[PATCH 0/5] spi: sh-msiof: Add DMA support

[PATCH 0/5] spi: sh-msiof: Add DMA support

[Geert Uytterhoeven]

This patch series refactors the Renesas MSIOF SPI driver and adds DMA
support.

  [1/5] spi: sh-msiof: Add more register documentation
  [2/5] spi: sh-msiof: Extract sh_msiof_spi_{start,stop}() helpers
  [3/5] spi: sh-msiof: Improve transfer error handling
  [4/5] spi: sh-msiof: Refactor sh_msiof_transfer_one()
  [5/5] spi: sh-msiof: Add DMA support

 drivers/spi/spi-sh-msiof.c   | 516 +++++++++++++++++++++++++++++++++++++++----
 include/linux/spi/sh_msiof.h |   2 +
 2 files changed, 469 insertions(+), 49 deletions(-)

Thanks!

Gr{oetje,eeting}s,

						Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
							    -- Linus Torvalds

[PATCH 1/5] spi: sh-msiof: Add more register documentation

[Geert Uytterhoeven]

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
---
 drivers/spi/spi-sh-msiof.c | 54 +++++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 53 insertions(+), 1 deletion(-)

diff --git a/drivers/spi/spi-sh-msiof.c b/drivers/spi/spi-sh-msiof.c
index 83c43707f093..7e5b15b48c04 100644
--- a/drivers/spi/spi-sh-msiof.c
+++ b/drivers/spi/spi-sh-msiof.c
@@ -113,9 +113,61 @@ struct sh_msiof_spi_priv {
 #define CTR_TXE		0x00000200 /* Transmit Enable */
 #define CTR_RXE		0x00000100 /* Receive Enable */
 
-/* STR and IER */
+/* FCTR */
+#define FCTR_TFWM_MASK	0xe0000000 /* Transmit FIFO Watermark */
+#define FCTR_TFWM_64	0x00000000 /*  Transfer Request when 64 empty stages */
+#define FCTR_TFWM_32	0x20000000 /*  Transfer Request when 32 empty stages */
+#define FCTR_TFWM_24	0x40000000 /*  Transfer Request when 24 empty stages */
+#define FCTR_TFWM_16	0x60000000 /*  Transfer Request when 16 empty stages */
+#define FCTR_TFWM_12	0x80000000 /*  Transfer Request when 12 empty stages */
+#define FCTR_TFWM_8	0xa0000000 /*  Transfer Request when 8 empty stages */
+#define FCTR_TFWM_4	0xc0000000 /*  Transfer Request when 4 empty stages */
+#define FCTR_TFWM_1	0xe0000000 /*  Transfer Request when 1 empty stage */
+#define FCTR_TFUA_MASK	0x07f00000 /* Transmit FIFO Usable Area */
+#define FCTR_TFUA_SHIFT		20
+#define FCTR_TFUA(i)	((i) << FCTR_TFUA_SHIFT)
+#define FCTR_RFWM_MASK	0x0000e000 /* Receive FIFO Watermark */
+#define FCTR_RFWM_1	0x00000000 /*  Transfer Request when 1 valid stages */
+#define FCTR_RFWM_4	0x00002000 /*  Transfer Request when 4 valid stages */
+#define FCTR_RFWM_8	0x00004000 /*  Transfer Request when 8 valid stages */
+#define FCTR_RFWM_16	0x00006000 /*  Transfer Request when 16 valid stages */
+#define FCTR_RFWM_32	0x00008000 /*  Transfer Request when 32 valid stages */
+#define FCTR_RFWM_64	0x0000a000 /*  Transfer Request when 64 valid stages */
+#define FCTR_RFWM_128	0x0000c000 /*  Transfer Request when 128 valid stages */
+#define FCTR_RFWM_256	0x0000e000 /*  Transfer Request when 256 valid stages */
+#define FCTR_RFUA_MASK	0x00001ff0 /* Receive FIFO Usable Area (0x40 = full) */
+#define FCTR_RFUA_SHIFT		 4
+#define FCTR_RFUA(i)	((i) << FCTR_RFUA_SHIFT)
+
+/* STR */
+#define STR_TFEMP	0x20000000 /* Transmit FIFO Empty */
+#define STR_TDREQ	0x10000000 /* Transmit Data Transfer Request */
 #define STR_TEOF	0x00800000 /* Frame Transmission End */
+#define STR_TFSERR	0x00200000 /* Transmit Frame Synchronization Error */
+#define STR_TFOVF	0x00100000 /* Transmit FIFO Overflow */
+#define STR_TFUDF	0x00080000 /* Transmit FIFO Underflow */
+#define STR_RFFUL	0x00002000 /* Receive FIFO Full */
+#define STR_RDREQ	0x00001000 /* Receive Data Transfer Request */
 #define STR_REOF	0x00000080 /* Frame Reception End */
+#define STR_RFSERR	0x00000020 /* Receive Frame Synchronization Error */
+#define STR_RFUDF	0x00000010 /* Receive FIFO Underflow */
+#define STR_RFOVF	0x00000008 /* Receive FIFO Overflow */
+
+/* IER */
+#define IER_TDMAE	0x80000000 /* Transmit Data DMA Transfer Req. Enable */
+#define IER_TFEMPE	0x20000000 /* Transmit FIFO Empty Enable */
+#define IER_TDREQE	0x10000000 /* Transmit Data Transfer Request Enable */
+#define IER_TEOFE	0x00800000 /* Frame Transmission End Enable */
+#define IER_TFSERRE	0x00200000 /* Transmit Frame Sync Error Enable */
+#define IER_TFOVFE	0x00100000 /* Transmit FIFO Overflow Enable */
+#define IER_TFUDFE	0x00080000 /* Transmit FIFO Underflow Enable */
+#define IER_RDMAE	0x00008000 /* Receive Data DMA Transfer Req. Enable */
+#define IER_RFFULE	0x00002000 /* Receive FIFO Full Enable */
+#define IER_RDREQE	0x00001000 /* Receive Data Transfer Request Enable */
+#define IER_REOFE	0x00000080 /* Frame Reception End Enable */
+#define IER_RFSERRE	0x00000020 /* Receive Frame Sync Error Enable */
+#define IER_RFUDFE	0x00000010 /* Receive FIFO Underflow Enable */
+#define IER_RFOVFE	0x00000008 /* Receive FIFO Overflow Enable */
 
