mmc: dw_mmc: handle unaligned buffers and sizes

Update functions for PIO pushing and pulling data to and from the FIFO
so that they can handle unaligned output buffers and unaligned buffer
lengths. This makes more of the tests in mmc_test pass.

Unaligned lengths in pulls are handled by reading the full FIFO item,
and storing the remaining bytes in a small internal buffer (part_buf).
The next data pull will copy data out of this buffer first before
accessing the FIFO again. Similarly, for pushes the final bytes that
don't fill a FIFO item are stored in the part_buf (or sent anyway if
it's the last transfer), and then the part_buf is included at the
beginning of the next buffer pushed.

Unaligned buffers in pulls are handled specially if the architecture
cannot do efficient unaligned accesses, by reading FIFO items into a
aligned local buffer, and memcpy'ing them into the output buffer, again
storing any remaining bytes in the internal buffer. Similarly for pushes
the buffer is memcpy'd into an aligned local buffer then written to the
FIFO.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Acked-by: Will Newton <will.newton@imgtec.com>
Signed-off-by: Chris Ball <cjb@laptop.org>

include/linux/mmc/dw_mmc.h

@@ -76,6 +76,9 @@ struct mmc_data;
  * @slot: Slots sharing this MMC controller.
  * @fifo_depth: depth of FIFO.
  * @data_shift: log2 of FIFO item size.
+ * @part_buf_start: Start index in part_buf.
+ * @part_buf_count: Bytes of partial data in part_buf.
+ * @part_buf: Simple buffer for partial fifo reads/writes.
  * @push_data: Pointer to FIFO push function.
  * @pull_data: Pointer to FIFO pull function.
  * @quirks: Set of quirks that apply to specific versions of the IP.
@@ -149,6 +152,13 @@ struct dw_mci {
 	/* FIFO push and pull */
 	int			fifo_depth;
 	int			data_shift;
+	u8			part_buf_start;
+	u8			part_buf_count;
+	union {
+		u16		part_buf16;
+		u32		part_buf32;
+		u64		part_buf;
+	};
 	void (*push_data)(struct dw_mci *host, void *buf, int cnt);
 	void (*pull_data)(struct dw_mci *host, void *buf, int cnt);