diff mbox series

[1/2] linux: show compression options only on some architectures

Message ID 20200924122222.3449188-1-thomas.petazzoni@bootlin.com
State Changes Requested
Headers show
Series [1/2] linux: show compression options only on some architectures | expand

Commit Message

Thomas Petazzoni Sept. 24, 2020, 12:22 p.m. UTC
Our Linux kernel compression options have the effect of enabling
CONFIG_KERNEL_{GZIP,LZMA,LZ4,...}. However, those configuration
options only exist on a subset of the CPU architectures, and it is
confusing to have such a choice visible when it in fact has no
effect. For example on ARM64, it has absolutely no effect, and the way
to create a compressed image is different, and the image is not
self-extractible.

So, we introduce a BR2_LINUX_KERNEL_ARCH_HAS_COMPRESSION to know which
architectures have compression support. Of course, it's never going to
be perfect as the set of architectures supporting compression might
change over time. But at least, it will allow us to hide this option
on architectures that do not at this point support compression.

Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
---
 linux/Config.in | 18 ++++++++++++++++++
 1 file changed, 18 insertions(+)

Comments

Yann E. MORIN Sept. 25, 2020, 7:37 a.m. UTC | #1
Thomas, All,

On 2020-09-24 14:22 +0200, Thomas Petazzoni spake thusly:
> Our Linux kernel compression options have the effect of enabling
> CONFIG_KERNEL_{GZIP,LZMA,LZ4,...}. However, those configuration
> options only exist on a subset of the CPU architectures, and it is
> confusing to have such a choice visible when it in fact has no
> effect. For example on ARM64, it has absolutely no effect, and the way
> to create a compressed image is different, and the image is not
> self-extractible.
> 
> So, we introduce a BR2_LINUX_KERNEL_ARCH_HAS_COMPRESSION to know which
> architectures have compression support. Of course, it's never going to
> be perfect as the set of architectures supporting compression might
> change over time. But at least, it will allow us to hide this option
> on architectures that do not at this point support compression.
> 
> Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
> ---
>  linux/Config.in | 18 ++++++++++++++++++
>  1 file changed, 18 insertions(+)
> 
> diff --git a/linux/Config.in b/linux/Config.in
> index 49da2b81b2..9f2af6da5b 100644
> --- a/linux/Config.in
> +++ b/linux/Config.in
> @@ -299,6 +299,22 @@ endchoice
>  # Kernel compression format
>  #
>  
> +# Indicates whether this architecture has support for built-in kernel
> +# compression, i.e architectures that select one of
> +# HAVE_KERNEL_{GZIP,LZMA,LZ4,LZO,XZ,UNCOMPRESSED}
> +config BR2_LINUX_KERNEL_ARCH_HAS_COMPRESSION
> +	bool
> +	default y if BR2_arcle || BR2_arceb
> +	default y if BR2_arm || BR2_armeb
> +	default y if BR2_csky
> +	default y if BR2_i386
> +	default y if BR2_mips || BR2_mipsel || BR2_mips64 || BR2_mips64el
> +	default y if BR2_powerpc || BR2_powerpc64 || BR2_powerpc64le
> +	default y if BR2_sh
> +	default y if BR2_x86_64
> +
> +if BR2_LINUX_KERNEL_ARCH_HAS_COMPRESSION

So, if the arch does not support kernel image compression, we will have
none of the compression symbols defined, which means that, in linux.mk,
we would have:

    104 LINUX_COMPRESSION_OPT_$(BR2_LINUX_KERNEL_GZIP) += CONFIG_KERNEL_GZIP
    105 LINUX_COMPRESSION_OPT_$(BR2_LINUX_KERNEL_LZ4) += CONFIG_KERNEL_LZ4
    106 LINUX_COMPRESSION_OPT_$(BR2_LINUX_KERNEL_LZMA) += CONFIG_KERNEL_LZMA
    107 LINUX_COMPRESSION_OPT_$(BR2_LINUX_KERNEL_LZO) += CONFIG_KERNEL_LZO
    108 LINUX_COMPRESSION_OPT_$(BR2_LINUX_KERNEL_XZ) += CONFIG_KERNEL_XZ
    109 LINUX_COMPRESSION_OPT_$(BR2_LINUX_KERNEL_ZSTD) += CONFIG_KERNEL_ZSTD
    [--SNIP--]
    319 define LINUX_KCONFIG_FIXUP_CMDS
    322     $(call KCONFIG_ENABLE_OPT,$(strip $(LINUX_COMPRESSION_OPT_y)))

