mmc: davinci_mmc: Use sg_miter for PIO

Use the scatterlist memory iterator instead of just
dereferencing virtual memory using sg_virt().
This make highmem references work properly.
Suggested-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/linux-mmc/20240122073423.GA25859@lst.de/Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Link: https://lore.kernel.org/r/20240127-mmc-proper-kmap-v2-1-d8e732aa97d1@linaro.orgSigned-off-by: Ulf Hansson <ulf.hansson@linaro.org>

drivers/mmc/host/davinci_mmc.c

@@ -180,12 +180,6 @@ struct mmc_davinci_host {
 #define DAVINCI_MMC_DATADIR_WRITE	2
 	unsigned char data_dir;
 
-	/* buffer is used during PIO of one scatterlist segment, and
-	 * is updated along with buffer_bytes_left.  bytes_left applies
-	 * to all N blocks of the PIO transfer.
-	 */
-	u8 *buffer;
-	u32 buffer_bytes_left;
 	u32 bytes_left;
 
 	struct dma_chan *dma_tx;
@@ -196,8 +190,8 @@ struct mmc_davinci_host {
 	bool active_request;
 
 	/* For PIO we walk scatterlists one segment at a time. */
+	struct sg_mapping_iter sg_miter;
 	unsigned int		sg_len;
-	struct scatterlist *sg;
 
 	/* Version of the MMC/SD controller */
 	u8 version;
@@ -213,30 +207,24 @@ struct mmc_davinci_host {
 static irqreturn_t mmc_davinci_irq(int irq, void *dev_id);
 
 /* PIO only */
-static void mmc_davinci_sg_to_buf(struct mmc_davinci_host *host)
-{
-	host->buffer_bytes_left = sg_dma_len(host->sg);
-	host->buffer = sg_virt(host->sg);
-	if (host->buffer_bytes_left > host->bytes_left)
-		host->buffer_bytes_left = host->bytes_left;
-}
-
 static void davinci_fifo_data_trans(struct mmc_davinci_host *host,
 					unsigned int n)
 {
+	struct sg_mapping_iter *sgm = &host->sg_miter;
+	size_t sglen;
 	u8 *p;
 	unsigned int i;
 
-	if (host->buffer_bytes_left == 0) {
-		host->sg = sg_next(host->data->sg);
-		mmc_davinci_sg_to_buf(host);
+	/*
+	 * By adjusting sgm->consumed this will give a pointer to the
+	 * current index into the sgm.
+	 */
+	if (!sg_miter_next(sgm)) {
+		dev_err(mmc_dev(host->mmc), "ran out of sglist prematurely\n");
+		return;
 	}
-
-	p = host->buffer;
-	if (n > host->buffer_bytes_left)
-		n = host->buffer_bytes_left;
-	host->buffer_bytes_left -= n;
-	host->bytes_left -= n;
+	p = sgm->addr;
+	sglen = sgm->length;
 
 	/* NOTE:  we never transfer more than rw_threshold bytes
 	 * to/from the fifo here; there's no I/O overlap.
@@ -261,7 +249,9 @@ static void davinci_fifo_data_trans(struct mmc_davinci_host *host,
 			p = p + (n & 3);
 		}
 	}
-	host->buffer = p;
+
+	sgm->consumed = n;
+	host->bytes_left -= n;
 }
 
 static void mmc_davinci_start_command(struct mmc_davinci_host *host,
@@ -517,6 +507,7 @@ mmc_davinci_prepare_data(struct mmc_davinci_host *host, struct mmc_request *req)
 	int fifo_lev = (rw_threshold == 32) ? MMCFIFOCTL_FIFOLEV : 0;
 	int timeout;
 	struct mmc_data *data = req->data;
+	unsigned int flags = SG_MITER_ATOMIC; /* Used from IRQ */
 
 	if (host->version == MMC_CTLR_VERSION_2)
 		fifo_lev = (rw_threshold == 64) ? MMCFIFOCTL_FIFOLEV : 0;
@@ -545,12 +536,14 @@ mmc_davinci_prepare_data(struct mmc_davinci_host *host, struct mmc_request *req)
 
 	/* Configure the FIFO */
 	if (data->flags & MMC_DATA_WRITE) {
+		flags |= SG_MITER_FROM_SG;
 		host->data_dir = DAVINCI_MMC_DATADIR_WRITE;
 		writel(fifo_lev | MMCFIFOCTL_FIFODIR_WR | MMCFIFOCTL_FIFORST,
 			host->base + DAVINCI_MMCFIFOCTL);
 		writel(fifo_lev | MMCFIFOCTL_FIFODIR_WR,
 			host->base + DAVINCI_MMCFIFOCTL);
 	} else {
+		flags |= SG_MITER_TO_SG;
 		host->data_dir = DAVINCI_MMC_DATADIR_READ;
 		writel(fifo_lev | MMCFIFOCTL_FIFODIR_RD | MMCFIFOCTL_FIFORST,
 			host->base + DAVINCI_MMCFIFOCTL);
@@ -558,7 +551,6 @@ mmc_davinci_prepare_data(struct mmc_davinci_host *host, struct mmc_request *req)
 			host->base + DAVINCI_MMCFIFOCTL);
 	}
 
-	host->buffer = NULL;
 	host->bytes_left = data->blocks * data->blksz;
 
 	/* For now we try to use DMA whenever we won't need partial FIFO
@@ -576,8 +568,7 @@ mmc_davinci_prepare_data(struct mmc_davinci_host *host, struct mmc_request *req)
 	} else {
 		/* Revert to CPU Copy */
 		host->sg_len = data->sg_len;
-		host->sg = host->data->sg;
-		mmc_davinci_sg_to_buf(host);
+		sg_miter_start(&host->sg_miter, data->sg, data->sg_len, flags);
 	}
 }
 
@@ -843,6 +834,8 @@ davinci_abort_data(struct mmc_davinci_host *host, struct mmc_data *data)
 {
 	mmc_davinci_reset_ctrl(host, 1);
 	mmc_davinci_reset_ctrl(host, 0);
+	if (!host->do_dma)
+		sg_miter_stop(&host->sg_miter);
 }
 
 static irqreturn_t mmc_davinci_sdio_irq(int irq, void *dev_id)
@@ -919,11 +912,13 @@ static irqreturn_t mmc_davinci_irq(int irq, void *dev_id)
 	if (qstatus & MMCST0_DATDNE) {
 		/* All blocks sent/received, and CRC checks passed */
 		if (data != NULL) {
-			if ((host->do_dma == 0) && (host->bytes_left > 0)) {
-				/* if datasize < rw_threshold
-				 * no RX ints are generated
-				 */
-				davinci_fifo_data_trans(host, host->bytes_left);
+			if (!host->do_dma) {
+				if (host->bytes_left > 0)
+					/* if datasize < rw_threshold
+					 * no RX ints are generated
+					 */
+					davinci_fifo_data_trans(host, host->bytes_left);
+				sg_miter_stop(&host->sg_miter);
 			}
 			end_transfer = 1;
 			data->bytes_xfered = data->blocks * data->blksz;