Re: [Phishing Risk] Re: [Phishing Risk] [External] Re: [PATCH] blk-iocost: fix very large vtime when iocg activate

[Chengming Zhou]

On 2022/5/17 09:03, Tejun Heo wrote:
> On Tue, May 17, 2022 at 08:57:55AM +0800, Chengming Zhou wrote:
>> #define time_after64(a,b)	\
>> 	(typecheck(__u64, a) &&	\
>> 	 typecheck(__u64, b) && \
>> 	 ((__s64)((b) - (a)) < 0))
>> #define time_before64(a,b)	time_after64(b,a)
>>
>> I still don't get why my changes are wrong. :-)
> 
> It's a wrapping timestamp where a lower value doesn't necessarily mean
> earlier. The before/after relationship is defined only in relation to each
> other. Imagine a cirle representing the whole value range and picking two
> spots in the circle, if one is in the clockwise half from the other, the
> former is said to be earlier than the latter and vice-versa. vtime runs way
> faster than nanosecs and wraps regularly, so we can't use absolute values to
> compare before/after.

Yes, thanks for the explanation. But the problem is not comparing two timestamp,
since ioc->margins.target is not a timestamp. This patch just fix a corner case
when now->vnow < ioc->margins.target:

u64 vtarget;

vtarget = now->vnow - ioc->margins.target; --> vtarget should be a timestamp earlier than vnow.

But when now->vnow < ioc->margins.target, vtarget would be a timestamp after vnow.

Thanks.

> 
> Thanks.
>