 
	
| [Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] RE: [Xen-users] PCI passtrough in HVM Domain ?
 PCI passthrough isn't supported for HVM guests yet. I don't 
know why - it shouldn't be too difficult to achieve, I would have thought, but 
on the other hand, for 90% of devices, it probably doesn't make much sense to 
just pass them on, since those 90% use direct memory accesses for the data 
transfers to/from the device, and thus you'd end up shipping out the wrong data 
- which at best means that the device just won't work, or at worst means that 
some data from another domain gets sent out to the public to see - the latter is 
worse becaue you wouldn't know it until you find that your bank-account has been 
emptied, because you sent all the details off onto the WWW and someone picked it 
up... ;-) The reason for this is that the HVM guest domain THINKS 
that memory starts at address zero (or thereabouts) and goes to whatever size 
we've assigned it. But in reality, the memory used by the HVM guest is living 
at, for example, 256-512MB. So when the network card is told to transmit memory 
from address 123456, it should really transmit what's at 256MB + 123456, not 
actual address 123456. Since the OS and it's drivers are completely unaware 
about this fact, they can't really fix it...  There are two solutions to this 
problem: 1. Use an IOMMU - unfortunately, there's none available for 
common x86 hardware today [IBM has one, which is used for large MP systems, I 
think - not sure if it's x86 or PowerPC based].  2. Use para-virtualized drivers, so the driver is capable 
of figuring out that the ACTUAL physical address of the data to be DMA'd is 
different from the OS's idea of the address.  -- Mats 
 _______________________________________________ Xen-users mailing list Xen-users@xxxxxxxxxxxxxxxxxxx http://lists.xensource.com/xen-users 
 
 | 
|  | Lists.xenproject.org is hosted with RackSpace, monitoring our |