Re: [Qemu-devel] [PATCH RFC v3 13/14] intel_iommu: allow dynamic switch of IOMMU region

[Peter Xu]

On Wed, Jan 18, 2017 at 03:49:44PM +0800, Peter Xu wrote:

[...]

> I was trying to invalidate the entire address space by sending a big
> IOTLB notification to vfio-pci, which looks like:
> 
>   IOMMUTLBEntry entry = {
>       .target_as = &address_space_memory,
>       .iova = 0,
>       .translated_addr = 0,
>       .addr_mask = (1 << 63) - 1,
>       .perm = IOMMU_NONE,     /* UNMAP */
>   };
> 
> Then I feed this entry to vfio-pci IOMMU notifier.
> 
> However, this was blocked in vfio_iommu_map_notify(), with error:
> 
>   qemu-system-x86_64: iommu has granularity incompatible with target AS
> 
> Since we have:
> 
>   /*
>    * The IOMMU TLB entry we have just covers translation through
>    * this IOMMU to its immediate target.  We need to translate
>    * it the rest of the way through to memory.
>    */
>   rcu_read_lock();
>   mr = address_space_translate(&address_space_memory,
>                                iotlb->translated_addr,
>                                &xlat, &len, iotlb->perm & IOMMU_WO);
>   if (!memory_region_is_ram(mr)) {
>       error_report("iommu map to non memory area %"HWADDR_PRIx"",
>                    xlat);
>       goto out;
>   }
>   /*
>    * Translation truncates length to the IOMMU page size,
>    * check that it did not truncate too much.
>    */
>   if (len & iotlb->addr_mask) {
>       error_report("iommu has granularity incompatible with target AS");
>       goto out;
>   }
> 
> In my case len == 0xa0000 (that's the translation result), and
> iotlb->addr_mask == (1<<63)-1. So looks like the translation above
> splitted the big region and a simple big UNMAP doesn't work for me.
> 
> Do you have any suggestion on how I can solve this? In what case will
> we need the above address_space_translate()?

Hmm... it should be checking that the translated address range is RAM.

However if with this, IOMMU notifiers won't be able to leverage the
vfio driver feature to unmap a very big region.

IMHO the check should only be meaningful for map operations. I'll try
to post a RFC patch for vfio-pci to allow unmap of very big regions,
to see whether that'll be a workable approach.

Thanks,

-- peterx