mmc: tmio-mmc: Improve DMA stability on sh-mobile

On some SDHI tmio implementations the order of DMA and command completion
interrupts swaps, which leads to malfunction. This patch postpones
DMA activation until the MMC command completion IRQ time.

Signed-off-by: Guennadi Liakhovetski <[e-mail]>
Acked-by: Paul Mundt <[e-mail]>
Signed-off-by: Chris Ball <[e-mail]>

drivers/mmc/host/tmio_mmc.c

···
 	unsigned int count;
 	unsigned long flags;
 
-	if (!data) {
+	if (host->chan_tx || host->chan_rx) {
+		pr_err("PIO IRQ in DMA mode!\n");
+		return;
+	} else if (!data) {
 		pr_debug("Spurious PIO IRQ\n");
 		return;
 	}
···
 		if (host->data->flags & MMC_DATA_READ) {
 			if (!host->chan_rx)
 				enable_mmc_irqs(host, TMIO_MASK_READOP);
+			else
+				tasklet_schedule(&host->dma_issue);
 		} else {
 			if (!host->chan_tx)
 				enable_mmc_irqs(host, TMIO_MASK_WRITEOP);
···
 #endif
 }
 
-static void tmio_dma_complete(void *arg)
-{
-	struct tmio_mmc_host *host = arg;
-
-	dev_dbg(&host->pdev->dev, "Command completed\n");
-
-	if (!host->data)
-		dev_warn(&host->pdev->dev, "NULL data in DMA completion!\n");
-	else
-		enable_mmc_irqs(host, TMIO_STAT_DATAEND);
-}
-
 static void tmio_mmc_start_dma_rx(struct tmio_mmc_host *host)
 {
 	struct scatterlist *sg = host->sg_ptr, *sg_tmp;
···
 		goto pio;
 	}
 
+	disable_mmc_irqs(host, TMIO_STAT_RXRDY);
+
 	/* The only sg element can be unaligned, use our bounce buffer then */
 	if (!aligned) {
 		sg_init_one(&host->bounce_sg, host->bounce_buf, sg->length);
···
 	ret = dma_map_sg(chan->device->dev, sg, host->sg_len, DMA_FROM_DEVICE);
 	if (ret > 0)
 		desc = chan->device->device_prep_slave_sg(chan, sg, ret,
-			DMA_FROM_DEVICE, DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
+			DMA_FROM_DEVICE, DMA_CTRL_ACK);
 
-	if (desc) {
-		desc->callback = tmio_dma_complete;
-		desc->callback_param = host;
+	if (desc)
 		cookie = dmaengine_submit(desc);
-		dma_async_issue_pending(chan);
-	}
+
 	dev_dbg(&host->pdev->dev, "%s(): mapped %d -> %d, cookie %d, rq %p\n",
 		__func__, host->sg_len, ret, cookie, host->mrq);
 
···
 		goto pio;
 	}
 
+	disable_mmc_irqs(host, TMIO_STAT_TXRQ);
+
 	/* The only sg element can be unaligned, use our bounce buffer then */
 	if (!aligned) {
 		unsigned long flags;
···
 	ret = dma_map_sg(chan->device->dev, sg, host->sg_len, DMA_TO_DEVICE);
 	if (ret > 0)
 		desc = chan->device->device_prep_slave_sg(chan, sg, ret,
-			DMA_TO_DEVICE, DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
+			DMA_TO_DEVICE, DMA_CTRL_ACK);
 
-	if (desc) {
-		desc->callback = tmio_dma_complete;
-		desc->callback_param = host;
+	if (desc)
 		cookie = dmaengine_submit(desc);
-	}
+
 	dev_dbg(&host->pdev->dev, "%s(): mapped %d -> %d, cookie %d, rq %p\n",
 		__func__, host->sg_len, ret, cookie, host->mrq);
 
···
 static void tmio_issue_tasklet_fn(unsigned long priv)
 {
 	struct tmio_mmc_host *host = (struct tmio_mmc_host *)priv;
-	struct dma_chan *chan = host->chan_tx;
+	struct dma_chan *chan = NULL;
 
-	dma_async_issue_pending(chan);
+	spin_lock_irq(&host->lock);
+
+	if (host && host->data) {
+		if (host->data->flags & MMC_DATA_READ)
+			chan = host->chan_rx;
+		else
+			chan = host->chan_tx;
+	}
+
+	spin_unlock_irq(&host->lock);
+
+	enable_mmc_irqs(host, TMIO_STAT_DATAEND);
+
+	if (chan)
+		dma_async_issue_pending(chan);
 }
 
 static void tmio_tasklet_fn(unsigned long arg)
 {
 	struct tmio_mmc_host *host = (struct tmio_mmc_host *)arg;
-	unsigned long flags;
 
-	spin_lock_irqsave(&host->lock, flags);
+	spin_lock_irq(&host->lock);
 
 	if (!host->data)
 		goto out;
···
 
 	tmio_mmc_do_data_irq(host);
 out:
-	spin_unlock_irqrestore(&host->lock, flags);
+	spin_unlock_irq(&host->lock);
 }
 
 /* It might be necessary to make filter MFD specific */