Re: [PATCH] Avoid that check_shl_overflow() triggers a compiler warning when building with W=1

[Bart Van Assche <bvanassche@acm.org>]

On 3/6/19 5:24 PM, Jason Gunthorpe wrote:
> On Wed, Mar 06, 2019 at 05:01:53PM -0800, Bart Van Assche wrote:
>> This patch avoids that the following warning is reported when building
>> the mlx5 driver with W=1:
>>
>> drivers/infiniband/hw/mlx5/qp.c: In function set_user_rq_size:
>> ./include/linux/overflow.h:230:6: warning: comparison of unsigned expression >= 0 is always true [-Wtype-limits]
>>     _s >= 0 && _s < 8 * sizeof(*d) ? _s : 0;  \
>>        ^
>> drivers/infiniband/hw/mlx5/qp.c:5820:6: note: in expansion of macro check_shl_overflow
>>    if (check_shl_overflow(rwq->wqe_count, rwq->wqe_shift, &rwq->buf_size))
>>        ^~~~~~~~~~~~~~~~~~
>>
>> Cc: Jason Gunthorpe <jgg@mellanox.com>
>> Cc: Leon Romanovsky <leonro@mellanox.com>
>> Cc: Rasmus Villemoes <linux@rasmusvillemoes.dk>
>> Fixes: 0c66847793d1 ("overflow.h: Add arithmetic shift helper") # v4.19
>> Signed-off-by: Bart Van Assche <bvanassche@acm.org>
>>   include/linux/overflow.h | 22 ++++++++++++++++++++--
>>   1 file changed, 20 insertions(+), 2 deletions(-)
>>
>> diff --git a/include/linux/overflow.h b/include/linux/overflow.h
>> index 40b48e2133cb..8afe0c0ada6f 100644
>> +++ b/include/linux/overflow.h
>> @@ -202,6 +202,24 @@
>>   
>>   #endif /* COMPILER_HAS_GENERIC_BUILTIN_OVERFLOW */
>>   
>> +/*
>> + * Evaluate a >= 0 without triggering a compiler warning if the type of a
>> + * is an unsigned type.
>> + */
>> +#define is_positive(a) ({					\
>> +	typeof(a) _minus_one = -1LL;				\
>> +	typeof((a) + 0U) _sign_mask = _minus_one > 0 ? 0 :	\
> 
> This is probably just is_signed_type(a)

Hi Jason,

I don't think that gcc accepts something like is_signed_type(typeof(a)) 
so I'm not sure that the is_signed_type() macro is useful in this context.

>> +				1ULL << (8 * sizeof(a) - 1);	\
>> +								\
>> +	((a) & _sign_mask) == 0;				\
>  
> This is the same sort of obfuscation that Leon was building, do you
> think the & is better than his ==, >  version?
> 
> Will gcc shortcircuit the warning if we write it as
> 
> (is_signed_type(a) && a < 0)
> 
> ?

I have tested this patch. With this patch applied no warnings are 
reported while building the mlx5 driver and the tests in 
lib/test_overflow.c pass.

Thanks,

Bart.