spi: uniphier: Add DMA transfer mode support
authorKunihiko Hayashi <hayashi.kunihiko@socionext.com>
Tue, 24 Dec 2019 00:58:27 +0000 (09:58 +0900)
committerMark Brown <broonie@kernel.org>
Wed, 25 Dec 2019 18:21:05 +0000 (18:21 +0000)
This adds DMA transfer mode support for UniPhier SPI controller.

Since this controller requires simulteaneous transmission and reception,
this indicates SPI_CONTROLLER_MUST_RX and SPI_CONTROLLER_MUST_TX.

Because the supported dma controller has alignment restiction,
there is also a restriction that 'maxburst' parameters in dma_slave_config
corresponds to one word width.

Signed-off-by: Kunihiko Hayashi <hayashi.kunihiko@socionext.com>
Link: https://lore.kernel.org/r/1577149107-30670-6-git-send-email-hayashi.kunihiko@socionext.com
Signed-off-by: Mark Brown <broonie@kernel.org>
drivers/spi/spi-uniphier.c

index c4e3b96b10385c3901fba4a048bdad4597efcb83..0fa50979644d448768e94c6d8c94a9f28a872321 100644 (file)
@@ -8,6 +8,7 @@
 #include <linux/bitops.h>
 #include <linux/clk.h>
 #include <linux/delay.h>
+#include <linux/dmaengine.h>
 #include <linux/interrupt.h>
 #include <linux/io.h>
 #include <linux/module.h>
@@ -23,6 +24,7 @@
 
 struct uniphier_spi_priv {
        void __iomem *base;
+       dma_addr_t base_dma_addr;
        struct clk *clk;
        struct spi_master *master;
        struct completion xfer_done;
@@ -32,6 +34,7 @@ struct uniphier_spi_priv {
        unsigned int rx_bytes;
        const u8 *tx_buf;
        u8 *rx_buf;
+       atomic_t dma_busy;
 
        bool is_save_param;
        u8 bits_per_word;
@@ -61,11 +64,16 @@ struct uniphier_spi_priv {
 #define   SSI_FPS_FSTRT                BIT(14)
 
 #define SSI_SR                 0x14
+#define   SSI_SR_BUSY          BIT(7)
 #define   SSI_SR_RNE           BIT(0)
 
 #define SSI_IE                 0x18
+#define   SSI_IE_TCIE          BIT(4)
 #define   SSI_IE_RCIE          BIT(3)
+#define   SSI_IE_TXRE          BIT(2)
+#define   SSI_IE_RXRE          BIT(1)
 #define   SSI_IE_RORIE         BIT(0)
+#define   SSI_IE_ALL_MASK      GENMASK(4, 0)
 
 #define SSI_IS                 0x1c
 #define   SSI_IS_RXRS          BIT(9)
@@ -87,6 +95,10 @@ struct uniphier_spi_priv {
 #define SSI_RXDR               0x24
 
 #define SSI_FIFO_DEPTH         8U
+#define SSI_FIFO_BURST_NUM     1
+
+#define SSI_DMA_RX_BUSY                BIT(1)
+#define SSI_DMA_TX_BUSY                BIT(0)
 
 static inline unsigned int bytes_per_word(unsigned int bits)
 {
@@ -334,6 +346,128 @@ static void uniphier_spi_set_cs(struct spi_device *spi, bool enable)
        writel(val, priv->base + SSI_FPS);
 }
 
