Re: stable/linux-3.16.y build: 178 builds: 1 failed, 177 passed, 2 errors, 57 warnings (v3.16.52)

[Ben Hutchings <ben@decadent.org.uk>]

[-- Attachment #1: Type: text/plain, Size: 1473 bytes --]

On Sat, 2018-01-13 at 19:51 +0100, Manfred Spraul wrote:
> Hi Arnd,
> 
> On 01/03/2018 12:15 AM, Arnd Bergmann wrote:
> > > 2 ipc/sem.c:377:6: warning: '___p1' may be used uninitialized in this function [-Wmaybe-uninitialized]
> > This code was last touched in 3.16 by the backport of commit
> > 5864a2fd3088 ("ipc/sem.c: fix complex_count vs. simple op race")
> > 
> > The warning is in "smp_load_acquire(&sma->complex_mode))", and I suspect
> > that commit 27d7be1801a4 ("ipc/sem.c: avoid using spin_unlock_wait()")
> > avoided the warning upstream by removing the smp_mb() before it.
> The smp_mb() pairs with spin_unlock_wait() in complexmode_enter()
> It is removed by commit 27d7be1801a4 ("ipc/sem.c: avoid using 
> spin_unlock_wait()").
> 
>  From what I see, it doesn't exist in any of the stable kernels 
> (intentionally, the above commit is a rewrite for better performance).
> 
> ___p1 is from smp_load_acquire()
>  >        typeof(*p) ___p1 = READ_ONCE(*p);                               \
> 
> I don't see how ___p1 could be used uninitialized. Perhaps a compiler issue?

On arm64 smp_load_acquire() was implemented in assembly that only
supported 4-byte and 8-byte words.  And complex_mode is a bool (1-byte) 
field.

So I believe the fix is:

878a84d5a8a1 arm64: add missing data types in smp_load_acquire/smp_store_release

Ben.

-- 
Ben Hutchings
Reality is just a crutch for people who can't handle science fiction.


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 833 bytes --]