 
 static u32 sh_msiof_read(struct sh_msiof_spi_priv *p, int reg_offs)
-- 
1.9.1

[PATCH 2/5] spi: sh-msiof: Extract sh_msiof_spi_{start,stop}() helpers

[Geert Uytterhoeven]

Based on an old patch by Guennadi Liakhovetski <g.liakhovetski@gmx.de>

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
---
 drivers/spi/spi-sh-msiof.c | 51 +++++++++++++++++++++++++++++++++-------------
 1 file changed, 37 insertions(+), 14 deletions(-)

diff --git a/drivers/spi/spi-sh-msiof.c b/drivers/spi/spi-sh-msiof.c
index 7e5b15b48c04..8bfa97e03502 100644
--- a/drivers/spi/spi-sh-msiof.c
+++ b/drivers/spi/spi-sh-msiof.c
@@ -509,6 +509,40 @@ static int sh_msiof_prepare_message(struct spi_master *master,
 	return 0;
 }
 
+static int sh_msiof_spi_start(struct sh_msiof_spi_priv *p, void *rx_buf)
+{
+	int ret;
+
+	/* setup clock and rx/tx signals */
+	ret = sh_msiof_modify_ctr_wait(p, 0, CTR_TSCKE);
+	if (rx_buf && !ret)
+		ret = sh_msiof_modify_ctr_wait(p, 0, CTR_RXE);
+	if (!ret)
+		ret = sh_msiof_modify_ctr_wait(p, 0, CTR_TXE);
+
+	/* start by setting frame bit */
+	if (!ret)
+		ret = sh_msiof_modify_ctr_wait(p, 0, CTR_TFSE);
+
+	return ret;
+}
+
+static int sh_msiof_spi_stop(struct sh_msiof_spi_priv *p, void *rx_buf)
+{
+	int ret;
+
+	/* shut down frame, rx/tx and clock signals */
+	ret = sh_msiof_modify_ctr_wait(p, CTR_TFSE, 0);
+	if (!ret)
+		ret = sh_msiof_modify_ctr_wait(p, CTR_TXE, 0);
+	if (rx_buf && !ret)
+		ret = sh_msiof_modify_ctr_wait(p, CTR_RXE, 0);
+	if (!ret)
+		ret = sh_msiof_modify_ctr_wait(p, CTR_TSCKE, 0);
+
+	return ret;
+}
+
 static int sh_msiof_spi_txrx_once(struct sh_msiof_spi_priv *p,
 				  void (*tx_fifo)(struct sh_msiof_spi_priv *,
 						  const void *, int, int),
@@ -536,15 +570,9 @@ static int sh_msiof_spi_txrx_once(struct sh_msiof_spi_priv *p,
 	if (tx_buf)
 		tx_fifo(p, tx_buf, words, fifo_shift);
 
-	/* setup clock and rx/tx signals */
-	ret = sh_msiof_modify_ctr_wait(p, 0, CTR_TSCKE);
-	if (rx_buf)
-		ret = ret ? ret : sh_msiof_modify_ctr_wait(p, 0, CTR_RXE);
-	ret = ret ? ret : sh_msiof_modify_ctr_wait(p, 0, CTR_TXE);
-
-	/* start by setting frame bit */
 	reinit_completion(&p->done);
-	ret = ret ? ret : sh_msiof_modify_ctr_wait(p, 0, CTR_TFSE);
+
+	ret = sh_msiof_spi_start(p, rx_buf);
 	if (ret) {
 		dev_err(&p->pdev->dev, "failed to start hardware\n");
 		goto err;
@@ -560,12 +588,7 @@ static int sh_msiof_spi_txrx_once(struct sh_msiof_spi_priv *p,
 	/* clear status bits */
 	sh_msiof_reset_str(p);
 
-	/* shut down frame, rx/tx and clock signals */
-	ret = sh_msiof_modify_ctr_wait(p, CTR_TFSE, 0);
-	ret = ret ? ret : sh_msiof_modify_ctr_wait(p, CTR_TXE, 0);
-	if (rx_buf)
-		ret = ret ? ret : sh_msiof_modify_ctr_wait(p, CTR_RXE, 0);
-	ret = ret ? ret : sh_msiof_modify_ctr_wait(p, CTR_TSCKE, 0);
+	ret = sh_msiof_spi_stop(p, rx_buf);
 	if (ret) {
 		dev_err(&p->pdev->dev, "failed to shut down hardware\n");
 		goto err;
-- 
1.9.1

[PATCH 3/5] spi: sh-msiof: Improve transfer error handling

[Geert Uytterhoeven]

  - Add a timeout when waiting for the transfer complete interrupt,
  - If sh_msiof_spi_stop() fails, there's no need to clear IER, as the
    interrupt handler has already done that,
  - Propagate transfer failures in sh_msiof_transfer_one().

