[PATCH 06/14] qcom: mtd: nand: add bam dma descriptor handling

[Abhishek Sahu]

1. prepare_bam_async_desc is the function which will call
   all the DMA APIâs. It will fetch the outstanding scatter gather
   list for passed channel and will do the DMA descriptor formation.
   The DMA flag is dependent upon the type of channel.

2. For ADM DMA, the descriptor is being formed for every DMA
   request so its sgl count will be always 1 while in BAM DMA, the
   clubbing of descriptor is being done to increase throughput.

3. ADM uses only one channel while in BAM, data descriptors
   will be submitted to tx channel (for write) or rx channel
   (for read) and all the registers read/write descriptors in
   command channel.

Signed-off-by: Abhishek Sahu <absahu@xxxxxxxxxxxxxx>
---
 drivers/mtd/nand/qcom_nandc.c | 119 ++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 114 insertions(+), 5 deletions(-)

diff --git a/drivers/mtd/nand/qcom_nandc.c b/drivers/mtd/nand/qcom_nandc.c
index f8d0bde..7042a65 100644
--- a/drivers/mtd/nand/qcom_nandc.c
+++ b/drivers/mtd/nand/qcom_nandc.c
@@ -206,14 +206,22 @@ struct bam_transaction {
  * This data type corresponds to the nand dma descriptor
  * @list - list for desc_info
  * @dir - DMA transfer direction
- * @sgl - sgl which will be used for single sgl dma descriptor
+ * @sgl - sgl which will be used for single sgl dma descriptor. Only used by ADM
+ * @bam_sgl - sgl which will be used for dma descriptor. Only used by BAM
+ * @sgl_cnt - number of SGL in bam_sgl. Only used by BAM
  * @dma_desc - low level dma engine descriptor
  */
 struct desc_info {
 	struct list_head node;
 
 	enum dma_data_direction dir;
-	struct scatterlist sgl;
+	union {
+			struct scatterlist sgl;
+			struct {
+				struct scatterlist *bam_sgl;
+				int sgl_cnt;
+			};
+	};
 	struct dma_async_tx_descriptor *dma_desc;
 };
 
@@ -564,6 +572,68 @@ static void update_rw_regs(struct qcom_nand_host *host, int num_cw, bool read)
 	nandc_set_reg(nandc, NAND_EXEC_CMD, 1);
 }
 
