[PATCH 0/2] spi: spi-ti-qspi: Support large NOR SPI flash

[PATCH 0/2] spi: spi-ti-qspi: Support large NOR SPI flash

[Jean Pihet]

v3 release: fix compilation warnings.

Large devices are bigger than >64MB in size.
- Since the TI QSPI IP block only maps 64MB of MMIO, use MMIO
  below the 64MB boundary and software generated transfers above.
- Optimize the software generated byte-transfers for dual and quad
  I/O read operations. The speed-up is 4.9x for quad I/O reads.

Note: depends on Tony's patches for hwmod cleanup, in order to get the
desired QSPI clk rate:
- [PATCH 1/2] ARM: dts: Configure interconnect target module for am4 qspi
- [PATCH 2/2] ARM: OMAP2+: Drop legacy platform data for am4 qspi

Tested using raw accesses (mtd_debug) and JFFS2 FS read/write/erase;
in single, dual and quad modes.
All accesses have been checked on the logic analyzer.

Jean Pihet (2):
  spi: spi-ti-qspi: support large flash devices
  spi: spi-ti-qspi: optimize byte-transfers

 drivers/spi/spi-ti-qspi.c | 84 +++++++++++++++++++++++++++++++++++++--
 1 file changed, 81 insertions(+), 3 deletions(-)

-- 
2.24.1

[PATCH 1/2] spi: spi-ti-qspi: support large flash devices

[Jean Pihet]

The TI QSPI IP has limitations:
- the MMIO region is 64MB in size
- in non-MMIO mode, the transfer can handle 4096 words max.

Add support for bigger devices.
Use MMIO and DMA transfers below the 64MB boundary, use
software generated transfers above.

Signed-off-by: Jean Pihet <jean.pihet-OTs2U3NB0ngRmelmmXo44Q@public.gmane.org>
Cc: Ryan Barnett <ryan.barnett-lFk7bPDcGtkY5TsXZYaR1UEOCMrvLtNR@public.gmane.org>
Cc: Conrad Ratschan <conrad.ratschan-lFk7bPDcGtkY5TsXZYaR1UEOCMrvLtNR@public.gmane.org>
Cc: Arnout Vandecappelle <arnout.vandecappelle-buIOx9BAs4sybS5Ee8rs3A@public.gmane.org>
---
 drivers/spi/spi-ti-qspi.c | 30 ++++++++++++++++++++++++++++++
 1 file changed, 30 insertions(+)

diff --git a/drivers/spi/spi-ti-qspi.c b/drivers/spi/spi-ti-qspi.c
index ad2942b3d0a9..0334e2926998 100644
--- a/drivers/spi/spi-ti-qspi.c
+++ b/drivers/spi/spi-ti-qspi.c
@@ -525,6 +525,35 @@ static void ti_qspi_setup_mmap_read(struct spi_device *spi, u8 opcode,
 		      QSPI_SPI_SETUP_REG(spi->chip_select));
 }
 
+static int ti_qspi_adjust_op_size(struct spi_mem *mem, struct spi_mem_op *op)
+{
+	struct ti_qspi *qspi = spi_controller_get_devdata(mem->spi->master);
+	size_t max_len;
+
+	if (op->data.dir == SPI_MEM_DATA_IN) {
+		if (op->addr.val < qspi->mmap_size) {
+			/* Limit MMIO to the mmaped region */
+			if (op->addr.val + op->data.nbytes > qspi->mmap_size) {
+				max_len = qspi->mmap_size - op->addr.val;
+				op->data.nbytes = min((size_t) op->data.nbytes,
+						      max_len);
+			}
+		} else {
+			/*
+			 * Use fallback mode (SW generated transfers) above the
+			 * mmaped region.
+			 * Adjust size to comply with the QSPI max frame length.
+			 */
+			max_len = QSPI_FRAME;
+			max_len -= 1 + op->addr.nbytes + op->dummy.nbytes;
+			op->data.nbytes = min((size_t) op->data.nbytes,
+					      max_len);
+		}
+	}
+
+	return 0;
+}
+
 static int ti_qspi_exec_mem_op(struct spi_mem *mem,
 			       const struct spi_mem_op *op)
 {
@@ -575,6 +604,7 @@ static int ti_qspi_exec_mem_op(struct spi_mem *mem,
 
