]> www.infradead.org Git - nvme.git/commitdiff
x86/i8253: Call clockevent_i8253_disable() with interrupts disabled
authorFernando Fernandez Mancera <ffmancera@riseup.net>
Tue, 1 Apr 2025 09:23:03 +0000 (11:23 +0200)
committerIngo Molnar <mingo@kernel.org>
Fri, 11 Apr 2025 05:28:20 +0000 (07:28 +0200)
There's a lockdep false positive warning related to i8253_lock:

  WARNING: HARDIRQ-safe -> HARDIRQ-unsafe lock order detected
  ...
  systemd-sleep/3324 [HC0[0]:SC0[0]:HE0:SE1] is trying to acquire:
  ffffffffb2c23398 (i8253_lock){+.+.}-{2:2}, at: pcspkr_event+0x3f/0xe0 [pcspkr]

  ...
  ... which became HARDIRQ-irq-unsafe at:
  ...
    lock_acquire+0xd0/0x2f0
    _raw_spin_lock+0x30/0x40
    clockevent_i8253_disable+0x1c/0x60
    pit_timer_init+0x25/0x50
    hpet_time_init+0x46/0x50
    x86_late_time_init+0x1b/0x40
    start_kernel+0x962/0xa00
    x86_64_start_reservations+0x24/0x30
    x86_64_start_kernel+0xed/0xf0
    common_startup_64+0x13e/0x141
  ...

Lockdep complains due pit_timer_init() using the lock in an IRQ-unsafe
fashion, but it's a false positive, because there is no deadlock
possible at that point due to init ordering: at the point where
pit_timer_init() is called there is no other possible usage of
i8253_lock because the system is still in the very early boot stage
with no interrupts.

But in any case, pit_timer_init() should disable interrupts before
calling clockevent_i8253_disable() out of general principle, and to
keep lockdep working even in this scenario.

Use scoped_guard() for that, as suggested by Thomas Gleixner.

[ mingo: Cleaned up the changelog. ]

Suggested-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Fernando Fernandez Mancera <ffmancera@riseup.net>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/r/Z-uwd4Bnn7FcCShX@gmail.com
arch/x86/kernel/i8253.c

index 80e262bb627fe1d28fa670c9c596696f7f57039d..cb9852ad6098935990dc0f77791eb8101d0c0fff 100644 (file)
@@ -46,7 +46,8 @@ bool __init pit_timer_init(void)
                 * VMMs otherwise steal CPU time just to pointlessly waggle
                 * the (masked) IRQ.
                 */
-               clockevent_i8253_disable();
+               scoped_guard(irq)
+                       clockevent_i8253_disable();
                return false;
        }
        clockevent_i8253_init(true);