Re: [PATCH v2 net-next] r8169: Add values missing in @get_stats64 from HW counters

[Corinna Vinschen <vinschen@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 1820 bytes --]

On Aug 22 13:23, Francois Romieu wrote:
> Corinna Vinschen <vinschen@redhat.com> :
> [...]
> > That won't happen with the current patch because only
> > rtl8169_reset_counters would print a log message, it's only called from
> > open, and open occurs rather seldom.  Atop of that the code only tries
> > to reset counters on HW supporting it, and only if resetting on the HW
> > fails, there will be a log message at all.  There's no reasonable chance
> > that failing to reset the counters will lead to log flooding.
> 
> Thanks for reformulating it. We are in violent agreement here.
> 
> [...]
> > I'm not trying to avoid work, I'm trying to understand.
> > 
> > As far as I see it failing to reset the counters has no impact on the
> > viability of the code.  It's still working with offsets and if the
> > offset is 0 or non-0, the user space won't see the difference in the
> > values returned by @get_stats64.  Successful resetting the counters is
> > just a bonus.
> 
> Sorry, my english was really bad:
> 
> the code should propagate failure when rtl8169_reset_counters and
> rtl8169_update_counters *simultaneously* fail.

Uhm... sorry, but that still doesn't answer the question.  As you can
see in my patch, the initalization at open time is already encapsulated
in a function rtl8169_init_counter_offsets. 

Assuming rtl8169_init_counter_offsets returns -1 if both functions,
rtl8169_reset_counters and rtl8169_update_counters fail.

Then... what?

Not being able to reset or update the counters is still not at all fatal
for the operation of the NIC as a whole and rtl_open in particular:

  rtl_open()
  {
    [...]

    /* This is non-fatal. */
    if (!rtl8169_init_counter_offsets ()) {

	  /* What to do here??? */

    }

    [...]
  }


Corinna

[-- Attachment #2: Type: application/pgp-signature, Size: 819 bytes --]