On Thu, 01 Feb 2007 18:59:55 +0100 Lionel Landwerlin wrote: > Hi, > we already had words on lkml about this bug with sky2 driver. > > I was having problems, and you told me to use the disable_msi=1 > parameter to see what happens. After a couple of hours of testing with > heavly ethernet load, I answered you it had fixed the problem. I was > wrong. Now, it takes much more time to crash. Most of time, I can't even > see what happens beacause the box is completly frozen. But after several > crashs, I only had my keyboard locked, usb unpowered, and ethernet > interface down, I finally had the possibility see that : > > Feb 1 18:35:06 cocoduo kernel: [59723.468000] NETDEV WATCHDOG: eth0: transmit time Transmit recovery logic is screwed up, I have a better version but working on testing it. See attachment 1 > > It's exactly the same error than before. What do you think of this > trace ? Is it related to sky2 driver or acpi ? Did you add debug output > since 2.6.19.2 (version of the kernel I'm using) that would help to fix > that bug ? What can I do to help to fix the bug ? > The problem is that some part of the transmit path (probably the DMA engine) is getting hung. There is a test patch to dump tx ring (cat /proc/net/sky2/eth0). See attachment 2 -- Stephen Hemminger