[PATCH] ARM: dts: socfpga: fine-tune L2 cache configuration

[Marek Vasut]

Enable double-linefill and increase prefetch offset, which gives
considerable read performance boost. The following numbers were
obtained using lmbench 3.0 bw_mem tool, for easier comparison, the
numbers are pasted in two columns. The test machine has Cyclone V
SoC running at 800MHz MPU clock and 512MiB 333MHz 16bit DDR3 DRAM.

Without patch	| With patch
$ for i in rd wr rdwr cp fwr frd fcp bzero bcopy ; do echo $i ; bw_mem 64M $i ; done
rd		| rd
64.00 526.46	| 64.00 1151.06
wr		| wr
64.00 329.95	| 64.00 346.14
rdwr		| rdwr
64.00 342.07	| 64.00 367.24
cp		| cp
64.00 239.79	| 64.00 322.47
fwr		| fwr
64.00 1027.90	| 64.00 1025.38
frd		| frd
64.00 322.36	| 64.00 641.89
fcp		| fcp
64.00 256.99	| 64.00 408.41
bzero		| bzero
64.00 1028.43	| 64.00 1025.07
bcopy		| bcopy
64.00 294.73	| 64.00 357.19

Signed-off-by: Marek Vasut <marex at denx.de>
Cc: Dinh Nguyen <dinguyen at opensource.altera.com>
---
 arch/arm/boot/dts/socfpga.dtsi | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/arch/arm/boot/dts/socfpga.dtsi b/arch/arm/boot/dts/socfpga.dtsi
index f2f74b5..d645bef 100644
--- a/arch/arm/boot/dts/socfpga.dtsi
+++ b/arch/arm/boot/dts/socfpga.dtsi
@@ -694,6 +694,11 @@
 			prefetch-data = <1>;
 			prefetch-instr = <1>;
 			arm,shared-override;
+			arm,double-linefill = <1>;
+			arm,double-linefill-incr = <0>;
+			arm,double-linefill-wrap = <1>;
+			arm,prefetch-drop = <0>;
+			arm,prefetch-offset = <7>;
 		};
 
 		mmc: dwmmc0 at ff704000 {
-- 
2.10.2