[vortex] 3c905C: Packet losses (a more precise description)

[Bogdan Costescu]

On Wed, 5 Jul 2000, Giuseppe Ciaccio wrote:

>   8K byte-wide RAM 5:3 Rx:Tx split, autoselect/Autonegotiate interface.
>   MII transceiver found at address 24, status 782d.

Everything seems to be normal.

> (NOTE: no "INVALID CHECKSUM" msg)

The printing of this message was supressed in recent drivers, as many
users complained about it....

>  Your link partner advertised 45e1: Flow-control 100baseTx-FD 100baseTx 10baseT-
> FD 10baseT, w/ 802.3X flow control.
> 
> (NOTE: the switch advertises flow-ctl although disabled!  A bug in mii-diag?
> A bug in the switch firmware?)

mii-diag reads the data from the card, so that's what the card knows about
the link partner. I'm sure that the driver/mii-diag handles this right
because my switch doesn't have flow control and such it's not reported.

> Basic registers of MII PHY #0:  0000 0000 0000 0000 0000 0000 0000 0000.
>  Basic mode control register 0x0000: Auto-negotiation disabled, with
>  Speed fixed at 10 mbps, half-duplex.
>  Basic mode status register 0x0000 ... 0000.
>    Link status: not established.
>  Link partner information information is not exchanged
> when in fixed speed mode.

Now this is strange. Why mii-diag cannot find the right registers when
autonegotiation is disabled ? Donald, do you have any theory ?

> Autoneg does not occur (it is switched off at the NICs).
> mii-diag reports "10 mbps"; however, the LEDs in the back are saying
> "100 mbps".  Who is right?

For some reason, mii-diag cannot read correctly the MII registers; the
interpretation (10 mbps, link not established and so on) is correct.

> Crossover cable, and repeater hub:  mii-diag as above, no packet losses.

Which one above: with all registers read as 0 (non-autoneg. case) or with
useful data (autoneg. case)?
If it's all registers 0, then we have identified a case when mii-diag
aparently fails to read the correct data: autonegotiation disabled. Can
you tell us exactly what mode have you forced ?

> Apparently, it is the switch' fault.  However, if I use other NICs, like the
> DEC DE500, the Intel EEPro 100 (one with 82557 and the other with 82558B),
> or even the old 3COM 3c905, in place of the 3c905C, no packet losses are
> observed through the very same switch.
> 
> A problem with flow ctl between my switch and the 3c905C adapters?

Might be! But are the other NICs supporting/using/reporting flow control ?
If they don't support and use it, the switch will not use it and you're on
the safe side (if flow control is the real cause). As 3C905C always
advertise flow control, if the switch also advertise it (not _really_
disabled) then it will be used.

> Is there any way for disabling flow ctl in the 3c905C?

No, as I wrote in my previous message, the docs say that the corresponding
bit is always set.

Sincerely,

Bogdan Costescu

IWR - Interdisziplinaeres Zentrum fuer Wissenschaftliches Rechnen
Universitaet Heidelberg, INF 368, D-69120 Heidelberg, GERMANY
Telephone: +49 6221 54 8869, Telefax: +49 6221 54 8868
E-mail: Bogdan.Costescu@IWR.Uni-Heidelberg.De