[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <4FEB7DC0.9090404@candelatech.com>
Date: Wed, 27 Jun 2012 14:40:16 -0700
From: Ben Greear <greearb@...delatech.com>
To: Eric Dumazet <eric.dumazet@...il.com>
CC: Stephen Hemminger <shemminger@...tta.com>,
Rick Jones <rick.jones2@...com>,
Tom Parkin <tparkin@...alix.com>, netdev@...r.kernel.org,
David.Laight@...LAB.COM, James Chapman <jchapman@...alix.com>
Subject: Re: [PATCH v2] l2tp: use per-cpu variables for u64_stats updates
On 06/27/2012 02:32 PM, Eric Dumazet wrote:
> On Wed, 2012-06-27 at 23:20 +0200, Eric Dumazet wrote:
>> On Wed, 2012-06-27 at 13:58 -0700, Ben Greear wrote:
>>
>>> It's worse than that: Even on 64-bit kernels, counters that are returned by
>>> netlink and /proc/net/dev as 64-bit may still wrap themselves at 32-bit
>>> intervals.
>>
>> Really ?
>>
>> Thats incredible you dont send a bug report then.
>
> A bug report to the application author, not the kernel.
Yeah, application author was me.
> /proc/net/dev is an ASCII file, and nothing gives the width of a field.
>
> Therefore, an application should cope with all cases (counter being a
> 32bit or 64bit integer), wrapping included.
>
> Note that this has little to do with the application or kernel being
> 32/64 bit code.
Notice that the netlink stats are claiming 64-bit and they are not
(always) 64-bit.
That is a nice binary API that is still wrapping before its time
in many cases. There may be good performance reasons for keeping
some counters at 32-bit, but it plays merry hell with anyone wanting
to optimize an application to poll less often for stats that are
supposedly 64-bit.
Thanks,
Ben
--
Ben Greear <greearb@...delatech.com>
Candela Technologies Inc http://www.candelatech.com
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists