Re: [PATCH v3 04/41] virtio: memory access APIs

On Mon, Nov 24, 2014 at 01:03:24PM +0100, Geert Uytterhoeven wrote:

On Mon, Nov 24, 2014 at 12:52 PM, Michael S. Tsirkin [off-list ref] wrote:

quoted

virtio 1.0 makes all memory structures LE, so
we need APIs to conditionally do a byteswap on BE
architectures.

To make it easier to check code statically,
add virtio specific types for multi-byte integers
in memory.

Add low level wrappers that do a byteswap conditionally, these will be
useful e.g. for vhost.  Add high level wrappers that
query device endian-ness and act accordingly.

quoted

diff --git a/include/linux/virtio_byteorder.h b/include/linux/virtio_byteorder.h
new file mode 100644
index 0000000..824ed0b
--- /dev/null
+++ b/include/linux/virtio_byteorder.h

quoted

+static inline u16 __virtio16_to_cpu(bool little_endian, __virtio16 val)
+{
+       if (little_endian)
+               return le16_to_cpu((__force __le16)val);
+       else
+               return (__force u16)val;
+}

What's wrong with just using le16-to_cpu() ...

le16-to_cpu() is simply wrong: virtio needs to be
LE or native endian, depending on whether it's running
in 0.9 or 1.0 mode.

quoted

--- a/include/uapi/linux/virtio_ring.h
+++ b/include/uapi/linux/virtio_ring.h

quoted

 /* Virtio ring descriptors: 16 bytes.  These can chain together via "next". */
 struct vring_desc {
        /* Address (guest-physical). */
-       __u64 addr;
+       __virtio64 addr;

... and __le64?

There's already lots of precedence or this, even in include/uapi/.

Gr{oetje,eeting}s,

                        Geert

__le would make people think they can use le16-to_cpu() which is wrong.

-- 
MST