No ethernet interface found after suspend-resume

[Dan Williams]

On Fri, 2010-04-30 at 21:54 +0530, Sriram V wrote:
> On Fri, Apr 30, 2010 at 9:40 PM, Chris Ball <cjb at laptop.org> wrote:
> > Hi,
> >
> >   > How can i reset the card? I use a gpio - If i do a reset - The
> >   > firmware again gets downloaded. I need to again configure my
> >   > ethernet interface and get an IP address. which i dont think is
> >   > the right way to do.
> >
> > Yes, that's what has to happen if the card loses power.  (And the card
> > losing power is what has to happen on vanilla 2.6.29.)
> >
> >   > Can you elaborate on this? I cant upgrade to .33. I would like to
> >   > back port to .29.
> >
> > Let me just check what you're trying to do -- you want the card to
> > retain power while the CPU's in suspend, and you want it to stay
> > associated during suspend.  Do you also want packets sent to the card
> > during suspend to wake up the system?
> 
> I think the connection should stay associated after comming back from
> suspend-resume. Becaue that was the state before going to suspend.

Remember that saying connected to the AP (by keeping the 8686 powered
and associated) while the host CPU is suspended only works reliably with
WEP and unencrypted modes.  When you're using WPA/WPA2 or Dynamic WEP,
you need to run the WPA supplicant process (for rekeying and other
things) which of course requires the host CPU.  This is not specific to
the 8686; all cards work this way.  Rekeying is a core feature of WPA
that greatly increases security.

You can probably get away with a short host suspend (like the OLPC does)
but anything more than 10 or 20 seconds and you're likely to get dropped
by the AP.  You can get dropped even if you're suspended for a single
second because the WPA rekey is triggered by two things: (a) time, and
(b) packet count.  If during suspend the rekey packet count threshold is
reached and your host CPU running the supplicant isn't awake to handle
the rekey, you'll get dropped by the AP when you come out of suspend.

So just because the state of the device was associated before you
suspend, don't expect that state to be the same when you exit suspend.

> I dont want the system to be woken up during suspend - i can live without it.
> 
> 
>  Or is it just the extra time
> > that it takes to bring the interface back up after suspend that you're
> > trying to get rid of?
> 
> 
> Do all - have to bring the interface backup after
> suspend-resume. Is this taken care by the user-space. But how does the
> user space know that the interface is back up and query for an IP everytime
> the device comes back from suspend and associate with an AP.

How the reassociation is handled on resume depends on what kind of
userspace you're running.  Any system needs something to configure the
network, whether that's scripts or ifplugd or NetworkManager or
whatever.  That thing also needs to handle the resume case, or really it
needs to handle any case where the card is disconnected from the AP,
which includes being dropped by the AP while suspended.

> Is this normal behaviour elsewhere?

Yes.  Normal laptop/desktop cards these days are "softmac" which means
that the card's firmware is very small and most operations are done  on
the host CPU.  THus if the machine goes to sleep, there's nothing to
handle the normal 802.11 management traffic (which the 8686's more
capable firmware will do for you) and clearly you're going to get
dropped by the AP while suspended with these types of cards.

> Not really sure this the default behaviour.

Yes, reconfiguring the interface on resume is the default behavior
because devices are almost always unpowered across suspend/resume.
There are some prominent exceptions here though, mostly in the embedded
space and of course OLPC.

> Besides, My device is externally powered and i just specify the ocr's
> ranges to the mmc drivers.
> So probably the state machine within libertas does not expect CMD5,
> CMD8 and so on....

Since the kernel just recently got the handling code for
powered-during-suspend MMC stuff, that's fairly necessary so that the
MMC bus (on which the libertas driver is based) knows about this
condition and doesn't try to re-initialize a powered card on resume.

Dan