c0 mm/slub.c:4236
 dup_mm kernel/fork.c:1466 [inline]
 copy_mm+0xdb/0x4b0 kernel/fork.c:1528
 copy_process+0x1706/0x3c00 kernel/fork.c:2168
 kernel_clone+0x21e/0x870 kernel/fork.c:2598
 __do_sys_clone kernel/fork.c:2741 [inline]
 __se_sys_clone kernel/fork.c:2725 [inline]
 __x64_sys_clone+0x18b/0x1e0 kernel/fork.c:2725
 do_syscall_x64 arch/x86/entry/syscall_64.c:63 [inline]
 do_syscall_64+0xfa/0x3b0 arch/x86/entry/syscall_64.c:94
 entry_SYSCALL_64_after_hwframe+0x77/0x7f
page last free pid 5719 tgid 5719 stack trace:
 reset_page_owner include/linux/page_owner.h:25 [inline]
 free_pages_prepare mm/page_alloc.c:1392 [inline]
 __free_frozen_pages+0xb80/0xd80 mm/page_alloc.c:2892
 __folio_put+0x21b/0x2c0 mm/swap.c:112
 put_netmem net/core/skbuff.c:7372 [inline]
 skb_page_unref include/linux/skbuff_ref.h:43 [inline]
 __skb_frag_unref include/linux/skbuff_ref.h:56 [inline]
 skb_release_data+0x49a/0x7c0 net/core/skbuff.c:1081
 skb_release_all net/core/skbuff.c:1152 [inline]
 napi_consume_skb+0x158/0x1e0 net/core/skbuff.c:1480
 skb_defer_free_flush net/core/dev.c:6632 [inline]
 net_rx_action+0x51b/0xe30 net/core/dev.c:7625
 handle_softirqs+0x286/0x870 kernel/softirq.c:579
 __do_softirq kernel/softirq.c:613 [inline]
 invoke_softirq kernel/softirq.c:453 [inline]
 __irq_exit_rcu+0xca/0x1f0 kernel/softirq.c:680
 irq_exit_rcu+0x9/0x30 kernel/softirq.c:696
 common_interrupt+0xbb/0xe0 arch/x86/kernel/irq.c:285
 asm_common_interrupt+0x26/0x40 arch/x86/include/asm/idtentry.h:693

Memory state around the buggy address:
 ffff88805f1dc980: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
 ffff88805f1dca00: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
>ffff88805f1dca80: fb fb fb fc fc fc fc fc fc fc fc fc fc fc fc fc
                   ^
 ffff88805f1dcb00: fc fc fc fc fc fc fc fc fa fb fb fb fb fb fb fb
 ffff88805f1dcb80: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
==================================================================
----------------
Code disassembly (best guess):
   0:	48 21 c3             	and    %rax,%rbx
   3:	0f 85 e9 01 00 00    	jne    0x1f2
   9:	e8 65 2d 1f 00       	call   0x1f2d73
   e:	48 8b 5c 24 20       	mov    0x20(%rsp),%rbx
  13:	4d 85 f6             	test   %r14,%r14
  16:	75 07                	jne    0x1f
  18:	e8 56 2d 1f 00       	call   0x1f2d73
  1d:	eb 06                	jmp    0x25
  1f:	e8 4f 2d 1f 00       	call   0x1f2d73
  24:	fb                   	sti
  25:	48 8b 44 24 28       	mov    0x28(%rsp),%rax
* 2a:	42 80 3c 20 00       	cmpb   $0x0,(%rax,%r12,1) <-- trapping instruction
  2f:	74 08                	je     0x39
  31:	48 89 df             	mov    %rbx,%rdi
  34:	e8 3a 41 83 00       	call   0x834173
  39:	48 8b 1b             	mov    (%rbx),%rbx
  3c:	48                   	rex.W
  3d:	8b                   	.byte 0x8b
  3e:	44                   	rex.R
  3f:	24                   	.byte 0x24

---
This report is generated by a bot. It may contain errors. See https://goo.gl/tpsmEJ for more information about syzbot. syzbot engineers can be reached at [email protected].

syzbot will keep track of this issue. See:
https://goo.gl/tpsmEJ#status for how to communicate with syzbot. For information about bisection process see: https://goo.gl/tpsmEJ#bisection

If the report is already addressed, let syzbot know by replying with:
#syz fix: exact-commit-title

If you want syzbot to run the reproducer, reply with:
#syz test: git://repo/address.git branch-or-commit-hash
If you attach or paste a git patch, syzbot will apply it before testing.

If you want to overwrite report's subsystems, reply with:
#syz set subsystems: new-subsystem
(See the list of subsystem names on the web dashboard)

If the report is a duplicate of another one, reply with:
#syz dup: exact-subject-of-another-report

If you want to undo deduplication, reply with:
#syz undup

#1 lore raw
Quoted 9 lines · syzbot

The issue was bisected to:

commit a12a498a9738db65152203467820bb15b6102bd2
Author: Yury Norov [NVIDIA] <[email protected]> Date:   Mon Jun 23 00:00:08 2025 +0000

    smp: Don't wait for remote work done if not needed in smp_call_function_many_cond()

While that change looks like it would increase the likelihood of hitting this issue, it does not look like the root cause.

Instead, the stack traces below show that the TLB flush code is being asked to flush the TLB for an mm that is exiting.

One CPU is running the TLB flush handler, while another CPU is freeing the mm_struct.

