Re: [PATCH] kbuild: Check for unknown options with cc-option and clang in Kbuild
From: Stephen Boyd
Date: Thu Jul 25 2019 - 14:07:44 EST
Quoting Nathan Chancellor (2019-07-24 22:18:57)
> Hi Stephen,
> Was the second Kbuild in the subject line supposed to be Kconfig?
Sure. I'll change it to Kconfig.
> On Wed, Jul 24, 2019 at 04:50:30PM -0700, Stephen Boyd wrote:
> > If the particular version of clang a user has doesn't enable
> > -Werror=unknown-warning-option by default, even though it is the
> > default, then make sure to pass the option to the Kconfig cc-option
> Hmmm interesting, I did not even know that was possible... Is that a
> clang configuration option or an out of tree patch? Looks like it has
> been on by default since clang 3.2: https://godbolt.org/z/mOmusu
I asked and it turns out that we force this flag off in the ChromeOS
toolchain so that we can compile the multitude of packages in our system
that assume various GCC specific warning flags. I guess this is easier
than patching all the Makefiles out there.
> > command so that testing options from Kconfig files works properly.
> > Otherwise, depending on the default values setup in the clang toolchain
> > we will silently assume options such as -Wmaybe-uninitialized are
> > supported by clang, when they really aren't.
> > This issue only started happening for me once commit 589834b3a009
> > ("kbuild: Add -Werror=unknown-warning-option to CLANG_FLAGS") was
> > applied on top of commit b303c6df80c9 ("kbuild: compute false-positive
> > -Wmaybe-uninitialized cases in Kconfig"). This leads kbuild to try and
> Prior to 589834b3a009, how did cc-option work at all if
> -Wunknown-warning-option wasn't enabled by default? I assume that clang
> would just eat any unknown flags while returning zero so it looked like
> the flag was supported?
Yes. But just warning options?
> > test for the existence of the -Wmaybe-uninitialized flag with the
> > cc-option command in scripts/Kconfig.include, and it doesn't see an
> > error returned from the option test so it sets the config value to Y.
> It might be worth explicitly saying somewhere in here that clang will
> not error on unknown flags without -Werror + -Wunknown-warning-option.
I think it warns on unknown flags, just not unknown warning options
(-Wfoo), so I didn't mention this.
> > Then the makefile tries to pass the unknown option on the command line
> > and -Werror=unknown-warning-option catches the invalid option and breaks
> > the build.
> > Note: this doesn't change the cc-option tests in Makefiles, because
> > those use a different rule that includes KBUILD_CFLAGS by default, and
> > the KBUILD_CFLAGS already has -Werror=unknown-warning-option. Thanks to
> > Doug for pointing out the different rule.
> >  https://clang.llvm.org/docs/DiagnosticsReference.html#wunknown-warning-option
> > Cc: Peter Smith <peter.smith@xxxxxxxxxx>
> > Cc: Nathan Chancellor <natechancellor@xxxxxxxxx>
> > Cc: Nick Desaulniers <ndesaulniers@xxxxxxxxxx>
> > Cc: Douglas Anderson <dianders@xxxxxxxxxxxx>
> > Signed-off-by: Stephen Boyd <swboyd@xxxxxxxxxxxx>
> > Change-Id: I3bb69d45bb062d1306acbf19bc0adfb60f706442
> I assume that shouldn't be there?
> Overall, seems okay to me (took me a sec to understand the bug,
> certainly a very specific one). It might make sense to explicitly add
> somewhere in the commit message that this syncs cc-option behavior
> between Kconfig and Kbuild as a whole, as I didn't understand that at
> first. Thanks for the triage and sorry for the breakage!
> Reviewed-by: Nathan Chancellor <natechancellor@xxxxxxxxx>
I reworded the commit text a bit now and I'll resend it soon. Thanks for