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
---
 drivers/spi/spi-sh-msiof.c | 18 +++++++++++++-----
 1 file changed, 13 insertions(+), 5 deletions(-)

diff --git a/drivers/spi/spi-sh-msiof.c b/drivers/spi/spi-sh-msiof.c
index 8bfa97e03502..29dee0436a9a 100644
--- a/drivers/spi/spi-sh-msiof.c
+++ b/drivers/spi/spi-sh-msiof.c
@@ -575,11 +575,16 @@ static int sh_msiof_spi_txrx_once(struct sh_msiof_spi_priv *p,
 	ret = sh_msiof_spi_start(p, rx_buf);
 	if (ret) {
 		dev_err(&p->pdev->dev, "failed to start hardware\n");
-		goto err;
+		goto stop_ier;
 	}
 
 	/* wait for tx fifo to be emptied / rx fifo to be filled */
-	wait_for_completion(&p->done);
+	ret = wait_for_completion_timeout(&p->done, HZ);
+	if (!ret) {
+		dev_err(&p->pdev->dev, "PIO timeout\n");
+		ret = -ETIMEDOUT;
+		goto stop_reset;
+	}
 
 	/* read rx fifo */
 	if (rx_buf)
@@ -591,12 +596,15 @@ static int sh_msiof_spi_txrx_once(struct sh_msiof_spi_priv *p,
 	ret = sh_msiof_spi_stop(p, rx_buf);
 	if (ret) {
 		dev_err(&p->pdev->dev, "failed to shut down hardware\n");
-		goto err;
+		return ret;
 	}
 
 	return words;
 
- err:
+stop_reset:
+	sh_msiof_reset_str(p);
+	sh_msiof_spi_stop(p, rx_buf);
+stop_ier:
 	sh_msiof_write(p, IER, 0);
 	return ret;
 }
@@ -679,7 +687,7 @@ static int sh_msiof_transfer_one(struct spi_master *master,
 					   rx_buf,
 					   words, bits);
 		if (n < 0)
-			break;
+			return n;
 
 		bytes_done += n * bytes_per_word;
 		words -= n;
-- 
1.9.1

[PATCH 4/5] spi: sh-msiof: Refactor sh_msiof_transfer_one()

[Geert Uytterhoeven]

  - Move buffer pointer and length setup to the top,
  - Make unsigned values unsigned,
  - Loop over words and increment pointers instead of recalculating them,
    which allows to kill bytes_done.

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
---
 drivers/spi/spi-sh-msiof.c | 43 +++++++++++++++++++++----------------------
 1 file changed, 21 insertions(+), 22 deletions(-)

diff --git a/drivers/spi/spi-sh-msiof.c b/drivers/spi/spi-sh-msiof.c
index 29dee0436a9a..ef831f1c1803 100644
--- a/drivers/spi/spi-sh-msiof.c
+++ b/drivers/spi/spi-sh-msiof.c
@@ -616,16 +616,17 @@ static int sh_msiof_transfer_one(struct spi_master *master,
 	struct sh_msiof_spi_priv *p = spi_master_get_devdata(master);
 	void (*tx_fifo)(struct sh_msiof_spi_priv *, const void *, int, int);
 	void (*rx_fifo)(struct sh_msiof_spi_priv *, void *, int, int);
-	int bits;
-	int bytes_per_word;
-	int bytes_done;
-	int words;
+	const void *tx_buf = t->tx_buf;
+	void *rx_buf = t->rx_buf;
+	unsigned int len = t->len;
+	unsigned int bits = t->bits_per_word;
+	unsigned int bytes_per_word;
+	unsigned int words;
 	int n;
 	bool swab;
 
-	bits = t->bits_per_word;
 
-	if (bits <= 8 && t->len > 15 && !(t->len & 3)) {
+	if (bits <= 8 && len > 15 && !(len & 3)) {
 		bits = 32;
 		swab = true;
 	} else {
@@ -639,34 +640,34 @@ static int sh_msiof_transfer_one(struct spi_master *master,
 		rx_fifo = sh_msiof_spi_read_fifo_8;
 	} else if (bits <= 16) {
 		bytes_per_word = 2;
-		if ((unsigned long)t->tx_buf & 0x01)
+		if ((unsigned long)tx_buf & 0x01)
 			tx_fifo = sh_msiof_spi_write_fifo_16u;
 		else
 			tx_fifo = sh_msiof_spi_write_fifo_16;
 
-		if ((unsigned long)t->rx_buf & 0x01)
+		if ((unsigned long)rx_buf & 0x01)
 			rx_fifo = sh_msiof_spi_read_fifo_16u;
 		else
 			rx_fifo = sh_msiof_spi_read_fifo_16;
 	} else if (swab) {
 		bytes_per_word = 4;
-		if ((unsigned long)t->tx_buf & 0x03)
+		if ((unsigned long)tx_buf & 0x03)
 			tx_fifo = sh_msiof_spi_write_fifo_s32u;
 		else
 			tx_fifo = sh_msiof_spi_write_fifo_s32;
 
