On 9/25/2018 10:04 AM, Thomas Monjalon wrote:
> 25/09/2018 10:03, Ferruh Yigit:
>> On 9/24/2018 5:59 PM, Thomas Monjalon wrote:
>>>>> --- a/drivers/net/ixgbe/ixgbe_rxtx.c
>>>>> +++ b/drivers/net/ixgbe/ixgbe_rxtx.c
>>>>> @@ -2025,7 +2025,7 @@ ixgbe_recv_pkts_lro(void *rx_queue, struct rte_mbuf
>>>>> **rx_pkts, uint16_t nb_pkts,
>>>>> struct ixgbe_rx_entry *next_rxe = NULL;
>>>>> struct rte_mbuf *first_seg;
>>>>> struct rte_mbuf *rxm;
>>>>> - struct rte_mbuf *nmb;
>>>>> + struct rte_mbuf *nmb = NULL;
>>>>
>>>> This change is unrelated. Can we separate this one?
>>>
>>> Yes it looks unrelated but it becomes necessary when including stdbool.h.
>>> I don't know the root cause, but yes, it may deserve a separate commit.
>>> Maybe an ixgbe maintainer can take care of it?
>>
>> Why becomes necessary? Does it give a build warning etc?
>> My concern is this is in data path, one extra assignment, it would be better
>> to
>> confirm it is really needed.
>
> Yes I had a compilation error.
> If you cannot reproduce it, I will try to re-run my compilation tests.
I got the error with gcc [1] but it seems false positive and only generated when
<stdbool.h> included in ixgbe_rxtx.c, so this is an odd one, I am not able to
find root cause.
But since it is false positive, what do you think adding compiler option to
disable this warning for this file?
[1]
.../drivers/net/ixgbe/ixgbe_rxtx.c:2139:14: error: ‘nmb’ may be used
uninitialized in this function [-Werror=maybe-uninitialized]
rxe->mbuf = nmb;
~~~~~~~~~~^~~~~
$ gcc --version
gcc (GCC) 8.1.1 20180712 (Red Hat 8.1.1-5)
