generic: 6.6: backport arm64 swiotlb default size reduction

Kernel 6.6 added dynamic SWIOTLB allocation, but with it also started
allocating 64MB of the SWIOTLB bounce buffer by default which is quite a
lot of memory on most OpenWrt devices.

Luckily in kernel 6.7 arm64 received an optimization that reduces that
default size to 1MB per 1GB of RAM if certain criteria was met.

So in order to reclaim back 63MB of RAM which brought some ipq807x devices
close to OOM under load lets backport the upstream commit.

Signed-off-by: Robert Marko <robimarko@gmail.com>

diff --git a/target/linux/generic/backport-6.6/300-v6.7-arm64-swiotlb-Reduce-the-default-size-if-no-ZONE_DMA.patch b/target/linux/generic/backport-6.6/300-v6.7-arm64-swiotlb-Reduce-the-default-size-if-no-ZONE_DMA.patch
new file mode 100644 (file)
index 0000000..6ede4d3
--- /dev/null
+++ b/target/linux/generic/backport-6.6/300-v6.7-arm64-swiotlb-Reduce-the-default-size-if-no-ZONE_DMA.patch
@@ -0,0 +1,51 @@
+From 65033574ade97afccba074d837fd269903a83a9a Mon Sep 17 00:00:00 2001
+From: Catalin Marinas <catalin.marinas@arm.com>
+Date: Thu, 5 Oct 2023 16:40:30 +0100
+Subject: [PATCH] arm64: swiotlb: Reduce the default size if no ZONE_DMA
+ bouncing needed
+
+With CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC enabled, the arm64 kernel still
+allocates the default SWIOTLB buffer (64MB) even if ZONE_DMA is disabled
+or all the RAM fits into this zone. However, this potentially wastes a
+non-negligible amount of memory on platforms with little RAM.
+
+Reduce the SWIOTLB size to 1MB per 1GB of RAM if only needed for
+kmalloc() buffer bouncing.
+
+Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
+Suggested-by: Ross Burton <ross.burton@arm.com>
+Cc: Ross Burton <ross.burton@arm.com>
+Cc: Will Deacon <will@kernel.org>
+Reviewed-by: Robin Murphy <robin.murphy@arm.com>
+---
+ arch/arm64/mm/init.c | 11 ++++++++++-
+ 1 file changed, 10 insertions(+), 1 deletion(-)
+
+--- a/arch/arm64/mm/init.c
++++ b/arch/arm64/mm/init.c
+@@ -16,6 +16,7 @@
+ #include <linux/nodemask.h>
+ #include <linux/initrd.h>
+ #include <linux/gfp.h>
++#include <linux/math.h>
+ #include <linux/memblock.h>
+ #include <linux/sort.h>
+ #include <linux/of.h>
+@@ -493,8 +494,16 @@ void __init mem_init(void)
+ {
+       bool swiotlb = max_pfn > PFN_DOWN(arm64_dma_phys_limit);
+ 
+-      if (IS_ENABLED(CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC))
++      if (IS_ENABLED(CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC) && !swiotlb) {
++              /*
++               * If no bouncing needed for ZONE_DMA, reduce the swiotlb
++               * buffer for kmalloc() bouncing to 1MB per 1GB of RAM.
++               */
++              unsigned long size =
++                      DIV_ROUND_UP(memblock_phys_mem_size(), 1024);
++              swiotlb_adjust_size(min(swiotlb_size_or_default(), size));
+               swiotlb = true;
++      }
+ 
+       swiotlb_init(swiotlb, SWIOTLB_VERBOSE);
+