Patch "Subject:[For 4.19/4.14] spi: fsl-cpm: Use 16 bit mode for large transfers with even size" has been added to the 4.14-stable tree

[gregkh at linuxfoundation.org]

This is a note to let you know that I've just added the patch titled

    Subject:[For 4.19/4.14] spi: fsl-cpm: Use 16 bit mode for large transfers with even size

to the 4.14-stable tree which can be found at:
    http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary

The filename of the patch is:
     spi-fsl-cpm-use-16-bit-mode-for-large-transfers-with-even-size.patch
and it can be found in the queue-4.14 subdirectory.

If you, or anyone else, feels it should not be added to the stable tree,
please let <stable at vger.kernel.org> know about it.


>From christophe.leroy at csgroup.eu Mon May 15 15:07:58 2023
From: Christophe Leroy <christophe.leroy at csgroup.eu>
Date: Mon, 15 May 2023 16:07:15 +0200
Subject:[For 4.19/4.14] spi: fsl-cpm: Use 16 bit mode for large transfers with even size
To: gregkh at linuxfoundation.org, stable at vger.kernel.org
Cc: Christophe Leroy <christophe.leroy at csgroup.eu>, linux-kernel at vger.kernel.org, linuxppc-dev at lists.ozlabs.org, Mark Brown <broonie at kernel.org>
Message-ID: <9363da33f54e9862b4b59c0ed97924ca7265f7a4.1684158520.git.christophe.leroy at csgroup.eu>

From: Christophe Leroy <christophe.leroy at csgroup.eu>

(cherry picked from upstream fc96ec826bced75cc6b9c07a4ac44bbf651337ab)

On CPM, the RISC core is a lot more efficiant when doing transfers
in 16-bits chunks than in 8-bits chunks, but unfortunately the
words need to be byte swapped as seen in a previous commit.

So, for large tranfers with an even size, allocate a temporary tx
buffer and byte-swap data before and after transfer.

This change allows setting higher speed for transfer. For instance
on an MPC 8xx (CPM1 comms RISC processor), the documentation tells
that transfer in byte mode at 1 kbit/s uses 0.200% of CPM load
at 25 MHz while a word transfer at the same speed uses 0.032%
of CPM load. This means the speed can be 6 times higher in
word mode for the same CPM load.

For the time being, only do it on CPM1 as there must be a
trade-off between the CPM load reduction and the CPU load required
to byte swap the data.

Signed-off-by: Christophe Leroy <christophe.leroy at csgroup.eu>
Link: https://lore.kernel.org/r/f2e981f20f92dd28983c3949702a09248c23845c.1680371809.git.christophe.leroy@csgroup.eu
Signed-off-by: Mark Brown <broonie at kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh at linuxfoundation.org>
---
 drivers/spi/spi-fsl-cpm.c |   23 +++++++++++++++++++++++
 drivers/spi/spi-fsl-spi.c |    3 +++
 2 files changed, 26 insertions(+)

--- a/drivers/spi/spi-fsl-cpm.c
+++ b/drivers/spi/spi-fsl-cpm.c
@@ -25,6 +25,7 @@
 #include <linux/spi/spi.h>
 #include <linux/types.h>
 #include <linux/platform_device.h>
+#include <linux/byteorder/generic.h>
 
 #include "spi-fsl-cpm.h"
 #include "spi-fsl-lib.h"
@@ -124,6 +125,21 @@ int fsl_spi_cpm_bufs(struct mpc8xxx_spi
 		mspi->rx_dma = mspi->dma_dummy_rx;
 		mspi->map_rx_dma = 0;
 	}
+	if (t->bits_per_word == 16 && t->tx_buf) {
+		const u16 *src = t->tx_buf;
+		u16 *dst;
+		int i;
+
+		dst = kmalloc(t->len, GFP_KERNEL);
+		if (!dst)
+			return -ENOMEM;
+
+		for (i = 0; i < t->len >> 1; i++)
+			dst[i] = cpu_to_le16p(src + i);
+
+		mspi->tx = dst;
+		mspi->map_tx_dma = 1;
+	}
 
 	if (mspi->map_tx_dma) {
 		void *nonconst_tx = (void *)mspi->tx; /* shut up gcc */
@@ -177,6 +193,13 @@ void fsl_spi_cpm_bufs_complete(struct mp
 	if (mspi->map_rx_dma)
 		dma_unmap_single(dev, mspi->rx_dma, t->len, DMA_FROM_DEVICE);
 	mspi->xfer_in_progress = NULL;
+
+	if (t->bits_per_word == 16 && t->rx_buf) {
+		int i;
+
+		for (i = 0; i < t->len; i += 2)
+			le16_to_cpus(t->rx_buf + i);
+	}
 }
 EXPORT_SYMBOL_GPL(fsl_spi_cpm_bufs_complete);
 
--- a/drivers/spi/spi-fsl-spi.c
+++ b/drivers/spi/spi-fsl-spi.c
@@ -366,6 +366,9 @@ static int fsl_spi_do_one_msg(struct spi
 				return -EINVAL;
 			if (t->bits_per_word == 16 || t->bits_per_word == 32)
 				t->bits_per_word = 8; /* pretend its 8 bits */
+			if (t->bits_per_word == 8 && t->len >= 256 &&
+			    (mpc8xxx_spi->flags & SPI_CPM1))
+				t->bits_per_word = 16;
 		}
 	}
 


Patches currently in stable-queue which might be from christophe.leroy at csgroup.eu are

queue-4.14/spi-fsl-cpm-use-16-bit-mode-for-large-transfers-with-even-size.patch
queue-4.14/spi-fsl-spi-re-organise-transfer-bits_per_word-adaptation.patch
queue-4.14/spi-spi-fsl-spi-automatically-adapt-bits-per-word-in-cpu-mode.patch