mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
* [BUG] timers/nohz: Oops in wake_up_nohz_cpu+0xb with cpumask index -1
@ 2026-11-17 10:48 Eryepa Yerepilov
  2026-11-17 15:59 ` Frederic Weisbecker
  0 siblings, 1 reply; 2+ messages in thread
From: Eryepa Yerepilov @ 2026-11-17 10:48 UTC (permalink / raw)
  To: linux-kernel; +Cc: Thomas Gleixner, Anna-Maria Behnsen, Frederic Weisbecker

Hi,

Summary: 7.3.4-arch1-1 panicked in idle on CPU 0 inside
wake_up_nohz_cpu(), reached from __mod_timer() by the clocksource
watchdog timer. The faulting instruction is a bit test against a static
cpumask, with the index in RAX = 0xffffffff, i.e. -1.

Full trace, copied from pstore (efi_pstore backend; the journal did not
survive the panic, the machine rebooted ~10 s later because of
panic_on_oops=1). Timestamps are seconds since boot, uptime 43040 s:

 [43040.869305] BUG: unable to handle page fault for address: ffffffffadef72d8
 [43040.869312] #PF: supervisor read access in kernel mode
 [43040.869315] #PF: error_code(0x0000) - not-present page
 [43040.869318] PGD 47ac25067 P4D 47ac25067 PUD 47ac26063 PMD 0
 [43040.869324] Oops: Oops: 0000 [#1] SMP NOPTI
 [43040.869329] CPU: 0 UID: 0 PID: 0 Comm: swapper/0 Tainted: G           OE       7.3.4-arch1-1 #1 PREEMPT(full)  6c50e2b86768c03b0a904a1018778cfad2de7f30
 [43040.869339] RIP: 0010:wake_up_nohz_cpu+0xb/0x160
 [43040.869346] Code: 07 ff ff ff 66 66 2e 0f 1f 84 00 00 00 00 00 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 0f 1f 40 d6 0f 1f 44 00 00 89 f8 <48> 0f a3 05 3d ed 72 02 0f 83 da 00 00 00 48 83 ec 10 89 fe 66 90
 [43040.869349] RSP: 0018:ffffcff5c0003dd0 EFLAGS: 00010082
 [43040.869352] RAX: 00000000ffffffff RBX: ffff8d0c0ea1dfc0 RCX: 0000000000000000
 [43040.869355] RDX: 000000001b000000 RSI: ffffffff8e9b66e0 RDI: 00000000ffffffff
 [43040.869357] RBP: ffff8d0c0ea1dfc0 R08: ffff8d0c0ea1dfc0 R09: 00000001028c3160
 [43040.869359] R10: 00001e45a76494c8 R11: 0000000000000009 R12: 0000000000000000
 [43040.869362] R13: 00000001028c315c R14: ffffffff8e9b66e0 R15: 00000000ffffffff
 [43040.869364] FS:  0000000000000000(0000) GS:ffff8d0c80110000(0000) knlGS:0000000000000000
 [43040.869367] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 [43040.869369] CR2: ffffffffadef72d8 CR3: 000000019bf59000 CR4: 0000000000f50ef0
 [43040.869372] PKRU: 55555554
 [43040.869374] Call Trace:
 [43040.869376]  <IRQ>
 [43040.869378]  __mod_timer+0x2b4/0x3a0
 [43040.869384]  ? __pfx_clocksource_watchdog+0x10/0x10
 [43040.869389]  clocksource_watchdog+0x559/0x590
 [43040.869393]  ? __pfx_clocksource_watchdog+0x10/0x10
 [43040.869398]  call_timer_fn+0x2a/0x140
 [43040.869402]  ? __pfx_clocksource_watchdog+0x10/0x10
 [43040.869405]  __run_timers+0x27c/0x330
 [43040.869412]  run_timer_softirq+0x46/0xf0
 [43040.869415]  ? irqtime_account_irq+0x33/0xd0
 [43040.869419]  handle_softirqs+0xde/0x290
 [43040.869429]  __irq_exit_rcu+0xb9/0x110
 [43040.869433]  sysvec_apic_timer_interrupt+0x71/0x90
 [43040.869438]  </IRQ>
 [43040.869440]  <TASK>
 [43040.869442]  asm_sysvec_apic_timer_interrupt+0x1a/0x20
 [43040.869446] RIP: 0010:cpuidle_enter_state+0xbb/0x400
 [43040.869473]  cpuidle_enter+0x31/0x50
 [43040.869477]  do_idle+0x1d3/0x2f0
 [43040.869483]  cpu_startup_entry+0x29/0x30
 [43040.869487]  rest_init+0xf7/0x100
 [43040.869497]  start_kernel+0xa0e/0xa20
 [43040.869502]  x86_64_start_reservations+0x24/0x30
 [43040.869507]  x86_64_start_kernel+0xda/0xe0
 [43040.869511]  common_startup_64+0x13e/0x151
 [43040.869519]  </TASK>
 [43040.912025] Kernel panic - not syncing: Fatal exception in interrupt

