On Wed, Aug 1, 2012 at 11:27 AM, Benjamin Herrenschmidt <b...@kernel.crashing.org> wrote: > On Wed, 2012-08-01 at 11:16 +0100, Stefan Hajnoczi wrote: > >> The len field is mostly informational. The virtio device driver >> inside the guest may make use of it. In many cases it doesn't so an >> incorrect len value has no effect. In >> drivers/block/virtio_blk.c:blk_done() the len variable is unused. > > Right but arguably the guest reading more than the len passed back into > the descriptor is itself a bug :-) > >> QEMU should call cpu_physical_memory_unmap() with the correct size >> value in hw/virtio.c:virtqueue_fill() so that the memory dirty bitmap >> is kept up-to-date. This is the only bad side-effect I can see here. > > With the current guest driver ... another one adhering strictly to the > spec might get bitten :-) Anyway, it's minor, but probably somebody > should fix. I don't have time right now, but if you don't beat me to it > I might try to give it a spin tomorrow.
Added to my TODO list but I probably won't get a chance before tomorrow either because I need to focus on QEMU 1.2 soft freeze. Stefan
