qede: Correctly map aggregation replacement pages
authorMintz, Yuval <Yuval.Mintz@cavium.com>
Wed, 2 Nov 2016 14:36:46 +0000 (16:36 +0200)
committerDavid S. Miller <davem@davemloft.net>
Thu, 3 Nov 2016 19:33:30 +0000 (15:33 -0400)
Driver allocates replacement buffers before-hand to make
sure whenever an aggregation begins there would be a replacement
for the Rx buffers, as we can't release the buffer until
aggregation is terminated and driver logic assumes the Rx rings
are always full.

For every other Rx page that's being allocated [I.e., regular]
the page is being completely mapped while for the replacement
buffers only the first portion of the page is being mapped.
This means that:
  a. Once replacement buffer replenishes the regular Rx ring,
assuming there's more than a single packet on page we'd post unmapped
memory toward HW [assuming mapping is actually done in granularity
smaller than page].
  b. Unmaps are being done for the entire page, which is incorrect.

Fixes: 55482edc25f06 ("qede: Add slowpath/fastpath support and enable hardware GRO")
Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/qlogic/qede/qede_main.c

index 7def29aaf65c962b63b483cb33cea9fd397b63bb..85f46dbecd5b57b5cdb537900a51dbd8cfa4dcc1 100644 (file)
@@ -2839,7 +2839,7 @@ static int qede_alloc_sge_mem(struct qede_dev *edev, struct qede_rx_queue *rxq)
                }
 
                mapping = dma_map_page(&edev->pdev->dev, replace_buf->data, 0,
-                                      rxq->rx_buf_size, DMA_FROM_DEVICE);
+                                      PAGE_SIZE, DMA_FROM_DEVICE);
                if (unlikely(dma_mapping_error(&edev->pdev->dev, mapping))) {
                        DP_NOTICE(edev,
                                  "Failed to map TPA replacement buffer\n");