On Tue, 09 Apr 2024 12:46:31 +0200 Thomas Gleixner <tglx@xxxxxxxxxxxxx> > On Mon, Apr 08 2024 at 19:46, syzbot wrote: > > ------------[ cut here ]------------ > > WARNING: CPU: 1 PID: 11239 at kernel/time/hrtimer.c:1053 hrtimer_forward+0x210/0x2d0 kernel/time/hrtimer.c:1053 > > RIP: 0010:hrtimer_forward+0x210/0x2d0 kernel/time/hrtimer.c:1053 > > Call Trace: > > <IRQ> > > hrtimer_forward_now include/linux/hrtimer.h:355 [inline] > > mac80211_hwsim_beacon+0x192/0x1f0 drivers/net/wireless/virtual/mac80211_hwsim.c:2335 > > __run_hrtimer kernel/time/hrtimer.c:1692 [inline] > > __hrtimer_run_queues+0x595/0xd00 kernel/time/hrtimer.c:1756 > > hrtimer_run_softirq+0x19a/0x2c0 kernel/time/hrtimer.c:1773 > > __do_softirq+0x2bc/0x943 kernel/softirq.c:554 > > invoke_softirq kernel/softirq.c:428 [inline] > > __irq_exit_rcu+0xf2/0x1c0 kernel/softirq.c:633 > > irq_exit_rcu+0x9/0x30 kernel/softirq.c:645 > > instr_sysvec_apic_timer_interrupt arch/x86/kernel/apic/apic.c:1043 [inline] > > sysvec_apic_timer_interrupt+0xa6/0xc0 arch/x86/kernel/apic/apic.c:1043 > > </IRQ> > > The code which arms the timer is not serialized against the callback so > the following can happen: > > CPU1 CPU2 > > __run_hrtimer(timer) > if (!hrtimer_queued(timer)) > hrtimer_start(timer); > > hrtimer_forward(timer) > WARN_ON(hrtimer_queued(timer) > > This really wants to use hrtimer_active() which takes the running > callback into account. > Then the race window is still open with hrtimer_active() and serialization outside hrtimer is needed. CPU1 CPU2 --- --- if (!hrtimer_is_active(timer)) if (!hrtimer_is_active(timer)) hrtimer_start(timer); __run_hrtimer(timer) hrtimer_start(timer); hrtimer_forward(timer) WARN_ON(hrtimer_queued(timer))
