Re: r8169 auto speed down issue

Re: r8169 auto speed down issue

[Francois Romieu]

hayeswang <hayeswang@realtek.com> :
[...]
> Do you have any suggestion about this?

Your description suggests that testing against the link partner ability
to work at 10M instead of testing for tp->link_ok could be good enough.

Does it make sense ?

-- 
Ueimor

RE: r8169 auto speed down issue

[hayeswang]

Francois Romieu [mailto:romieu@fr.zoreil.com] 
[...]
> Your description suggests that testing against the link 
> partner ability
> to work at 10M instead of testing for tp->link_ok could be 
> good enough.
> 
> Does it make sense ?
> 

Furthermore, should it not speed down without linking, even though the cable
would be plugged after suspending or shutdowning?

Best Regards,
Hayes

Re: r8169 auto speed down issue

[Francois Romieu]

hayeswang <hayeswang@realtek.com> :
> Francois Romieu [mailto:romieu@fr.zoreil.com] 
> [...]
> > Your description suggests that testing against the link 
> > partner ability to work at 10M instead of testing for
                                   ^^^^^^^^^^ -> "and"
> > tp->link_ok could be good enough.

As a policy we may assume that capabilities of the link partner won't
change after a link loss event - and there is currently no way we can
change this policy - but it won't always work as expected (especially
if "expected == automagically").

[...]
> Furthermore, should it not speed down without linking, even though the cable
> would be plugged after suspending or shutdowning?

I don't get your point. Can you reformulate ?

-- 
Ueimor

RE: r8169 auto speed down issue

[hayeswang]

Francois Romieu [mailto:romieu@fr.zoreil.com] 
> Sent: Friday, March 29, 2013 3:21 PM
> To: Hayeswang
> Cc: netdev@vger.kernel.org; linux-kernel@vger.kernel.org; 
> bowgotsai@google.com; Ryankao
> Subject: Re: r8169 auto speed down issue
> 
[...]
> 
> I don't get your point. Can you reformulate ?

Sorry for my unclear descriptor. I just think a case that the nic suspends or
shutdowns without cable plugging. Then, the cable is plugged again. If the nic
speed down to 10M and the link partner force 100M, the issue appears again. If
the nic doesn't speed down for normal link partner, it requires more power when
the linking recovers. Finally, I determine to set the speed to 10M when the link
partner supports 10M. And for the other case, setting the speed to 100M. This
avoids the giga nic to keep the speed to 1000M, and could fix this issue.
However, I wonder if there is a switch which forces the speed to giga.

Best Regards,
Hayes

Re: r8169 auto speed down issue

[Francois Romieu]

hayeswang <hayeswang@realtek.com> :
[...]
> Sorry for my unclear descriptor. I just think a case that the nic suspends or
> shutdowns without cable plugging. Then, the cable is plugged again. If the nic
> speed down to 10M and the link partner force 100M, the issue appears again. If
> the nic doesn't speed down for normal link partner, it requires more power
> when the linking recovers.

Yes. The nic can not guess what the user values most: power saving in a
stable link layer configuration vs ability to detect and adapt to a yet
unseen configuration.

It could be a reason why EEE exists.

> Finally, I determine to set the speed to 10M when the link partner supports
> 10M. And for the other case, setting the speed to 100M. This avoids the giga
> nic to keep the speed to 1000M, and could fix this issue.
> However, I wonder if there is a switch which forces the speed to giga.

We can check if the partner supports 100M too. It's almost free and it will
avoid this problem.

-- 
Ueimor