On 09/01/2011 09:25 PM, Anthony Liguori wrote: > I think this is the wrong approach given the introduction of the memory API. > > I think we should have a generic memory access function that takes a > MemoryRegion as it's first argument. > > The PCI bus should then expose one memory region for each device (that's how > it can figure out where the access is coming from).
A MemoryRegion is the wrong abstraction for the device's view back into system memory. The new memory API really has nothing to do with DMA or an IOMMU. Think about it: 99.999% of the time we're writing to the one MemoryRegion that is RAM (with the other remaining fraction being pci-pci dma tricks), except in a non-contiguous manner. Obviously we need something else to represent the non-contiguous-ness. I suppose you could abuse Avi's AddressSpace, with hundreds of MemoryRegion aliases, but that would be a horrible interface to actually use in this situation. I think David's patch is fine, as far as it goes for the majority of the PCI-only devices that we emulate. As for the devices attached to more than one bus, they'll just have to wait until we actually hash out the underlying dma api. r~
