Empirical testing shows that rcar-dmac engine can handle byte-aligned
addresses without issues. Relax the driver by reducing
RCAR_DMAC_MEMCPY_XFER_SIZE from 4 to 1 byte so that memcpy can run on
unaligned buffers.
One practical user is ntb_netdev + ntb_transport. On its TX path,
skb->data often ends up 2 bytes off a 4-byte boundary due to the 16-byte
headroom alignment (HH_DATA_MOD) combined with a 14-byte Ethernet header
push, yielding a 4n+2 start address, falling back to CPU memcpy. With
this change, rcar-dmac can be used there as well.
In local testing, hacking skb->data to be 4-byte aligned while keeping
RCAR_DMAC_MEMCPY_XFER_SIZE = 4 did not produce a measurable throughput
difference compared to the unaligned case with this change, suggesting
the previous 4-byte constraint was not performance-driven on the tested
systems.
Signed-off-by: Koichiro Den <den@valinux.co.jp>
---
drivers/dma/sh/rcar-dmac.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/dma/sh/rcar-dmac.c b/drivers/dma/sh/rcar-dmac.c
index 0c45ce8c74aa..c4d6c2f9a26b 100644
--- a/drivers/dma/sh/rcar-dmac.c
+++ b/drivers/dma/sh/rcar-dmac.c
@@ -303,7 +303,7 @@ struct rcar_dmac_of_data {
#define RCAR_GEN4_DMACHCLR 0x0100
/* Hardcode the MEMCPY transfer size to 4 bytes. */
-#define RCAR_DMAC_MEMCPY_XFER_SIZE 4
+#define RCAR_DMAC_MEMCPY_XFER_SIZE 1
/* -----------------------------------------------------------------------------
* Device access
--
2.48.1