Skip to contentSkip to conversation

sis190

DISCUSSION Pascal CHAPPERON 2 participants 2 messages Jul 6, 2005 – Jul 7, 2005

lore
Share & tools
lore
#0 lore raw
Quoted 2 lines

Message du 05/07/05 01:34
De : "Francois Romieu" <[email protected]>

[...]

Quoted 3 lines

Can you check if there is a regression in sis190-000.patch available at
http://www.zoreil.com/~romieu/sis190/20050704-2.6.13-rc1/patches ?

That one works perfectly; i tried it in the same conditions as the previous patch, and i don't notice a regression.

[...]

Quoted 3 lines

If it works and you want some entertainment, you can apply sis190-010.patch
+ sis190-020.patch and experience with ethtool/mii-tool.

[...]

sis190-010.patch does not compile properly :

# make clean;make;make install
rm -f *.o *.ko .*.cmd .*.flags *.mod.c
make -C /lib/modules/2.6.12.1/build SUBDIRS=/data/src/sis190  modules
make[1]: Entering directory `/usr/src/linux-2.6.12.1-custom'
  CC [M]  /data/src/sis190/sis190.o
/data/src/sis190/sis190.c: In function `sis190_init_one':
/data/src/sis190/sis190.c:1318: error: structure has no member named `poll_controller'
# diff -puN sis190-20050704-10.c sis190.c
--- sis190-20050704-10.c        2005-07-06 14:49:19.000000000 +0200
+++ sis190.c    2005-07-06 15:41:49.000000000 +0200

@@ -639,8 +640,6 @@ static void sis190_netpoll(struct net_de

        sis190_interrupt(pdev->irq, dev, NULL);
        enable_irq(pdev->irq);

}
-#else
-#define sis190_netpoll NULL
#endif

static void sis190_free_rx_skb(struct sis190_private *tp,
@@ -1314,7 +1313,9 @@ static int __devinit sis190_init_one(str

        dev->tx_timeout = sis190_tx_timeout;
        dev->watchdog_timeo = SIS190_TX_TIMEOUT;
        dev->hard_start_xmit = sis190_start_xmit;
+#ifdef CONFIG_NET_POLL_CONTROLLER
        dev->poll_controller = sis190_netpoll;
+#endif
        dev->set_multicast_list = sis190_set_rx_mode;
        SET_ETHTOOL_OPS(dev, &sis190_ethtool_ops);
        dev->irq = pdev->irq;

After that, it compiles and works as expected; i can use netconsole and get local messages on the other box.

There is something wrong whith sis190-020.patch;
it works, i can use the link as usual, i can use
ethtool :

# 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

But i get those traces in syslog, when the driver is loaded and every time i use ethtool :

[...]
scheduling while atomic: mii-tool/0x00000001/4699

Call Trace:<ffffffff8030358a>{schedule+122} <ffffffff801ce6de>{avc_has_perm_noaudit+1134}
       <ffffffff80191290>{update_atime+64} <ffffffff80139109>{__mod_timer+425}
       <ffffffff803047fe>{schedule_timeout+158} <ffffffff80139b70>{process_timeout+0}
       <ffffffff80139f68>{msleep+56} <ffffffff880370c8>{:sis190:__mdio_cmd+24}
       <ffffffff8803712f>{:sis190:mdio_read+15} <ffffffff88034868>{:mii:generic_mii_ioctl+152}
       <ffffffff880387dd>{:sis190:sis190_ioctl+77} <ffffffff802a30db>{dev_ifsioc+971}
       <ffffffff802a33e5>{dev_ioctl+741} <ffffffff802e47da>{inet_ioctl+138}
       <ffffffff80297fdc>{sock_ioctl+572} <ffffffff8018967a>{do_ioctl+58}
       <ffffffff8018998d>{vfs_ioctl+685} <ffffffff80189a2a>{sys_ioctl+106}
       <ffffffff8010d866>{system_call+126}

[...]
scheduling while atomic: ethtool/0x00000001/4703

Call Trace:<ffffffff8030358a>{schedule+122} <ffffffff80139109>{__mod_timer+425}
       <ffffffff803047fe>{schedule_timeout+158} <ffffffff80139b70>{process_timeout+0}
       <ffffffff80139f68>{msleep+56} <ffffffff880370c8>{:sis190:__mdio_cmd+24}
       <ffffffff8803712f>{:sis190:mdio_read+15} <ffffffff880340df>{:mii:mii_ethtool_gset+223}
       <ffffffff880385de>{:sis190:sis190_get_settings+46} <ffffffff802a43d2>{dev_ethtool+290}
       <ffffffff80159862>{buffered_rmqueue+690} <ffffffff80159a32>{__alloc_pages+194}
       <ffffffff801ce6de>{avc_has_perm_noaudit+1134} <ffffffff801650f3>{do_no_page+483}
       <ffffffff8015be4a>{__do_page_cache_readahead+362} <ffffffff8018f5eb>{__d_lookup+363}
       <ffffffff801ce6de>{avc_has_perm_noaudit+1134} <ffffffff801ce6de>{avc_has_perm_noaudit+1134}
       <ffffffff80191290>{update_atime+64} <ffffffff801ce76a>{avc_has_perm+90}
       <ffffffff80159862>{buffered_rmqueue+690} <ffffffff801cf57a>{inode_has_perm+106}
       <ffffffff801561fa>{filemap_nopage+394} <ffffffff802a33b5>{dev_ioctl+693}
       <ffffffff802e47da>{inet_ioctl+138} <ffffffff80297fdc>{sock_ioctl+572}
       <ffffffff8018967a>{do_ioctl+58} <ffffffff8018998d>{vfs_ioctl+685}
       <ffffffff80189a2a>{sys_ioctl+106} <ffffffff8010d866>{system_call+126}

[...]

I also tried to set sis190 mode (eg: 10 full autoneg off), but the driver froze the box.

As for now mii support seems unstable, i do not play further more with the media management.

Regards
Pascal

#1 lore raw

Pascal CHAPPERON <[email protected]> :
[...]

Quoted 2 lines

That one works perfectly; i tried it in the same conditions as the previous patch, and i don't notice a regression.

Nice.

[...]

Quoted 1 line

sis190-010.patch does not compile properly :

Point taken. Thanks.

[...]

Quoted 5 lines

But i get those traces in syslog, when the driver is loaded and every time i use ethtool :

[...]
scheduling while atomic: mii-tool/0x00000001/4699

spinlock + msleep in __mdio_cmd()... Completely stupid.

The patchkit of the day should fix these issues:
http://www.zoreil.com/~romieu/sis190/20050706-2.6.13-rc1/patches

Tarball:
http://www.zoreil.com/~romieu/sis190/20050706-2.6.13-rc1.tar.bz2

Please shake it. You should be able to modify the set of displayed messages with ethtool.

I'd appreciate if you could check the allowed frame size range, say ping -s 1468 ... ping -s 1473 to compare with Lars's results.

You can add something like a ping -q -l 48 -s 64 -f to your tests and increase the 48 and NUM_{RX/TX}_DESC but I am not sure that the remote 8139 will be able to go terribly far. If it performs well, pktgen could be useful too.

Lars, can you describe the second host that you have used so far for the tests:
- which nic ?
- which kernel ?

(on an unrelated note, something enabled the "send mail as html" checkbox in your mail user agent: you may consider removing it).

--
Ueimor