[RFC PATCH] x86/boot: use zstd -19 if building for 32-bit

[RFC PATCH] x86/boot: use zstd -19 if building for 32-bit

[Mingcong Bai]

When using native build environments (that is, 32-bit x86 chroot/container
as build environment for 32-bit x86 binaries), `zstd -22' requests more
virtual memory than that could be allowed on 32-bit hosts.

Introduce `cmd_zstd_with_size' for CONFIG_X86_32 builds.

This is quite possibly just stupid... But there doesn't seem to be a way
to detect the host environment. So I'm sending this patch as a means to
ask for your opinions.

Any help would be greated appreciated.

Signed-off-by: Mingcong Bai <jeffbai@aosc.io>
---
 arch/x86/boot/compressed/Makefile | 5 +++++
 scripts/Makefile.lib              | 3 +++
 2 files changed, 8 insertions(+)

diff --git a/arch/x86/boot/compressed/Makefile b/arch/x86/boot/compressed/Makefile
index 07e0e64b9a98..2261fb0d4d1d 100644
--- a/arch/x86/boot/compressed/Makefile
+++ b/arch/x86/boot/compressed/Makefile
@@ -143,8 +143,13 @@ $(obj)/vmlinux.bin.lzo: $(vmlinux.bin.all-y) FORCE
 	$(call if_changed,lzo_with_size)
 $(obj)/vmlinux.bin.lz4: $(vmlinux.bin.all-y) FORCE
 	$(call if_changed,lz4_with_size)
+ifdef CONFIG_X86_32
+$(obj)/vmlinux.bin.zst: $(vmlinux.bin.all-y) FORCE
+	$(call if_changed,zstd_with_size)
+else
 $(obj)/vmlinux.bin.zst: $(vmlinux.bin.all-y) FORCE
 	$(call if_changed,zstd22_with_size)
+endif
 
 suffix-$(CONFIG_KERNEL_GZIP)	:= gz
 suffix-$(CONFIG_KERNEL_BZIP2)	:= bz2
diff --git a/scripts/Makefile.lib b/scripts/Makefile.lib
index 0718e39cedda..09910dc57af3 100644
--- a/scripts/Makefile.lib
+++ b/scripts/Makefile.lib
@@ -464,6 +464,9 @@ quiet_cmd_xzmisc = XZMISC  $@
 quiet_cmd_zstd = ZSTD    $@
       cmd_zstd = cat $(real-prereqs) | $(ZSTD) -19 > $@
 
+quiet_cmd_zstd = ZSTD    $@
+      cmd_zstd_with_size = { cat $(real-prereqs) | $(ZSTD) -19; $(size_append); } > $@
+
 quiet_cmd_zstd22 = ZSTD22  $@
       cmd_zstd22 = cat $(real-prereqs) | $(ZSTD) -22 --ultra > $@
 
-- 
2.55.0

Re: [RFC PATCH] x86/boot: use zstd -19 if building for 32-bit

[Xi Ruoyao]

On Tue, 2026-07-28 at 22:31 +0800, Mingcong Bai wrote:
> When using native build environments (that is, 32-bit x86 chroot/container
> as build environment for 32-bit x86 binaries), `zstd -22' requests more
> virtual memory than that could be allowed on 32-bit hosts.
> 
> Introduce `cmd_zstd_with_size' for CONFIG_X86_32 builds.
> 
> This is quite possibly just stupid... But there doesn't seem to be a way
> to detect the host environment. So I'm sending this patch as a means to
> ask for your opinions.

"readelf -d $(which zstd) -h | grep ELF64" ?

But it's also stupid, if someone is using a script wrapping the actual
zstd executable or cross-compiling from a non-ELF platform things will
still blow up.

-- 
Xi Ruoyao <xry111@xry111.site>

Re: [RFC PATCH] x86/boot: use zstd -19 if building for 32-bit

[Ard Biesheuvel]

On Tue, 28 Jul 2026, at 17:31, Mingcong Bai wrote:
> When using native build environments (that is, 32-bit x86 chroot/container
> as build environment for 32-bit x86 binaries), `zstd -22' requests more
> virtual memory than that could be allowed on 32-bit hosts.
>
> Introduce `cmd_zstd_with_size' for CONFIG_X86_32 builds.
>
> This is quite possibly just stupid... But there doesn't seem to be a way
> to detect the host environment. So I'm sending this patch as a means to
> ask for your opinions.
>
> Any help would be greated appreciated.
>
> Signed-off-by: Mingcong Bai <jeffbai@aosc.io>
> ---
>  arch/x86/boot/compressed/Makefile | 5 +++++
>  scripts/Makefile.lib              | 3 +++
>  2 files changed, 8 insertions(+)
>

Doesn't this mean ZSTD -22 is generally broken on 32-bit? How is this
specific to the kernel?