What the registers say

The faulting instruction at wake_up_nohz_cpu+0xb is

  48 0f a3 05 3d ed 72 02    bt %rax,0x272ed3d(%rip)

i.e. a bit test on a static cpumask with the bit index in RAX. RAX and
RDI both hold 0xffffffff, so the index is -1 and the effective address is
the mask minus one 8-byte word: the faulting address 0xffffffffadef72d8
is exactly 8 bytes below the tested mask at 0xffffffffadef72e0. That
address is above the end of the image (link base 0xffffffff81000000,
boot offset 0xac00000) and unmapped - hence error_code 0 with PMD 0.

The mask test therefore behaves as written; the wrong value is the CPU
number, which arrives at wake_up_nohz_cpu() as -1. The call chain is the
clocksource watchdog timer (re-armed from its own callback) ->
__mod_timer() -> wake_up_nohz_cpu().

Where a -1 CPU can come from on that path is what I could not work out
from the code I have at hand, so I would appreciate a hint. A timer that
re-arms itself from its callback on an idle CPU does not look like an
exotic pattern, so this may be reachable more often than this single
report suggests.

Environment

 - 7.3.4-arch1-1 (Arch build), PREEMPT(full), SMP NOPTI,
   CONFIG_NO_HZ_FULL=y (tick_nohz_full_mask exists; no nohz_full= on the
   command line)
 - the kernel is tainted G OE: out-of-tree unsigned modules are loaded
   (NVIDIA 620.70.06 open modules, msi_psu). There was no WARN or BUG in
   that boot before the Oops, and no MCE/EDAC entries in it at all.
 - AMD Ryzen 9 5950X, ASUS ROG CROSSHAIR VIII DARK HERO, BIOS 4402
 - a desktop that suspends and resumes routinely, which is why the
   uptime counter is large (suspend time is not counted)
 - "clocksource: Watchdog remote CPU N read timed out" is logged once per
   boot on this machine, 1-70 s after start-up (N varies). I mention it in
   case it is related; it may just be the watchdog probing before all CPUs
   are up.

Caveat: the same machine also hits a separate mm/shmem problem on 7.3.4 -
"BUG: Bad page state ... refcount:0 mapcount:-1" from shmem_undo_range,
pages evidently released while still mapped - so memory corruption cannot
be ruled out completely. What makes me read this trace as a logic bug
rather than as corruption is that RAX/RDI hold a well-formed -1 and the
tested address is exactly mask - 8, which is what a bit index of -1
produces; a corrupted index would not land one word before a mask.

Happy to test patches, gather more data, or rerun with tracing enabled.

Thanks,
Eryepa Yerepilov

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2026-11-17 15:59 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-11-17 10:48 [BUG] timers/nohz: Oops in wake_up_nohz_cpu+0xb with cpumask index -1 Eryepa Yerepilov
2026-11-17 15:59 ` Frederic Weisbecker

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®