But then, LINUX_COMPRESSION_OPT_y would be empty, so we'd call KCONFIG_ENABLE_OPT
with an empty string, which would expand to:

    $(call KCONFIG_MUNGE_DOT_CONFIG, $(1), $(1)=y, $(2))

... and replacing the parameters would give (I'm pretty sure you'll see
where we're going now):

    $(call KCONFIG_MUNGE_DOT_CONFIG, , =y, $(2))

When KCONFIG_MUNGE_DOT_CONFIG is expanded, we'd get:

    $(SED) "/\\<$(strip $(1))\\>/d" $(call KCONFIG_DOT_CONFIG,$(3))
    echo '$(strip $(2))' >> $(call KCONFIG_DOT_CONFIG,$(3))

... which would eventualy lead us to:

    $(SED) "/\\<)\\>/d" /some/.config-file
    echo '=y' >> /some/.config-file

And to be honest, I am not sure that last line would generate a fully
legit kconfig setting... ;-)

Of course, running 'make oldconfig' on that modified .config would drop
invalid lines, so you would not notivce the issue, but still this is not
correct...

Regards,
Yann E. MORIN.

>  choice
>  	prompt "Kernel compression format"
>  	help
> @@ -326,6 +342,8 @@ config BR2_LINUX_KERNEL_ZSTD
>  
>  endchoice
>  
> +endif
> +
>  config BR2_LINUX_KERNEL_IMAGE_TARGET_NAME
>  	string "Kernel image target name"
>  	depends on BR2_LINUX_KERNEL_IMAGE_TARGET_CUSTOM
> -- 
> 2.26.2
> 
> _______________________________________________
> buildroot mailing list
> buildroot@busybox.net
> http://lists.busybox.net/mailman/listinfo/buildroot
Thomas Petazzoni Sept. 25, 2020, 7:45 a.m. UTC | #2
Hello,

On Fri, 25 Sep 2020 09:37:58 +0200
"Yann E. MORIN" <yann.morin.1998@free.fr> wrote:

> So, if the arch does not support kernel image compression, we will have
> none of the compression symbols defined, which means that, in linux.mk,
> we would have:
> 
>     104 LINUX_COMPRESSION_OPT_$(BR2_LINUX_KERNEL_GZIP) += CONFIG_KERNEL_GZIP
>     105 LINUX_COMPRESSION_OPT_$(BR2_LINUX_KERNEL_LZ4) += CONFIG_KERNEL_LZ4
>     106 LINUX_COMPRESSION_OPT_$(BR2_LINUX_KERNEL_LZMA) += CONFIG_KERNEL_LZMA
>     107 LINUX_COMPRESSION_OPT_$(BR2_LINUX_KERNEL_LZO) += CONFIG_KERNEL_LZO
>     108 LINUX_COMPRESSION_OPT_$(BR2_LINUX_KERNEL_XZ) += CONFIG_KERNEL_XZ
>     109 LINUX_COMPRESSION_OPT_$(BR2_LINUX_KERNEL_ZSTD) += CONFIG_KERNEL_ZSTD
>     [--SNIP--]
>     319 define LINUX_KCONFIG_FIXUP_CMDS
>     322     $(call KCONFIG_ENABLE_OPT,$(strip $(LINUX_COMPRESSION_OPT_y)))
> 
> But then, LINUX_COMPRESSION_OPT_y would be empty, so we'd call KCONFIG_ENABLE_OPT
> with an empty string, which would expand to:
> 
>     $(call KCONFIG_MUNGE_DOT_CONFIG, $(1), $(1)=y, $(2))
> 
> ... and replacing the parameters would give (I'm pretty sure you'll see
> where we're going now):
> 
>     $(call KCONFIG_MUNGE_DOT_CONFIG, , =y, $(2))
> 
> When KCONFIG_MUNGE_DOT_CONFIG is expanded, we'd get:
> 
>     $(SED) "/\\<$(strip $(1))\\>/d" $(call KCONFIG_DOT_CONFIG,$(3))
>     echo '$(strip $(2))' >> $(call KCONFIG_DOT_CONFIG,$(3))
> 
> ... which would eventualy lead us to:
> 
>     $(SED) "/\\<)\\>/d" /some/.config-file
>     echo '=y' >> /some/.config-file
> 
> And to be honest, I am not sure that last line would generate a fully
> legit kconfig setting... ;-)
> 
> Of course, running 'make oldconfig' on that modified .config would drop
> invalid lines, so you would not notivce the issue, but still this is not
> correct...

