Re: [dpdk-dev] error: value computed is not used

["Ananyev, Konstantin" <konstantin.ananyev@intel.com>]

> -----Original Message-----
> From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Thomas Monjalon
> Sent: Monday, December 15, 2014 2:17 PM
> To: Wodkowski, PawelX
> Cc: dev@dpdk.org
> Subject: Re: [dpdk-dev] error: value computed is not used
> 
> 2014-12-15 13:47, Wodkowski, PawelX:
> > From: Thomas Monjalon [mailto:thomas.monjalon@6wind.com]
> > > 2014-12-15 11:27, Wodkowski, PawelX:
> > > > From: Thomas Monjalon [mailto:thomas.monjalon@6wind.com]
> > > > > 2014-12-08 15:26, Wodkowski, PawelX:
> > > > > > From: Qiu, Michael
> > > > > > > On 2014/12/8 19:00, Wodkowski, PawelX wrote:
> > > > > > > >> lib/librte_pmd_enic/enic_main.c: In function 'enic_set_rsskey':
> > > > > > > >> lib/librte_pmd_enic/enic_main.c:862:2: error: value computed is not
> > > used
> > > > > > > >>
> > > > > > > >> I dig out that, it was ome issue of  the macros rte_memcpy()
> > > > > > > >> #define rte_memcpy(dst, src, n)              \
> > > > > > > >>         ((__builtin_constant_p(n)) ?          \
> > > > > > > >>         memcpy((dst), (src), (n)) :          \
> > > > > > > >>         rte_memcpy_func((dst), (src), (n)))
> > > > > > > >>
> > > > > > > >> When I use only (n) instead of (__builtin_constant_p(n), it will pass( I
> > > > > > > >> know that it was incorrect, just a experiment).
> > > > > > > >>
> > > > > > > >> But I try to use inline function instead of macros:
> > > > > > > >> static inline void * rte_memcpy(void *dst, const void *src, size_t n)
> > > > > > > >> {
> > > > > > > >>         return __builtin_constant_p(n) ? memcpy(dst, src, n) :
> > > > > > > >>                                          rte_memcpy_func(dst, src, n);
> > > > > > > >> }
> > > > > > > >>
> > > > > > > >> It will pass:), and works, this could be one potential workaround fix.
> > > > > > > >>
> > > > > > > >> Who knows why? The root cause is what?
> > > > > > > >>
> > > > > > > >> I've no idea about this.
> > > > > > > >>
> > > > > > > > I got the same issue while ago. I don't remember exactly everything
> > > > > > > > but my conclusion was that there was some bug in compiler. I think,
> > > > > > > > when 'n' I constant and/or small compiler is inlining memcpy and
> > > throwing
> > > > > > > > everything else (including returned value). In that case error is not
> > > > > > > > produced (I think this is a bug in compiler). In other case it is computing
> > > > > > > > some value calling memcpy or rte_ memcpy and you should at least
> > > > > > > > explicitly throw it away by casting to void. I like solution with static
> > > > > > >
> > > > > > > Actually, I try to pass "n" as a Int value like 4, it still report this
> > > > > > > error :)
> > > > > >
> > > > > > My workaround was:
> > > > > > (void) rte_memcpy(...);
> > > > > >
> > > > > > But this is only a workaround.
> > > > >
> > > > > It's not so bad.
> > > > >
> > > > > > > > inline but someone else should spoke about possible side effects.
> > > > > > >
> > > > > > > Yes, but as I know inline is better than macros.
> > > > >
> > > > > From the GCC manual:
> > > > > "
> > > > > You may use this built-in function in either a macro or an inline function.
> > > > > However, if you use it in an inlined function and pass an argument of the
> > > > > function as the argument to the built-in, GCC never returns 1 when you call
> > > > > the inline function with a string constant or compound literal and does not
> > > > > return 1 when you pass a constant numeric value to the inline function unless
> > > > > you specify the -O option.
> > > > > "
> > > > >
> > > > > It seems the "inline fix" cannot be used.
> > > > >
> > > > > I'm going to send a patch with Pawel's workaround.
> > > >
> > > > And something like this?
> > > >
> > > >  #define rte_memcpy(dst, src, n)              \
> > > > -	((__builtin_constant_p(n)) ?          \
> > > > +	({ (__builtin_constant_p(n)) ?          \
> > > >  	memcpy((dst), (src), (n)) :          \
> > > > -	rte_memcpy_func((dst), (src), (n)))
> > > > +	rte_memcpy_func((dst), (src), (n)); })
> > >
> > > What happens to the returned value after this change?
> > > ptr = rte_memcpy(dst, src, n) + offset:
> > >
> > https://gcc.gnu.org/onlinedocs/gcc/Statement-Exprs.html#Statement-Exprs
> >
> > Whole expression should be 'void *' type (like *memcpy()) and it should work
> > as usual (see maxint() example in above link). It is GCC extension.
> 
> OK nice.
> I didn't test it on SUSE 11 SP3. I assume you did it?
> Please Pawel, could you send a proper patch quickly?
> If nobody disagree, it'll be merged in RC5 today.
> 
> > > > Thomas, can you check build with EXTRA_CFLAG='-Wunused-value'.
> > >
> > > You mean EXTRA_CFLAGS (with a S).
> > > It fails in many locations.
> > > What's your point?
> >
> > I am just asking if this is an typo, error or intend to do statements with no effects like bellow.
> >
> > ixgbe_common.c:4429:3: error: statement with no effect [-Werror=unused-value]
> >
> > 4426:	/* first pull in the header so we know the buffer length */
> > 4427:	for (bi = 0; bi < dword_len; bi++) {
> > 4428:		buffer[bi] = IXGBE_READ_REG_ARRAY(hw, IXGBE_FLEX_MNG, bi);
> > 4429:		IXGBE_LE32_TO_CPUS(&buffer[bi]); // <------ here
> > 4430	}
> 
> It's an intent. On big endian CPU, this has an effect.

Hmm, I think there is a bug in lib/librte_pmd_ixgbe/ixgbe/ixgbe_osdep.h:
#define IXGBE_LE32_TO_CPUS(_i) rte_le_to_cpu_32(_i)

It probably should be:
#define IXGBE_LE32_TO_CPUS(_i) rte_le_to_cpu_32(*(_i))

Not much point to do byte swapping for the pointer.
And that what ixgbe BSD driver is doing.

Though I still not sure why it is needed here, as the computed value is not used anyway.
What the author probably meant to do:
buffer[bi] = rte_le_to_cpu_32 (buffer[bi]);
To achieve that we need:
#define IXGBE_LE32_TO_CPUS(x) (*(x) = rte_le_to_cpu_32(*(x)))
Correct?

Konstantin

> 
> > > Do you to support -Wunused-value?
> >
> > No, I just turned this on to check above change and was surprised what happened.
> 
> Honestly, I don't know if there is a good fix for this warning.
> 
> --
> Thomas