-		if ((unsigned long)t->rx_buf & 0x03)
+		if ((unsigned long)rx_buf & 0x03)
 			rx_fifo = sh_msiof_spi_read_fifo_s32u;
 		else
 			rx_fifo = sh_msiof_spi_read_fifo_s32;
 	} else {
 		bytes_per_word = 4;
-		if ((unsigned long)t->tx_buf & 0x03)
+		if ((unsigned long)tx_buf & 0x03)
 			tx_fifo = sh_msiof_spi_write_fifo_32u;
 		else
 			tx_fifo = sh_msiof_spi_write_fifo_32;
 
-		if ((unsigned long)t->rx_buf & 0x03)
+		if ((unsigned long)rx_buf & 0x03)
 			rx_fifo = sh_msiof_spi_read_fifo_32u;
 		else
 			rx_fifo = sh_msiof_spi_read_fifo_32;
@@ -676,20 +677,18 @@ static int sh_msiof_transfer_one(struct spi_master *master,
 	sh_msiof_spi_set_clk_regs(p, clk_get_rate(p->clk), t->speed_hz);
 
 	/* transfer in fifo sized chunks */
-	words = t->len / bytes_per_word;
-	bytes_done = 0;
-
-	while (bytes_done < t->len) {
-		void *rx_buf = t->rx_buf ? t->rx_buf + bytes_done : NULL;
-		const void *tx_buf = t->tx_buf ? t->tx_buf + bytes_done : NULL;
-		n = sh_msiof_spi_txrx_once(p, tx_fifo, rx_fifo,
-					   tx_buf,
-					   rx_buf,
+	words = len / bytes_per_word;
+
+	while (words > 0) {
+		n = sh_msiof_spi_txrx_once(p, tx_fifo, rx_fifo, tx_buf, rx_buf,
 					   words, bits);
 		if (n < 0)
 			return n;
 
-		bytes_done += n * bytes_per_word;
+		if (tx_buf)
+			tx_buf += n * bytes_per_word;
+		if (rx_buf)
+			rx_buf += n * bytes_per_word;
 		words -= n;
 	}
 
-- 
1.9.1

[PATCH 5/5] spi: sh-msiof: Add DMA support

[Geert Uytterhoeven]

Add DMA support to the MSIOF driver using platform data.

As MSIOF DMA is limited to 32-bit words (requiring byte/wordswapping for
smaller wordsizes), and the group length is limited to 256 words, DMA is
performed on two fixed pages, allocated and mapped at driver initialization
time.

Performance figures (in Mbps) on r8a7791/koelsch at different SPI clock
frequencies for 1024-byte and 4096-byte transfers:

                   1024 bytes           4096 bytes
  -  3.25 MHz: PIO  2.1, DMA  2.6 | PIO  2.8, DMA  3.1
  -  6.5  MHz: PIO  3.2, DMA  4.4 | PIO  5.0, DMA  5.9
  - 13    MHz: PIO  4.2, DMA  6.6 | PIO  8.2, DMA 10.7
  - 26    MHz: PIO  5.9, DMA 10.4 | PIO 12.4, DMA 18.4

Note that DMA is only faster than PIO for transfers that exceed the FIFO
size (typically 64 words / 256 bytes).

Also note that large transfers (larger than the group length for DMA, or
larger than the FIFO size for PIO), should use cs-gpio (with the
appropriate pinmux setup), as the hardware chipselect will be deasserted in
between chunks.

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
---
 drivers/spi/spi-sh-msiof.c   | 350 ++++++++++++++++++++++++++++++++++++++++++-
 include/linux/spi/sh_msiof.h |   2 +
 2 files changed, 345 insertions(+), 7 deletions(-)

diff --git a/drivers/spi/spi-sh-msiof.c b/drivers/spi/spi-sh-msiof.c
index ef831f1c1803..9922ed3a4441 100644
--- a/drivers/spi/spi-sh-msiof.c
+++ b/drivers/spi/spi-sh-msiof.c
@@ -2,6 +2,7 @@
  * SuperH MSIOF SPI Master Interface
  *
  * Copyright (c) 2009 Magnus Damm
+ * Copyright (C) 2014 Glider bvba
  *
  * This program is free software; you can redistribute it and/or modify
  * it under the terms of the GNU General Public License version 2 as
@@ -13,6 +14,8 @@
 #include <linux/clk.h>
 #include <linux/completion.h>
 #include <linux/delay.h>
+#include <linux/dma-mapping.h>
+#include <linux/dmaengine.h>
 #include <linux/err.h>
 #include <linux/gpio.h>
 #include <linux/interrupt.h>
@@ -23,6 +26,7 @@
 #include <linux/of_device.h>
 #include <linux/platform_device.h>
 #include <linux/pm_runtime.h>
+#include <linux/sh_dma.h>
 
 #include <linux/spi/sh_msiof.h>
 #include <linux/spi/spi.h>
@@ -37,6 +41,7 @@ struct sh_msiof_chipdata {
 };
 
 struct sh_msiof_spi_priv {
+	struct spi_master *master;
 	void __iomem *mapbase;
 	struct clk *clk;
 	struct platform_device *pdev;
@@ -45,6 +50,10 @@ struct sh_msiof_spi_priv {
 	struct completion done;
 	int tx_fifo_size;
 	int rx_fifo_size;
+	void *tx_dma_page;
+	void *rx_dma_page;
+	dma_addr_t tx_dma_addr;
+	dma_addr_t rx_dma_addr;
 };
 