+/*
+ * Maps the scatter gather list for DMA transfer and forms the DMA descriptor
+ * for BAM. This descriptor will be added in the NAND DMA descriptor queue
+ * which will be submitted to DMA engine.
+ */
+static int prepare_bam_async_desc(struct qcom_nand_controller *nandc,
+				  struct dma_chan *chan,
+				  unsigned long flags)
+{
+	struct desc_info *desc;
+	struct scatterlist *sgl;
+	unsigned int sgl_cnt;
+	struct bam_transaction *bam_txn = nandc->bam_txn;
+	enum dma_transfer_direction dir_eng;
+	struct dma_async_tx_descriptor *dma_desc;
+
+	desc = kzalloc(sizeof(*desc), GFP_KERNEL);
+	if (!desc)
+		return -ENOMEM;
+
+	if (chan == nandc->cmd_chan) {
+		sgl = &bam_txn->cmd_sgl[bam_txn->cmd_sgl_start];
+		sgl_cnt = bam_txn->cmd_sgl_pos - bam_txn->cmd_sgl_start;
+		bam_txn->cmd_sgl_start = bam_txn->cmd_sgl_pos;
+		dir_eng = DMA_MEM_TO_DEV;
+		desc->dir = DMA_TO_DEVICE;
+	} else if (chan == nandc->tx_chan) {
+		sgl = &bam_txn->data_sg[bam_txn->tx_sgl_start];
+		sgl_cnt = bam_txn->tx_sgl_pos - bam_txn->tx_sgl_start;
+		bam_txn->tx_sgl_start = bam_txn->tx_sgl_pos;
+		dir_eng = DMA_MEM_TO_DEV;
+		desc->dir = DMA_TO_DEVICE;
+	} else {
+		sgl = &bam_txn->data_sg[bam_txn->rx_sgl_start];
+		sgl_cnt = bam_txn->rx_sgl_pos - bam_txn->rx_sgl_start;
+		bam_txn->rx_sgl_start = bam_txn->rx_sgl_pos;
+		desc->dir = DMA_FROM_DEVICE;
+		dir_eng = DMA_DEV_TO_MEM;
+	}
+
+	sg_mark_end(sgl + sgl_cnt - 1);
+	dma_map_sg(nandc->dev, sgl, sgl_cnt, desc->dir);
+
+	desc->sgl_cnt = sgl_cnt;
+	desc->bam_sgl = sgl;
+
+	dma_desc = dmaengine_prep_slave_sg(chan, sgl, sgl_cnt, dir_eng,
+					   flags);
+
+	if (!dma_desc) {
+		dev_err(nandc->dev, "failure in prep desc\n");
+		kfree(desc);
+		return -EINVAL;
+	}
+
+	desc->dma_desc = dma_desc;
+
+	list_add_tail(&desc->node, &nandc->desc_list);
+
+	return 0;
+}
+
 static int prep_dma_desc(struct qcom_nand_controller *nandc, bool read,
 			 int reg_off, const void *vaddr, int size,
 			 bool flow_control)
@@ -891,12 +961,44 @@ static int submit_descs(struct qcom_nand_controller *nandc)
 {
 	struct desc_info *desc;
 	dma_cookie_t cookie = 0;
+	struct bam_transaction *bam_txn = nandc->bam_txn;
+	int r;
+
+	if (nandc->dma_bam_enabled) {
+		if (bam_txn->rx_sgl_pos - bam_txn->rx_sgl_start) {
+			r = prepare_bam_async_desc(nandc, nandc->rx_chan, 0);
+			if (r)
+				return r;
+		}
+
+		if (bam_txn->tx_sgl_pos - bam_txn->tx_sgl_start) {
+			r = prepare_bam_async_desc(nandc, nandc->tx_chan,
+						   DMA_PREP_INTERRUPT);
+			if (r)
+				return r;
+		}
+
+		if (bam_txn->cmd_sgl_pos - bam_txn->cmd_sgl_start) {
+			r = prepare_bam_async_desc(nandc, nandc->cmd_chan,
+						   DMA_PREP_CMD);
+			if (r)
+				return r;
+		}
+	}
 
 	list_for_each_entry(desc, &nandc->desc_list, node)
 		cookie = dmaengine_submit(desc->dma_desc);
 
-	if (dma_sync_wait(nandc->chan, cookie) != DMA_COMPLETE)
-		return -ETIMEDOUT;
+	if (nandc->dma_bam_enabled) {
+		dma_async_issue_pending(nandc->tx_chan);
+		dma_async_issue_pending(nandc->rx_chan);
+
+		if (dma_sync_wait(nandc->cmd_chan, cookie) != DMA_COMPLETE)
+			return -ETIMEDOUT;
+	} else {
+		if (dma_sync_wait(nandc->chan, cookie) != DMA_COMPLETE)
+			return -ETIMEDOUT;
+	}
 
 	return 0;
 }
@@ -907,7 +1009,14 @@ static void free_descs(struct qcom_nand_controller *nandc)
 
 	list_for_each_entry_safe(desc, n, &nandc->desc_list, node) {
 		list_del(&desc->node);
-		dma_unmap_sg(nandc->dev, &desc->sgl, 1, desc->dir);
+
+		if (nandc->dma_bam_enabled)
+			dma_unmap_sg(nandc->dev, desc->bam_sgl,
+				     desc->sgl_cnt, desc->dir);
+		else
+			dma_unmap_sg(nandc->dev, &desc->sgl, 1,
+				     desc->dir);
+
 		kfree(desc);
 	}
 }
-- 
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted by The Linux Foundation