Pascal CHAPPERON <pascal.chapperon@wanadoo.fr> :
[...]
> I don't receive Lars's results, but here are my results.

Lars can apparently not reproduce it any more. I'll consider this
issue fixed until it returns. Thanks for checking.

[ping test removed]
> # cat /proc/net/pktgen/eth0
> Params: count 10000000  min_pkt_size: 60  max_pkt_size: 60
>      frags: 0  delay: 0  clone_skb: 1000000  ifname: eth0
>      flows: 0 flowlen: 0
>      dst_min: 10.169.21.1  dst_max:
>      src_min:   src_max:
>      src_mac: 00:11:2F:E9:42:70  dst_mac: 00:40:F4:A8:70:BC
>      udp_src_min: 9  udp_src_max: 9  udp_dst_min: 9  udp_dst_max: 9
>      src_mac_count: 0  dst_mac_count: 0
>      Flags: IPDST_RND
> Current:
>      pkts-sofar: 8699475  errors: 0
>      started: 1120913419217947us  stopped: 1120913484497743us idle: 0us
>      seq_num: 8699485  cur_dst_mac_offset: 0  cur_src_mac_offset: 0
>      cur_saddr: 0x1515a90a  cur_daddr: 0x115a90a
>      cur_udp_dst: 9  cur_udp_src: 9
>      flows: 0
> Result: OK: 65279796(c65279796+d0) usec, 8699475 (60byte,0frags)
>   133264pps 63Mb/sec (63966720bps) errors: 0

Not too bad: it trashes data fast enough :o)

Can you do the same test but send the traffic from the host which
embeds the r8169 ?

The sis190 should not be responsive during the flow. I expect that it
will happily return to a normal state once the traffic stops.

If it behaves correctly, I'll send a first version for inclusion.

> "ethtool -s eth0 ..." does not freeze the station anymore, but
> "autoneg off" does not work properly :
> 
> # ethtool -s eth0 speed 10 duplex half autoneg off
> # ethtool eth0
> Settings for eth0:
>         Supported ports: [ TP MII ]
>         Supported link modes:   10baseT/Half 10baseT/Full
>                                 100baseT/Half 100baseT/Full
>         Supports auto-negotiation: Yes
>         Advertised link modes:  10baseT/Half 10baseT/Full
>                                 100baseT/Half 100baseT/Full
>         Advertised auto-negotiation: Yes
>         Speed: 10Mb/s
>         Duplex: Half
>         Port: MII
>         PHYAD: 0
>         Transceiver: internal
>         Auto-negotiation: on
>         Current message level: 0x00000037 (55)
>         Link detected: yes
> # ethtool eth0
> Settings for eth0:
>         Supported ports: [ TP MII ]
>         Supported link modes:   10baseT/Half 10baseT/Full
>                                 100baseT/Half 100baseT/Full
>         Supports auto-negotiation: Yes
>         Advertised link modes:  10baseT/Half 10baseT/Full
>                                 100baseT/Half 100baseT/Full
>         Advertised auto-negotiation: Yes
>         Speed: 100Mb/s
>         Duplex: Full
>         Port: MII
>         PHYAD: 0
>         Transceiver: internal
>         Auto-negotiation: on
>         Current message level: 0x00000037 (55)
>         Link detected: yes
> 
> The values written in StationControl are probaly wrong,

One should expect some messages to appear in the log after you forced
the link at 10Mb/s. Do you notice anything in dmesg ?

I would be interested to know if the attached patch makes a difference
(the patch applies on top of the current driver).

Can you issue a simple 'ethtool -s eth0 autoneg off' and report what
happens ?

[html]

Sorry, it was just a notice for Lars

--
Ueimor