 #define TMDR1	0x00	/* Transmit Mode Register 1 */
@@ -84,6 +93,8 @@ struct sh_msiof_spi_priv {
 #define MDR2_WDLEN1(i)	(((i) - 1) << 16) /* Word Count (1-64/256 (SH, A1))) */
 #define MDR2_GRPMASK1	0x00000001 /* Group Output Mask 1 (SH, A1) */
 
+#define MAX_WDLEN	256U
+
 /* TSCR and RSCR */
 #define SCR_BRPS_MASK	    0x1f00 /* Prescaler Setting (1-32) */
 #define SCR_BRPS(i)	(((i) - 1) << 8)
@@ -282,8 +293,6 @@ static void sh_msiof_spi_set_pin_regs(struct sh_msiof_spi_priv *p,
 	 *    1    0         11     11    0    0
 	 *    1    1         11     11    1    1
 	 */
-	sh_msiof_write(p, FCTR, 0);
-
 	tmp = MDR1_SYNCMD_SPI | 1 << MDR1_FLD_SHIFT | MDR1_XXSTP;
 	tmp |= !cs_high << MDR1_SYNCAC_SHIFT;
 	tmp |= lsb_first << MDR1_BITLSB_SHIFT;
@@ -319,8 +328,6 @@ static void sh_msiof_spi_set_mode_regs(struct sh_msiof_spi_priv *p,
 
 	if (rx_buf)
 		sh_msiof_write(p, RMDR2, dr2);
-
-	sh_msiof_write(p, IER, STR_TEOF | STR_REOF);
 }
 
 static void sh_msiof_reset_str(struct sh_msiof_spi_priv *p)
@@ -563,8 +570,12 @@ static int sh_msiof_spi_txrx_once(struct sh_msiof_spi_priv *p,
 	/* the fifo contents need shifting */
 	fifo_shift = 32 - bits;
 
+	/* default FIFO watermarks for PIO */
+	sh_msiof_write(p, FCTR, 0);
+
 	/* setup msiof transfer mode registers */
 	sh_msiof_spi_set_mode_regs(p, tx_buf, rx_buf, bits, words);
+	sh_msiof_write(p, IER, IER_TEOFE | IER_REOFE);
 
 	/* write tx fifo */
 	if (tx_buf)
@@ -609,11 +620,170 @@ stop_ier:
 	return ret;
 }
 
