Date:   Sun, 14 Mar 2021 04:26:57 -0400
From:   "Michael S. Tsirkin" <mst@...hat.com>
To:     Laurent Vivier <laurent@...ier.eu>
Cc:     linux-kernel@...r.kernel.org,
        virtualization@...ts.linux-foundation.org,
        Jason Wang <jasowang@...hat.com>
Subject: Re: [PATCH] virtio-mmio: read[wl]()/write[wl] are already
 little-endian

On Sat, Mar 13, 2021 at 06:10:29PM +0100, Laurent Vivier wrote:
> 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) {

Sounds reasonable ...


> And we need the same thing in virtio_pci_config_read() (and write).

We already have it, see below.

> 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

virtio pci does this: modern accesses use:

	virtio_pci_device_read

static uint64_t virtio_pci_device_read(void *opaque, hwaddr addr,
                                       unsigned size)
{
    VirtIOPCIProxy *proxy = opaque;
    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
    uint64_t val = 0;

    if (vdev == NULL) {
        return val;
    }

    switch (size) {
    case 1:
        val = virtio_config_modern_readb(vdev, addr);
        break;
    case 2:
        val = virtio_config_modern_readw(vdev, addr);
        break;
    case 4:
        val = virtio_config_modern_readl(vdev, addr);
        break;
    }
    return val;
}


while legacy accesses use:

	virtio_pci_config_read

static uint64_t virtio_pci_config_read(void *opaque, hwaddr addr,
                                       unsigned size)
{
    VirtIOPCIProxy *proxy = opaque;
    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
    uint32_t config = VIRTIO_PCI_CONFIG_SIZE(&proxy->pci_dev);
    uint64_t val = 0;
    if (addr < config) {
        return virtio_ioport_read(proxy, addr);
    }
    addr -= config;

    switch (size) {
    case 1:
        val = virtio_config_readb(vdev, addr);
        break;
    case 2:
        val = virtio_config_readw(vdev, addr);
        if (virtio_is_big_endian(vdev)) {
            val = bswap16(val);
        }
        break;
    case 4:
        val = virtio_config_readl(vdev, addr);
        if (virtio_is_big_endian(vdev)) {
            val = bswap32(val);
        }
        break;
    }
    return val;
}


the naming is configing but there you are.

virtio_pci_config_read is also calling virtio_is_big_endian.

        
static inline bool virtio_is_big_endian(VirtIODevice *vdev)
{       
    if (!virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
        assert(vdev->device_endian != VIRTIO_DEVICE_ENDIAN_UNKNOWN);
        return vdev->device_endian == VIRTIO_DEVICE_ENDIAN_BIG;
    }   
    /* Devices conforming to VIRTIO 1.0 or later are always LE. */
    return false;
}       
            

I note that
virtio_is_big_endian is kind of wrong for modern config accesses since it
checks guest feature bits - config accesses are done before features are
acknowledged.
Luckily this function is never called for a modern guest ...
It would be nice to add virtio_legacy_is_big_endian and call that
when we know it's a legacy interface.


-- 
MST

Powered by blists - more mailing lists