On Wed, Mar 30, 2022 at 11:04:24AM +0100, Stefan Hajnoczi wrote: > This makes me wonder whether there is a deeper issue with the > pci_setup_iommu() API: the lack of per-device cleanup callbacks. > Per-device IOMMU resources should be freed when a device is hot > unplugged. > > From what I can tell this is not the case today: > > - hw/i386/intel_iommu.c:vtd_find_add_as() allocates and adds device > address spaces but I can't find where they are removed and freed. > VTDAddressSpace instances pointed to from vtd_bus->dev_as[] are leaked. > > - hw/i386/amd_iommu.c has similar leaks.
AFAICT it's because there's no device-specific data cached in the per-device IOMMU address space, at least so far. IOW, all the data structures allocated here can be re-used when a new device is plugged in after the old device unplugged. It's definitely not ideal since after unplug (and before a new device plugged in) the resource is not needed at all so it's kind of wasted, but it should work functionally. If to achieve that, some iommu_unplug() or iommu_cleanup() hook sounds reasonable. One thing I'm not sure is these iommu ops are per-bus not per-device. So I'm not sure whether that's what we wanted here because remote device cleanup seems to be per-device only. Thanks, -- Peter Xu