+static void sh_msiof_dma_complete(void *arg)
+{
+	struct sh_msiof_spi_priv *p = arg;
+
+	sh_msiof_write(p, IER, 0);
+	complete(&p->done);
+}
+
+static int sh_msiof_dma_once(struct sh_msiof_spi_priv *p, const void *tx,
+			     void *rx, unsigned int len)
+{
+	u32 ier_bits = 0;
+	struct dma_async_tx_descriptor *desc_tx = NULL, *desc_rx = NULL;
+	dma_cookie_t cookie;
+	int ret;
+
+	/* 1 stage FIFO watermarks for DMA */
+	sh_msiof_write(p, FCTR, FCTR_TFWM_1 | FCTR_RFWM_1);
+
+	/* setup msiof transfer mode registers (32-bit words) */
+	sh_msiof_spi_set_mode_regs(p, tx, rx, 32, len / 4);
+
+	if (tx) {
+		ier_bits |= IER_TDREQE | IER_TDMAE;
+		dma_sync_single_for_device(&p->pdev->dev, p->tx_dma_addr, len,
+					   DMA_TO_DEVICE);
+		desc_tx = dmaengine_prep_slave_single(p->master->dma_tx,
+					p->tx_dma_addr, len, DMA_TO_DEVICE,
+					DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
+		if (!desc_tx)
+			return -EIO;
+	}
+
+	if (rx) {
+		ier_bits |= IER_RDREQE | IER_RDMAE;
+		desc_rx = dmaengine_prep_slave_single(p->master->dma_rx,
+					p->rx_dma_addr, len, DMA_FROM_DEVICE,
+					DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
+		if (!desc_rx)
+			return -EIO;
+	}
+	sh_msiof_write(p, IER, ier_bits);
+
+	reinit_completion(&p->done);
+
+	if (rx) {
+		desc_rx->callback = sh_msiof_dma_complete;
+		desc_rx->callback_param = p;
+		cookie = dmaengine_submit(desc_rx);
+		if (dma_submit_error(cookie)) {
+			ret = cookie;
+			goto stop_ier;
+		}
+		dma_async_issue_pending(p->master->dma_rx);
+	}
+
+	if (tx) {
+		if (rx) {
+			/* No callback */
+			desc_tx->callback = NULL;
+		} else {
+			desc_tx->callback = sh_msiof_dma_complete;
+			desc_tx->callback_param = p;
+		}
+		cookie = dmaengine_submit(desc_tx);
+		if (dma_submit_error(cookie)) {
+			ret = cookie;
+			goto stop_rx;
+		}
+		dma_async_issue_pending(p->master->dma_tx);
+	}
+
+	ret = sh_msiof_spi_start(p, rx);
+	if (ret) {
+		dev_err(&p->pdev->dev, "failed to start hardware\n");
+		goto stop_tx;
+	}
+
+	/* wait for tx fifo to be emptied / rx fifo to be filled */
+	ret = wait_for_completion_timeout(&p->done, HZ);
+	if (!ret) {
+		dev_err(&p->pdev->dev, "DMA timeout\n");
+		ret = -ETIMEDOUT;
+		goto stop_reset;
+	}
+
+	/* clear status bits */
+	sh_msiof_reset_str(p);
+
+	ret = sh_msiof_spi_stop(p, rx);
+	if (ret) {
+		dev_err(&p->pdev->dev, "failed to shut down hardware\n");
+		return ret;
+	}
+
+	if (rx)
+		dma_sync_single_for_cpu(&p->pdev->dev, p->rx_dma_addr, len,
+					DMA_FROM_DEVICE);
+
+	return 0;
+
+stop_reset:
+	sh_msiof_reset_str(p);
+	sh_msiof_spi_stop(p, rx);
+stop_tx:
+	if (tx)
+		dmaengine_terminate_all(p->master->dma_tx);
+stop_rx:
+	if (rx)
+		dmaengine_terminate_all(p->master->dma_rx);
+stop_ier:
+	sh_msiof_write(p, IER, 0);
+	return ret;
+}
+
+static void copy_bswap32(u32 *dst, const u32 *src, unsigned int words)
+{
+	/* src or dst can be unaligned, but not both */
+	if ((unsigned long)src & 3) {
+		while (words--) {
+			*dst++ = swab32(get_unaligned(src));
+			src++;
+		}
+	} else if ((unsigned long)dst & 3) {
+		while (words--) {
+			put_unaligned(swab32(*src++), dst);
+			dst++;
+		}
+	} else {
+		while (words--)
+			*dst++ = swab32(*src++);
+	}
+}
+
+static void copy_wswap32(u32 *dst, const u32 *src, unsigned int words)
+{
+	/* src or dst can be unaligned, but not both */
+	if ((unsigned long)src & 3) {
+		while (words--) {
+			*dst++ = swahw32(get_unaligned(src));
+			src++;
+		}
+	} else if ((unsigned long)dst & 3) {
+		while (words--) {
+			put_unaligned(swahw32(*src++), dst);
+			dst++;
+		}
+	} else {
+		while (words--)
+			*dst++ = swahw32(*src++);
+	}
+}
+
+static void copy_plain32(u32 *dst, const u32 *src, unsigned int words)
+{
+	memcpy(dst, src, words * 4);
+}
+
 static int sh_msiof_transfer_one(struct spi_master *master,
 				 struct spi_device *spi,
 				 struct spi_transfer *t)
 {
 	struct sh_msiof_spi_priv *p = spi_master_get_devdata(master);
+	void (*copy32)(u32 *, const u32 *, unsigned int);
 	void (*tx_fifo)(struct sh_msiof_spi_priv *, const void *, int, int);
 	void (*rx_fifo)(struct sh_msiof_spi_priv *, void *, int, int);
 	const void *tx_buf = t->tx_buf;
@@ -624,7 +794,48 @@ static int sh_msiof_transfer_one(struct spi_master *master,
 	unsigned int words;
 	int n;
 	bool swab;
+	int ret;
+
+	/* setup clocks (clock already enabled in chipselect()) */
+	sh_msiof_spi_set_clk_regs(p, clk_get_rate(p->clk), t->speed_hz);
+
+	while (master->dma_tx && len > 15) {
+		/*
+		 *  DMA supports 32-bit words only, hence pack 8-bit and 16-bit
+		 *  words, with byte resp. word swapping.
+		 */
+		unsigned int l = min(len, MAX_WDLEN * 4);
+
+		if (bits <= 8) {
+			if (l & 3)
+				break;
+			copy32 = copy_bswap32;
+		} else if (bits <= 16) {
+			if (l & 1)
+				break;
+			copy32 = copy_wswap32;
+		} else {
+			copy32 = copy_plain32;
+		}
+
+		if (tx_buf)
+			copy32(p->tx_dma_page, tx_buf, l / 4);
 
+		ret = sh_msiof_dma_once(p, tx_buf, rx_buf, l);
+		if (ret)
+			return ret;
+
+		if (rx_buf) {
+			copy32(rx_buf, p->rx_dma_page, l / 4);
+			rx_buf += l;
+		}
+		if (tx_buf)
+			tx_buf += l;
+
+		len -= l;
+		if (!len)
+			return 0;
+	}
 
 	if (bits <= 8 && len > 15 && !(len & 3)) {
 		bits = 32;
@@ -673,9 +884,6 @@ static int sh_msiof_transfer_one(struct spi_master *master,
 			rx_fifo = sh_msiof_spi_read_fifo_32;
 	}
 
-	/* setup clocks (clock already enabled in chipselect()) */
-	sh_msiof_spi_set_clk_regs(p, clk_get_rate(p->clk), t->speed_hz);
-
 	/* transfer in fifo sized chunks */
 	words = len / bytes_per_word;
 
@@ -745,6 +953,123 @@ static struct sh_msiof_spi_info *sh_msiof_spi_parse_dt(struct device *dev)
 }
 #endif
 
