On 05/27/2014 09:57 PM, Mark Rutland wrote:

On Tue, May 27, 2014 at 01:44:27PM +0100, Zhangfei Gao wrote:

quoted

Add support for the hix5hd2 XGMAC 1Gb ethernet device.
The controller requires two queues for tx and two queues for rx.
Controller fetch buffer from free queue and then push to used queue.
Diver should prepare free queue and free buffer from used queue.

Signed-off-by: Zhangfei Gao <zhangfei.gao@linaro.org>
---
  drivers/net/ethernet/Kconfig                  |    1 +
  drivers/net/ethernet/Makefile                 |    1 +
  drivers/net/ethernet/hisilicon/Kconfig        |   27 +
  drivers/net/ethernet/hisilicon/Makefile       |    5 +
  drivers/net/ethernet/hisilicon/hix5hd2_gmac.c | 1057 +++++++++++++++++++++++++
  5 files changed, 1091 insertions(+)
  create mode 100644 drivers/net/ethernet/hisilicon/Kconfig
  create mode 100644 drivers/net/ethernet/hisilicon/Makefile
  create mode 100644 drivers/net/ethernet/hisilicon/hix5hd2_gmac.c

[...]

quoted

+struct hix5hd2_desc {
+       __le32 buff_addr;
+       __le32 cmd;
+} __aligned(32);

[...]

quoted

+static void hix5hd2_rx_refill(struct hix5hd2_priv *priv)
+{
+       struct hix5hd2_desc *desc;
+       struct sk_buff *skb;
+       u32 start, end, num, pos, i;
+
+       /* software write pointer */
+       start = dma_cnt(readl_relaxed(priv->base + RX_FQ_WR_ADDR));
+       /* logic read pointer */
+       end = dma_cnt(readl_relaxed(priv->base + RX_FQ_RD_ADDR));
+       num = CIRC_SPACE(start, end, RX_DESC_NUM);
+
+       for (i = 0, pos = start; i < num; i++) {
+               if (priv->rx_skb[pos])
+                       break;
+               else {
+                       skb = netdev_alloc_skb_ip_align(priv->netdev,
+                               MAC_MAX_FRAME_SIZE);
+                       if (unlikely(skb == NULL))
+                               break;
+               }
+
+               desc = priv->rx_fq.desc + pos;
+               desc->buff_addr =
+                       dma_map_single(priv->dev, skb->data,
+                                       MAC_MAX_FRAME_SIZE, DMA_FROM_DEVICE);

buff_addr was declared as an __le32, but here we're pasting values of
the kernel's native endianness (which might not be little). You will
need to carefully convert the endianness of this value when dealing with
it.

quoted

+               if (dma_mapping_error(priv->dev, desc->buff_addr)) {
+                       dev_kfree_skb_any(skb);
+                       break;
+               }
+
+               priv->rx_skb[pos] = skb;
+               desc->cmd = (MAC_MAX_FRAME_SIZE - 1) | DESC_VLD_FREE;

Likewise cmd was described as an __le32 and needs to be converted
appropriately.

Got it, will use le32_to_cpu & cpu_to_le32 for the conversion.
Thanks for the info.