net: mvpp2: fix DMA address calculation in mvpp2_txq_inc_put()
commitb57ffb2a8466a7628b316c8e3a5ce34b8ee4e519
authorThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Tue, 21 Feb 2017 10:28:01 +0000 (21 11:28 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 15 Mar 2017 01:57:13 +0000 (15 09:57 +0800)
treef7d5b7035465723ac3ecb10729d79c0c3395d63c
parent376a12eb7f608fad96b13fab3f151eb4c9b40c7c
net: mvpp2: fix DMA address calculation in mvpp2_txq_inc_put()

commit 239a3b663647869330955ec59caac0100ef9b60a upstream.

When TX descriptors are filled in, the buffer DMA address is split
between the tx_desc->buf_phys_addr field (high-order bits) and
tx_desc->packet_offset field (5 low-order bits).

However, when we re-calculate the DMA address from the TX descriptor in
mvpp2_txq_inc_put(), we do not take tx_desc->packet_offset into
account. This means that when the DMA address is not aligned on a 32
bytes boundary, we end up calling dma_unmap_single() with a DMA address
that was not the one returned by dma_map_single().

This inconsistency is detected by the kernel when DMA_API_DEBUG is
enabled. We fix this problem by properly calculating the DMA address in
mvpp2_txq_inc_put().

Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/net/ethernet/marvell/mvpp2.c