+static struct dma_chan *sh_msiof_request_dma_chan(struct device *dev,
+	enum dma_transfer_direction dir, unsigned int id, dma_addr_t port_addr)
+{
+	dma_cap_mask_t mask;
+	struct dma_chan *chan;
+	struct dma_slave_config cfg;
+	int ret;
+
+	dma_cap_zero(mask);
+	dma_cap_set(DMA_SLAVE, mask);
+
+	chan = dma_request_channel(mask, shdma_chan_filter,
+				  (void *)(unsigned long)id);
+	if (!chan) {
+		dev_warn(dev, "dma_request_channel failed\n");
+		return NULL;
+	}
+
+	memset(&cfg, 0, sizeof(cfg));
+	cfg.slave_id = id;
+	cfg.direction = dir;
+	if (dir == DMA_MEM_TO_DEV)
+		cfg.dst_addr = port_addr;
+	else
+		cfg.src_addr = port_addr;
+
+	ret = dmaengine_slave_config(chan, &cfg);
+	if (ret) {
+		dev_warn(dev, "dmaengine_slave_config failed %d\n", ret);
+		dma_release_channel(chan);
+		return NULL;
+	}
+
+	return chan;
+}
+
+static int sh_msiof_request_dma(struct sh_msiof_spi_priv *p)
+{
+	struct platform_device *pdev = p->pdev;
+	struct device *dev = &pdev->dev;
+	const struct sh_msiof_spi_info *info = dev_get_platdata(dev);
+	const struct resource *res;
+	struct spi_master *master;
+
+	if (!info || !info->dma_tx_id || !info->dma_rx_id)
+		return 0;	/* The driver assumes no error */
+
+	/* The DMA engine uses the second register set, if present */
+	res = platform_get_resource(pdev, IORESOURCE_MEM, 1);
+	if (!res)
+		res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+
+	master = p->master;
+	master->dma_tx = sh_msiof_request_dma_chan(dev, DMA_MEM_TO_DEV,
+						   info->dma_tx_id,
+						   res->start + TFDR);
+	if (!master->dma_tx)
+		return -ENODEV;
+
+	master->dma_rx = sh_msiof_request_dma_chan(dev, DMA_DEV_TO_MEM,
+						   info->dma_rx_id,
+						   res->start + RFDR);
+	if (!master->dma_rx)
+		goto free_tx_chan;
+
+	p->tx_dma_page = (void *)__get_free_page(GFP_KERNEL | GFP_DMA);
+	if (!p->tx_dma_page)
+		goto free_rx_chan;
+
+	p->rx_dma_page = (void *)__get_free_page(GFP_KERNEL | GFP_DMA);
+	if (!p->rx_dma_page)
+		goto free_tx_page;
+
+	p->tx_dma_addr = dma_map_single(dev, p->tx_dma_page, PAGE_SIZE,
+					DMA_TO_DEVICE);
+	if (dma_mapping_error(dev, p->tx_dma_addr))
+		goto free_rx_page;
+
+	p->rx_dma_addr = dma_map_single(dev, p->rx_dma_page, PAGE_SIZE,
+					DMA_FROM_DEVICE);
+	if (dma_mapping_error(dev, p->rx_dma_addr))
+		goto unmap_tx_page;
+
+	dev_info(dev, "DMA available");
+	return 0;
+
+unmap_tx_page:
+	dma_unmap_single(dev, p->tx_dma_addr, PAGE_SIZE, DMA_TO_DEVICE);
+free_rx_page:
+	free_page((unsigned long)p->rx_dma_page);
+free_tx_page:
+	free_page((unsigned long)p->tx_dma_page);
+free_rx_chan:
+	dma_release_channel(master->dma_rx);
+free_tx_chan:
+	dma_release_channel(master->dma_tx);
+	master->dma_tx = NULL;
+	return -ENODEV;
+}
+
+static void sh_msiof_release_dma(struct sh_msiof_spi_priv *p)
+{
+	struct spi_master *master = p->master;
+	struct device *dev;
+
+	if (!master->dma_tx)
+		return;
+
+	dev = &p->pdev->dev;
+	dma_unmap_single(dev, p->rx_dma_addr, PAGE_SIZE, DMA_FROM_DEVICE);
+	dma_unmap_single(dev, p->tx_dma_addr, PAGE_SIZE, DMA_TO_DEVICE);
+	free_page((unsigned long)p->rx_dma_page);
+	free_page((unsigned long)p->tx_dma_page);
+	dma_release_channel(master->dma_rx);
+	dma_release_channel(master->dma_tx);
+}
+
 static int sh_msiof_spi_probe(struct platform_device *pdev)
 {
 	struct resource	*r;
@@ -762,6 +1087,9 @@ static int sh_msiof_spi_probe(struct platform_device *pdev)
 
 	p = spi_master_get_devdata(master);
 
+	platform_set_drvdata(pdev, p);
+	p->master = master;
+
 	of_id = of_match_device(sh_msiof_match, &pdev->dev);
 	if (of_id) {
 		p->chipdata = of_id->data;
@@ -831,6 +1159,10 @@ static int sh_msiof_spi_probe(struct platform_device *pdev)
 	master->auto_runtime_pm = true;
 	master->transfer_one = sh_msiof_transfer_one;
 
+	ret = sh_msiof_request_dma(p);
+	if (ret < 0)
+		dev_warn(&pdev->dev, "DMA not available, using PIO\n");
+
 	ret = devm_spi_register_master(&pdev->dev, master);
 	if (ret < 0) {
 		dev_err(&pdev->dev, "spi_register_master error.\n");
@@ -840,6 +1172,7 @@ static int sh_msiof_spi_probe(struct platform_device *pdev)
 	return 0;
 
  err2:
+	sh_msiof_release_dma(p);
 	pm_runtime_disable(&pdev->dev);
  err1:
 	spi_master_put(master);
@@ -848,6 +1181,9 @@ static int sh_msiof_spi_probe(struct platform_device *pdev)
 
 static int sh_msiof_spi_remove(struct platform_device *pdev)
 {
+	struct sh_msiof_spi_priv *p = platform_get_drvdata(pdev);
+
+	sh_msiof_release_dma(p);
 	pm_runtime_disable(&pdev->dev);
 	return 0;
 }
diff --git a/include/linux/spi/sh_msiof.h b/include/linux/spi/sh_msiof.h
index 2e8db3d2d2e5..88a14d81c49e 100644
--- a/include/linux/spi/sh_msiof.h
+++ b/include/linux/spi/sh_msiof.h
@@ -5,6 +5,8 @@ struct sh_msiof_spi_info {
 	int tx_fifo_override;
 	int rx_fifo_override;
 	u16 num_chipselect;
+	unsigned int dma_tx_id;
+	unsigned int dma_rx_id;
 };
 
