Al Chu wrote: >> Deployed successfully for a number of months until we realized it >> couldn't quite do for us what we wanted (generate netdumps via SMI >> interrupts). >> > > Hmm. For the record, the issue was an interrupt issue in the kernel. > Not anything having to do with the BMC's ability to generate the > interrupt. > > Al SMIs and NMIs are particularly nasty to deal with in the kernel. I don't know much about SMIs. The IPMI spec doesn't say anything about them or how you identify the BMC watchdog as the source of the SMI. I'm sure there's another spec I would have to read and digest for that. NMIs have no way to reasonably identify the BMC as the source of the NMI. I say "reasonably" because you could send a message to the BMC and await a response. However, that's just not something you can feasibly do in an NMI; if something else generated the NMI (like an ECC error) then you have just spent 30-40ms in an NMI. Not a very good thing to do.
The Linux IPMI driver is able to use NMIs in certain circumstances. Basically, it puts itself last in line and says "if no one else handled this NMI, it must be mine". It tries to detect if there is another NMI source that cannot be identified (nmi_watchdog=1) and won't enable itself in that situation. Not ideal, but it seems to work ok in practice as other sources can be identified. -Corey ------------------------------------------------------------------------- Using Tomcat but need to do more? Need to support web services, security? Get stuff done quickly with pre-integrated technology to make your job easier Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642 _______________________________________________ Ipmitool-devel mailing list Ipmitool-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/ipmitool-devel
