Le 11/03/2021 à 16:44, Michael S. Tsirkin a écrit :
> On Tue, Mar 09, 2021 at 11:43:13PM +0100, Laurent Vivier wrote:
>> read[wl]()/write[wl] already access memory in little-endian mode.
>
> But then they convert it to CPU right? We just convert it back ...
In fact the problem is in QEMU.
On a big-endian guest, the readw() returns a byte-swapped value, This means
QEMU doesn't provide a
little-endian value.
I found in QEMU virtio_mmio_read() provides a value with byte-swapped bytes.
The problem comes from virtio_config_readX() functions that read the value
using ldX_p accessors.
Is it normal not to use the modern variant here if we are not in legacy mode?
I think we should have something like this in virtio_mmio_read (and write):
--- a/hw/virtio/virtio-mmio.c
+++ b/hw/virtio/virtio-mmio.c
@@ -112,15 +112,28 @@ static uint64_t virtio_mmio_read(void *opaque, hwaddr
offset, unsigned size)
if (offset >= VIRTIO_MMIO_CONFIG) {
offset -= VIRTIO_MMIO_CONFIG;
- switch (size) {
- case 1:
- return virtio_config_readb(vdev, offset);
- case 2:
- return virtio_config_readw(vdev, offset);
- case 4:
- return virtio_config_readl(vdev, offset);
- default:
- abort();
+ if (proxy->legacy) {
+ switch (size) {
+ case 1:
+ return virtio_config_readb(vdev, offset);
+ case 2:
+ return virtio_config_readw(vdev, offset);
+ case 4:
+ return virtio_config_readl(vdev, offset);
+ default:
+ abort();
+ }
+ } else {
+ switch (size) {
+ case 1:
+ return virtio_config_modern_readb(vdev, offset);
+ case 2:
+ return virtio_config_modern_readw(vdev, offset);
+ case 4:
+ return virtio_config_modern_readl(vdev, offset);
+ default:
+ abort();
+ }
}
}
if (size != 4) {
And we need the same thing in virtio_pci_config_read() (and write).
And this could explain why it works with virtio-pci and not with virtio-mmio
with the big-endian guest:
with virtio-pci the bytes are swapped twice (once in virtio-mmio and then in
virtio-pci), so they
are restored to the initial value, whereas with direct virtio-mmio they are
swapped only once.
Thanks,
Laurent