The CPU that sent the simultaneous TLB flush is not visible in the stack traces below, but we seem to have various places around the MM where we flush the TLB for another mm, without taking any measures to protect against that mm being freed while the flush is ongoing.

I wonder if an easy, and low overhead way to fix that is just to make the mm_struct RCU freed, and hold the rcu_read_lock() in any place where we zap the TLB for another mm.

Are there any places where that wouldn't work?

Do we need to take refcounts on the mm_struct, instead?

Quoted 247 lines

==================================================================

BUG: KASAN: slab-use-after-free in instrument_atomic_write
include/linux/instrumented.h:82 [inline]
BUG: KASAN: slab-use-after-free in clear_bit include/asm-
generic/bitops/instrumented-atomic.h:41 [inline]
BUG: KASAN: slab-use-after-free in cpumask_clear_cpu
include/linux/cpumask.h:628 [inline]
BUG: KASAN: slab-use-after-free in flush_tlb_func+0x23d/0x6c0
arch/x86/mm/tlb.c:1132
Write of size 8 at addr ffff88805f1dca80 by task kworker/1:1/43

CPU: 1 UID: 0 PID: 43 Comm: kworker/1:1 Not tainted 6.16.0-rc4-next-
20250630-syzkaller #0 PREEMPT(full) 
Hardware name: Google Google Compute Engine/Google Compute Engine,
BIOS Google 05/07/2025
Workqueue: usb_hub_wq hub_event
Call Trace:
 <IRQ>
 dump_stack_lvl+0x189/0x250 lib/dump_stack.c:120
 print_address_description mm/kasan/report.c:408 [inline]
 print_report+0xd2/0x2b0 mm/kasan/report.c:521
 kasan_report+0x118/0x150 mm/kasan/report.c:634
 check_region_inline mm/kasan/generic.c:-1 [inline]
 kasan_check_range+0x2b0/0x2c0 mm/kasan/generic.c:189
 instrument_atomic_write include/linux/instrumented.h:82 [inline]
 clear_bit include/asm-generic/bitops/instrumented-atomic.h:41
[inline]
 cpumask_clear_cpu include/linux/cpumask.h:628 [inline]
 flush_tlb_func+0x23d/0x6c0 arch/x86/mm/tlb.c:1132
 csd_do_func kernel/smp.c:134 [inline]
 __flush_smp_call_function_queue+0x370/0xaa0 kernel/smp.c:540
 __sysvec_call_function_single+0xa8/0x3d0 arch/x86/kernel/smp.c:271
 instr_sysvec_call_function_single arch/x86/kernel/smp.c:266 [inline]
 sysvec_call_function_single+0x9e/0xc0 arch/x86/kernel/smp.c:266
 </IRQ>
 <TASK>
 asm_sysvec_call_function_single+0x1a/0x20
arch/x86/include/asm/idtentry.h:709
RIP: 0010:console_flush_all+0x7f7/0xc40 kernel/printk/printk.c:3227
Code: 48 21 c3 0f 85 e9 01 00 00 e8 65 2d 1f 00 48 8b 5c 24 20 4d 85

f6 75 07 e8 56 2d 1f 00 eb 06 e8 4f 2d 1f 00 fb 48 8b 44 24 28 <42> 80 3c 20 00 74 08 48 89 df e8 3a 41 83 00 48 8b 1b 48 8b 44 24
RSP: 0018:ffffc90000b36fc0 EFLAGS: 00000293
RAX: 1ffffffff1d36a63 RBX: ffffffff8e9b5318 RCX: ffff88801faa9e00
RDX: 0000000000000000 RSI: 0000000000000000 RDI: 0000000000000000
RBP: ffffc90000b37110 R08: ffffffff8fa17437 R09: 1ffffffff1f42e86
R10: dffffc0000000000 R11: fffffbfff1f42e87 R12: dffffc0000000000
R13: 0000000000000001 R14: 0000000000000200 R15: ffffffff8e9b52c0

 __console_flush_and_unlock kernel/printk/printk.c:3285 [inline]
 console_unlock+0xc4/0x270 kernel/printk/printk.c:3325
 vprintk_emit+0x5b7/0x7a0 kernel/printk/printk.c:2450
 dev_vprintk_emit+0x337/0x3f0 drivers/base/core.c:4916
 dev_printk_emit+0xe0/0x130 drivers/base/core.c:4927
 _dev_info+0x10a/0x160 drivers/base/core.c:4985
 announce_device+0x117/0x2c0 drivers/usb/core/hub.c:2417
 usb_new_device+0x4ef/0x16f0 drivers/usb/core/hub.c:2680
 hub_port_connect drivers/usb/core/hub.c:5571 [inline]
 hub_port_connect_change drivers/usb/core/hub.c:5711 [inline]
 port_event drivers/usb/core/hub.c:5871 [inline]
 hub_event+0x2941/0x4a00 drivers/usb/core/hub.c:5953
 process_one_work kernel/workqueue.c:3239 [inline]
 process_scheduled_works+0xae1/0x17b0 kernel/workqueue.c:3322
 worker_thread+0x8a0/0xda0 kernel/workqueue.c:3403
 kthread+0x70e/0x8a0 kernel/kthread.c:463
 ret_from_fork+0x3fc/0x770 arch/x86/kernel/process.c:148
 ret_from_fork_asm+0x1a/0x30 arch/x86/entry/entry_64.S:245
 </TASK>