 static const struct spi_controller_mem_ops ti_qspi_mem_ops = {
 	.exec_op = ti_qspi_exec_mem_op,
+	.adjust_op_size = ti_qspi_adjust_op_size,
 };
 
 static int ti_qspi_start_transfer_one(struct spi_master *master,
-- 
2.24.1

[PATCH 2/2] spi: spi-ti-qspi: optimize byte-transfers

[Jean Pihet]

Optimize the 8-bit based transfers, as used by the SPI flash
devices, by reading the data registers by 32 and 128 bits when
possible and copy the contents to the receive buffer.

The speed improvement is 4.9x using quad read.

Signed-off-by: Jean Pihet <jean.pihet-OTs2U3NB0ngRmelmmXo44Q@public.gmane.org>
Cc: Ryan Barnett <ryan.barnett-lFk7bPDcGtkY5TsXZYaR1UEOCMrvLtNR@public.gmane.org>
Cc: Conrad Ratschan <conrad.ratschan-lFk7bPDcGtkY5TsXZYaR1UEOCMrvLtNR@public.gmane.org>
Cc: Arnout Vandecappelle <arnout.vandecappelle-buIOx9BAs4sybS5Ee8rs3A@public.gmane.org>
---
 drivers/spi/spi-ti-qspi.c | 54 ++++++++++++++++++++++++++++++++++++---
 1 file changed, 51 insertions(+), 3 deletions(-)

diff --git a/drivers/spi/spi-ti-qspi.c b/drivers/spi/spi-ti-qspi.c
index 0334e2926998..858fda8ac73e 100644
--- a/drivers/spi/spi-ti-qspi.c
+++ b/drivers/spi/spi-ti-qspi.c
@@ -314,6 +314,8 @@ static int qspi_read_msg(struct ti_qspi *qspi, struct spi_transfer *t,
 {
 	int wlen;
 	unsigned int cmd;
+	u32 rx;
+	u8 rxlen, rx_wlen;
 	u8 *rxbuf;
 
 	rxbuf = t->rx_buf;
@@ -336,14 +338,60 @@ static int qspi_read_msg(struct ti_qspi *qspi, struct spi_transfer *t,
 		if (qspi_is_busy(qspi))
 			return -EBUSY;
 
