Message-ID: <20070630231734.GF2553@xi.wantstofly.org>
Date:	Sun, 1 Jul 2007 01:17:34 +0200
From:	Lennert Buytenhek <buytenh@...tstofly.org>
To:	Michael Buesch <mb@...sch.de>
Cc:	Török Edvin <edwintorok@...il.com>,
	zambrano@...adcom.com, netdev@...r.kernel.org,
	linux-kernel@...r.kernel.org, power@...host.org
Subject: Re: [PATCH] b44: power down PHY when interface down

On Sun, Jul 01, 2007 at 12:24:40AM +0200, Michael Buesch wrote:

> > > Hm, I was going to measure the real power advantage with a
> > > PCI-extender card. But my B44B0 card doesn't seem to work in
> > > that extender card. It works perfectly fine sticked directly into
> > > the motherboard, though, and other cards like a BCM4318 work in
> > > the extender, too.
> > > Not sure what this is.
> > > The extender has an application note about nonworking cards in the
> > > extender and a too big resistor on the board IDSEL pin being the
> > > cause of this.
> > 
> > Does the card show up in lspci at all?
> 
> No it doesn't.

Right, so it sounds like it might be this issue.


> > Does the extender board have a PCI-PCI bridge on it?  (If not,
> > there's not really any reason to resistively couple the IDSEL
> > line to the host, since the host should take care of that.)
> 
> There's no bridge. It just decouples all voltage lines, so you can
> drive it from external supply and/or measure voltages and current.
> On the PCB it looks like the the IDSEL line is rather directly
> routed to the host IDSEL. It just goes through one of the bus
> isolation chips. So I guess (just my guess) that this chip has some
> resistance and if the total resistance of the chip + the IDSEL
> resistor on the mainboard goes above some threshold it doesn't work
> anymore for some cards. In the application note they write
> about trouble for IDSEL resistors >51ohms.

More or less.  You can't add the resistances like that, since the
bus isolation chip buffers the IDSEL signal, but it is correct that
if the host's IDSEL resistor is larger than a certain value, the
combination of the resistive coupling of IDSEL plus the extra buffer
in the isolator might be causing the IDSEL input on the 'guest' PCI
board to assert too late (or not assert at all), causing config
accesses to fail.

(This also depends on the specific 'guest' PCI board used, as you
noted, due to differing IDSEL trace lengths/capacitances and input
pin capacitances on different PCI boards.  Also, it might work at
33 MHz but not work at 66 MHz, etc.)

If you feel adventurous, you could try to hack around this by
figuring out which AD[31:16] line this PCI slot's IDSEL line is
resistively coupled to (depends on the slot), and then adding
another parallel resistor on the board itself to make the bus
isolator's input buffer charge faster.  Note that this does
increase the load on that specific AD[] line, which might cause
other funny effects.


> > > Maybe I can try with another machine tomorrow.
> > 
> > That would only make a difference if there is no PCI-PCI bridge on
> > the extender board.
> 
> Well, they suggest it in the application note as a possible fix. ;)

The bus isolation chip doesn't count as a PCI-PCI bridge. :)  I'm just
saying that you wouldn't see the issue you are seeing now if the
extender board had a real PCI-PCI bridge on it, since in that case the
type 0 config access to the guest PCI board would be generated by the
bridge instead of by the host.
-
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