Re: [dpdk-dev] [PATCH] sched: add 64-bit counter retrieval API

[Thomas Monjalon <thomas@monjalon.net>]

Is this old patch still relevant?


25/07/2018 11:02, Alan Dewar:
> Hi Stephen,
> 
> Sorry about the delay in responding to your comment.
> 
> Maybe I'm missing something but I don't consider 64-bit counter
> rollover to be a problem.
> The DPDK QoS stats are just 32-bits wide and are zeroed when read.  So
> if we read them frequently enough the 32-bit counters won't wrap.
> 
> This new API is just a convenient way to read the 32-bit counters into
> variables that are 64-bit wide which allows Vyatta to have a common
> API slightly above the DPDK.
> 
> I'd hope that at some point in time the DPDK will move to 64-bit
> counters, but there may be technical reasons why this can't happen
> that I'm not aware of (support for 32-bit CPUs??)
> 
> Regards
> Alan
> 
> On Mon, Jul 23, 2018 at 5:52 PM, Stephen Hemminger
> <stephen@networkplumber.org> wrote:
> > On Wed, 18 Jul 2018 15:51:39 +0100
> > alangordondewar@gmail.com wrote:
> >
> >> From: Alan Dewar <alan.dewar@att.com>
> >>
> >> Add new APIs to retrieve counters in 64-bit wide fields.
> >>
> >> Signed-off-by: Alan Dewar <alan.dewar@att.com>
> >
> > Do you want to consider 64 bit counter roll over on 32 bit platform?
> > The problem is that incrementing an 64 bit value is not atomic on
> > 32 bit cpu. The carry operation can race with reading.
> >
> > The kernel has special case code to do restartable sequence for
> > accessing 64 bit counter on 32 bit CPU. These functions become
> > nop's on 64bit.
>