Allocated by task 5962:
 kasan_save_stack mm/kasan/common.c:47 [inline]
 kasan_save_track+0x3e/0x80 mm/kasan/common.c:68
 unpoison_slab_object mm/kasan/common.c:319 [inline]
 __kasan_slab_alloc+0x6c/0x80 mm/kasan/common.c:345
 kasan_slab_alloc include/linux/kasan.h:250 [inline]
 slab_post_alloc_hook mm/slub.c:4180 [inline]
 slab_alloc_node mm/slub.c:4229 [inline]
 kmem_cache_alloc_noprof+0x1c1/0x3c0 mm/slub.c:4236
 dup_mm kernel/fork.c:1466 [inline]
 copy_mm+0xdb/0x4b0 kernel/fork.c:1528
 copy_process+0x1706/0x3c00 kernel/fork.c:2168
 kernel_clone+0x21e/0x870 kernel/fork.c:2598
 __do_sys_clone kernel/fork.c:2741 [inline]
 __se_sys_clone kernel/fork.c:2725 [inline]
 __x64_sys_clone+0x18b/0x1e0 kernel/fork.c:2725
 do_syscall_x64 arch/x86/entry/syscall_64.c:63 [inline]
 do_syscall_64+0xfa/0x3b0 arch/x86/entry/syscall_64.c:94
 entry_SYSCALL_64_after_hwframe+0x77/0x7f

Freed by task 8084:
 kasan_save_stack mm/kasan/common.c:47 [inline]
 kasan_save_track+0x3e/0x80 mm/kasan/common.c:68
 kasan_save_free_info+0x46/0x50 mm/kasan/generic.c:576
 poison_slab_object mm/kasan/common.c:247 [inline]
 __kasan_slab_free+0x62/0x70 mm/kasan/common.c:264
 kasan_slab_free include/linux/kasan.h:233 [inline]
 slab_free_hook mm/slub.c:2417 [inline]
 slab_free mm/slub.c:4680 [inline]
 kmem_cache_free+0x18f/0x400 mm/slub.c:4782
 exit_mm+0x1da/0x2c0 kernel/exit.c:581
 do_exit+0x648/0x2300 kernel/exit.c:947
 do_group_exit+0x21c/0x2d0 kernel/exit.c:1100
 __do_sys_exit_group kernel/exit.c:1111 [inline]
 __se_sys_exit_group kernel/exit.c:1109 [inline]
 __x64_sys_exit_group+0x3f/0x40 kernel/exit.c:1109
 x64_sys_call+0x21ba/0x21c0
arch/x86/include/generated/asm/syscalls_64.h:232
 do_syscall_x64 arch/x86/entry/syscall_64.c:63 [inline]
 do_syscall_64+0xfa/0x3b0 arch/x86/entry/syscall_64.c:94
 entry_SYSCALL_64_after_hwframe+0x77/0x7f

The buggy address belongs to the object at ffff88805f1dc080

 which belongs to the cache mm_struct of size 2584 The buggy address is located 2560 bytes inside of freed 2584-byte region [ffff88805f1dc080, ffff88805f1dca98)

The buggy address belongs to the physical page:
page: refcount:0 mapcount:0 mapping:0000000000000000

index:0xffff88805f1dcb40 pfn:0x5f1d8
head: order:3 mapcount:0 entire_mapcount:0 nr_pages_mapped:0 pincount:0
memcg:ffff88805f612801
anon flags: 0xfff00000000040(head|node=0|zone=1|lastcpupid=0x7ff)
page_type: f5(slab)

raw: 00fff00000000040 ffff88801a44bb40 0000000000000000
dead000000000001
raw: ffff88805f1dcb40 00000000000b0009 00000000f5000000

ffff88805f612801
head: 00fff00000000040 ffff88801a44bb40 0000000000000000 dead000000000001
head: ffff88805f1dcb40 00000000000b0009 00000000f5000000 ffff88805f612801
head: 00fff00000000003 ffffea00017c7601 00000000ffffffff 00000000ffffffff
head: ffffffffffffffff 0000000000000000 00000000ffffffff 0000000000000008

page dumped because: kasan: bad access detected
page_owner tracks the page as allocated
page last allocated via order 3, migratetype Unmovable, gfp_mask
0xd20c0(__GFP_IO|__GFP_FS|__GFP_NOWARN|__GFP_NORETRY|__GFP_COMP|__GFP
_NOMEMALLOC), pid 5719, tgid 5719 (dhcpcd-run-hook), ts 61767719991,
free_ts 61767008311
 set_page_owner include/linux/page_owner.h:32 [inline]
 post_alloc_hook+0x240/0x2a0 mm/page_alloc.c:1848
 prep_new_page mm/page_alloc.c:1856 [inline]
 get_page_from_freelist+0x21e4/0x22c0 mm/page_alloc.c:3855
 __alloc_frozen_pages_noprof+0x181/0x370 mm/page_alloc.c:5145
 alloc_pages_mpol+0x232/0x4a0 mm/mempolicy.c:2419
 alloc_slab_page mm/slub.c:2487 [inline]
 allocate_slab+0x8a/0x370 mm/slub.c:2655
 new_slab mm/slub.c:2709 [inline]
 ___slab_alloc+0xbeb/0x1410 mm/slub.c:3891
 __slab_alloc mm/slub.c:3981 [inline]
 __slab_alloc_node mm/slub.c:4056 [inline]
 slab_alloc_node mm/slub.c:4217 [inline]
 kmem_cache_alloc_noprof+0x283/0x3c0 mm/slub.c:4236
 dup_mm kernel/fork.c:1466 [inline]
 copy_mm+0xdb/0x4b0 kernel/fork.c:1528
 copy_process+0x1706/0x3c00 kernel/fork.c:2168
 kernel_clone+0x21e/0x870 kernel/fork.c:2598
 __do_sys_clone kernel/fork.c:2741 [inline]
 __se_sys_clone kernel/fork.c:2725 [inline]
 __x64_sys_clone+0x18b/0x1e0 kernel/fork.c:2725
 do_syscall_x64 arch/x86/entry/syscall_64.c:63 [inline]
 do_syscall_64+0xfa/0x3b0 arch/x86/entry/syscall_64.c:94
 entry_SYSCALL_64_after_hwframe+0x77/0x7f
page last free pid 5719 tgid 5719 stack trace:
 reset_page_owner include/linux/page_owner.h:25 [inline]
 free_pages_prepare mm/page_alloc.c:1392 [inline]
 __free_frozen_pages+0xb80/0xd80 mm/page_alloc.c:2892
 __folio_put+0x21b/0x2c0 mm/swap.c:112
 put_netmem net/core/skbuff.c:7372 [inline]
 skb_page_unref include/linux/skbuff_ref.h:43 [inline]
 __skb_frag_unref include/linux/skbuff_ref.h:56 [inline]
 skb_release_data+0x49a/0x7c0 net/core/skbuff.c:1081
 skb_release_all net/core/skbuff.c:1152 [inline]
 napi_consume_skb+0x158/0x1e0 net/core/skbuff.c:1480
 skb_defer_free_flush net/core/dev.c:6632 [inline]
 net_rx_action+0x51b/0xe30 net/core/dev.c:7625
 handle_softirqs+0x286/0x870 kernel/softirq.c:579
 __do_softirq kernel/softirq.c:613 [inline]
 invoke_softirq kernel/softirq.c:453 [inline]
 __irq_exit_rcu+0xca/0x1f0 kernel/softirq.c:680
 irq_exit_rcu+0x9/0x30 kernel/softirq.c:696
 common_interrupt+0xbb/0xe0 arch/x86/kernel/irq.c:285
 asm_common_interrupt+0x26/0x40 arch/x86/include/asm/idtentry.h:693

Memory state around the buggy address:
 ffff88805f1dc980: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
 ffff88805f1dca00: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
Quoted 1 line

ffff88805f1dca80: fb fb fb fc fc fc fc fc fc fc fc fc fc fc fc fc

                   ^
 ffff88805f1dcb00: fc fc fc fc fc fc fc fc fa fb fb fb fb fb fb fb
 ffff88805f1dcb80: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
==================================================================
----------------

Code disassembly (best guess):
   0:	48 21 c3             	and    %rax,%rbx
   3:	0f 85 e9 01 00 00    	jne    0x1f2
   9:	e8 65 2d 1f 00       	call   0x1f2d73
   e:	48 8b 5c 24 20       	mov    0x20(%rsp),%rbx
  13:	4d 85 f6             	test   %r14,%r14
  16:	75 07                	jne    0x1f
  18:	e8 56 2d 1f 00       	call   0x1f2d73
  1d:	eb 06                	jmp    0x25
  1f:	e8 4f 2d 1f 00       	call   0x1f2d73
  24:	fb                   	sti
  25:	48 8b 44 24 28       	mov    0x28(%rsp),%rax
* 2a:	42 80 3c 20 00       	cmpb   $0x0,(%rax,%r12,1) <--

trapping instruction
  2f: 74 08                 je     0x39
  31: 48 89 df              mov    %rbx,%rdi
  34: e8 3a 41 83 00        call   0x834173
  39: 48 8b 1b              mov    (%rbx),%rbx
  3c: 48                    rex.W
  3d: 8b                    .byte 0x8b
  3e: 44                    rex.R
  3f: 24                    .byte 0x24

---
This report is generated by a bot. It may contain errors. See https://goo.gl/tpsmEJ for more information about syzbot. syzbot engineers can be reached at [email protected].

syzbot will keep track of this issue. See:
https://goo.gl/tpsmEJ#status for how to communicate with syzbot. For information about bisection process see:
https://goo.gl/tpsmEJ#bisection

If the report is already addressed, let syzbot know by replying with:
#syz fix: exact-commit-title

If you want syzbot to run the reproducer, reply with:
#syz test: git://repo/address.git branch-or-commit-hash If you attach or paste a git patch, syzbot will apply it before testing.

If you want to overwrite report's subsystems, reply with:
#syz set subsystems: new-subsystem
(See the list of subsystem names on the web dashboard)

If the report is a duplicate of another one, reply with:
#syz dup: exact-subject-of-another-report

If you want to undo deduplication, reply with:
#syz undup

Signature · 1 line
All Rights Reversed.
#2 lore raw
Quoted 23 lines · Rik van Riel
Quoted 9 lines · syzbot

The issue was bisected to:

commit a12a498a9738db65152203467820bb15b6102bd2

Author: Yury Norov [NVIDIA] <[email protected]>
Date:   Mon Jun 23 00:00:08 2025 +0000

smp: Don't wait for remote work done if not needed in
smp_call_function_many_cond()

While that change looks like it would increase the likelihood of hitting this issue, it does not look like the root cause.