+		switch (wlen) {
+		case 1:
+			/*
+			 * Optimize the 8-bit words transfers, as used by
+			 * the SPI flash devices.
+			 */
+			if (count >= QSPI_WLEN_MAX_BYTES) {
+				rxlen = QSPI_WLEN_MAX_BYTES;
+			} else {
+				rxlen = min(count, 4);
+			}
+			rx_wlen = rxlen << 3;
+			cmd &= ~QSPI_WLEN_MASK;
+			cmd |= QSPI_WLEN(rx_wlen);
+			break;
+		default:
+			rxlen = wlen;
+			break;
+		}
+
 		ti_qspi_write(qspi, cmd, QSPI_SPI_CMD_REG);
 		if (ti_qspi_poll_wc(qspi)) {
 			dev_err(qspi->dev, "read timed out\n");
 			return -ETIMEDOUT;
 		}
+
 		switch (wlen) {
 		case 1:
-			*rxbuf = readb(qspi->base + QSPI_SPI_DATA_REG);
+			/*
+			 * Optimize the 8-bit words transfers, as used by
+			 * the SPI flash devices.
+			 */
+			if (count >= QSPI_WLEN_MAX_BYTES) {
+				u32 *rxp = (u32 *) rxbuf;
+				rx = readl(qspi->base + QSPI_SPI_DATA_REG_3);
+				*rxp++ = be32_to_cpu(rx);
+				rx = readl(qspi->base + QSPI_SPI_DATA_REG_2);
+				*rxp++ = be32_to_cpu(rx);
+				rx = readl(qspi->base + QSPI_SPI_DATA_REG_1);
+				*rxp++ = be32_to_cpu(rx);
+				rx = readl(qspi->base + QSPI_SPI_DATA_REG);
+				*rxp++ = be32_to_cpu(rx);
+			} else {
+				u8 *rxp = rxbuf;
+				rx = readl(qspi->base + QSPI_SPI_DATA_REG);
+				if (rx_wlen >= 8)
+					*rxp++ = rx >> (rx_wlen - 8);
+				if (rx_wlen >= 16)
+					*rxp++ = rx >> (rx_wlen - 16);
+				if (rx_wlen >= 24)
+					*rxp++ = rx >> (rx_wlen - 24);
+				if (rx_wlen >= 32)
+					*rxp++ = rx;
+			}
 			break;
 		case 2:
 			*((u16 *)rxbuf) = readw(qspi->base + QSPI_SPI_DATA_REG);
@@ -352,8 +400,8 @@ static int qspi_read_msg(struct ti_qspi *qspi, struct spi_transfer *t,
 			*((u32 *)rxbuf) = readl(qspi->base + QSPI_SPI_DATA_REG);
 			break;
 		}
-		rxbuf += wlen;
-		count -= wlen;
+		rxbuf += rxlen;
+		count -= rxlen;
 	}
 
 	return 0;
-- 
2.24.1

Applied "spi: spi-ti-qspi: optimize byte-transfers" to the spi tree

[Mark Brown]

The patch

   spi: spi-ti-qspi: optimize byte-transfers

has been applied to the spi tree at

   https://git.kernel.org/pub/scm/linux/kernel/git/broonie/spi.git for-5.6

All being well this means that it will be integrated into the linux-next
tree (usually sometime in the next 24 hours) and sent to Linus during
the next merge window (or sooner if it is a bug fix), however if
problems are discovered then the patch may be dropped or reverted.  

You may get further e-mails resulting from automated or manual testing
and review of the tree, please engage with people reporting problems and
send followup patches addressing any issues that are reported if needed.

If any updates are required or you are submitting further changes they
should be sent as incremental updates against current git, existing
patches will not be replaced.

Please add any relevant lists and maintainers to the CCs when replying
to this mail.

Thanks,
Mark

>From e7cc5cfbea4c9bd2c452cb81d0829e9259dd84d8 Mon Sep 17 00:00:00 2001
From: Jean Pihet <jean.pihet@newoldbits.com>
Date: Tue, 14 Jan 2020 13:41:25 +0100
Subject: [PATCH] spi: spi-ti-qspi: optimize byte-transfers

Optimize the 8-bit based transfers, as used by the SPI flash
devices, by reading the data registers by 32 and 128 bits when
possible and copy the contents to the receive buffer.

The speed improvement is 4.9x using quad read.

Signed-off-by: Jean Pihet <jean.pihet@newoldbits.com>
Cc: Ryan Barnett <ryan.barnett@rockwellcollins.com>
Cc: Conrad Ratschan <conrad.ratschan@rockwellcollins.com>
Cc: Arnout Vandecappelle <arnout.vandecappelle@essensium.com>
Link: https://lore.kernel.org/r/20200114124125.361429-3-jean.pihet@newoldbits.com
Signed-off-by: Mark Brown <broonie@kernel.org>
---
 drivers/spi/spi-ti-qspi.c | 54 ++++++++++++++++++++++++++++++++++++---
 1 file changed, 51 insertions(+), 3 deletions(-)

