mtd: rawnand: denali: fix a race condition when DMA is kicked

I thought the read-back of the DMA_ENABLE register was unnecessary
(at least it is working on my boards), then deleted it in commit
586a2c52909d ("mtd: nand: denali: squash denali_enable_dma() helper
into caller").  Sorry, I was wrong - it caused a timing issue on
Cyclone5 SoCFPGAs.

Revive the register read-back, commenting why this is necessary.

Fixes: 586a2c52909d ("mtd: nand: denali: squash denali_enable_dma() helper into caller")
Cc: <stable@vger.kernel.org>
Reported-by: Steffen Trumtrar <s.trumtrar@pengutronix.de>
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: Miquel Raynal <miquel.raynal@bootlin.com>
Signed-off-by: Boris Brezillon <boris.brezillon@bootlin.com>

drivers/mtd/nand/raw/denali.c

@@ -596,6 +596,12 @@ static int denali_dma_xfer(struct denali_nand_info *denali, void *buf,
 	}
 
 	iowrite32(DMA_ENABLE__FLAG, denali->reg + DMA_ENABLE);
+	/*
+	 * The ->setup_dma() hook kicks DMA by using the data/command
+	 * interface, which belongs to a different AXI port from the
+	 * register interface.  Read back the register to avoid a race.
+	 */
+	ioread32(denali->reg + DMA_ENABLE);
 
 	denali_reset_irq(denali);
 	denali->setup_dma(denali, dma_addr, page, write);