Instead, the stack traces below show that the TLB flush code is being asked to flush the TLB for an mm that is exiting.

One CPU is running the TLB flush handler, while another CPU is freeing the mm_struct.

Quoted 6 lines

The CPU that sent the simultaneous TLB flush is not visible in the stack traces below, but we seem to have various places around the MM where we flush the TLB for another mm, without taking any measures to protect against that mm being freed while the flush is ongoing.

TLB flushes via IPIs on x86 are always synchronous, right? flush_tlb_func is only referenced from native_flush_tlb_multi() in calls to on_each_cpu_mask() (with wait=true) or on_each_cpu_cond_mask() (with wait=1).
So I think this is not an issue, unless you're claiming that we call native_flush_tlb_multi() with an already-freed info->mm?

And I think the bisected commit really is the buggy one: It looks at "nr_cpus", which tracks *how many CPUs we have to IPI*, but assumes that "nr_cpus" tracks *how many CPUs we posted work to*. Those numbers are not the same: If we post work to a CPU that already had IPI work pending, we just add a list entry without sending another IPI.

#3 lore raw
Quoted 39 lines · Jann Horn
Quoted 23 lines · Rik van Riel
Quoted 9 lines · syzbot

The issue was bisected to:

commit a12a498a9738db65152203467820bb15b6102bd2

Author: Yury Norov [NVIDIA] <[email protected]>
Date:   Mon Jun 23 00:00:08 2025 +0000

smp: Don't wait for remote work done if not needed in
smp_call_function_many_cond()

While that change looks like it would increase the likelihood of hitting this issue, it does not look like the root cause.

Instead, the stack traces below show that the TLB flush code is being asked to flush the TLB for an mm that is exiting.

One CPU is running the TLB flush handler, while another CPU is freeing the mm_struct.

Quoted 6 lines

The CPU that sent the simultaneous TLB flush is not visible in the stack traces below, but we seem to have various places around the MM where we flush the TLB for another mm, without taking any measures to protect against that mm being freed while the flush is ongoing.

TLB flushes via IPIs on x86 are always synchronous, right? flush_tlb_func is only referenced from native_flush_tlb_multi() in calls to on_each_cpu_mask() (with wait=true) or on_each_cpu_cond_mask() (with wait=1).
So I think this is not an issue, unless you're claiming that we call native_flush_tlb_multi() with an already-freed info->mm?

Quoted 5 lines

And I think the bisected commit really is the buggy one: It looks at "nr_cpus", which tracks *how many CPUs we have to IPI*, but assumes that "nr_cpus" tracks *how many CPUs we posted work to*. Those numbers are not the same: If we post work to a CPU that already had IPI work pending, we just add a list entry without sending another IPI.

Or in other words: After that blamed commit, if CPU 1 posts a TLB flush to CPU 3, and then CPU 2 also quickly posts a TLB flush to CPU 3, then CPU 2 will erroneously not wait for the TLB flush to complete before reporting flush completion, which AFAICS means we can get both stale TLB entries and (less often) UAF.

I think the correct version of that commit would be to revert that commit and instead just move the "run_remote = true;" line down, below the cond_func() check.

#4 lore raw
Quoted 14 lines · Jann Horn
Quoted 7 lines · Jann Horn

TLB flushes via IPIs on x86 are always synchronous, right? flush_tlb_func is only referenced from native_flush_tlb_multi() in calls to on_each_cpu_mask() (with wait=true) or on_each_cpu_cond_mask() (with wait=1).
So I think this is not an issue, unless you're claiming that we call native_flush_tlb_multi() with an already-freed info->mm?

Quoted 5 lines

And I think the bisected commit really is the buggy one: It looks at "nr_cpus", which tracks *how many CPUs we have to IPI*, but assumes that "nr_cpus" tracks *how many CPUs we posted work to*. Those numbers are not the same: If we post work to a CPU that already had IPI work pending, we just add a list entry without sending another IPI.

Quoted 5 lines

Or in other words: After that blamed commit, if CPU 1 posts a TLB flush to CPU 3, and then CPU 2 also quickly posts a TLB flush to CPU 3, then CPU 2 will erroneously not wait for the TLB flush to complete before reporting flush completion, which AFAICS means we can get both stale TLB entries and (less often) UAF.

Right you are. Well analyzed and I missed it when taking the lot.

Quoted 3 lines

I think the correct version of that commit would be to revert that commit and instead just move the "run_remote = true;" line down, below the cond_func() check.

I remove it from the relevant tip branch

Thanks,

tglx

#5 lore raw
Quoted 21 lines · Thomas Gleixner
Quoted 14 lines · Jann Horn
Quoted 7 lines · Jann Horn

TLB flushes via IPIs on x86 are always synchronous, right? flush_tlb_func is only referenced from native_flush_tlb_multi() in calls to on_each_cpu_mask() (with wait=true) or on_each_cpu_cond_mask() (with wait=1).
So I think this is not an issue, unless you're claiming that we call native_flush_tlb_multi() with an already-freed info->mm?

Quoted 5 lines

And I think the bisected commit really is the buggy one: It looks at "nr_cpus", which tracks *how many CPUs we have to IPI*, but assumes that "nr_cpus" tracks *how many CPUs we posted work to*. Those numbers are not the same: If we post work to a CPU that already had IPI work pending, we just add a list entry without sending another IPI.

