[PATCH 05/25] treewide: Correct use of long help

Tom Rini trini at konsulko.com
Fri Oct 6 18:55:04 CEST 2023


On Fri, Oct 06, 2023 at 07:03:17AM -0600, Simon Glass wrote:
> Hi Tom,
> 
> On Thu, 5 Oct 2023 at 20:16, Tom Rini <trini at konsulko.com> wrote:
> >
> > On Thu, Oct 05, 2023 at 07:41:49PM -0600, Simon Glass wrote:
> > > Hi Tom,
> > >
> > > On Thu, 5 Oct 2023 at 08:53, Tom Rini <trini at konsulko.com> wrote:
> > > >
> > > > On Wed, Oct 04, 2023 at 07:23:47PM -0600, Simon Glass wrote:
> > > > > Hi Tom,
> > > > >
> > > > > On Sun, 24 Sept 2023 at 17:27, Tom Rini <trini at konsulko.com> wrote:
> > > > > >
> > > > > > On Sun, Sep 24, 2023 at 02:39:23PM -0600, Simon Glass wrote:
> > > > > > > Some commands assume that CONFIG_SYS_LONGHELP is always defined.
> > > > > > > Declaration of long help should be bracketed by an #ifdef to avoid an
> > > > > > > 'unused variable' warning.
> > > > > > >
> > > > > > > Fix this treewide.
> > > > > > >
> > > > > > > Signed-off-by: Simon Glass <sjg at chromium.org>
> > > > > > [snip]
> > > > > > > diff --git a/arch/arm/mach-imx/cmd_dek.c b/arch/arm/mach-imx/cmd_dek.c
> > > > > > > index 6fa5b41fcd38..25ea7d3b37da 100644
> > > > > > > --- a/arch/arm/mach-imx/cmd_dek.c
> > > > > > > +++ b/arch/arm/mach-imx/cmd_dek.c
> > > > > > > @@ -393,11 +393,12 @@ static int do_dek_blob(struct cmd_tbl *cmdtp, int flag, int argc,
> > > > > > >       return blob_encap_dek(src_addr, dst_addr, len);
> > > > > > >  }
> > > > > > >
> > > > > > > -/***************************************************/
> > > > > > > +#if IS_ENABLED(CONFIG_SYS_LONGHELP)
> > > > > > >  static char dek_blob_help_text[] =
> > > > > > >       "src dst len            - Encapsulate and create blob of data\n"
> > > > > > >       "                         $len bits long at address $src and\n"
> > > > > > >       "                         store the result at address $dst.\n";
> > > > > > > +#endif
> > > > > > >
> > > > > > >  U_BOOT_CMD(
> > > > > > >       dek_blob, 4, 1, do_dek_blob,
> > > > > >
> > > > > > This really doesn't read nicely.  I would rather (globally and fix
> > > > > > existing users) __maybe_unused this instead.  I think there's just one
> > > > > > example today that isn't "foo_help_text".
> > > > >
> > > > > Hmm, what do you think about adding a __longhelp symbol to cause the
> > > > > linker to discard it when not needed?
> > > >
> > > > Well, I don't think we need linker list magic when __maybe_unused will
> > > > just have them be discarded normally.
> > >
> > > Yes, perhaps things are in a better state than they used to be, but
> > > there is a linker discard for commands at present.
> >
> > Yes, but __maybe_unused means we don't get a warning about it, and it's
> > literally discarded as part of --gc-sections as it done everywhere with
> > maybe 3 exceptions?
> 
> Actually with this series we get a lot closer to that. The problem
> with the status quo is that disabling CMDLINE doesn't disable most
> commands, relying instead on discarding them at link time.

I don't follow you here.  We're talking only about the long help.
There's already an option to enable/disable it.  When disabled all of
the long help text should be discarded, because we reference it via
U_BOOT_CMD macro which in turn cares about it, or not.  What's missing
is a U_BOOT_LONGHELP macro so that instead of:
#ifdef CONFIG_SYS_LONGHELP
static char cat_help_text[] =
        "<interface> <dev[:part]> <file>\n"
        "  - Print file from 'dev' on 'interface' to standard output\n";
#endif

We do:
U_BOOT_LONGHELP(cat,
        "<interface> <dev[:part]> <file>\n"
        "  - Print file from 'dev' on 'interface' to standard output\n"
);

Which then does:
static __maybe_unused char cat_help_text[] =
...
;

And we discard the text automatically due to --gc-sections.  We possibly
haven't already been doing this since back when we first started using
--gc-sections there was a bug in binutils that caused text like the
above to still get combined in to other objects and not discarded.
That's been fixed for ages.

And the above macro would also let us clean up U_BOOT_CMD macro slightly
to just omit the longhelp option and instead always do _CMD_HELP(_name)
inside the macros.  It'll also make it harder to add new commands
without a long help by accident.

> With this series, it looks like we can in fact do that, so I should
> remove the discards as well.
> 
> The one proviso is that this does drop a lot of things we want...e.g.
> BOOTSTD_DEFAULTS cannot be enabled due to its use of 'select', meaning
> that common filesystems are dropped. So we'll need more effort after
> this, to allow (for example) bootmeths that don't need commands to
> work correctly. But I think this series at least provides a better
> starting point for teasing things apart.
> 
> So OK I'll go with __maybe_unused for the ones that need it, which
> isn't too many given that many of the strings are just included
> directly in the macro. It is considerably easier than trying to be to
> clever about things.

Yes, this series itself has a lot of problems that need to be addressed
before reposting because I don't like "hiding" that network stuff no
longer works, and I also saw that DFU also silently failing.

-- 
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 659 bytes
Desc: not available
URL: <https://lists.denx.de/pipermail/u-boot/attachments/20231006/0d89198c/attachment.sig>


More information about the U-Boot mailing list