Thanks for spotting this, I'll fix in v2. That being said, one can
notice that you're pointing out that this bogus =y line will be
dropped... just as equally as the bogus CONFIG_KERNEL_GZIP=y that we
have today gets dropped on arm64 and other architectures for which
those CONFIG_KERNEL_{GZIP,LZMA,LZ4,...} options don't exist :-)

Thomas
Yann E. MORIN Sept. 25, 2020, 8:14 a.m. UTC | #3
Thomas, All,

On 2020-09-25 09:45 +0200, Thomas Petazzoni spake thusly:
> On Fri, 25 Sep 2020 09:37:58 +0200
> "Yann E. MORIN" <yann.morin.1998@free.fr> wrote:
[--SNIP--]
> >     $(SED) "/\\<)\\>/d" /some/.config-file
> >     echo '=y' >> /some/.config-file
> > And to be honest, I am not sure that last line would generate a fully
> > legit kconfig setting... ;-)
> > Of course, running 'make oldconfig' on that modified .config would drop
> > invalid lines, so you would not notivce the issue, but still this is not
> > correct...
> Thanks for spotting this, I'll fix in v2. That being said, one can
> notice that you're pointing out that this bogus =y line will be
> dropped... just as equally as the bogus CONFIG_KERNEL_GZIP=y that we
> have today gets dropped on arm64 and other architectures for which
> those CONFIG_KERNEL_{GZIP,LZMA,LZ4,...} options don't exist :-)

Yes, but an unknown symbol assignment is still valid syntax, and
silently discarded by kconfig, because that is the goal of running
oldconfig: assigning new symbols and dropping old ones.

But the '=y' is not even syntactically valid. kconfig is still lenient
about that, though (but it does warn, doesn't it?), but this is not a
reason to generate an invalid file.

Regards,
Yann E. MORIN.
diff mbox series

Patch

diff --git a/linux/Config.in b/linux/Config.in
index 49da2b81b2..9f2af6da5b 100644
--- a/linux/Config.in
+++ b/linux/Config.in
@@ -299,6 +299,22 @@  endchoice
 # Kernel compression format
 #
 
+# Indicates whether this architecture has support for built-in kernel
+# compression, i.e architectures that select one of
+# HAVE_KERNEL_{GZIP,LZMA,LZ4,LZO,XZ,UNCOMPRESSED}
+config BR2_LINUX_KERNEL_ARCH_HAS_COMPRESSION
+	bool
+	default y if BR2_arcle || BR2_arceb
+	default y if BR2_arm || BR2_armeb
+	default y if BR2_csky
+	default y if BR2_i386
+	default y if BR2_mips || BR2_mipsel || BR2_mips64 || BR2_mips64el
+	default y if BR2_powerpc || BR2_powerpc64 || BR2_powerpc64le
+	default y if BR2_sh
+	default y if BR2_x86_64
+
+if BR2_LINUX_KERNEL_ARCH_HAS_COMPRESSION
+
 choice
 	prompt "Kernel compression format"
 	help
@@ -326,6 +342,8 @@  config BR2_LINUX_KERNEL_ZSTD
 
 endchoice
 
+endif
+
 config BR2_LINUX_KERNEL_IMAGE_TARGET_NAME
 	string "Kernel image target name"
 	depends on BR2_LINUX_KERNEL_IMAGE_TARGET_CUSTOM