Quoted 5 lines

Or in other words: After that blamed commit, if CPU 1 posts a TLB flush to CPU 3, and then CPU 2 also quickly posts a TLB flush to CPU 3, then CPU 2 will erroneously not wait for the TLB flush to complete before reporting flush completion, which AFAICS means we can get both stale TLB entries and (less often) UAF.

Quoted 7 lines

Right you are. Well analyzed and I missed it when taking the lot.

Quoted 3 lines

I think the correct version of that commit would be to revert that commit and instead just move the "run_remote = true;" line down, below the cond_func() check.

I remove it from the relevant tip branch

Thank you guys for explaining that and sorry for the buggy patch. I was actually under impression that run_remote duplicates nr_cpus !=0, and even have a patch that removes run_remote.

Maybe worth to add a comment on what run_remote and nr_cpus track?

Thanks,
Yury

#6 lore raw

PATCH 1/1 smp: Wait for enqueued work regardless of IPI sent

Quoted 5 lines · On Wed, 2 Jul 2025 13:44:34 -0400 Yury Norov <[email protected]> wrote:

Thank you guys for explaining that and sorry for the buggy patch. I was actually under impression that run_remote duplicates nr_cpus !=0, and even have a patch that removes run_remote.

Maybe worth to add a comment on what run_remote and nr_cpus track?

This thread did surface some useful content, and Jann also pointed out a good optimization that can be made, by not setting run_remote if "func" tells us to skip remote CPUs.

Thomas, please let me know if you already reverted Yury's patch, and want me to re-send this without the last hunk.

---8<---
From 2ae6417fa7ce16f1bfa574cbabba572436adbed9 Mon Sep 17 00:00:00 2001
From: Rik van Riel <[email protected]>
Date: Wed, 2 Jul 2025 13:52:54 -0400
Subject: [PATCH] smp: Wait for enqueued work regardless of IPI sent

Whenever work is enqueued with a remote CPU, smp_call_function_many_cond() may need to wait for that work to be completed, regardless of whether or not the remote CPU needed to be woken up with an IPI, or the work was being added to the queue of an already woken up CPU.

However, if no work is enqueued with a remote CPU, because "func" told us to skip all CPUs, do not wait.

Document the difference between "work enqueued", and "CPU needs to be woken up"

Reported-by [email protected]
Fixes a12a498a9738 ("smp: Don't wait for remote work done if not needed in smp_call_function_many_cond()")
---
 kernel/smp.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)
kernel/smp.c +7 -3
index 84561258cd22..c5e1da7a88da 100644
@@ -802,7 +802,6 @@ static void smp_call_function_many_cond(const struct cpumask *mask,
 
 	/* Check if we need remote execution, i.e., any CPU excluding this one. */
 	if (cpumask_any_and_but(mask, cpu_online_mask, this_cpu) < nr_cpu_ids) {
-		run_remote = true;
 		cfd = this_cpu_ptr(&cfd_data);
 		cpumask_and(cfd->cpumask, mask, cpu_online_mask);
 		__cpumask_clear_cpu(this_cpu, cfd->cpumask);
@@ -816,6 +815,9 @@ static void smp_call_function_many_cond(const struct cpumask *mask,
 				continue;
 			}
 
+			/* Work is enqueued on a remote CPU. */
+			run_remote = true;
+
 			csd_lock(csd);
 			if (wait)
 				csd->node.u_flags |= CSD_TYPE_SYNC;
@@ -827,6 +829,10 @@ static void smp_call_function_many_cond(const struct cpumask *mask,
 #endif
 			trace_csd_queue_cpu(cpu, _RET_IP_, func, csd);
 
+			/*
+			 * Kick the remote CPU if this is the first work
+			 * item enqueued.
+			 */
 			if (llist_add(&csd->node.llist, &per_cpu(call_single_queue, cpu))) {
 				__cpumask_set_cpu(cpu, cfd->cpumask_ipi);
 				nr_cpus++;
@@ -843,8 +849,6 @@ static void smp_call_function_many_cond(const struct cpumask *mask,
 			send_call_function_single_ipi(last_cpu);
 		else if (likely(nr_cpus > 1))
 			send_call_function_ipi_mask(cfd->cpumask_ipi);
-		else
-			run_remote = false;
 	}
 
 	/* Check if we need local execution. */
Signature · 1 line
2.49.0
Reviewed-by on v1 1/1
#7 lore raw

Re: [PATCH] smp: Wait for enqueued work regardless of IPI sent

Quoted 54 lines · Rik van Riel
Quoted 5 lines · On Wed, 2 Jul 2025 13:44:34 -0400 Yury Norov <[email protected]> wrote:

Thank you guys for explaining that and sorry for the buggy patch. I was actually under impression that run_remote duplicates nr_cpus !=0, and even have a patch that removes run_remote.

Maybe worth to add a comment on what run_remote and nr_cpus track?

This thread did surface some useful content, and Jann also pointed out a good optimization that can be made, by not setting run_remote if "func" tells us to skip remote CPUs.

Thomas, please let me know if you already reverted Yury's patch, and want me to re-send this without the last hunk.

