Re: 2.6.35-rc1 regression with pvclock and smp guests

[Avi Kivity <avi@redhat.com>]

  On 07/28/2010 12:00 AM, Arjan Koers wrote:
> On 2010-07-26 20:59, Arjan Koers wrote:
>
>> I ran into the same problem. 2.6.34.1 and 2.6.35-rc6 SMP guest
>> kernels hang during boot.
>
> It appears that last is way ahead of ret twice.
> The kernel boots with this debug patch that makes the clock go
> backwards if the difference is big:
>
>   	last = atomic64_read(&last_value);
>   	do {
> -		if (ret<  last)
> -			return last;
> +		if (ret<  last) {
> +			if ( last - ret<  25000000 )
> +				return last;
> +			else
> +				printk("pvclock backwards: ret = %llx; last = %llx\n", ret, last);
> +		}
>   		last = atomic64_cmpxchg(&last_value, last, ret);
>   	} while (unlikely(last != ret));
>
>
>
> [    0.037122] Total of 2 processors activated (11198.08 BogoMIPS).
> [    0.037118] x86 PAT enabled: cpu 1, old 0x0, new 0x7010600070106
> [    0.040000] pvclock backwards: ret = 108373705fe2; last = 210aff61470a

Zaaaacchhhh?!

-- 
I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.