Hi,
Sorry for bothering you people but i can't stand this anymore!
Trying to net-install a linux distribution or even get the
kenel to hack it and change these damn timeouts is impossible
since kernel 2.2 or so.
Should we throw these old network cards away and buy new ones?
Or we should probably install Windows to be able to download the
linux kernel and then to hack it?
I had to borrow another network card to upgrade my linux
after trying for 1 week to download it with xircom one.
Why can't you make the timeouts as parameters?
Now I need to restart my network card from 2 and 2 minutes to
be able to download more from the damn kernel.
It isn't a problem only for xirc2ps_cs, it is a problem for
ne drivers too and probably more others.
If you don't want your card to reset after
flooding your kernel log with tx timeouts you have to change the
timeouts by hand. Of course that is possible if you are EVER able to
download the hole kernel with the shitty driver you already have.
Well, that's about all.
Sorry for this email, i just reached the limit of my patience.
Now back to that "pushing the patience to the limits" game called
"downloading the kernel".
Regards,
Adi/
On Sat, Oct 26, 2002 at 05:07:01AM +0200, Adrian Pop wrote:
>
> Hi,
>
> Sorry for bothering you people but i can't stand this anymore!
<...much pointless wanking snipped...>
If you want the problem solved, try supplying us with some useful
information rather than just p*ssing and moaning about it. What kernel
are you running? What's your architecture? Motherboard chipset? What
specific NIC(s) are you using? How do we reproduce the problem? What,
for $DEITY's sake, is the exact error message you're getting?
Show a little effort.
--Adam, who has no patience left by the end of the week
On Sat, 26 Oct 2002, Adam Kropelin wrote:
> Show a little effort.
Here it comes some little effort :)
Doesn't matter the kernel version,
if is bigger than 2.2 (better to say: the version
when the timeouts handlers were introduced in
the network drivers)
Doesn't matter the computer even.
Error:
NETDEV WATCHDOG: eth0: transmit timed out
eth0: transmit timed out
And after this the card resets, and it
takes quite a while for that.
My workaround for NE2000 cards: in 8390.h replaced
#define TX_TIMEOUT (20*HZ/100)
with
#define TX_TIMEOUT (100*HZ/100)
I used the NE2000 cards in 2 computers:
Computer 1: K6 4xx Mhz (i don't remeber more because i don't have it
anymore)
Computer 2: Cyrix 486DX2 66Mhz
Network cards: NE2000 ISA
Both gived the errors under medium load with
the original timeout: TX_TIMEOUT (20*HZ/100).
They reported no error and worked perfectly after
changing TX_TIMEOUT to (100*HZ/100)
Now the story repeats itself with Dell Latitude
laptop and with the pcmcia credit card Xircom
that uses xirc2ps_cs driver. If i download
with more than 20Kbps for 2 minutes the network
goes down and i have to "pcmcia restart" to make it
work again.
In xirc2ps_cs.c the TX_TIMEOUT is (400*HZ/1000).
I just finally downloaded the kernel and i'll see
what improvements i have if i make the timeout
bigger.
The error happens with older
NE2K-PCI cards too, and
it really doesn't matter
the computer (Dual Pentium - 233Mhz or Athlon - 1Ghz).
The most annoying thing is when you try to
install linux from the net. It just takes
forever because of the timeouts/card resets.
Or it just stalles.
I don't know what it should be done concering these
problems. To give TX_TIMEOUT as a boot parameter
would be ok, or at least in the bootnet.img and pcmciadd.img
to make the timeout bigger for these drivers.
Regards,
/Adi, who still has some patience left.
Hi !
On Sat, Oct 26, 2002 at 07:58:36AM +0200, Adrian Pop wrote:
> NETDEV WATCHDOG: eth0: transmit timed out
> eth0: transmit timed out
> And after this the card resets, and it
> takes quite a while for that.
>
> My workaround for NE2000 cards: in 8390.h replaced
> #define TX_TIMEOUT (20*HZ/100)
> with
> #define TX_TIMEOUT (100*HZ/100)
you understand that this means that your card often needds more than 0.2s to
send a frame ? Are all your cards connected to a defective hub, or a network
with lots of collisions ? Is the yellow led on your hub constantly lit ?
I have several ISA cards here (3c509, hp, ...) on a 386sx and a 486DX2-66 and
I have used NE2K for years on these machines under 2.2 and 2.4 kernels
without even one problem. The 486 is a firewall which generates lots of
collisions on the hub when uploading files to the local web server, but I
never had the problem you mention on these machines.
In fact, the only PC on which I see it is my laptop. When I send lots of
traffic to a 3c575 (cardbus), I get lots of these. This is not because of
the card (it works well in other notebooks, and other cards do the same),
but the notebook itself. When it's getting hot, I think it looses interrupts!
And please, give us an example of kernel version you use. >2.2 is not an
answer. How do you want people to read the code if they don't know *exactly*
which one certainly induces the problem ? We don't even know if it's a vanilla
kernel or a patched one.
Cheers,
Willy
> Both gived the errors under medium load with
> the original timeout: TX_TIMEOUT (20*HZ/100).
> They reported no error and worked perfectly after
> changing TX_TIMEOUT to (100*HZ/100)
Your ethernet is on fire.
Either that, or your machine is somehow not quite healthy.
Considering your "it really doesn't matter the computer",
I'd vote for the former, though.
T.