 #endif /* __SPI_SH_MSIOF_H__ */
-- 
1.9.1

Re: [PATCH 1/5] spi: sh-msiof: Add more register documentation

[Mark Brown]

[-- Attachment #1: Type: text/plain, Size: 148 bytes --]

On Fri, Jun 20, 2014 at 12:16:16PM +0200, Geert Uytterhoeven wrote:
> Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>

Applied, thanks.

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH 2/5] spi: sh-msiof: Extract sh_msiof_spi_{start,stop}() helpers

[Mark Brown]

[-- Attachment #1: Type: text/plain, Size: 159 bytes --]

On Fri, Jun 20, 2014 at 12:16:17PM +0200, Geert Uytterhoeven wrote:
> Based on an old patch by Guennadi Liakhovetski <g.liakhovetski@gmx.de>

Applied, thanks.

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH 3/5] spi: sh-msiof: Improve transfer error handling

[Mark Brown]

[-- Attachment #1: Type: text/plain, Size: 338 bytes --]

On Fri, Jun 20, 2014 at 12:16:18PM +0200, Geert Uytterhoeven wrote:
>   - Add a timeout when waiting for the transfer complete interrupt,
>   - If sh_msiof_spi_stop() fails, there's no need to clear IER, as the
>     interrupt handler has already done that,
>   - Propagate transfer failures in sh_msiof_transfer_one().

Applied, thanks.

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH 4/5] spi: sh-msiof: Refactor sh_msiof_transfer_one()

[Mark Brown]

[-- Attachment #1: Type: text/plain, Size: 293 bytes --]

On Fri, Jun 20, 2014 at 12:16:19PM +0200, Geert Uytterhoeven wrote:
>   - Move buffer pointer and length setup to the top,
>   - Make unsigned values unsigned,
>   - Loop over words and increment pointers instead of recalculating them,
>     which allows to kill bytes_done.

Applied, thanks.

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH 5/5] spi: sh-msiof: Add DMA support

[Mark Brown]

[-- Attachment #1: Type: text/plain, Size: 202 bytes --]

On Fri, Jun 20, 2014 at 12:16:20PM +0200, Geert Uytterhoeven wrote:
> Add DMA support to the MSIOF driver using platform data.

This doesn't apply against current code, can you please check and
resend?

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH 5/5] spi: sh-msiof: Add DMA support

[Geert Uytterhoeven]

Hi Mark,

On Sat, Jun 21, 2014 at 12:19 PM, Mark Brown <broonie@kernel.org> wrote:
> On Fri, Jun 20, 2014 at 12:16:20PM +0200, Geert Uytterhoeven wrote:
>> Add DMA support to the MSIOF driver using platform data.
>
> This doesn't apply against current code, can you please check and
> resend?

My patches were based on your for-next branch, which has
commit 1e8231b79ccbb3e6e7b3f6af9f998dca16b00efe
Author: Jingoo Han <jg1.han@samsung.com>
Date:   Tue Apr 29 17:21:25 2014 +0900

    spi: sh-msiof: remove unnecessary OOM messages

(from topic/cleanup), but topic/sh-msiof hasn't.

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds

Re: [PATCH 5/5] spi: sh-msiof: Add DMA support

[Geert Uytterhoeven]

Hi Mark,

On Sun, Jun 22, 2014 at 10:06 AM, Geert Uytterhoeven
<geert@linux-m68k.org> wrote:
> On Sat, Jun 21, 2014 at 12:19 PM, Mark Brown <broonie@kernel.org> wrote:
>> On Fri, Jun 20, 2014 at 12:16:20PM +0200, Geert Uytterhoeven wrote:
>>> Add DMA support to the MSIOF driver using platform data.
>>
>> This doesn't apply against current code, can you please check and
>> resend?
>
> My patches were based on your for-next branch, which has
> commit 1e8231b79ccbb3e6e7b3f6af9f998dca16b00efe
> Author: Jingoo Han <jg1.han@samsung.com>
> Date:   Tue Apr 29 17:21:25 2014 +0900
>
>     spi: sh-msiof: remove unnecessary OOM messages

Sorry, wrong copy and paste, I meant:

commit b8fa0abd47f3f2fa47c1b2b06182c422d0c2350d
Author: Alexander Shiyan <shc_work@mail.ru>
Date:   Wed Jun 4 18:17:09 2014 +0400

    spi: sh-msiof: Remove unnecessary platform_set_drvdata()

    Drop call to platform_set_drvdata() as driver data is not used
    anywhere in the driver.

    Signed-off-by: Alexander Shiyan <shc_work@mail.ru>
    Signed-off-by: Mark Brown <broonie@linaro.org>

> (from topic/cleanup), but topic/sh-msiof hasn't.

Do you want me to resend? Against which branch?

Basically I had to revert the above commit, as I do use the driver data.
So if I would rebase and resend against topic/sh-msiof, you'll get a merge
conflict when merging this into for-next.

Thanks!

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds