Re: [rcu:rcu/next 30/45] include/linux/compiler.h:343:2: error: implicit declaration of function 'smp_read_barrier_depends'

[Paul E. McKenney]

Hello, Will,

Good point -- I should have removed that as soon as you posted the
update.  I have removed it now.

I am happy to take the patches, but let's make sure that I am up to
speed on the current state and dependencies.  Here is my current
scorecard, please double-check:

1.	Your patcheset from October 12th for nuking lockless_dereference():
	lkml.kernel.org/r/1507818377-7546-1-git-send-email-will.deacon@xxxxxxx

2.	Mark Rutland's prepatory patchset for nuking ACCESS_ONCE():
	-rcu, v4.14-rc4..251e52a951b0 ("rcutorture: formal: prepare for
	ACCESS_ONCE() removal").  Depends on #1.

3.	My mop-up patchset for two remaining occurrences of
	ACCESS_ONCE() in documentation and a comment.  No real urgency
	or dependencies here. -rcu, 11721220e6bf ("treewide: Kill off
	remaining ACCESS_ONCE()".

4.	Mark's scripted patchset for nuking ACCESS_ONCE(), which will
	be run my Linus, hopefully at the end of the merge window that
	takes #1 and #2.

5.	My patchset for removing most smp_read_barrier_depends()
	instances.  -rcu, 11721220e6bf..b7a74661caeb ("keyring: Remove
	now-redundant smp_read_barrier_depends()").  These depend on
	#1, and many of them are non-trivial, so they will likely
	straggle in over time as they accumulate sufficient testing
	and/or acks.  Three of them are ready to go in.

6.	Removing smp_read_barrier_depends() from the InfiniBand drivers.
	These use cases are a bit obscure, so may take some time.
	Andrea Parri kindly volunteered to chase these down, but could
	use responses to his queries to the InfiniBand maintainers.
	These will likely depend on #1, though as Peter Zijlstra pointed
	out, there is no record of any Alpha systems using InfiniBand,
	so maybe they can be treated independently.

Did I get that right?  If I have the wrong patches or am missing some
dependencies, please let me know.  Otherwise, I will create a branch
including available patches from 1-3 and 5 above.

Are people comfortable with my pushing the straightforward stuff
(that is, excluding #5 and #6) into the next merge window?

							Thanx, Paul

On Tue, Oct 17, 2017 at 05:14:58PM +0100, Will Deacon wrote:
> Hi Paul,
> 
> It looks like the breakage below is from a version of the patches that
> existed before I split compiler.h in half. What's the plan with these
> patches? I'd be happier for you to take them, but you'll want to take the
> most recent version in that case.
> 
> Cheers,
> 
> Will
> 
> On Sun, Oct 15, 2017 at 07:29:36AM +0800, kbuild test robot wrote:
> > tree:   https://git.kernel.org/pub/scm/linux/kernel/git/paulmck/linux-rcu.git rcu/next
> > head:   ea788e7da4886dff9782ad61d4c5e6ebadfa8260
> > commit: c3a030152f67ef977129c11e5b37a8e6071d4b6f [30/45] locking/barriers: Kill lockless_dereference
> > config: sparc-alldefconfig (attached as .config)
> > compiler: sparc-linux-gcc (GCC) 6.2.0
> > reproduce:
> >         wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross
> >         chmod +x ~/bin/make.cross
> >         git checkout c3a030152f67ef977129c11e5b37a8e6071d4b6f
> >         # save the attached .config to linux build tree
> >         make.cross ARCH=sparc 
> > 
> > All errors (new ones prefixed by >>):
> > 
> >    In file included from include/uapi/linux/stddef.h:1:0,
> >                     from include/linux/stddef.h:4,
> >                     from include/uapi/linux/posix_types.h:4,
> >                     from include/uapi/linux/types.h:13,
> >                     from include/linux/types.h:5,
> >                     from include/linux/thread_info.h:10,
> >                     from arch/sparc/include/asm/current.h:14,
> >                     from include/linux/sched.h:11,
> >                     from arch/sparc/kernel/asm-offsets.c:13:
> >    include/linux/list.h: In function 'list_empty':
> > >> include/linux/compiler.h:343:2: error: implicit declaration of function 'smp_read_barrier_depends' [-Werror=implicit-function-declaration]
> >      smp_read_barrier_depends(); /* Enforce dependency ordering from x */ \
> >      ^
> >    include/linux/compiler.h:346:22: note: in expansion of macro '__READ_ONCE'
> >     #define READ_ONCE(x) __READ_ONCE(x, 1)
> >                          ^~~~~~~~~~~
> >    include/linux/list.h:202:9: note: in expansion of macro 'READ_ONCE'
> >      return READ_ONCE(head->next) == head;
> >             ^~~~~~~~~
> >    cc1: some warnings being treated as errors
> >    make[2]: *** [arch/sparc/kernel/asm-offsets.s] Error 1
> >    make[2]: Target '__build' not remade because of errors.
> >    make[1]: *** [prepare0] Error 2
> >    make[1]: Target 'prepare' not remade because of errors.
> >    make: *** [sub-make] Error 2
> > 
> > vim +/smp_read_barrier_depends +343 include/linux/compiler.h
> > 
> >    312	
> >    313	/*
> >    314	 * Prevent the compiler from merging or refetching reads or writes. The
> >    315	 * compiler is also forbidden from reordering successive instances of
> >    316	 * READ_ONCE, WRITE_ONCE and ACCESS_ONCE (see below), but only when the
> >    317	 * compiler is aware of some particular ordering.  One way to make the
> >    318	 * compiler aware of ordering is to put the two invocations of READ_ONCE,
> >    319	 * WRITE_ONCE or ACCESS_ONCE() in different C statements.
> >    320	 *
> >    321	 * In contrast to ACCESS_ONCE these two macros will also work on aggregate
> >    322	 * data types like structs or unions. If the size of the accessed data
> >    323	 * type exceeds the word size of the machine (e.g., 32 bits or 64 bits)
> >    324	 * READ_ONCE() and WRITE_ONCE() will fall back to memcpy(). There's at
> >    325	 * least two memcpy()s: one for the __builtin_memcpy() and then one for
> >    326	 * the macro doing the copy of variable - '__u' allocated on the stack.
> >    327	 *
> >    328	 * Their two major use cases are: (1) Mediating communication between
> >    329	 * process-level code and irq/NMI handlers, all running on the same CPU,
> >    330	 * and (2) Ensuring that the compiler does not  fold, spindle, or otherwise
> >    331	 * mutilate accesses that either do not require ordering or that interact
> >    332	 * with an explicit memory barrier or atomic instruction that provides the
> >    333	 * required ordering.
> >    334	 */
> >    335	
> >    336	#define __READ_ONCE(x, check)						\
> >    337	({									\
> >    338		union { typeof(x) __val; char __c[1]; } __u;			\
> >    339		if (check)							\
> >    340			__read_once_size(&(x), __u.__c, sizeof(x));		\
> >    341		else								\
> >    342			__read_once_size_nocheck(&(x), __u.__c, sizeof(x));	\
> >  > 343		smp_read_barrier_depends(); /* Enforce dependency ordering from x */ \
> >    344		__u.__val;							\
> >    345	})
> >    346	#define READ_ONCE(x) __READ_ONCE(x, 1)
> >    347	
> > 
> > ---
> > 0-DAY kernel test infrastructure                Open Source Technology Center
> > https://lists.01.org/pipermail/kbuild-all                   Intel Corporation
> 
>