On 2007/02/21 23:44, Andreas Bihlmaier wrote: > After uping wpi0 the machine completely freezes for a couple of seconds, > then accepts input every couple of seconds. There is no unusual > interrupt load coming from wpi. > The "funny" thing: If there is traffic on the ethernet (bce) the machine > becomes usable again.
I worked it out like this: (I think this is correct but am not certain). wpi(4) is actually generating interrupts on 11 (note this changed when you enabled ACPI), but the OS doesn't know about this (maybe <probably?> due to broken tables in the BIOS). Instead it thinks that when wpi needs attention it will generate an interrupt on 4. By causing activity on bce you generate interrupts on 4 causing the service routines for all the devices the OS _thinks_ are on 4 to be run, and making the wpi work. You only see high interrupt load when interrupts are broken in a certain way (I suspect it's to do with a mismatch between different types of triggering e.g. level/edge, but I haven't thought it through that far yet).