diff --git a/drivers/spi/spi-ti-qspi.c b/drivers/spi/spi-ti-qspi.c
index 0334e2926998..858fda8ac73e 100644
--- a/drivers/spi/spi-ti-qspi.c
+++ b/drivers/spi/spi-ti-qspi.c
@@ -314,6 +314,8 @@ static int qspi_read_msg(struct ti_qspi *qspi, struct spi_transfer *t,
 {
 	int wlen;
 	unsigned int cmd;
+	u32 rx;
+	u8 rxlen, rx_wlen;
 	u8 *rxbuf;
 
 	rxbuf = t->rx_buf;
@@ -336,14 +338,60 @@ static int qspi_read_msg(struct ti_qspi *qspi, struct spi_transfer *t,
 		if (qspi_is_busy(qspi))
 			return -EBUSY;
 
+		switch (wlen) {
+		case 1:
+			/*
+			 * Optimize the 8-bit words transfers, as used by
+			 * the SPI flash devices.
+			 */
+			if (count >= QSPI_WLEN_MAX_BYTES) {
+				rxlen = QSPI_WLEN_MAX_BYTES;
+			} else {
+				rxlen = min(count, 4);
+			}
+			rx_wlen = rxlen << 3;
+			cmd &= ~QSPI_WLEN_MASK;
+			cmd |= QSPI_WLEN(rx_wlen);
+			break;
+		default:
+			rxlen = wlen;
+			break;
+		}
+
 		ti_qspi_write(qspi, cmd, QSPI_SPI_CMD_REG);
 		if (ti_qspi_poll_wc(qspi)) {
 			dev_err(qspi->dev, "read timed out\n");
 			return -ETIMEDOUT;
 		}
+
 		switch (wlen) {
 		case 1:
-			*rxbuf = readb(qspi->base + QSPI_SPI_DATA_REG);
+			/*
+			 * Optimize the 8-bit words transfers, as used by
+			 * the SPI flash devices.
+			 */
+			if (count >= QSPI_WLEN_MAX_BYTES) {
+				u32 *rxp = (u32 *) rxbuf;
+				rx = readl(qspi->base + QSPI_SPI_DATA_REG_3);
+				*rxp++ = be32_to_cpu(rx);
+				rx = readl(qspi->base + QSPI_SPI_DATA_REG_2);
+				*rxp++ = be32_to_cpu(rx);
+				rx = readl(qspi->base + QSPI_SPI_DATA_REG_1);
+				*rxp++ = be32_to_cpu(rx);
+				rx = readl(qspi->base + QSPI_SPI_DATA_REG);
+				*rxp++ = be32_to_cpu(rx);
+			} else {
+				u8 *rxp = rxbuf;
+				rx = readl(qspi->base + QSPI_SPI_DATA_REG);
+				if (rx_wlen >= 8)
+					*rxp++ = rx >> (rx_wlen - 8);
+				if (rx_wlen >= 16)
+					*rxp++ = rx >> (rx_wlen - 16);
+				if (rx_wlen >= 24)
+					*rxp++ = rx >> (rx_wlen - 24);
+				if (rx_wlen >= 32)
+					*rxp++ = rx;
+			}
 			break;
 		case 2:
 			*((u16 *)rxbuf) = readw(qspi->base + QSPI_SPI_DATA_REG);
@@ -352,8 +400,8 @@ static int qspi_read_msg(struct ti_qspi *qspi, struct spi_transfer *t,
 			*((u32 *)rxbuf) = readl(qspi->base + QSPI_SPI_DATA_REG);
 			break;
 		}
-		rxbuf += wlen;
-		count -= wlen;
+		rxbuf += rxlen;
+		count -= rxlen;
 	}
 
 	return 0;
-- 
2.20.1

Applied "spi: spi-ti-qspi: support large flash devices" to the spi tree

[Mark Brown]

The patch

   spi: spi-ti-qspi: support large flash devices

has been applied to the spi tree at

   https://git.kernel.org/pub/scm/linux/kernel/git/broonie/spi.git for-5.6

All being well this means that it will be integrated into the linux-next
tree (usually sometime in the next 24 hours) and sent to Linus during
the next merge window (or sooner if it is a bug fix), however if
problems are discovered then the patch may be dropped or reverted.  

