Re: [dpdk-dev] [PATCH v2] lpm: fix unchecked return value

["Medvedkin, Vladimir" <vladimir.medvedkin@intel.com>]

Hi Ruifeng,

On 18/07/2020 10:22, Ruifeng Wang wrote:
> 
>> -----Original Message-----
>> From: Medvedkin, Vladimir <vladimir.medvedkin@intel.com>
>> Sent: Saturday, July 18, 2020 1:12 AM
>> To: Ruifeng Wang <Ruifeng.Wang@arm.com>; Bruce Richardson
>> <bruce.richardson@intel.com>
>> Cc: dev@dpdk.org; nd <nd@arm.com>; Honnappa Nagarahalli
>> <Honnappa.Nagarahalli@arm.com>; Phil Yang <Phil.Yang@arm.com>
>> Subject: Re: [PATCH v2] lpm: fix unchecked return value
>>
>> Hi Ruifeng,
>>
> Hi Vladimir,
> 
>> On 16/07/2020 16:49, Ruifeng Wang wrote:
>>> Coverity complains about unchecked return value of
>> rte_rcu_qsbr_dq_enqueue.
>>> By default, defer queue size is big enough to hold all tbl8 groups.
>>> When enqueue fails, return error to the user to indicate system issue.
>>>
>>> Coverity issue: 360832
>>> Fixes: 8a9f8564e9f9 ("lpm: implement RCU rule reclamation")
>>>
>>> Signed-off-by: Ruifeng Wang <ruifeng.wang@arm.com>
>>> ---
>>> v2:
>>> Converted return value to conform to LPM API convention. (Vladimir)
>>>
>>>    lib/librte_lpm/rte_lpm.c | 19 +++++++++++++------
>>>    1 file changed, 13 insertions(+), 6 deletions(-)
>>>
>>> diff --git a/lib/librte_lpm/rte_lpm.c b/lib/librte_lpm/rte_lpm.c index
>>> 2db9e16a2..757436f49 100644
>>> --- a/lib/librte_lpm/rte_lpm.c
>>> +++ b/lib/librte_lpm/rte_lpm.c
>>> @@ -532,11 +532,12 @@ tbl8_alloc(struct rte_lpm *lpm)
>>>    	return group_idx;
>>>    }
>>>
>>> -static void
>>> +static int32_t
>>>    tbl8_free(struct rte_lpm *lpm, uint32_t tbl8_group_start)
>>>    {
>>>    	struct rte_lpm_tbl_entry zero_tbl8_entry = {0};
>>>    	struct __rte_lpm *internal_lpm;
>>> +	int status;
>>>
>>>    	internal_lpm = container_of(lpm, struct __rte_lpm, lpm);
>>>    	if (internal_lpm->v == NULL) {
>>> @@ -552,9 +553,15 @@ tbl8_free(struct rte_lpm *lpm, uint32_t
>> tbl8_group_start)
>>>    				__ATOMIC_RELAXED);
>>>    	} else if (internal_lpm->rcu_mode == RTE_LPM_QSBR_MODE_DQ) {
>>>    		/* Push into QSBR defer queue. */
>>> -		rte_rcu_qsbr_dq_enqueue(internal_lpm->dq,
>>> +		status = rte_rcu_qsbr_dq_enqueue(internal_lpm->dq,
>>>    				(void *)&tbl8_group_start);
>>> +		if (status == 1) {
>>> +			RTE_LOG(ERR, LPM, "Failed to push QSBR FIFO\n");
>>> +			return -rte_errno;
>>> +		}
>>>    	}
>>> +
>>> +	return 0;
>>>    }
>>>
>>>    static __rte_noinline int32_t
>>> @@ -1040,7 +1047,7 @@ delete_depth_big(struct rte_lpm *lpm, uint32_t
>> ip_masked,
>>>    #define group_idx next_hop
>>>    	uint32_t tbl24_index, tbl8_group_index, tbl8_group_start,
>> tbl8_index,
>>>    			tbl8_range, i;
>>> -	int32_t tbl8_recycle_index;
>>> +	int32_t tbl8_recycle_index, status = 0;
>>>
>>>    	/*
>>>    	 * Calculate the index into tbl24 and range. Note: All depths
>>> larger @@ -1097,7 +1104,7 @@ delete_depth_big(struct rte_lpm *lpm,
>> uint32_t ip_masked,
>>>    		 */
>>>    		lpm->tbl24[tbl24_index].valid = 0;
>>>    		__atomic_thread_fence(__ATOMIC_RELEASE);
>>> -		tbl8_free(lpm, tbl8_group_start);
>>> +		status = tbl8_free(lpm, tbl8_group_start);
>>>    	} else if (tbl8_recycle_index > -1) {
>>>    		/* Update tbl24 entry. */
>>>    		struct rte_lpm_tbl_entry new_tbl24_entry = { @@ -1113,10
>> +1120,10
>>> @@ delete_depth_big(struct rte_lpm *lpm, uint32_t ip_masked,
>>>    		__atomic_store(&lpm->tbl24[tbl24_index],
>> &new_tbl24_entry,
>>>    				__ATOMIC_RELAXED);
>>>    		__atomic_thread_fence(__ATOMIC_RELEASE);
>>> -		tbl8_free(lpm, tbl8_group_start);
>>> +		status = tbl8_free(lpm, tbl8_group_start);
>>>    	}
>>>    #undef group_idx
>>> -	return 0;
>>> +	return status;
>>
>> This will change rte_lpm_delete API. As a suggestion, you can leave it as it
>> was before ("return 0"), and send separate patch (with "return status)"
>> which will be targeted to 20.11.
>>
> 
> Is the change of API  because a variable is returned instead of constant?
> The patch passed ABI check on Travis: http://mails.dpdk.org/archives/test-report/2020-July/144864.html
> So I didn't know there is API/ABI issue.


Because new error status codes are returned. At the moment 
rte_lpm_delete() returns only -EINVAL. After patches it will also 
returns -ENOSPC. The user's code may not handle this returned error status.

On the other hand, from documentation about returned value:
"0 on success, negative value otherwise",
and given the fact that this behavior is only after calling 
rte_lpm_rcu_qsbr_add(), I think we can accept this patch.
Bruce, please correct me.

> 
> Thanks.
> /Ruifeng
>>>    }
>>>
>>>    /*
>>>
>>
>> --
>> Regards,
>> Vladimir

Acked-by: Vladimir Medvedkin <vladimir.medvedkin@intel.com>

-- 
Regards,
Vladimir