Fealnx watchdog timeouts issue

From: SuD (Alex)
Date: Thu Aug 19 2004 - 20:09:53 EST

I changed my pc (but kept the card) and watchdog timeouts are less frequent, but they eventually happen along with packet corruption.
With my earlier pc and kernel i could produce it just by doing a "ping -f" in the lan, but now it happens more randomly. The symptoms are that 2 bytes of data are inserted before the rest of data in the frame (i captured it with ethereal) or maybe that the pointer is wrong by 2 bytes.

While doing a flood ping i got normal ping replies:
0000 00 40 d0 21 96 b2 00 02 44 1f 5f 59 08 00 45 00 .@.!.... D._Y..E.
0010 00 54 e3 ef 00 00 40 01 4b d7 d9 d9 09 84 d9 d8 .T....@. K.......
0020 8d ac 00 00 21 12 8e 09 a3 01 40 88 65 ca 00 07 ....!... ..@.e...
0030 1c 86 08 09 0a 0b 0c 0d 0e 0f 10 11 12 13 14 15 ........ ........
0040 16 17 18 19 1a 1b 1c 1d 1e 1f 20 21 22 23 24 25 ........ .. !"#$%
0050 26 27 28 29 2a 2b 2c 2d 2e 2f 30 31 32 33 34 35 &'()*+,- ./012345

and also a wrong one, note there is extra "00 00" at the begining but there are 2 bytes missing at the end:
0000 00 00 00 40 d0 21 96 b2 00 02 44 1f 5f 59 08 00 ...@.!.. ..D._Y..
0010 45 00 00 54 e3 f0 00 00 40 01 4b d6 d9 d9 09 84 E..T.... @.K.....
0020 d9 d8 8d ac 00 00 f8 9f 8e 09 a4 01 40 88 65 ca ........ ....@.e.
0030 00 07 43 f8 08 09 0a 0b 0c 0d 0e 0f 10 11 12 13 ..C..... ........
0040 14 15 16 17 18 19 1a 1b 1c 1d 1e 1f 20 21 22 23 ........ .... !"#
0050 24 25 26 27 28 29 2a 2b 2c 2d 2e 2f 30 31 32 33 $%&'()*+ ,-./0123

Today i got this error:

eth0: Promiscuous mode enabled.
device eth0 left promiscuous mode
NETDEV WATCHDOG: eth0: transmit timed out
eth0: Transmit timed out, status 00000000, resetting...
Rx ring cf6b4000: 80000000 80000000 80000000 80000000 80000000 80000000 80000000 80000000 80000000 80000000 80000000 80000000
Tx ring cf6b5000: 80000000 80000000 0000 80000000 80000000 80000000
Badness in local_bh_enable at kernel/softirq.c:136
[<c011c319>] local_bh_enable+0x89/0x90
[<c0334f64>] ip_ct_find_proto+0x34/0x50
[<c0335350>] destroy_conntrack+0x10/0x100
[<c02db108>] __kfree_skb+0x98/0x120
[<d088735c>] reset_tx_descriptors+0x5c/0xb0 [fealnx]
[<d0887018>] tx_timeout+0xc8/0x130 [fealnx]
[<c02e9db0>] dev_watchdog+0x0/0xc0
[<c02e9e5e>] dev_watchdog+0xae/0xc0
[<c011fe4a>] run_timer_softirq+0xca/0x190
[<c011ffe7>] do_timer+0xc7/0xd0
[<c011c259>] __do_softirq+0x79/0x80
[<c011c287>] do_softirq+0x27/0x30
[<c0105b39>] do_IRQ+0xf9/0x130
[<c0104028>] common_interrupt+0x18/0x20

PS: nice my patch for mac address got in the official tree but, do you think it'd be better if it the driver implemented it's own "set_mac_address" function so only then it would update the card's registers and try to mess the less possible?
--- Begin Message --- I recently was given a surecom 10/100 ethernet card and found that i could not change Mac address for it as other driver/devices allow.
I tried to implement the missing feature and noticed that the device is quite peculiar (or either my system is broken), when trying to fill mac address registers (no matter whether io_ops is set):
- If I write a byte (writeb) to an even i/o address it seems like actually a word was written (the next byte is set to 0).
- If I write a byte to an odd i/o address my pc gets bad freezed.

That made think of writing 16bit words (writew) for the memory address, as opposed to what most driver examples do. It works for me (I hope i set the lines at the right place in device_open function after the device is reset). I hope the code is clean enough and does not mess with byte ordering. This is the patch (this time against 2.6.5):


--- linux-2.6.5.orig/drivers/net/fealnx.c 2004-04-09 05:04:20.000000000 +0200
+++ linux/drivers/net/fealnx.c 2004-04-23 01:47:38.000000000 +0200
@@ -882,12 +882,17 @@
{
struct netdev_private *np = dev->priv;
long ioaddr = dev->base_addr;
-
+ int i;
+
writel(0x00000001, ioaddr + BCR); /* Reset */

if (request_irq(dev->irq, &intr_handler, SA_SHIRQ, dev->name, dev))
return -EAGAIN;

+ for (i = 0; i < 3; i++) {
+ writew(((unsigned short*)dev->dev_addr)[i], ioaddr + PAR0 + i*2);
+ }
+
init_ring(dev);

writel(np->rx_ring_dma, ioaddr + RXLBA);


----
By the way, i get some Watchdog Timeouts while transmitting every 10 minutes or so (also with 2.4 and original drivers).
This is what i get:
NETDEV WATCHDOG: eth0: transmit timed out
eth0: Transmit timed out, status 00000000, resetting...
Rx ring cf674000: 80000000 80000000 80000000 80000000 80000000 80000000 80000000 80000000 80000000 80000000 80000000 80000000
Tx ring cf675000: 80000000 80000000 80000000 80000000 80000000 0000

I have also recently experienced problems with my cablemodem Motorola SB5100e. It has registered this mac address:
D4:62:00:02:44:1F
which does not exist but is a mixture of my surecom card (fealnx) 00:02:44:1F:5F:59 and the network gateway 00:05:00:E2:D4:62.
I am not sure who's fault this is (i would not trust my cablemodem much anyway).

Thank you for reading.


--- End Message ---