> Is it possible that the ip address assigned to the > system's hostname was bound to the bge0 interface; > that is, after disabling the bge driver, connections > to the local machine using the machine' `hostname` > failed ? I suspect that the X screen blank needs > such > a connection to wake up from screen blank mode...
No, bge0 came in later. That was when it was running wpi0, with bge0 disabled in the kernel. > That is, when there is no cable connected to the > bge NIC hardware, the machine starts to consume > lots of kernel cpu time after a few minutes, and > eventually hangs the system? Correct. > And when a cable is connected, there is no > excessive kernel cpu time usage, and the machine > doesn't hang? Aside of those many mwaiti86 (or so), correct. At least, the machine doesn't hang (as before, let's leave out the non-return after some 2 hours of my absence), and has a beautiful load less 0.10. > Yep; I'd say something is broken in the bge > driver... And in the kernel architecture, I'd add. I don't consider it proper for the kernel to shoot itself by trying to wake up a little NIC at the periphery.? > Maybe the BIOS has configured the nic hardware to > enter > a power saving state after five minutes with no > activity; > and the Solaris bge driver is confused when the > device > enters that power saving state (it tries to recover > by > reseting the bge hardware, but fails to wake up the > hardware, > and tries to wait forever for the firmware to become > ready) ? As I mentioned before, the BIOS is most ugly in this machine, it doesn't allow much, many settings seem to be inaccessible by the user. As for your theory above, we have two phases to consider: 1. For some reason, the kernel tries hard to wake up bge0 after some 5 minutes, consuming all of CPU0 2. After a few minutes more, the system is completely dead. What else does it do, what does it try to achieve a few minutes later that kills it completely? > I think a possible workaround is to disable > svc:/network/physical:nwam, enable > svc:/network/physical:default, and manually > configure the wpi0 interface (and not use the > bge interface for now). My workaround is much more elegant: Boot to Ubuntu, OpenBSD or XP. They all work pretty well on this machine. :( Nevertheless, if there is anything more required from my side to help debugging this situation, let me know! Uwe -- This message posted from opensolaris.org
