[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [PATCH 15/17] pci: use memory core for iommu support
From: |
Paolo Bonzini |
Subject: |
Re: [Qemu-devel] [PATCH 15/17] pci: use memory core for iommu support |
Date: |
Wed, 01 May 2013 18:07:23 +0200 |
User-agent: |
Mozilla/5.0 (X11; Linux x86_64; rv:17.0) Gecko/20130311 Thunderbird/17.0.4 |
Il 01/05/2013 07:06, David Gibson ha scritto:
>>> + + /* FIXME: inherit memory region from bus creator */ +
>>> memory_region_init_alias(mr, "iommu-nop", get_system_memory(),
>>> 0, INT64_MAX); + return mr;
> I don't see the reason for creating a new alias for each PCI
> device. Can't pci_dev->iommu just point directly to
> get_system_memory() in the normal case?
>
> In addition to creating additional objects, having these aliases
> makes it much les obvious how to tell if two PCI devices share an
> IOMMU address space. We have to be able to determine that for
> VFIO, since devices which share an address space in the host
> clearly can't be assigned to different address spaces in the
> guest.
Right, the alias that is needed to enable/disable bus-mastering is
created already below, so pci_dev->iommu can be shared (as it is in the
spapr case).
Paolo