On Wed, Sep 23, 2009 at 02:09:34PM -0700, Mark E Mason wrote: > Did we ever figure out why the kernel was attempting to set the affinity to more than one CPU? The hardware certainly supports doing that (but we're not at the moment). The hardware semantics is a bit ususual. On Sibyte the hardware will route an interrupt to all CPUs set in the affinity mask. On most other systems that I'm aware of it will route the interrupt to only one of CPUs selected by irq_chip->set_affinity(). We don't want such an stampede so the Sibyte interrupt code clears all but the lowest set bit from the set_affinity() argument. Either way, attempting to set the mask to an arbitrary non-empty set is entirely legal so the warning itself was a bug and the volume potencially made them a real problem for a few users. Ralf
