net: mediatek: fix off by one in the TX ring allocation
authorJohn Crispin <john@phrozen.org>
Fri, 10 Jun 2016 11:28:06 +0000 (13:28 +0200)
committerDavid S. Miller <davem@davemloft.net>
Sat, 11 Jun 2016 06:29:30 +0000 (23:29 -0700)
The TX ring setup has an off by one error causing it to not utilise all
descriptors. This has the side effect that we need to reset the next
pointer at runtime to make it work. Fix the off by one and remove the
code fixing the ring at runtime.

Signed-off-by: John Crispin <john@phrozen.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/mediatek/mtk_eth_soc.c

index 3b3ba2e55bdce0be1ed76a674b6aa747bac720c2..c097e9e3926c634872cbfd4191987176af5aec6a 100644 (file)
@@ -926,7 +926,6 @@ static int mtk_poll_tx(struct mtk_eth *eth, int budget, bool *tx_again)
                }
                mtk_tx_unmap(eth->dev, tx_buf);
 
-               ring->last_free->txd2 = next_cpu;
                ring->last_free = desc;
                atomic_inc(&ring->free_count);
 
@@ -1032,7 +1031,7 @@ static int mtk_tx_alloc(struct mtk_eth *eth)
 
        atomic_set(&ring->free_count, MTK_DMA_SIZE - 2);
        ring->next_free = &ring->dma[0];
-       ring->last_free = &ring->dma[MTK_DMA_SIZE - 2];
+       ring->last_free = &ring->dma[MTK_DMA_SIZE - 1];
        ring->thresh = MAX_SKB_FRAGS;
 
        /* make sure that all changes to the dma ring are flushed before we