Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Wed, 5 May 2021 17:34:54 +0200
From:   Peter Zijlstra <peterz@infradead.org>
To:     Andi Kleen <ak@linux.intel.com>
Cc:     linux-kernel@vger.kernel.org, Andi Kleen <andi@firstfloor.org>,
        linux-toolchains@vger.kernel.org
Subject: Re: [PATCH] sched: Work around undefined behavior in sched class
 checking
Message-ID: <YJK7HlQYAIFVj9fL@hirez.programming.kicks-ass.net>
References: <20210505033945.1282851-1-ak@linux.intel.com>
 <YJI/OwoflyY2IXvf@hirez.programming.kicks-ass.net>
 <20210505143442.GR4032392@tassilo.jf.intel.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20210505143442.GR4032392@tassilo.jf.intel.com>
Precedence: bulk

On Wed, May 05, 2021 at 07:34:42AM -0700, Andi Kleen wrote:
> > > Use RELOC_HIDE to make this work. This hides the symbols from gcc,
> > > so the optimizer won't make these assumption. I also split
> > > the BUG_ONs in multiple.
> > 
> > Urgh, that insanity again :/ Can't we pretty please get a GCC flag to
> > disable that?
> 
> Even if that was done (I could totally see the gcc people pushing back on this;
> why should they add special flags just for Linux developers not understanding
> ISO-C?)

I understand C fine, I just don't agree with it. I also want to
explicitly define as much UB as is possible, because UB is just utter
garbage.

So just like we do with -fwrapv and others, add more knobs that
explictly define away UB. Less UB is more better. This being C it's
unlikely we'll ever get to no UB, but we should damn well try :-)

> you would still need the fix for already shipping compilers.

Yes, there is that.