Re: transmit timed out, tx_status 00 status e000
Donald Becker <[email protected]> Mon, 3 Jun 2002 18:45:55 -0400 (EDT)
| Newsgroups | gmane.linux.drivers.vortex.user |
|---|---|
| Message-ID | <Pine.LNX.4.33.0206031822110.1329-100000@presario> |
On Mon, 3 Jun 2002, Gerhard Hintermayer wrote: > A while ago I read some postings about this error, but found no > solution for that error. I had a 3c905B Cyclone running into that > *every* time I put heavy load onto the NIC, checked the source and > changed TX_TIMEOUT for 2*HZ to 6*HZ, and the error was gone. Is that a > the right way to go ? No. This has been reported various times. It's a hack that can mask certain problems (and it might be a good idea just for that reason), but it doesn't fix a problem. The reason for the 2 second delay is that a properly functioning Ethernet should allow a packet to be sent well less than a second. The driver "watchdog" doubles this and assumes that Something Bad has happened that might be fixed by a reset. The only only ill effects from increasing the watchdog delay to 6 seconds are: If there is a problem that a reset will fix, you are off the air for 4 seconds longer. You don't count some transient network problems as errors, leading to a false sense of reliability. > I also have a 3C905B-FX, running for more than 2 years without any > problem and then *bang* - the same error. Three of your friends throw up after eating chicken salad. Do you think "I should find more robust friends" or "we should check that refrigerator". I suspect that you have a network-wide problem. > I can't imagine, that this is not a known and not yet solved issue. I don't consider messages reporting errors to be a bug. It's only a bug if the driver fails to continue operation when it could safely do so. -- Donald Becker [email protected] Scyld Computing Corporation http://www.scyld.com 410 Severn Ave. Suite 210 Second Generation Beowulf Clusters Annapolis MD 21403 410-990-9993