From 12c4efe3509b8018e76ea3ebda8227cb53bf5887 Mon Sep 17 00:00:00 2001
From: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Date: Sun, 8 May 2022 18:55:41 +0100
Subject: [PATCH] iio: core: Fix IIO_ALIGN and rename as it was not sufficiently large
Mime-version: 1.0
Content-type: text/plain; charset=UTF-8
Content-transfer-encoding: 8bit
Git-commit: 12c4efe3509b8018e76ea3ebda8227cb53bf5887
Patch-mainline: v6.0-rc1
References: git-fixes

Discussion of the series:
https://lore.kernel.org/all/20220405135758.774016-1-catalin.marinas@arm.com/
mm, arm64: Reduce ARCH_KMALLOC_MINALIGN brought to my attention that
our current IIO usage of L1CACHE_ALIGN is insufficient as their are Arm
platforms out their with non coherent DMA and larger cache lines at
at higher levels of their cache hierarchy.

Rename the define to make it's purpose more explicit. It will be used
much more widely going forwards (to replace incorrect ____cacheline_aligned
markings.

Note this patch will greatly reduce the padding on some architectures
that have smaller requirements for DMA safe buffers.

The history of changing values of ARCH_KMALLOC_MINALIGN via
ARCH_DMA_MINALIGN on arm64 is rather complex. I'm not tagging this
as fixing a particular patch from that route as it's not clear what to tag.

Most recently a change to bring them back inline was reverted because
of some Qualcomm Kryo cores with an L2 cache with 128-byte lines
sitting above the point of coherency.

c1132702c71f Revert "arm64: cache: Lower ARCH_DMA_MINALIGN to 64 (L1_CACHE_BYTES)"
That reverts:
65688d2a05de arm64: cache: Lower ARCH_DMA_MINALIGN to 64 (L1_CACHE_BYTES) which
refers to the change originally being motivated by Thunder x1 performance
rather than correctness.

Fixes: 6f7c8ee585e9d ("staging:iio: Add ability to allocate private data space to iio_allocate_device")
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Acked-by: Nuno Sá <nuno.sa@analog.com>
Link: https://lore.kernel.org/r/20220508175712.647246-2-jic23@kernel.org
Acked-by: Takashi Iwai <tiwai@suse.de>

---
 drivers/iio/industrialio-core.c |    4 ++--
 include/linux/iio/iio.h         |   14 ++++++++++----
 2 files changed, 12 insertions(+), 6 deletions(-)

--- a/drivers/iio/industrialio-core.c
+++ b/drivers/iio/industrialio-core.c
@@ -1437,11 +1437,11 @@ struct iio_dev *iio_device_alloc(int siz
 
 	alloc_size = sizeof(struct iio_dev);
 	if (sizeof_priv) {
-		alloc_size = ALIGN(alloc_size, IIO_ALIGN);
+		alloc_size = ALIGN(alloc_size, IIO_DMA_MINALIGN);
 		alloc_size += sizeof_priv;
 	}
 	/* ensure 32-byte alignment of whole construct ? */
-	alloc_size += IIO_ALIGN - 1;
+	alloc_size += IIO_DMA_MINALIGN - 1;
 
 	dev = kzalloc(alloc_size, GFP_KERNEL);
 
--- a/include/linux/iio/iio.h
+++ b/include/linux/iio/iio.h
@@ -9,6 +9,7 @@
 
 #include <linux/device.h>
 #include <linux/cdev.h>
+#include <linux/slab.h>
 #include <linux/iio/types.h>
 #include <linux/of.h>
 /* IIO TODO LIST */
@@ -674,19 +675,24 @@ static inline void *iio_device_get_drvda
 	return dev_get_drvdata(&indio_dev->dev);
 }
 
-/* Can we make this smaller? */
-#define IIO_ALIGN L1_CACHE_BYTES
+/*
+ * Used to ensure the iio_priv() structure is aligned to allow that structure
+ * to in turn include IIO_DMA_MINALIGN'd elements such as buffers which
+ * must not share  cachelines with the rest of the structure, thus making
+ * them safe for use with non-coherent DMA.
+ */
+#define IIO_DMA_MINALIGN ARCH_KMALLOC_MINALIGN
 struct iio_dev *iio_device_alloc(int sizeof_priv);
 
 static inline void *iio_priv(const struct iio_dev *indio_dev)
 {
-	return (char *)indio_dev + ALIGN(sizeof(struct iio_dev), IIO_ALIGN);
+	return (char *)indio_dev + ALIGN(sizeof(struct iio_dev), IIO_DMA_MINALIGN);
 }
 
 static inline struct iio_dev *iio_priv_to_dev(void *priv)
 {
 	return (struct iio_dev *)((char *)priv -
-				  ALIGN(sizeof(struct iio_dev), IIO_ALIGN));
+				  ALIGN(sizeof(struct iio_dev), IIO_DMA_MINALIGN));
 }
 
 void iio_device_free(struct iio_dev *indio_dev);