---8<---
From 2ae6417fa7ce16f1bfa574cbabba572436adbed9 Mon Sep 17 00:00:00 2001
From: Rik van Riel <[email protected]>
Date: Wed, 2 Jul 2025 13:52:54 -0400
Subject: [PATCH] smp: Wait for enqueued work regardless of IPI sent

Whenever work is enqueued with a remote CPU, smp_call_function_many_cond() may need to wait for that work to be completed, regardless of whether or not the remote CPU needed to be woken up with an IPI, or the work was being added to the queue of an already woken up CPU.

However, if no work is enqueued with a remote CPU, because "func" told us to skip all CPUs, do not wait.

Document the difference between "work enqueued", and "CPU needs to be woken up"

Reported-by [email protected]
Fixes a12a498a9738 ("smp: Don't wait for remote work done if not needed in smp_call_function_many_cond()")
---
 kernel/smp.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)
kernel/smp.c +0 -1
index 84561258cd22..c5e1da7a88da 100644
@@ -802,7 +802,6 @@ static void smp_call_function_many_cond(const struct cpumask *mask,
 
 	/* Check if we need remote execution, i.e., any CPU excluding this one. */
 	if (cpumask_any_and_but(mask, cpu_online_mask, this_cpu) < nr_cpu_ids) {
-		run_remote = true;
 		cfd = this_cpu_ptr(&cfd_data);
 		cpumask_and(cfd->cpumask, mask, cpu_online_mask);
 		__cpumask_clear_cpu(this_cpu, cfd->cpumask);

kernel/smp.c static void smp_call_function_many_cond(const struct cpumask *mask, · line 820

Quoted 6 lines
kernel/smp.c +3 -0
@@ -816,6 +815,9 @@ static void smp_call_function_many_cond(const struct cpumask *mask,
 				continue;
 			}
 
+			/* Work is enqueued on a remote CPU. */
+			run_remote = true;
+

I actually ended up with the same on my cratch branch:

https://github.com/norov/linux/commit/8a32ca4b60dc68ac54f3b70b4be7a5863dc3934e

So,

Quoted 26 lines
kernel/smp.c +4 -2
 			csd_lock(csd);
 			if (wait)
 				csd->node.u_flags |= CSD_TYPE_SYNC;
@@ -827,6 +829,10 @@ static void smp_call_function_many_cond(const struct cpumask *mask,
 #endif
 			trace_csd_queue_cpu(cpu, _RET_IP_, func, csd);
 
+			/*
+			 * Kick the remote CPU if this is the first work
+			 * item enqueued.
+			 */
 			if (llist_add(&csd->node.llist, &per_cpu(call_single_queue, cpu))) {
 				__cpumask_set_cpu(cpu, cfd->cpumask_ipi);
 				nr_cpus++;
@@ -843,8 +849,6 @@ static void smp_call_function_many_cond(const struct cpumask *mask,
 			send_call_function_single_ipi(last_cpu);
 		else if (likely(nr_cpus > 1))
 			send_call_function_ipi_mask(cfd->cpumask_ipi);
-		else
-			run_remote = false;
 	}
 
 	/* Check if we need local execution. */
Signature · 1 line
2.49.0
#8 lore raw

Re: [PATCH] smp: Wait for enqueued work regardless of IPI sent

In reply to Rik van Riel
Quoted 2 lines · Rik van Riel

Thomas, please let me know if you already reverted Yury's patch, and want me to re-send this without the last hunk.

I did so immediately after saying so in my previous reply. It's gone in tip and next.

Thanks,

tglx

#9 lore raw

PATCH v2 1/1 smp: Wait for enqueued work regardless of IPI sent

Quoted 6 lines · On Thu, 03 Jul 2025 18:56:11 +0200 Thomas Gleixner <[email protected]> wrote:
Quoted 2 lines · Rik van Riel

Thomas, please let me know if you already reverted Yury's patch, and want me to re-send this without the last hunk.

I did so immediately after saying so in my previous reply. It's gone in tip and next.

Here is v2 of the patch, with the last hunk removed, and the changelog adjusted to match the new context.

---8<---
From 2ae6417fa7ce16f1bfa574cbabba572436adbed9 Mon Sep 17 00:00:00 2001
From: Rik van Riel <[email protected]>
Date: Wed, 2 Jul 2025 13:52:54 -0400
Subject: [PATCH] smp: Wait only if work was enqueued

Whenever work is enqueued with a remote CPU, smp_call_function_many_cond() may need to wait for that work to be completed. However, if no work is enqueued with a remote CPU, because "func" told us to skip all CPUs, there is no need to wait.

Set run_remote only if work was enqueued on remote CPUs.

Document the difference between "work enqueued", and "CPU needs to be woken up"

---
 kernel/smp.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletions(-)
kernel/smp.c +7 -1
index 84561258cd22..c5e1da7a88da 100644
@@ -802,7 +802,6 @@ static void smp_call_function_many_cond(const struct cpumask *mask,
 
 	/* Check if we need remote execution, i.e., any CPU excluding this one. */
 	if (cpumask_any_and_but(mask, cpu_online_mask, this_cpu) < nr_cpu_ids) {
-		run_remote = true;
 		cfd = this_cpu_ptr(&cfd_data);
 		cpumask_and(cfd->cpumask, mask, cpu_online_mask);
 		__cpumask_clear_cpu(this_cpu, cfd->cpumask);
@@ -816,6 +815,9 @@ static void smp_call_function_many_cond(const struct cpumask *mask,
 				continue;
 			}
 
