RFC: Use of "jiffies" vs "jiffies64" in the neighbour system.

RFC: Use of "jiffies" vs "jiffies64" in the neighbour system.

[Ulf Samuelsson]

If you run HZ = 1000, then jiffies will wrap around after 49,71 days.
This means that all time compares in the neighbour system will fail.

 From what I can see from "jiffies,h" there is no attempt to detect the 
wrap-around.

#define time_after(a,b)        \
     (typecheck(unsigned long, a) && \
      typecheck(unsigned long, b) && \
      ((long)(b) - (long)(a) < 0))
#define time_before(a,b)    time_after(b,a)

#define time_after_eq(a,b)    \
     (typecheck(unsigned long, a) && \
      typecheck(unsigned long, b) && \
      ((long)(a) - (long)(b) >= 0))
#define time_before_eq(a,b)    time_after_eq(b,a)

While the problem is not big for most users, I suspect it will
affect quality at the time of wrap-around.

It looks like garbage collection will stop in neigh_alloc,
since time has not passed time of last flush.

reachable time will no longer be recomputed in periodic_work

When you are in REACHABLE state, you will stay there,
since "now" will be before "confirmed" +"reachable_time",
so the entry will not be moved to DELAY or STALE,

If you are in delay state, you will move the entry
into REACHABLE even if you did not get a response.

and so on....

-- 
Best Regards,
Ulf Samuelsson

Re: RFC: Use of "jiffies" vs "jiffies64" in the neighbour system.

[Michal Kubecek]

On Wed, Mar 25, 2015 at 11:11:45AM +0100, Ulf Samuelsson wrote:
> If you run HZ = 1000, then jiffies will wrap around after 49,71 days.
> This means that all time compares in the neighbour system will fail.
> 
> From what I can see from "jiffies,h" there is no attempt to detect
> the wrap-around.
> 
> #define time_after(a,b)        \
>     (typecheck(unsigned long, a) && \
>      typecheck(unsigned long, b) && \
>      ((long)(b) - (long)(a) < 0))
> #define time_before(a,b)    time_after(b,a)
> 
> #define time_after_eq(a,b)    \
>     (typecheck(unsigned long, a) && \
>      typecheck(unsigned long, b) && \
>      ((long)(a) - (long)(b) >= 0))
> #define time_before_eq(a,b)    time_after_eq(b,a)

I might have misunderstood you but those macros do handle the
wrap-around. For example, with 64-bit long and a = 1,
b = 0xffffffffffffffff, you get

  ((long)(b) - (long)(a)) = (-1) - (1) = -2 < 0

so that time_after(1, 0xffffffffffffffff) is true. It will give correct
results as long as the difference is less than half of range of the
type.

                                                         Michal Kubecek

Re: RFC: Use of "jiffies" vs "jiffies64" in the neighbour system.

[Ulf Samuelsson]

On 03/25/2015 11:27 AM, Michal Kubecek wrote:
> On Wed, Mar 25, 2015 at 11:11:45AM +0100, Ulf Samuelsson wrote:
>> If you run HZ = 1000, then jiffies will wrap around after 49,71 days.
>> This means that all time compares in the neighbour system will fail.
>>
>>  From what I can see from "jiffies,h" there is no attempt to detect
>> the wrap-around.
>>
>> #define time_after(a,b)        \
>>      (typecheck(unsigned long, a) && \
>>       typecheck(unsigned long, b) && \
>>       ((long)(b) - (long)(a) < 0))
>> #define time_before(a,b)    time_after(b,a)
>>
>> #define time_after_eq(a,b)    \
>>      (typecheck(unsigned long, a) && \
>>       typecheck(unsigned long, b) && \
>>       ((long)(a) - (long)(b) >= 0))
>> #define time_before_eq(a,b)    time_after_eq(b,a)
> I might have misunderstood you but those macros do handle the
> wrap-around. For example, with 64-bit long and a = 1,
> b = 0xffffffffffffffff, you get
>
>    ((long)(b) - (long)(a)) = (-1) - (1) = -2 < 0
>
> so that time_after(1, 0xffffffffffffffff) is true. It will give correct
> results as long as the difference is less than half of range of the
> type.
>
>                                                           Michal Kubecek
>

OK, Now I see why it works.

Best Regards,
Ulf Samuelsson