From: Jonathan F. <jon...@gm...> - 2009-01-22 19:09:18
|
On Wed, Jan 21, 2009 at 6:10 PM, Brandeburg, Jesse <jes...@in...> wrote: > Jonathan Fournier wrote: >>> so what is the actual connection type on your ethernet, is it a >>> directly connected fiber port? Is it a serdes blade? forgive me >>> I've never tested one of the ATCA platforms. >> >> I just found out more documentation for that platform, since most ATCA >> switch are different, you can't always rely on the standard ;) >> >> I included 2 diagrams for reference, 5.3 is the base interface >> switching and 5.4 is the fabric interface switching. >> >> My ATCA board has 6 ports, 2 base interfaces, 2 fabric interfaces, 2 >> rtm interfaces. Only the fabric was reported as fiber, because the >> base switch contains a TXVRs & Magnetics to 10/100/1000 Base T. >> >> I think I now know why it's reported as "fiber" on the fabric, because >> of the 1000 Base BX. But it's not a "real" fibre, since it's a >> backplane... > > If its connected over SERDES (BX) it should be using the serdes device IDs > and then the code paths taken would probably be more sane, we have several > other designs with SERDES connected 82546 parts that work okay with our > drivers. After re digesting all the info I collected regarding 1000Base-BX and SerDes I re read: Designing SERDES-SERDES Interfaces with the 82546GB http://download.intel.com/design/network/applnots/ap466.pdf And I guess Sections 2.1.2, 2.1.4.x, and 3.2 explain it well. > >> I've been told by an ATCA backplane expert that the base interface is >> four-pairs like UTP, the fabric interface is eight pairs, and you >> chose how many pairs to use to create various options (1 , 2, 4, 10 >> GigE). I've also been told that the fabric is just supposed to be the >> MAC (SERDES) interface and doesn`t use the PHY. Would it explain a >> potential confusion on the media type? >> >> I read in this AN from Intel: >> Designing SERDES Applications-- >> 82545/82546, 82571/82572 & >> 631xESB/632xESB >> Application Note--AN498 >> >> That the Device ID should be forced for SERDES operation (in my case >> 82546GB 107Bh I assume?) > > I think that is better than 1079 (copper)! > >> Do you see potential problem not using SERDES? What about the patch >> below, I'm not sure I fully understand what it's doing with the M88 >> registers... > > The M88 stuff might be messing with the internal phy to make sure it is > off. Or more likely it may be trying to drive MDIO information out > to the SERDES fabric, but I'm guessing. Is there a place I could find what the M88 registers mean? I can't find Marvell Datasheets :( > > If your ATCA board had the correct eeprom image it should be disabling > the internal phy and also have the correct device ID for serdes > > Have you tried our 8.0.6 or newer driver from sourceforge just to > see if it works? I just tried 8.0.9 and ethtool now reports ports as Twisted Pair instead of Fiber ;) Of course link won't come up Sep 19 08:14:43 localhost kernel: ADDRCONF(NETDEV_UP): eth3: link is not ready Sep 19 08:16:54 localhost kernel: ADDRCONF(NETDEV_UP): eth2: link is not ready I'll retry with bare 2.6.27.11 (7.3.20-k3-NAPI), I think the link was at least coming up. But I now have a strong feeling that the eeprom wasn't shipped correctly, that this patch was hacking around to make it work. But now with latest driver it could be just worst since the media type isn't what the hack is expecting! Cheers, /jonathan > > We've also been doing lots of recent work fixing some bugs in the serdes > link detection code, might make a difference for you if you have some link > issues with our current drivers. |