You may get further e-mails resulting from automated or manual testing
and review of the tree, please engage with people reporting problems and
send followup patches addressing any issues that are reported if needed.

If any updates are required or you are submitting further changes they
should be sent as incremental updates against current git, existing
patches will not be replaced.

Please add any relevant lists and maintainers to the CCs when replying
to this mail.

Thanks,
Mark

>From e97f491450805fe6cbfd482b97b5427b21dec575 Mon Sep 17 00:00:00 2001
From: Jean Pihet <jean.pihet@newoldbits.com>
Date: Tue, 14 Jan 2020 13:41:24 +0100
Subject: [PATCH] spi: spi-ti-qspi: support large flash devices

The TI QSPI IP has limitations:
- the MMIO region is 64MB in size
- in non-MMIO mode, the transfer can handle 4096 words max.

Add support for bigger devices.
Use MMIO and DMA transfers below the 64MB boundary, use
software generated transfers above.

Signed-off-by: Jean Pihet <jean.pihet@newoldbits.com>
Cc: Ryan Barnett <ryan.barnett@rockwellcollins.com>
Cc: Conrad Ratschan <conrad.ratschan@rockwellcollins.com>
Cc: Arnout Vandecappelle <arnout.vandecappelle@essensium.com>
Link: https://lore.kernel.org/r/20200114124125.361429-2-jean.pihet@newoldbits.com
Signed-off-by: Mark Brown <broonie@kernel.org>
---
 drivers/spi/spi-ti-qspi.c | 30 ++++++++++++++++++++++++++++++
 1 file changed, 30 insertions(+)

diff --git a/drivers/spi/spi-ti-qspi.c b/drivers/spi/spi-ti-qspi.c
index ad2942b3d0a9..0334e2926998 100644
--- a/drivers/spi/spi-ti-qspi.c
+++ b/drivers/spi/spi-ti-qspi.c
@@ -525,6 +525,35 @@ static void ti_qspi_setup_mmap_read(struct spi_device *spi, u8 opcode,
 		      QSPI_SPI_SETUP_REG(spi->chip_select));
 }
 
+static int ti_qspi_adjust_op_size(struct spi_mem *mem, struct spi_mem_op *op)
+{
+	struct ti_qspi *qspi = spi_controller_get_devdata(mem->spi->master);
+	size_t max_len;
+
+	if (op->data.dir == SPI_MEM_DATA_IN) {
+		if (op->addr.val < qspi->mmap_size) {
+			/* Limit MMIO to the mmaped region */
+			if (op->addr.val + op->data.nbytes > qspi->mmap_size) {
+				max_len = qspi->mmap_size - op->addr.val;
+				op->data.nbytes = min((size_t) op->data.nbytes,
+						      max_len);
+			}
+		} else {
+			/*
+			 * Use fallback mode (SW generated transfers) above the
+			 * mmaped region.
+			 * Adjust size to comply with the QSPI max frame length.
+			 */
+			max_len = QSPI_FRAME;
+			max_len -= 1 + op->addr.nbytes + op->dummy.nbytes;
+			op->data.nbytes = min((size_t) op->data.nbytes,
+					      max_len);
+		}
+	}
+
+	return 0;
+}
+
 static int ti_qspi_exec_mem_op(struct spi_mem *mem,
 			       const struct spi_mem_op *op)
 {
@@ -575,6 +604,7 @@ static int ti_qspi_exec_mem_op(struct spi_mem *mem,
 
 static const struct spi_controller_mem_ops ti_qspi_mem_ops = {
 	.exec_op = ti_qspi_exec_mem_op,
+	.adjust_op_size = ti_qspi_adjust_op_size,
 };
 
 static int ti_qspi_start_transfer_one(struct spi_master *master,
-- 
2.20.1