On Sat, Oct 08, 2016 at 02:14:09PM +0800, Peter Xu wrote: > On Fri, Oct 07, 2016 at 06:24:15PM +0200, Radim Krčmář wrote: > > [...] > > > KVM accepts the address in host endianess and QEMU/VTD code also uses > > host endianess for internal representation of memory addresses, so this > > hunk should be fine. > > > > It is confusing, because the VTD is definitely broken with respect to > > endianess -- it is even trying to swap the order of bits in a byte in > > the definition of VTD_MSIMessage. > > I don't believe that dma_memory_write() accepted LE address on BE hosts, > > so the existing code for filling the address is wrong: > > > > msg.__addr_head = cpu_to_le32(0xfee); > > Yeah. This is my fault. Sorry for the troubles. > > I have a patch (as well...) to fix this in my local tree, but not > posted (as mst suggested). Maybe it's time to post some of them now (I > tried to make patches more into a bunch so that they won't be lost in > mailing list in case maintainer missed it).
I'll send them after your series to avoid unecessary conflicts. -- peterx
