[9fans] Problems with igbe

[9fans] Problems with igbe

[Philipp]

Hi!

This time, when I'm writing you, I've got problems
with another machine then the machine where I still
have the same problem and i mailed about it a few days
ago.
It's now about my eth card and igbe, which seems failing
init the card or sth. like that.
At the end of this mail, pci -v output and my kmesg output (which prompts
the error to me) are attached.

Btw.: The card did well under FreeBSD and does well with Win XP,
so I don't think the EEPROM is really screwed up.

Best regards,
Philipp


---
term% pci -v
# some other devices here
2.1.0: net   02.00.00 8086/1019   5 0:fe9e0000 131072 2:0000bc01 32
    Intel Corporation 82547EI Gigabit Ethernet Controller (LOM)
---

---
term% cat /dev/kmesg
Plan 9
#some lines containing E820: + hexnumbers + cpu0 output, followed by
ELCR: 0C20
igbe: bad EEPROM checksum - 0xFFC0
igbe: bad EEPROM checksum - 0xFFC0 # yeah, the message is displayed twice and both are _identical_
# + some usb stuff
---

Re: [9fans] Problems with igbe

[erik quanstrom]

the datasheet for this part says that the eeprom checksum
for this part must be 0xbaba.  its calculated at the end of
/sys/src/9/pc/etherigbe.c:^/at93c46r.

the reason for two bad checksum messages is likely that you
have a two port card.  both ports share the same eeprom.
if this is true, you may need to change :1775 from
	if ((ctlr->id == i82546gb || ctlr->id == i82546eb) &&
to
	if ((ctlr->id == i82546gb || ctlr->id == i82546eb || ctlr->id == i82547ei) &&
(this isn't really the correct test, but i don't have the datasheet
for the 82547 handy.  close enough.)

it is possible that the eeprom does have an invalid signature
but will still work.  you can attempt this by commenting out
/sys/src/9/pc/etherigbe.c:1767.

- erik

Re: [9fans] Problems with igbe

[Philipp]

On 16:27 Sat 24 Nov     , erik quanstrom wrote:
> the datasheet for this part says that the eeprom checksum
> for this part must be 0xbaba.  its calculated at the end of
> /sys/src/9/pc/etherigbe.c:^/at93c46r.
> 
> the reason for two bad checksum messages is likely that you
> have a two port card.  both ports share the same eeprom.
> if this is true, you may need to change :1775 from
> 	if ((ctlr->id == i82546gb || ctlr->id == i82546eb) &&
> to
> 	if ((ctlr->id == i82546gb || ctlr->id == i82546eb || ctlr->id == i82547ei) &&
> (this isn't really the correct test, but i don't have the datasheet
> for the 82547 handy.  close enough.)
> 
> it is possible that the eeprom does have an invalid signature
> but will still work.  you can attempt this by commenting out
> /sys/src/9/pc/etherigbe.c:1767.
> 
> - erik

Hi!
Commenting out /sys/src/9/pc/etherigbe.c:1767 worked for me, at
least, for a while. It was kind of obvious, so maybe my mail was
kind of spam. But commenting it out is not all. #l is set up now,
but /net/ether0/addr gives me now sth. like ffffffff, so maybe
that's why i never get a connection to my dhcpd. I think that
has sth. to do with some function which sets up/reads the mac
address and isn't (correctly) implemented for my controller,
but I don't know much about nic drivers.

Best regards,
Philipp

Re: [9fans] Problems with igbe

[erik quanstrom]

> Hi!
> Commenting out /sys/src/9/pc/etherigbe.c:1767 worked for me, at
> least, for a while. It was kind of obvious, so maybe my mail was
> kind of spam. But commenting it out is not all. #l is set up now,
> but /net/ether0/addr gives me now sth. like ffffffff, so maybe
> that's why i never get a connection to my dhcpd. I think that
> has sth. to do with some function which sets up/reads the mac
> address and isn't (correctly) implemented for my controller,
> but I don't know much about nic drivers.
> 
> Best regards,
> Philipp

okay, the reason the eeprom checksum is wrong is that the data
coming from your card.  this is where the station address aka mac is stored.
and other important defaults. so either we're doing it wrong or it really is corrupt.

the first option is more attactive.  in at93c46r() there's a switch statement
on what type of ctlr you have.  try adding your controller type after
	case i82546eb:
perhaps igbe is using the wrong method.

- erik

Re: [9fans] Problems with igbe

[Philipp]

On 07:25 Sun 25 Nov     , erik quanstrom wrote:
> > Hi!
> > Commenting out /sys/src/9/pc/etherigbe.c:1767 worked for me, at
> > least, for a while. It was kind of obvious, so maybe my mail was
> > kind of spam. But commenting it out is not all. #l is set up now,
> > but /net/ether0/addr gives me now sth. like ffffffff, so maybe
> > that's why i never get a connection to my dhcpd. I think that
> > has sth. to do with some function which sets up/reads the mac
> > address and isn't (correctly) implemented for my controller,
> > but I don't know much about nic drivers.
> > 
> > Best regards,
> > Philipp
> 
> okay, the reason the eeprom checksum is wrong is that the data
> coming from your card.  this is where the station address aka mac is stored.
> and other important defaults. so either we're doing it wrong or it really is corrupt.
> 
> the first option is more attactive.  in at93c46r() there's a switch statement
> on what type of ctlr you have.  try adding your controller type after
> 	case i82546eb:
> perhaps igbe is using the wrong method.
> 
> - erik
Hi!
Yeah, exactly, that's it. And because igbe used the default case here (btw. we
had a discussion in irc about default cases at the beginning of the switch statement)
the some EEPROM stuff failed (there's also no bad EEPROM checksum message anymore),
resulting in a invalid mac address.
Thanks for this great remote diagnostics and keeping patient with me.

Best regards,
Philipp

Re: [9fans] Problems with igbe

[erik quanstrom]

> Hi!
> Yeah, exactly, that's it. And because igbe used the default case here (btw. we
> had a discussion in irc about default cases at the beginning of the switch statement)
> the some EEPROM stuff failed (there's also no bad EEPROM checksum message anymore),
> resulting in a invalid mac address.
> Thanks for this great remote diagnostics and keeping patient with me.
> 

you're welcome.

i didn't quite grok where you are now.  did changing the switch statement
fix your problem entirely?

also, i didn't get if you have a two port card or not.

(btw: a mac address of ffffffffffff is not the default.  this is the eeprom
returning read errors.  i would imagine all bits in the eeprom block are
set.)

- erik