Re: via-rhine v1.05: Oversized Ethernet frames when in promisc mode

From: Urban Widmark (urban@svenskatest.se)
Date: Fri Jun 09 2000 - 20:31:40 EST


On Fri, 9 Jun 2000, Konrad Podloucky wrote:

> OK, here are a few:
> eth0: Oversized Ethernet frame spanned multiple buffers, entry 0x6db67b
> length 0 status 0600!
> ^- ?
> eth0: Oversized Ethernet frame c2fcc8b0 vs c2fcc8b0.
> and then immediately after that:
> eth0: Oversized Ethernet frame spanned multiple buffers, entry 0x6db67c
> length 78 status 8d00!
> eth0: Oversized Ethernet frame c2fcc8c0 vs c2fcc8c0.
> As far as I can see every message with status 0600 is immediately followed
> by one with status 8d00. Those with status 0600 alway have length 0

Except for the zero length, that the driver doesn't attempt to handle it,
that the driver tried to set up the card not to use more than one
descriptor per frame, it doesn't look *that* wrong ... without knowing if
the 78 bytes are all it's supposed to be.

0x6db67b is followed by 0x6db67c, take the 0 bytes from 0x6db67b and add
the 78 from 0x6db67c and we have something that the card thinks is ok. I
wonder if it contains something useful.

> Here's a 0d09 between a 0600 and a 8d00:
> eth0: Oversized Ethernet frame spanned multiple buffers, entry 0x97bad9
> length 0 status 0600!
> eth0: Oversized Ethernet frame spanned multiple buffers, entry 0x97bada
> length 1282 status 0d09!

"buffer error" and "crc error". This 0600 was not followed by a "clean"
8d00 ... the errors most likely clears the RxOK flag.

> eth0: Oversized Ethernet frame spanned multiple buffers, entry 0x97bbad
> length 186 status 8d00!

This one is later (211 packets).

> And here's an example of a 9500:

Just a broadcast version of 8d00, I think. Yep, all that needs to be done
is solve the 8d00's, then the rest will follow ...

silly suggestions/questions

+ which compiler? gcc-2.95.183567-latest-cvs-thingy or some more common
  (egcs-1.1.2 myself), you could try compiling it with an older
  "safer" compiler (failing that, try a newer, more improved one :)

+ do you have the possibility to test 2.4.0-test1-ac12?
  (or whatever the latest may be). The 2.3/2.4 driver does the locking
  (and PCI setup magic) differently, probably doesn't matter but it might
  be worth trying.

/Urban

-
To unsubscribe from this list: send the line "unsubscribe linux-net" in
the body of a message to majordomo@vger.rutgers.edu



This archive was generated by hypermail 2b29 : Thu Jun 15 2000 - 21:00:40 EST