mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
From: Nathan Chancellor <nathan@kernel.org>
To: Calvin Owens <calvin@wbinvd.org>
Cc: linux-kernel@vger.kernel.org,
	Masahiro Yamada <masahiroy@kernel.org>,
	Nicolas Schier <nicolas@fjasle.eu>,
	linux-kbuild@vger.kernel.org
Subject: Re: [PATCH] kbuild: buildtar: Add arm support
Date: Fri, 12 Apr 2024 14:47:47 -0700	[thread overview]
Message-ID: <20240412214747.GE2252629@dev-arch.thelio-3990X> (raw)
In-Reply-To: <ZhmKPlMl_WJOPIXs@mozart.vkv.me>

On Fri, Apr 12, 2024 at 12:23:42PM -0700, Calvin Owens wrote:
> On Wednesday 04/10 at 15:56 -0700, Calvin Owens wrote:
> > On Wednesday 04/10 at 10:04 -0700, Nathan Chancellor wrote:
> > > Hi Calvin,
> > > 
> > > Thanks for the patch!
> > > 
> > > On Tue, Apr 09, 2024 at 10:17:07AM -0700, Calvin Owens wrote:
> > > > Make 'make tar-pkg' and friends work on 32-bit arm.
> > > > 
> > > > Signed-off-by: Calvin Owens <calvin@wbinvd.org>
> > > 
> > > Technically speaking, buildtar works for 32-bit ARM right now (I use it
> > > almost daily), this is just explicitly adding it to the supported list
> > > to avoid the warning and putting zImage at vmlinuz-${KERNELRELEASE}
> > > instead of vmlinux-kbuild-${KERNELRELEASE}, right?
> > 
> > Exactly. I assumed (maybe incorrectly?) the vmlinux-kbuild-* name was
> > generic "unimplemented" filler that was meant to be replaced. It seems
> > like the vmlinuz-* naming has sort of become a de facto standard in the
> > tar-pkgs.

I think your first assumption is likely correct although I have not
looked back at the history to confirm that. I am not as sure on the
second statement, mainly just because not all kernel images are
compressed so they wouldn't necessarily make sense as vmlinuz. I think
it just happens that many of the most popular architectures have default
compressed kernel images.

> > The context for me is a pile of scripts that build kernels and boot them
> > with QEMU on arm and arm64: it's convenient if the tar-pkg structure is
> > consistent between the two (and across other architectures too).

Yes, I think including that reasoning in the commit message makes sense,
since it is justification for changing the status quo.

> > > That said, looks mostly fine to me, one comment below.
> > > 
> > > Before:
> > > 
> > >   './System.map' -> 'tar-install/boot/System.map-6.9.0-rc3-00023-g2c71fdf02a95'
> > >   '.config' -> 'tar-install/boot/config-6.9.0-rc3-00023-g2c71fdf02a95'
> > >   './vmlinux' -> 'tar-install/boot/vmlinux-6.9.0-rc3-00023-g2c71fdf02a95'
> > >   'arch/arm/boot/zImage' -> 'tar-install/boot/vmlinux-kbuild-6.9.0-rc3-00023-g2c71fdf02a95'
> > > 
> > >   ** ** **  WARNING  ** ** **
> > > 
> > >   Your architecture did not define any architecture-dependent files
> > >   to be placed into the tarball. Please add those to scripts/package/buildtar ...
> > > 
> > > After:
> > > 
> > >   './System.map' -> 'tar-install/boot/System.map-6.9.0-rc3-00023-g2c71fdf02a95-dirty'
> > >   '.config' -> 'tar-install/boot/config-6.9.0-rc3-00023-g2c71fdf02a95-dirty'
> > >   './vmlinux' -> 'tar-install/boot/vmlinux-6.9.0-rc3-00023-g2c71fdf02a95-dirty'
> > >   './arch/arm/boot/zImage' -> 'tar-install/boot/vmlinuz-6.9.0-rc3-00023-g2c71fdf02a95-dirty'
> > > 
> > > and the location of zImage is the only thing that changes as expected.
> > > 
> > > > ---
> > > >  scripts/package/buildtar | 3 +++
> > > >  1 file changed, 3 insertions(+)
> > > > 
> > > > diff --git a/scripts/package/buildtar b/scripts/package/buildtar
> > > > index 72c91a1b832f..0939f9eabbf2 100755
> > > > --- a/scripts/package/buildtar
> > > > +++ b/scripts/package/buildtar
> > > > @@ -101,6 +101,9 @@ case "${ARCH}" in
> > > >  			fi
> > > >  		done
> > > >  		;;
> > > > +	arm)
> > > > +		[ -f "${objtree}/arch/arm/boot/zImage" ] && cp -v -- "${objtree}/arch/arm/boot/zImage" "${tmpdir}/boot/vmlinuz-${KERNELRELEASE}"
> > > 
> > > While it probably does not matter too much, it would be more proper to
> > > make this
> > > 
> > >   [ -f "${KBUILD_IMAGE}" ] && cp -v -- "${KBUILD_IMAGE}" "${tmpdir}/boot/vmlinuz-${KERNELRELEASE}"
> > > 
> > > as the current line does not work with CONFIG_XIP_KERNEL=y, since zImage
> > > does not exist (KBUILD_IMAGE is arch/arm/boot/xipImage with this
> > > configuration)
> > > 
> > >   $ ls arch/arm/boot
> > >   compressed  dts  xipImage
> > > 
> > > resulting in buildtar failing because
> > > 
> > >   [ -f "${objtree}/arch/arm/boot/zImage" ]
> > > 
> > > fails and is the last statement that runs in the script (and the tar
> > > package is not really complete in this configuration anyways).
> > > 
> > > Prior to this change, the correct image would get placed into the
> > > tarball.
> > > 
> > >   'arch/arm/boot/xipImage' -> 'tar-install/boot/vmlinux-kbuild-6.9.0-rc3-00023-g2c71fdf02a95'
> > 
> > Makes sense, thanks. Although...
> > 
> > > > +		;;
> > > >  	*)
> > > >  		[ -f "${KBUILD_IMAGE}" ] && cp -v -- "${KBUILD_IMAGE}" "${tmpdir}/boot/vmlinux-kbuild-${KERNELRELEASE}"
> > > >  		echo "" >&2
> > 
> > ...it ends up looking almost identical to the default case. Does it make
> > make more sense to change the destination in the default case and remove
> > the warning? I'm not sure if anything might rely on the current
> > behavior, it goes all the way back (git sha 6d983feab809).
> 
> What I'm trying to say is: using KBUILD_IMAGE like you suggest allows
> more of the existing cases to be combined, like the below (and probably
> alpha too, at least).

