Re: [Qemu-devel] [PATCH] exec: don't return int64_t in address_space_cache_init()

[Jason Wang]

On 2017年03月30日 05:26, Michael S. Tsirkin wrote:
> On Wed, Mar 29, 2017 at 02:12:50PM +0800, Jason Wang wrote:
> > We return int64_t as the length of region cache but accept hwaddr as
> > the required length. This is wrong and may confuse the caller since
> > the it can lead comparison between signed and unsigned types. The
> > caller can not catch the failure in this case. Fixing this by
> > returning hwaddr and return zero on failure.
> >
> > Fixes: 5eba0404b9829 ("virtio: use MemoryRegionCache to access descriptors")
> > Fixes: e45da65322386 ("virtio: validate address space cache during init")
> > Cc: Cornelia Huck<address@hidden>
> > Cc: Paolo Bonzini<address@hidden>
> > Signed-off-by: Jason Wang<address@hidden>
> Can you be more specific about the symptoms this fixes in the
> commit log?
> E.g. "This actually triggers on XYZ when using ABC".

I want do this, but in fact this was triggered by a bug of qemu (see the 
thread of iommu reset vs region cache).

In that case, when used map fails, then check

     if (len < size) {
        virtio_error(vdev, "Cannot map used");
        goto err_used;
    }

can not catch the -EFAULT, since len is converted to unsigned.

Thanks