Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Wed, 15 Jun 2022 18:26:29 -0700
From:   "Paul E. McKenney" <paulmck@kernel.org>
To:     Stephen Rothwell <sfr@canb.auug.org.au>
Cc:     Nathan Chancellor <nathan@kernel.org>,
        Frederic Weisbecker <frederic@kernel.org>,
        Sami Tolvanen <samitolvanen@google.com>,
        Kees Cook <keescook@chromium.org>, llvm@lists.linux.dev,
        linux-kernel@vger.kernel.org, linux-next@vger.kernel.org
Subject: Re: Contextual conflict between kspp and rcu trees
Message-ID: <20220616012629.GL1790663@paulmck-ThinkPad-P17-Gen-1>
Reply-To: paulmck@kernel.org
References: <Yqo5SequJuC2qX6S@dev-arch.thelio-3990X>
 <20220616091634.55ebbdb0@canb.auug.org.au>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20220616091634.55ebbdb0@canb.auug.org.au>
Precedence: bulk

On Thu, Jun 16, 2022 at 09:16:34AM +1000, Stephen Rothwell wrote:
> Hi Nathan,
> 
> On Wed, 15 Jun 2022 12:55:53 -0700 Nathan Chancellor <nathan@kernel.org> wrote:
> >
> > There is a contextual conflict between commit e1d337335207 ("cfi: Fix
> > __cfi_slowpath_diag RCU usage with cpuidle") in the kspp tree and commit
> > dcc0c11aa87b ("rcu/context-tracking: Remove rcu_irq_enter/exit()") in
> > the rcu tree, which is visible when building ARCH=arm64 defconfig +
> > CONFIG_LTO_CLANG_THIN=y + CONFIG_CFI_CLANG=y with clang:
> > 
> >   kernel/cfi.c:298:3: error: call to undeclared function 'rcu_irq_enter'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration]
> >                   rcu_irq_enter();
> >                   ^
> >   kernel/cfi.c:298:3: note: did you mean 'ct_irq_enter'?
> >   ./include/linux/context_tracking_irq.h:6:6: note: 'ct_irq_enter' declared here
> >   void ct_irq_enter(void);
> >        ^
> >   kernel/cfi.c:307:3: error: call to undeclared function 'rcu_irq_exit'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration]
> >                   rcu_irq_exit();
> >                   ^
> >   kernel/cfi.c:307:3: note: did you mean 'ct_irq_exit'?
> >   ./include/linux/context_tracking_irq.h:7:6: note: 'ct_irq_exit' declared here
> >   void ct_irq_exit(void);
> >        ^
> >   2 errors generated.
> > 
> > 
> > Per the above RCU commit and commit 6c5218715286 ("context_tracking:
> > Take IRQ eqs entrypoints over RCU"), it appears that the following diff
> > is the proper fix up. Would you mind applying it to the merge of
> > whichever tree comes second if possible? I did build and boot test it
> > but it would not be a bad idea for Sami and Frederic to verify that it
> > is correct so that Kees/Paul can mention it to Linus :)
> > 
> > Cheers,
> > Nathan
> > 
> > diff --git a/kernel/cfi.c b/kernel/cfi.c
> > index 08102d19ec15..2046276ee234 100644
> > --- a/kernel/cfi.c
> > +++ b/kernel/cfi.c
> > @@ -295,7 +295,7 @@ static inline cfi_check_fn find_check_fn(unsigned long ptr)
> >  	rcu_idle = !rcu_is_watching();
> >  	if (rcu_idle) {
> >  		local_irq_save(flags);
> > -		rcu_irq_enter();
> > +		ct_irq_enter();
> >  	}
> >  
> >  	if (IS_ENABLED(CONFIG_CFI_CLANG_SHADOW))
> > @@ -304,7 +304,7 @@ static inline cfi_check_fn find_check_fn(unsigned long ptr)
> >  		fn = find_module_check_fn(ptr);
> >  
> >  	if (rcu_idle) {
> > -		rcu_irq_exit();
> > +		ct_irq_exit();
> >  		local_irq_restore(flags);
> >  	}
> >  
> 
> I will apply that to the merge of the rcu tree today (unless Paul finds
> time to update (and test :-) ) the rcu tree before I get to it) as the
> CFI fix is now in Linus' tree.

Please!

In theory, I could rebase -rcu to linus/master now, but in practice the
odds of it producing something useful are all too low.  I will take care
of it tomorrow (Thursday) morning, Pacific Time.

							Thanx, Paul