Re: [PATCH] timer tsc ensure we allow for initial tsc and tsc sync

[Andrew Morton <akpm@osdl.org>]

Andy Whitcroft <apw@shadowen.org> wrote:
>
> From: John Stultz <johnstul@us.ibm.com>
> 
> Suppress lost tick detection until we are fully initialised.
> This prevents any modifications to the high resolution timers
> from causing non-linearities in the flow of time.  For example on
> an SMP system we resyncronise the TSC values for all processors.
> This results in a TSC reset which will be seen as a huge apparent
> tick loss.  This can cause premature expiry of timers and in extreme
> cases can cause the soft lockup detection to fire.
> 
> Acked-by: Andy Whitcroft <apw@shadowen.org>
> 
> diff --git a/arch/i386/kernel/timers/timer_tsc.c b/arch/i386/kernel/timers/timer_tsc.c
> --- a/arch/i386/kernel/timers/timer_tsc.c
> +++ b/arch/i386/kernel/timers/timer_tsc.c
> @@ -45,6 +45,15 @@ static unsigned long last_tsc_high; /* m
>  static unsigned long long monotonic_base;
>  static seqlock_t monotonic_lock = SEQLOCK_UNLOCKED;
>  
> +/* Avoid compensating for lost ticks before TSCs are synched */
> +static int detect_lost_ticks;
> +static int __init start_lost_tick_compensation(void)
> +{
> +	detect_lost_ticks = 1;
> +	return 0;
> +}
> +late_initcall(start_lost_tick_compensation);
> +
>  /* convert from cycles(64bits) => nanoseconds (64bits)
>   *  basic equation:
>   *		ns = cycles / (freq / ns_per_sec)
> @@ -196,7 +205,8 @@ static void mark_offset_tsc_hpet(void)
>  
>  	/* lost tick compensation */
>  	offset = hpet_readl(HPET_T0_CMP) - hpet_tick;
> -	if (unlikely(((offset - hpet_last) > hpet_tick) && (hpet_last != 0))) {
> +	if (unlikely(((offset - hpet_last) > hpet_tick) && (hpet_last != 0))
> +					&& detect_lost_ticks) {

Simple enough.  John, so you feel that this is 2.6.16 material?

Note that the time changes in -mm will blow this change away, so I'd be
needing a fresh version of this patch against next-mm, please.