On Wed, 2011-11-02 at 13:26 +0800, Kai Huang wrote: > Hi, > > In case of direct io, without the interrupt remapping in IOMMU (intel > VT-d or AMD IOMMU), hypervisor needs to inject interrupt for guest > when the guest is scheduled to specific CPU. At the beginning I > thought with IOMMU's interrupt remapping, the hardware can directly > forward the interrupt to guest without trapping into hypervisor when > the interrupt happens, but after reading the Intel VT-d's manual, I > found the interrupt mapping feature just add another mapping which > allows software to control (mainly) the destination and vector, and we > still need hypervisor to inject the interrupt when the guest is > scheduled as only after the guest is scheduled, the target CPU can be > known. If my understanding is correct, seems the interrupt remapping > does not bring any performance improvement. So what's the benefit of > IOMMU's interrupt remapping? Can someone explain the usage model of > interrupt remapping in IOMMU? Interrupt remapping provides isolation and compatibility, not performance. The hypervisor being able to direct interrupts to a target CPU also allows it the ability to filter interrupts and prevent the device from signaling spurious or malicious interrupts. This is particularly important with message signaled interrupts since any device capable of DMA is able to inject random MSIs into the host. The compatibility side is a feature of Intel platforms supporting x2apic. The interrupt remapper provides a translation layer to allow xapic aware hardware, such as ioapics, to function when the processors are switched to x2apic mode. Thanks, Alex -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
