Re: [PATCH RFC bootconfig] Allow forcing unconditional bootconfig processing

From: Paul E. McKenney
Date: Mon Jan 09 2023 - 19:07:58 EST


On Tue, Jan 10, 2023 at 08:56:36AM +0900, Masami Hiramatsu wrote:
> On Sun, 8 Jan 2023 20:25:01 -0800
> "Paul E. McKenney" <paulmck@xxxxxxxxxx> wrote:
>
> > On Sun, Jan 08, 2023 at 03:04:25PM +0900, Masami Hiramatsu wrote:
> > > On Sat, 7 Jan 2023 08:22:02 -0800
> > > "Paul E. McKenney" <paulmck@xxxxxxxxxx> wrote:
> > >
> > > > On Sun, Jan 08, 2023 at 12:22:15AM +0900, Masami Hiramatsu wrote:
> > > > > On Wed, 4 Jan 2023 16:58:38 -0800
> > > > > "Paul E. McKenney" <paulmck@xxxxxxxxxx> wrote:
> > > > >
> > > > > > The BOOT_CONFIG family of Kconfig options allows a bootconfig file
> > > > > > containing kernel boot parameters to be embedded into an initrd or into
> > > > > > the kernel itself. This can be extremely useful when deploying kernels
> > > > > > in cases where some of the boot parameters depend on the kernel version
> > > > > > rather than on the server hardware, firmware, or workload.
> > > > > >
> > > > > > Unfortunately, the "bootconfig" kernel parameter must be specified in
> > > > > > order to cause the kernel to look for the embedded bootconfig file,
> > > > > > and it clearly does not help to embed this "bootconfig" kernel parameter
> > > > > > into that file.
> > > > > >
> > > > > > Therefore, provide a new BOOT_CONFIG_FORCE Kconfig option that causes the
> > > > > > kernel to act as if the "bootconfig" kernel parameter had been specified.
> > > > > > In other words, kernels built with CONFIG_BOOT_CONFIG_FORCE=y will look
> > > > > > for the embedded bootconfig file even when the "bootconfig" kernel
> > > > > > parameter is omitted. This permits kernel-version-dependent kernel
> > > > > > boot parameters to be embedded into the kernel image without the need to
> > > > > > (for example) update large numbers of boot loaders.
> > > > >
> > > > > I like this because this is a simple solution. We have another option
> > > > > to specify "bootconfig" in CONFIG_CMDLINE, but it can be overwritten by
> > > > > bootloader. Thus, it is better to have this option so that user can
> > > > > always enable bootconfig.
> > > >
> > > > Glad you like it!
> > > >
> > > > In addition, if the help text is accurate, another shortcoming of
> > > > CONFIG_CMDLINE is that its semantics vary from one architecture to
> > > > another. Some have CONFIG_CMDLINE override the boot-loader supplied
> > > > parameters, and others differ in the order in which the parameters
> > > > are processed.
> > >
> > > Yes, that differences confuse us...
> >
> > I am glad that it is not just me. ;-)
> >
> > I will add words to that effect to the commit log.
> >
> > > > > Acked-by: Masami Hiramatsu (Google) <mhiramat@xxxxxxxxxx>
> > > >
> > > > Thank you!
> > > >
> > > > > BTW, maybe CONFIG_BOOT_CONFIG_EMBED is better to select this.
> > > > > (or at least recommend to enable this)
> > > >
> > > > Like this?
> > >
> > > Yes! Thanks.
> > >
> > > >
> > > > Thanx, Paul
> > > >
> > > > ------------------------------------------------------------------------
> > > >
> > > > commit d09a1505c51a70da38b34ac38062977299aef742
> > > > Author: Paul E. McKenney <paulmck@xxxxxxxxxx>
> > > > Date: Sat Jan 7 08:09:22 2023 -0800
> > > >
> > > > bootconfig: Default BOOT_CONFIG_FORCE to y if BOOT_CONFIG_EMBED
> > > >
> > > > When a kernel is built with CONFIG_BOOT_CONFIG_EMBED=y, the intention
> > > > will normally be to unconditionally provide the specified kernel-boot
> > > > arguments to the kernel, as opposed to requiring a separately provided
> > > > bootconfig parameter. Therefore, make the BOOT_CONFIG_FORCE Kconfig
> > > > option default to y in kernels built with CONFIG_BOOT_CONFIG_EMBED=y.
> > > >
> > > > The old semantics may be obtained by manually overriding this default.
> > > >
> > > > Suggested-by: Masami Hiramatsu <mhiramat@xxxxxxxxxx>
> > > > Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxx>
> > >
> > > Acked-by: Masami Hiramatsu (Google) <mhiramat@xxxxxxxxxx>
> >
> > Applied, thank you!
>
> Paul, just for confirmation, have you picked these patches on your tree?

I have, but if you would prefer to take them, just let me know when you
have pulled them in. It is easy for me to drop them.

Here they are in the -rcu tree:

3d9ccc4a8b56e bootconfig: Allow forcing unconditional bootconfig processing
68b920592ff67 bootconfig: Default BOOT_CONFIG_FORCE to y if BOOT_CONFIG_EMBED

git://git.kernel.org/pub/scm/linux/kernel/git/paulmck/linux-rcu.git

Thanx, Paul

> Thank you,
>
> >
> > > Thank you!
> > >
> > > >
> > > > diff --git a/init/Kconfig b/init/Kconfig
> > > > index 0fb19fa0edba9..97a0f14d9020d 100644
> > > > --- a/init/Kconfig
> > > > +++ b/init/Kconfig
> > > > @@ -1379,6 +1379,7 @@ config BOOT_CONFIG
> > > > config BOOT_CONFIG_FORCE
> > > > bool "Force unconditional bootconfig processing"
> > > > depends on BOOT_CONFIG
> > > > + default y if BOOT_CONFIG_EMBED
> > > > help
> > > > With this Kconfig option set, BOOT_CONFIG processing is carried
> > > > out even when the "bootconfig" kernel-boot parameter is omitted.
> > >
> > >
> > > --
> > > Masami Hiramatsu (Google) <mhiramat@xxxxxxxxxx>
>
>
> --
> Masami Hiramatsu (Google) <mhiramat@xxxxxxxxxx>