+			/* Work is enqueued on a remote CPU. */
+			run_remote = true;
+
 			csd_lock(csd);
 			if (wait)
 				csd->node.u_flags |= CSD_TYPE_SYNC;
@@ -827,6 +829,10 @@ static void smp_call_function_many_cond(const struct cpumask *mask,
 #endif
 			trace_csd_queue_cpu(cpu, _RET_IP_, func, csd);
 
+			/*
+			 * Kick the remote CPU if this is the first work
+			 * item enqueued.
+			 */
 			if (llist_add(&csd->node.llist, &per_cpu(call_single_queue, cpu))) {
 				__cpumask_set_cpu(cpu, cfd->cpumask_ipi);
 				nr_cpus++;
Signature · 1 line
2.49.0
merge bot Applied
#10 lore raw

[tip: smp/core] smp: Wait only if work was enqueued

The following commit has been merged into the smp/core branch of tip:

The following commit has been merged into the smp/core branch of tip:

Commit-ID:     946a7281982530d333eaee62bd1726f25908b3a9
Gitweb:        https://git.kernel.org/tip/946a7281982530d333eaee62bd1726f25908b3a9
Author:        Rik van Riel <[email protected]>
AuthorDate:    Wed, 02 Jul 2025 13:52:54 -04:00
Committer:     Thomas Gleixner <[email protected]>
CommitterDate: Sun, 06 Jul 2025 11:57:39 +02:00

smp: Wait only if work was enqueued

Whenever work is enqueued for a remote CPU, smp_call_function_many_cond() may need to wait for that work to be completed. However, if no work is enqueued for a remote CPU, because the condition func() evaluated to false for all CPUs, there is no need to wait.

Set run_remote only if work was enqueued on remote CPUs.

Document the difference between "work enqueued", and "CPU needs to be woken up"

---
 kernel/smp.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
kernel/smp.c +7 -1
index 99d1fd0..c5e1da7 100644
@@ -802,7 +802,6 @@ static void smp_call_function_many_cond(const struct cpumask *mask,
 
 	/* Check if we need remote execution, i.e., any CPU excluding this one. */
 	if (cpumask_any_and_but(mask, cpu_online_mask, this_cpu) < nr_cpu_ids) {
-		run_remote = true;
 		cfd = this_cpu_ptr(&cfd_data);
 		cpumask_and(cfd->cpumask, mask, cpu_online_mask);
 		__cpumask_clear_cpu(this_cpu, cfd->cpumask);
@@ -816,6 +815,9 @@ static void smp_call_function_many_cond(const struct cpumask *mask,
 				continue;
 			}
 
+			/* Work is enqueued on a remote CPU. */
+			run_remote = true;
+
 			csd_lock(csd);
 			if (wait)
 				csd->node.u_flags |= CSD_TYPE_SYNC;
@@ -827,6 +829,10 @@ static void smp_call_function_many_cond(const struct cpumask *mask,
 #endif
 			trace_csd_queue_cpu(cpu, _RET_IP_, func, csd);
 
+			/*
+			 * Kick the remote CPU if this is the first work
+			 * item enqueued.
+			 */
 			if (llist_add(&csd->node.llist, &per_cpu(call_single_queue, cpu))) {
 				__cpumask_set_cpu(cpu, cfd->cpumask_ipi);
 				nr_cpus++;
#11 lore raw
In reply to Jann Horn
Quoted 9 lines · Jann Horn

TLB flushes via IPIs on x86 are always synchronous, right? flush_tlb_func is only referenced from native_flush_tlb_multi() in calls to on_each_cpu_mask() (with wait=true) or on_each_cpu_cond_mask() (with wait=1).
So I think this is not an issue, unless you're claiming that we call native_flush_tlb_multi() with an already-freed info->mm?

It looks like there are a least some cases where try_to_unmap() can call flush_tlb_range() with an mm that belongs to some other process.

I don't know whether that is an issue.

Quoted 7 lines

And I think the bisected commit really is the buggy one: It looks at "nr_cpus", which tracks *how many CPUs we have to IPI*, but assumes that "nr_cpus" tracks *how many CPUs we posted work to*. Those numbers
are not the same: If we post work to a CPU that already had IPI work pending, we just add a list entry without sending another IPI.

Good point, we do need to wait when we enqueue work, even if we do not send an IPI anywhere!

You are right that the bisected commit is buggy and should be reverted.

Signature · 1 line
All Rights Reversed.
#12 lore raw
Quoted 8 lines · Rik van Riel
Quoted 7 lines · Jann Horn

TLB flushes via IPIs on x86 are always synchronous, right? flush_tlb_func is only referenced from native_flush_tlb_multi() in calls to on_each_cpu_mask() (with wait=true) or on_each_cpu_cond_mask() (with wait=1).
So I think this is not an issue, unless you're claiming that we call native_flush_tlb_multi() with an already-freed info->mm?

Quoted 5 lines

It looks like there are a least some cases where
try_to_unmap() can call flush_tlb_range() with
an mm that belongs to some other process.

I don't know whether that is an issue.

try_to_unmap() relies on read-locking either the anon_vma (for anonymous pages) or the address_space (for file pages) throughout the entire rmap walk to ensure that the list of VMAs attached to the anon_vma/address_space stays stable during the operation, which guarantees that those VMAs can't go away, which guarantees that the associated MMs can't go away.

If the caller passes in TTU_RMAP_LOCKED, they promise that they've already taken care of this rmap locking; otherwise, rmap_walk() will do it internally.