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

[Paul E. McKenney]

On Sat, Feb 25, 2023 at 09:58:11AM +0900, Masami Hiramatsu wrote:
> On Fri, 24 Feb 2023 08:33:07 -0800
> "Paul E. McKenney" <paulmck@xxxxxxxxxx> wrote:
> 
> > On Sat, Feb 25, 2023 at 01:13:06AM +0900, Masami Hiramatsu wrote:
> > > Hi Geert,
> > > 
> > > On Fri, 24 Feb 2023 09:31:50 +0100
> > > Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote:
> > > 
> > > > Hi Paul,
> > > > 
> > > > On Sat, Jan 7, 2023 at 5:33 PM Paul E. McKenney <paulmck@xxxxxxxxxx> wrote:
> > > > > On Sun, Jan 08, 2023 at 12:22:15AM +0900, Masami Hiramatsu wrote:
> > > > > > BTW, maybe CONFIG_BOOT_CONFIG_EMBED is better to select this.
> > > > > > (or at least recommend to enable this)
> > > > >
> > > > > Like this?
> > > > >
> > > > >                                                         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>
> > > > >
> > > > > 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.
> > > > 
> > > > Thanks for your patch, which is now commit 6ded8a28ed80e4cc
> > > > ("bootconfig: Default BOOT_CONFIG_FORCE to y if BOOT_CONFIG_EMBED").
> > > > 
> > > > After this change, an all{mod,yes}config kernel has:
> > > > 
> > > >     CONFIG_BOOT_CONFIG_FORCE=y
> > > >     CONFIG_BOOT_CONFIG_EMBED=y
> > > >     CONFIG_BOOT_CONFIG_EMBED_FILE=""
> > > > 
> > > > Will this actually work? I haven't tried booting such a kernel yet.
> > > 
> > > Yeah, good question. It is same as when you boot the kernel with 'bootconfig'
> > > but do not add the bootconfig file to initrd. You may see below message
> > > on boot log, but kernel boots normally. :)
> > > 
> > >  'bootconfig' found on command line, but no bootconfig found
> > > 
> > > (Maybe it is better to fix the message, because if BOOT_CONFIG_FORCE=y, this
> > > will be shown without 'bootconfig' on command line.)
> > 
> > I just tried it again, and for me it just silently ignores the bootconfig
> > setup.  Which is what I recall happening when I tried it when creating
> > the patch.
> > 
> > Here is the .config file pieces of interest:
> > 
> > CONFIG_BOOT_CONFIG=y
> > CONFIG_BOOT_CONFIG_FORCE=y
> > CONFIG_BOOT_CONFIG_EMBED=y
> > CONFIG_BOOT_CONFIG_EMBED_FILE=""
> > 
> > Anyone else seeing something different?
> 
> Hmm, from the code, I think you'll see that message in early console log.
> 
> In init/main.c:
> 
> ----
> #ifdef CONFIG_BOOT_CONFIG
> /* Is bootconfig on command line? */
> static bool bootconfig_found = IS_ENABLED(CONFIG_BOOT_CONFIG_FORCE);
> static size_t initargs_offs;
> #else
> ----
> And
> ----
> static void __init setup_boot_config(void)
> {
> ...
>         strscpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
>         err = parse_args("bootconfig", tmp_cmdline, NULL, 0, 0, 0, NULL,
>                          bootconfig_params);
> 
>         if (IS_ERR(err) || !bootconfig_found)
>                 return;
> 
>         /* parse_args() stops at the next param of '--' and returns an address */
>         if (err)
>                 initargs_offs = err - tmp_cmdline;
> 
>         if (!data) {
>                 pr_err("'bootconfig' found on command line, but no bootconfig found\n");
>                 return;
>         }
> ----
> 
> Thus, if CONFIG_BOOT_CONFIG_FORCE=y, the process passes the below check
> 
>         if (IS_ERR(err) || !bootconfig_found)
>                 return;
> 
> But since we have an empty 'data', the error should be printed.

And you are quite right, the runs without data files did get me this:

'bootconfig' found on command line, but no bootconfig found

Please accept my apologies for my confusion.

							Thanx, Paul

> Thank you,
> 
> > 
> > 							Thanx, Paul
> > 
> > > Thank you!
> > > 
> > > > 
> > > > Gr{oetje,eeting}s,
> > > > 
> > > >                         Geert
> > > > 
> > > > -- 
> > > > Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx
> > > > 
> > > > In personal conversations with technical people, I call myself a hacker. But
> > > > when I'm talking to journalists I just say "programmer" or something like that.
> > > >                                 -- Linus Torvalds
> > > 
> > > 
> > > -- 
> > > Masami Hiramatsu (Google) <mhiramat@xxxxxxxxxx>
> 
> 
> -- 
> Masami Hiramatsu (Google) <mhiramat@xxxxxxxxxx>