Re: [RFC PATCH] x86/boot: use zstd -19 if building for 32-bit

[H. Peter Anvin]

On July 28, 2026 7:31:35 AM PDT, Mingcong Bai <jeffbai@aosc.io> wrote:
>When using native build environments (that is, 32-bit x86 chroot/container
>as build environment for 32-bit x86 binaries), `zstd -22' requests more
>virtual memory than that could be allowed on 32-bit hosts.
>
>Introduce `cmd_zstd_with_size' for CONFIG_X86_32 builds.
>
>This is quite possibly just stupid... But there doesn't seem to be a way
>to detect the host environment. So I'm sending this patch as a means to
>ask for your opinions.
>
>Any help would be greated appreciated.
>
>Signed-off-by: Mingcong Bai <jeffbai@aosc.io>
>---
> arch/x86/boot/compressed/Makefile | 5 +++++
> scripts/Makefile.lib              | 3 +++
> 2 files changed, 8 insertions(+)
>
>diff --git a/arch/x86/boot/compressed/Makefile b/arch/x86/boot/compressed/Makefile
>index 07e0e64b9a98..2261fb0d4d1d 100644
>--- a/arch/x86/boot/compressed/Makefile
>+++ b/arch/x86/boot/compressed/Makefile
>@@ -143,8 +143,13 @@ $(obj)/vmlinux.bin.lzo: $(vmlinux.bin.all-y) FORCE
> 	$(call if_changed,lzo_with_size)
> $(obj)/vmlinux.bin.lz4: $(vmlinux.bin.all-y) FORCE
> 	$(call if_changed,lz4_with_size)
>+ifdef CONFIG_X86_32
>+$(obj)/vmlinux.bin.zst: $(vmlinux.bin.all-y) FORCE
>+	$(call if_changed,zstd_with_size)
>+else
> $(obj)/vmlinux.bin.zst: $(vmlinux.bin.all-y) FORCE
> 	$(call if_changed,zstd22_with_size)
>+endif
> 
> suffix-$(CONFIG_KERNEL_GZIP)	:= gz
> suffix-$(CONFIG_KERNEL_BZIP2)	:= bz2
>diff --git a/scripts/Makefile.lib b/scripts/Makefile.lib
>index 0718e39cedda..09910dc57af3 100644
>--- a/scripts/Makefile.lib
>+++ b/scripts/Makefile.lib
>@@ -464,6 +464,9 @@ quiet_cmd_xzmisc = XZMISC  $@
> quiet_cmd_zstd = ZSTD    $@
>       cmd_zstd = cat $(real-prereqs) | $(ZSTD) -19 > $@
> 
>+quiet_cmd_zstd = ZSTD    $@
>+      cmd_zstd_with_size = { cat $(real-prereqs) | $(ZSTD) -19; $(size_append); } > $@
>+
> quiet_cmd_zstd22 = ZSTD22  $@
>       cmd_zstd22 = cat $(real-prereqs) | $(ZSTD) -22 --ultra > $@
> 

This is really just stupid. Basing it on the config setting is just plain wrong.

Re: [RFC PATCH] x86/boot: use zstd -19 if building for 32-bit

[Mingcong Bai]

Hi Ard,

在 2026/7/28 23:58, Ard Biesheuvel 写道:
> 
> 
> On Tue, 28 Jul 2026, at 17:31, Mingcong Bai wrote:
>> When using native build environments (that is, 32-bit x86 chroot/container
>> as build environment for 32-bit x86 binaries), `zstd -22' requests more
>> virtual memory than that could be allowed on 32-bit hosts.
>>
>> Introduce `cmd_zstd_with_size' for CONFIG_X86_32 builds.
>>
>> This is quite possibly just stupid... But there doesn't seem to be a way
>> to detect the host environment. So I'm sending this patch as a means to
>> ask for your opinions.
>>
>> Any help would be greated appreciated.
>>
>> Signed-off-by: Mingcong Bai <jeffbai@aosc.io>
>> ---
>>   arch/x86/boot/compressed/Makefile | 5 +++++
>>   scripts/Makefile.lib              | 3 +++
>>   2 files changed, 8 insertions(+)
>>
> 
> Doesn't this mean ZSTD -22 is generally broken on 32-bit? How is this
> specific to the kernel?
It's not (and remains unresolved)[1] - as I mentioned in the email, I 
sent this patch out as a means to spark some discussion on how to 
approach this problem.

Nick Terrel mentioned in commit 48f7ddf785af ("init: Add support for 
zstd compressed kernel") that "zstd22 is used for kernel compression 
because it is decompressed in a single pass, so zstd doesn't need to 
allocate a window buffer." But I wonder if it resulted in tangible 
change in compression ratio - but I can do more testing.

[1]: https://github.com/facebook/zstd/issues/1964

Best Regards,
Mingcong Bai