virtio: fix PCI accesses for ppc64 in legacy mode
authorDavid Marchand <david.marchand@6wind.com>
Tue, 17 May 2016 09:59:51 +0000 (11:59 +0200)
committerThomas Monjalon <thomas.monjalon@6wind.com>
Wed, 15 Jun 2016 17:06:53 +0000 (19:06 +0200)
Although ppc supports both endianesses, qemu supposes that the cpu is
big endian and enforces this for the virtio-net stuff.

Fix PCI accesses in legacy mode. Only ppc64le is supported at the moment.

Signed-off-by: David Marchand <david.marchand@6wind.com>
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
drivers/net/virtio/virtio_pci.c

index 845141b..9c9a3dd 100644 (file)
  */
 #define VIRTIO_PCI_CONFIG(hw) (((hw)->use_msix) ? 24 : 20)
 
+/*
+ * Since we are in legacy mode:
+ * http://ozlabs.org/~rusty/virtio-spec/virtio-0.9.5.pdf
+ *
+ * "Note that this is possible because while the virtio header is PCI (i.e.
+ * little) endian, the device-specific region is encoded in the native endian of
+ * the guest (where such distinction is applicable)."
+ *
+ * For powerpc which supports both, qemu supposes that cpu is big endian and
+ * enforces this for the virtio-net stuff.
+ */
+
 static void
 legacy_read_dev_config(struct virtio_hw *hw, size_t offset,
                       void *dst, int length)
 {
+#ifdef RTE_ARCH_PPC_64
+       int size;
+
+       while (length > 0) {
+               if (length >= 4) {
+                       size = 4;
+                       rte_eal_pci_ioport_read(&hw->io, dst, size,
+                               VIRTIO_PCI_CONFIG(hw) + offset);
+                       *(uint32_t *)dst = rte_be_to_cpu_32(*(uint32_t *)dst);
+               } else if (length >= 2) {
+                       size = 2;
+                       rte_eal_pci_ioport_read(&hw->io, dst, size,
+                               VIRTIO_PCI_CONFIG(hw) + offset);
+                       *(uint16_t *)dst = rte_be_to_cpu_16(*(uint16_t *)dst);
+               } else {
+                       size = 1;
+                       rte_eal_pci_ioport_read(&hw->io, dst, size,
+                               VIRTIO_PCI_CONFIG(hw) + offset);
+               }
+
+               dst = (char *)dst + size;
+               offset += size;
+               length -= size;
+       }
+#else
        rte_eal_pci_ioport_read(&hw->io, dst, length,
                                VIRTIO_PCI_CONFIG(hw) + offset);
+#endif
 }
 
 static void
 legacy_write_dev_config(struct virtio_hw *hw, size_t offset,
                        const void *src, int length)
 {
+#ifdef RTE_ARCH_PPC_64
+       union {
+               uint32_t u32;
+               uint16_t u16;
+       } tmp;
+       int size;
+
+       while (length > 0) {
+               if (length >= 4) {
+                       size = 4;
+                       tmp.u32 = rte_cpu_to_be_32(*(const uint32_t *)src);
+                       rte_eal_pci_ioport_write(&hw->io, &tmp.u32, size,
+                               VIRTIO_PCI_CONFIG(hw) + offset);
+               } else if (length >= 2) {
+                       size = 2;
+                       tmp.u16 = rte_cpu_to_be_16(*(const uint16_t *)src);
+                       rte_eal_pci_ioport_write(&hw->io, &tmp.u16, size,
+                               VIRTIO_PCI_CONFIG(hw) + offset);
+               } else {
+                       size = 1;
+                       rte_eal_pci_ioport_write(&hw->io, src, size,
+                               VIRTIO_PCI_CONFIG(hw) + offset);
+               }
+
+               src = (const char *)src + size;
+               offset += size;
+               length -= size;
+       }
+#else
        rte_eal_pci_ioport_write(&hw->io, src, length,
                                 VIRTIO_PCI_CONFIG(hw) + offset);
+#endif
 }
 
 static uint64_t