[dpdk-dev] [PATCH] kni: fix unit test segfault

Ferruh Yigit ferruh.yigit at intel.com
Thu May 4 16:22:19 CEST 2017


On 5/4/2017 6:23 AM, gowrishankar muthukrishnan wrote:
> Hi Ferruh,
> Even w/o this patch, I mostly find mbufs returned by rte_kni_rx_burst() 
> always 0
> (or sometimes 1) where as more than 32000 mbufs approx created in 
> ingress side
> for this test.
> 
>    dpdk/test/test $ ./test -l 0,1,2 --socket-mem 1024
> 
> Am I missing something required for this unit test ?.

Hi Gowrishankar,

KNI module should be inserted with one of the loopback options [1] for
the unit test.

[1]
There are two loopback options supported, both loopbacks Rx/Tx path in
kernel side, this is easy way to test KNI.

options are: lo_mode_fifo and lo_mode_fifo_skb

usage:
insmod build/kmod/rte_kni.ko lo_mode=lo_mode_fifo
or
insmod build/kmod/rte_kni.ko lo_mode=lo_mode_fifo_skb


Regards,
ferruh

> 
> Thanks,
> Gowrishankar
> 
> On Wednesday 03 May 2017 09:40 PM, Ferruh Yigit wrote:
>> To clean alloc_q, which has physicall addresses of the mbufs, kni lib
>> free the pkt_mempool, but this leads a crash in kni unit test.
>>
>> KNI library shouldn't free the pkt_mempool.
>>
>> Implementation updated to find the mbufs in the alloc_q and return them
>> back to mempool.
>>
>> Fixes: 8eba5ebd1811 ("kni: fix possible memory leak")
>>
>> Signed-off-by: Ferruh Yigit <ferruh.yigit at intel.com>
>> ---
>>   lib/librte_kni/rte_kni.c | 34 ++++++++++++++++++++++------------
>>   1 file changed, 22 insertions(+), 12 deletions(-)
>>
>> diff --git a/lib/librte_kni/rte_kni.c b/lib/librte_kni/rte_kni.c
>> index 52fcd4b..c3f9208 100644
>> --- a/lib/librte_kni/rte_kni.c
>> +++ b/lib/librte_kni/rte_kni.c
>> @@ -451,17 +451,35 @@ kni_free_fifo(struct rte_kni_fifo *fifo)
>>   	} while (ret);
>>   }
>>
>> +static void *
>> +va2pa(struct rte_mbuf *m)
>> +{
>> +	return (void *)((unsigned long)m -
>> +			((unsigned long)m->buf_addr -
>> +			 (unsigned long)m->buf_physaddr));
>> +}
>> +
>>   static void
>> -kni_free_fifo_phy(struct rte_mempool *pktmbuf_pool, struct rte_kni_fifo *fifo)
>> +obj_free(struct rte_mempool *mp __rte_unused, void *opaque, void *obj,
>> +		unsigned obj_idx __rte_unused)
>> +{
>> +	struct rte_mbuf *m = obj;
>> +	void *mbuf_phys = opaque;
>> +
>> +	if (va2pa(m) == mbuf_phys)
>> +		rte_pktmbuf_free(m);
>> +}
>> +
>> +static void
>> +kni_free_fifo_phy(struct rte_mempool *mp, struct rte_kni_fifo *fifo)
>>   {
>>   	void *mbuf_phys;
>>   	int ret;
>>
>> -	rte_mempool_free(pktmbuf_pool);
>> -
>> -	/* All mbufs alredy freed with rte_mempoll_free, just free the fifo */
>>   	do {
>>   		ret = kni_fifo_get(fifo, &mbuf_phys, 1);
>> +		if (ret)
>> +			rte_mempool_obj_iter(mp, obj_free, mbuf_phys);
>>   	} while (ret);
>>   }
>>
>> @@ -557,14 +575,6 @@ rte_kni_handle_request(struct rte_kni *kni)
>>   	return 0;
>>   }
>>
>> -static void *
>> -va2pa(struct rte_mbuf *m)
>> -{
>> -	return (void *)((unsigned long)m -
>> -			((unsigned long)m->buf_addr -
>> -			 (unsigned long)m->buf_physaddr));
>> -}
>> -
>>   unsigned
>>   rte_kni_tx_burst(struct rte_kni *kni, struct rte_mbuf **mbufs, unsigned num)
>>   {
> 
> 



More information about the dev mailing list