I see you already sent v2, which I will review shortly, but doing this
change certainly seems reasonable to me. We could add a comment above it
like

  # Architectures with just a compressed KBUILD_IMAGE

> ---8<---
> 
> diff --git a/scripts/package/buildtar b/scripts/package/buildtar
> index 72c91a1b832f..66b4d8d308b6 100755
> --- a/scripts/package/buildtar
> +++ b/scripts/package/buildtar
> @@ -54,8 +54,8 @@ cp -v -- "${objtree}/vmlinux" "${tmpdir}/boot/vmlinux-${KERNELRELEASE}"
>  # Install arch-specific kernel image(s)
>  #
>  case "${ARCH}" in
> -	x86|i386|x86_64)
> -		[ -f "${objtree}/arch/x86/boot/bzImage" ] && cp -v -- "${objtree}/arch/x86/boot/bzImage" "${tmpdir}/boot/vmlinuz-${KERNELRELEASE}"
> +	x86|i386|x86_64|arm)
> +		[ -f "${KBUILD_IMAGE}" ] && cp -v -- "${KBUILD_IMAGE}" "${tmpdir}/boot/vmlinuz-${KERNELRELEASE}"
>  		;;
>  	alpha)
>  		[ -f "${objtree}/arch/alpha/boot/vmlinux.gz" ] && cp -v -- "${objtree}/arch/alpha/boot/vmlinux.gz" "${tmpdir}/boot/vmlinuz-${KERNELRELEASE}"

  reply	other threads:[~2024-04-12 21:47 UTC|newest]

Thread overview: 10+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2024-04-09 17:17 Calvin Owens
2024-04-10 17:04 ` Nathan Chancellor
2024-04-10 22:56   ` Calvin Owens
2024-04-12 19:23     ` Calvin Owens
2024-04-12 21:47       ` Nathan Chancellor [this message]
2024-04-12 19:26     ` [PATCH v2] kbuild: buildtar: Add explicit 32-bit " Calvin Owens
2024-04-12 21:48       ` Nathan Chancellor
2024-06-21  2:25       ` [RESEND PATCH " Calvin Owens
2024-06-21 14:55         ` Nathan Chancellor
2024-06-22  2:55           ` Calvin Owens

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20240412214747.GE2252629@dev-arch.thelio-3990X \
    --to=nathan@kernel.org \
    --cc=calvin@wbinvd.org \
    --cc=linux-kbuild@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=masahiroy@kernel.org \
    --cc=nicolas@fjasle.eu \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®