On 10.01.2012, at 23:52, Alexander Graf wrote: > > On 10.01.2012, at 23:49, Benjamin Herrenschmidt wrote: > >> On Tue, 2012-01-10 at 23:41 +0100, Alexander Graf wrote: >>> >>> No. Libhw shouldn't be able to know anything about target endianness. >>> If a device is as brokenly spec'ed as virtio and is coupled to the >>> "main CPU endianness", it clearly belongs with the CPU, not into >>> libhw. >> >> Ok, can you guys solve this and tell me what I should do ? :-) > > Your patch is fine. I also wrote up another patch that fixes the memcpy() in > virtio.c so you patch actually works on x86 too. I sent it to you, so please > include it in your next submission. > > I would heavily veto against keeping virtio-pci in libhw though if it depends > on the target endianness. PCI devices don't know about CPUs. Period.
The only thing I could imagine doable would be basically 2 different variants of virtio-pci that happen to expose the same PCI ID, but are initialized differently by a qdev property. We could then have the alias to normal virtio-pci device names in target specific code, but the actual device code in libhw. Alex