+static bool uniphier_spi_can_dma(struct spi_master *master,
+                                struct spi_device *spi,
+                                struct spi_transfer *t)
+{
+       struct uniphier_spi_priv *priv = spi_master_get_devdata(master);
+       unsigned int bpw = bytes_per_word(priv->bits_per_word);
+
+       if ((!master->dma_tx && !master->dma_rx)
+           || (!master->dma_tx && t->tx_buf)
+           || (!master->dma_rx && t->rx_buf))
+               return false;
+
+       return DIV_ROUND_UP(t->len, bpw) > SSI_FIFO_DEPTH;
+}
+
+static void uniphier_spi_dma_rxcb(void *data)
+{
+       struct spi_master *master = data;
+       struct uniphier_spi_priv *priv = spi_master_get_devdata(master);
+       int state = atomic_fetch_andnot(SSI_DMA_RX_BUSY, &priv->dma_busy);
+
+       uniphier_spi_irq_disable(priv, SSI_IE_RXRE);
+
+       if (!(state & SSI_DMA_TX_BUSY))
+               spi_finalize_current_transfer(master);
+}
+
+static void uniphier_spi_dma_txcb(void *data)
+{
+       struct spi_master *master = data;
+       struct uniphier_spi_priv *priv = spi_master_get_devdata(master);
+       int state = atomic_fetch_andnot(SSI_DMA_TX_BUSY, &priv->dma_busy);
+
+       uniphier_spi_irq_disable(priv, SSI_IE_TXRE);
+
+       if (!(state & SSI_DMA_RX_BUSY))
+               spi_finalize_current_transfer(master);
+}
+
+static int uniphier_spi_transfer_one_dma(struct spi_master *master,
+                                        struct spi_device *spi,
+                                        struct spi_transfer *t)
+{
+       struct uniphier_spi_priv *priv = spi_master_get_devdata(master);
+       struct dma_async_tx_descriptor *rxdesc = NULL, *txdesc = NULL;
+       int buswidth;
+
+       atomic_set(&priv->dma_busy, 0);
+
+       uniphier_spi_set_fifo_threshold(priv, SSI_FIFO_BURST_NUM);
+
+       if (priv->bits_per_word <= 8)
+               buswidth = DMA_SLAVE_BUSWIDTH_1_BYTE;
+       else if (priv->bits_per_word <= 16)
+               buswidth = DMA_SLAVE_BUSWIDTH_2_BYTES;
+       else
+               buswidth = DMA_SLAVE_BUSWIDTH_4_BYTES;
+
+       if (priv->rx_buf) {
+               struct dma_slave_config rxconf = {
+                       .direction = DMA_DEV_TO_MEM,
+                       .src_addr = priv->base_dma_addr + SSI_RXDR,
+                       .src_addr_width = buswidth,
+                       .src_maxburst = SSI_FIFO_BURST_NUM,
+               };
+
+               dmaengine_slave_config(master->dma_rx, &rxconf);
+
+               rxdesc = dmaengine_prep_slave_sg(
+                       master->dma_rx,
+                       t->rx_sg.sgl, t->rx_sg.nents,
+                       DMA_DEV_TO_MEM, DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
+               if (!rxdesc)
+                       goto out_err_prep;
+
+               rxdesc->callback = uniphier_spi_dma_rxcb;
+               rxdesc->callback_param = master;
+
+               uniphier_spi_irq_enable(priv, SSI_IE_RXRE);
+               atomic_or(SSI_DMA_RX_BUSY, &priv->dma_busy);
+
+               dmaengine_submit(rxdesc);
+               dma_async_issue_pending(master->dma_rx);
+       }
+
+       if (priv->tx_buf) {
+               struct dma_slave_config txconf = {
+                       .direction = DMA_MEM_TO_DEV,
+                       .dst_addr = priv->base_dma_addr + SSI_TXDR,
+                       .dst_addr_width = buswidth,
+                       .dst_maxburst = SSI_FIFO_BURST_NUM,
+               };
+
+               dmaengine_slave_config(master->dma_tx, &txconf);
+
+               txdesc = dmaengine_prep_slave_sg(
+                       master->dma_tx,
+                       t->tx_sg.sgl, t->tx_sg.nents,
+                       DMA_MEM_TO_DEV, DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
+               if (!txdesc)
+                       goto out_err_prep;
+
+               txdesc->callback = uniphier_spi_dma_txcb;
+               txdesc->callback_param = master;
+
+               uniphier_spi_irq_enable(priv, SSI_IE_TXRE);
+               atomic_or(SSI_DMA_TX_BUSY, &priv->dma_busy);
+
+               dmaengine_submit(txdesc);
+               dma_async_issue_pending(master->dma_tx);
+       }
+
+       /* signal that we need to wait for completion */
+       return (priv->tx_buf || priv->rx_buf);
+
+out_err_prep:
+       if (rxdesc)
+               dmaengine_terminate_sync(master->dma_rx);
+
+       return -EINVAL;
+}
+
 static int uniphier_spi_transfer_one_irq(struct spi_master *master,
                                         struct spi_device *spi,
                                         struct spi_transfer *t)
@@ -395,6 +529,7 @@ static int uniphier_spi_transfer_one(struct spi_master *master,
 {
        struct uniphier_spi_priv *priv = spi_master_get_devdata(master);
        unsigned long threshold;
+       bool use_dma;
 
        /* Terminate and return success for 0 byte length transfer */
        if (!t->len)
@@ -402,6 +537,10 @@ static int uniphier_spi_transfer_one(struct spi_master *master,
 
        uniphier_spi_setup_transfer(spi, t);
 
