Messages in this thread | | | Date | Sat, 20 Dec 2008 21:10:33 +0100 (CET) | From | Thomas Gleixner <> | Subject | Re: [PATCH] posix-timers: CLOCK_MONOTONIC_RAW: fix the usage of ->it_clock |
| |
On Sat, 20 Dec 2008, Oleg Nesterov wrote:
> (compile tested) > > common_timer_create() and common_timer_set() blindly pass ->it_clock to > hrtimer_init() as clock_id. This used to work until CLOCK_MONOTONIC_RAW > was introduced, now we should be careful. > > Perhaps it makes sense to add BUG_ON(clock_id >= HRTIMER_MAX_CLOCK_BASES) > to __hrtimer_init(), the wrong clock_id leads to catastrophe. > > Reported-by: Eric Sesterhenn <snakebyte@gmx.de> > Signed-off-by: Oleg Nesterov <oleg@redhat.com> > > --- K-28/kernel/posix-timers.c~CLOCK_MONOTONIC_RAW 2008-12-02 17:12:40.000000000 +0100 > +++ K-28/kernel/posix-timers.c 2008-12-20 18:23:28.000000000 +0100 > @@ -191,12 +191,20 @@ static inline int common_clock_set(const > return do_sys_settimeofday(tp, NULL); > } > > -static int common_timer_create(struct k_itimer *new_timer) > +static inline int > +__common_timer_init(struct k_itimer *timer, enum hrtimer_mode mode) > { > - hrtimer_init(&new_timer->it.real.timer, new_timer->it_clock, 0); > + clockid_t clock_id = timer->it_clock ? > + CLOCK_MONOTONIC : CLOCK_REALTIME; > + hrtimer_init(&timer->it.real.timer, clock_id, mode); > return 0; > }
No, this is wrong. We do not want to create a timer for CLOCK_MONOTONIC_RAW.
tglx
| |