2021-11-19 12:07:37

by Jubin Zhong

[permalink] [raw]
Subject: [PATCH] kbuild: Enable armthumb BCJ filter for Thumb-2 kernel

xz_wrap.sh use $SRCARCH to detect the BCJ filter. However, assigning
arm BCJ filter to Thumb-2 kernel is not optimal. In my case, about 5%
decrease of image size is observed with armthumb BCJ filter:

Test results:
hardware: QEMU emulator version 3.1.0
config: vexpress_defconfig with THUMB2_KERNEL & KERNEL_XZ on
arm BCJ: 4029808
armthumb BCJ: 3827280

Choose armthumb BCJ filter for Thumb-2 kernel to make smaller images.

Signed-off-by: Jubin Zhong <[email protected]>
---
lib/decompress_unxz.c | 3 +++
scripts/xz_wrap.sh | 5 +++++
2 files changed, 8 insertions(+)

diff --git a/lib/decompress_unxz.c b/lib/decompress_unxz.c
index 9f4262e..7d6b952 100644
--- a/lib/decompress_unxz.c
+++ b/lib/decompress_unxz.c
@@ -131,6 +131,9 @@
#ifdef CONFIG_ARM
# define XZ_DEC_ARM
#endif
+#ifdef CONFIG_THUMB2_KERNEL
+# define XZ_DEC_ARMTHUMB
+#endif
#ifdef CONFIG_IA64
# define XZ_DEC_IA64
#endif
diff --git a/scripts/xz_wrap.sh b/scripts/xz_wrap.sh
index 76e9cbc..47409bb 100755
--- a/scripts/xz_wrap.sh
+++ b/scripts/xz_wrap.sh
@@ -8,6 +8,7 @@
# This file has been put into the public domain.
# You can do whatever you want with this file.
#
+. include/config/auto.conf

BCJ=
LZMA2OPTS=
@@ -20,4 +21,8 @@ case $SRCARCH in
sparc) BCJ=--sparc ;;
esac

+if [ -n "${CONFIG_THUMB2_KERNEL}" ];then
+ BCJ=--armthumb
+fi
+
exec $XZ --check=crc32 $BCJ --lzma2=$LZMA2OPTS,dict=32MiB
--
1.8.5.6



2021-11-19 19:56:48

by Lasse Collin

[permalink] [raw]
Subject: Re: [PATCH] kbuild: Enable armthumb BCJ filter for Thumb-2 kernel

On 2021-11-19 Jubin Zhong wrote:
> xz_wrap.sh use $SRCARCH to detect the BCJ filter. However, assigning
> arm BCJ filter to Thumb-2 kernel is not optimal. In my case, about 5%
> decrease of image size is observed with armthumb BCJ filter:
>
> Test results:
> hardware: QEMU emulator version 3.1.0
> config: vexpress_defconfig with THUMB2_KERNEL & KERNEL_XZ on
> arm BCJ: 4029808
> armthumb BCJ: 3827280
>
> Choose armthumb BCJ filter for Thumb-2 kernel to make smaller images.

I didn't test the patch but it looks reasonable to me. Below are a small
optimization idea and two very minor style suggestions.

> --- a/lib/decompress_unxz.c
> +++ b/lib/decompress_unxz.c
> @@ -131,6 +131,9 @@
> #ifdef CONFIG_ARM
> # define XZ_DEC_ARM
> #endif
> +#ifdef CONFIG_THUMB2_KERNEL
> +# define XZ_DEC_ARMTHUMB
> +#endif
> #ifdef CONFIG_IA64
> # define XZ_DEC_IA64
> #endif

If a Thumb-2 kernel will always use the ARM-Thumb BCJ filter, one can
save a few bytes from the pre-boot code by omitting the ARM BCJ filter:

--- a/lib/decompress_unxz.c
+++ b/lib/decompress_unxz.c
@@ -129,7 +129,11 @@
# define XZ_DEC_POWERPC
#endif
#ifdef CONFIG_ARM
-# define XZ_DEC_ARM
+# ifdef CONFIG_THUMB2_KERNEL
+# define XZ_DEC_ARMTHUMB
+# else
+# define XZ_DEC_ARM
+# endif
#endif
#ifdef CONFIG_IA64
# define XZ_DEC_IA64

> --- a/scripts/xz_wrap.sh
> +++ b/scripts/xz_wrap.sh
> @@ -8,6 +8,7 @@
> # This file has been put into the public domain.
> # You can do whatever you want with this file.
> #
> +. include/config/auto.conf

I suggest adding an empty line before this new line so that it is
clearly separated from the header comment.

> +if [ -n "${CONFIG_THUMB2_KERNEL}" ];then

I suggest adding a space after the semi-colon: ]; then

With or without the above modifications:

Acked-by: Lasse Collin <[email protected]>

--
Lasse Collin

2021-11-20 03:18:51

by Jubin Zhong

[permalink] [raw]
Subject: Re: [PATCH] kbuild: Enable armthumb BCJ filter for Thumb-2 kernel

> If a Thumb-2 kernel will always use the ARM-Thumb BCJ filter, one can
> save a few bytes from the pre-boot code by omitting the ARM BCJ filter:
>
> --- a/lib/decompress_unxz.c
> +++ b/lib/decompress_unxz.c
> @@ -129,7 +129,11 @@
> # define XZ_DEC_POWERPC
> #endif
> #ifdef CONFIG_ARM
> -# define XZ_DEC_ARM
> +# ifdef CONFIG_THUMB2_KERNEL
> +# define XZ_DEC_ARMTHUMB
> +# else
> +# define XZ_DEC_ARM
> +# endif
> #endif
> #ifdef CONFIG_IA64
> # define XZ_DEC_IA64
>
>> --- a/scripts/xz_wrap.sh
>> +++ b/scripts/xz_wrap.sh
>> @@ -8,6 +8,7 @@
>> # This file has been put into the public domain.
>> # You can do whatever you want with this file.
>> #
>> +. include/config/auto.conf
>
> I suggest adding an empty line before this new line so that it is
> clearly separated from the header comment.
>
>> +if [ -n "${CONFIG_THUMB2_KERNEL}" ];then
>
> I suggest adding a space after the semi-colon: ]; then
>
> With or without the above modifications:

Thanks for your advices. I will incorporate the above modifications and send patch v2.