+       use_dma = master->can_dma ? master->can_dma(master, spi, t) : false;
+       if (use_dma)
+               return uniphier_spi_transfer_one_dma(master, spi, t);
+
        /*
         * If the transfer operation will take longer than
         * SSI_POLL_TIMEOUT_US, it should use irq.
@@ -445,7 +584,17 @@ static void uniphier_spi_handle_err(struct spi_master *master,
        val = SSI_FC_TXFFL | SSI_FC_RXFFL;
        writel(val, priv->base + SSI_FC);
 
-       uniphier_spi_irq_disable(priv, SSI_IE_RCIE | SSI_IE_RORIE);
+       uniphier_spi_irq_disable(priv, SSI_IE_ALL_MASK);
+
+       if (atomic_read(&priv->dma_busy) & SSI_DMA_TX_BUSY) {
+               dmaengine_terminate_async(master->dma_tx);
+               atomic_andnot(SSI_DMA_TX_BUSY, &priv->dma_busy);
+       }
+
+       if (atomic_read(&priv->dma_busy) & SSI_DMA_RX_BUSY) {
+               dmaengine_terminate_async(master->dma_rx);
+               atomic_andnot(SSI_DMA_RX_BUSY, &priv->dma_busy);
+       }
 }
 
 static irqreturn_t uniphier_spi_handler(int irq, void *dev_id)
@@ -493,6 +642,9 @@ static int uniphier_spi_probe(struct platform_device *pdev)
 {
        struct uniphier_spi_priv *priv;
        struct spi_master *master;
+       struct resource *res;
+       struct dma_slave_caps caps;
+       u32 dma_tx_burst = 0, dma_rx_burst = 0;
        unsigned long clk_rate;
        int irq;
        int ret;
@@ -507,11 +659,13 @@ static int uniphier_spi_probe(struct platform_device *pdev)
        priv->master = master;
        priv->is_save_param = false;
 
-       priv->base = devm_platform_ioremap_resource(pdev, 0);
+       res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+       priv->base = devm_ioremap_resource(&pdev->dev, res);
        if (IS_ERR(priv->base)) {
                ret = PTR_ERR(priv->base);
                goto out_master_put;
        }
+       priv->base_dma_addr = res->start;
 
        priv->clk = devm_clk_get(&pdev->dev, NULL);
        if (IS_ERR(priv->clk)) {
@@ -555,7 +709,44 @@ static int uniphier_spi_probe(struct platform_device *pdev)
        master->unprepare_transfer_hardware
                                = uniphier_spi_unprepare_transfer_hardware;
        master->handle_err = uniphier_spi_handle_err;
+       master->can_dma = uniphier_spi_can_dma;
+
        master->num_chipselect = 1;
+       master->flags = SPI_CONTROLLER_MUST_RX | SPI_CONTROLLER_MUST_TX;
+
+       master->dma_tx = dma_request_chan(&pdev->dev, "tx");
+       if (IS_ERR_OR_NULL(master->dma_tx)) {
+               if (PTR_ERR(master->dma_tx) == -EPROBE_DEFER)
+                       goto out_disable_clk;
+               master->dma_tx = NULL;
+               dma_tx_burst = INT_MAX;
+       } else {
+               ret = dma_get_slave_caps(master->dma_tx, &caps);
+               if (ret) {
+                       dev_err(&pdev->dev, "failed to get TX DMA capacities: %d\n",
+                               ret);
+                       goto out_disable_clk;
+               }
+               dma_tx_burst = caps.max_burst;
+       }
+
+       master->dma_rx = dma_request_chan(&pdev->dev, "rx");
+       if (IS_ERR_OR_NULL(master->dma_rx)) {
+               if (PTR_ERR(master->dma_rx) == -EPROBE_DEFER)
+                       goto out_disable_clk;
+               master->dma_rx = NULL;
+               dma_rx_burst = INT_MAX;
+       } else {
+               ret = dma_get_slave_caps(master->dma_rx, &caps);
+               if (ret) {
+                       dev_err(&pdev->dev, "failed to get RX DMA capacities: %d\n",
+                               ret);
+                       goto out_disable_clk;
+               }
+               dma_rx_burst = caps.max_burst;
+       }
+
+       master->max_dma_len = min(dma_tx_burst, dma_rx_burst);
 
        ret = devm_spi_register_master(&pdev->dev, master);
        if (ret)
@@ -575,6 +766,11 @@ static int uniphier_spi_remove(struct platform_device *pdev)
 {
        struct uniphier_spi_priv *priv = platform_get_drvdata(pdev);
 
+       if (priv->master->dma_tx)
+               dma_release_channel(priv->master->dma_tx);
+       if (priv->master->dma_rx)
+               dma_release_channel(priv->master->dma_rx);
+
        clk_disable_unprepare(priv->clk);
 
        return 0;