Oh, sorry, my mistakes, it is in the tx_release_mbuf
static void
ixgbe_tx_queue_release_mbufs(struct ixgbe_tx_queue *txq)
{
unsigned i;
if (txq->sw_ring != NULL) {
for (i = 0; i < txq->nb_tx_desc; i++) {
if (txq->sw_ring[i].mbuf != NULL) {
rte_pktmbuf_free_seg(txq->sw_ring[i].mbuf);
txq->sw_ring[i].mbuf = NULL;
}
}
}
}
So the real patch should be added here.
> ? 2015?8?3????4:10?Lu, Wenzhuo <wenzhuo.lu at intel.com> ???
>
> Hi Peng,
>
>> -----Original Message-----
>> From: HePeng [mailto:xnhp0320 at icloud.com <mailto:xnhp0320 at icloud.com>]
>> Sent: Monday, August 3, 2015 3:09 PM
>> To: Lu, Wenzhuo
>> Cc: dev at dpdk.org <mailto:dev at dpdk.org>
>> Subject: Re: [dpdk-dev] [PATCH] [new]ixgbe:set txep.mbuf to NULL when calling
>> ixgbe_tx_free_bufs
>>
>> Hi Wenzhuo,
>> The issue is in the function *ixgbe_dev_free_queues* called in the
>> *ixgbe_dev_close*.
>> The *ixgbe_dev_free_queues* will call *ixgbe_rx_queue_release_mbuf*
>> to recycle all the mbuf on the queues. If some mbufs have already been
>> recycled
>> by the *ixgbe_tx_free_bufs*, their ref cnt is 0.
>>
>> However, since the pointers are not set to NULL,
>> *ixgbe_rx_queue_release_mbuf* will also check the mbufs whose ref cnt is 0,
>> then if one enables *CONFIG_RTE_MBUF_DEBUG*, the sanity checks will warn
>> that the ref cnt is bad, and the program will bail out.
>>
>> As you said if this is a designed behavior, you need to fix the code in
>> *ixgbe_rx_queue_release_mbuf* to skip the mbuf that already been recycled.
> But I think it's skipped, like this,
>
> if (rxq->sw_ring[i].mbuf != NULL &&
>
> rte_mbuf_refcnt_read(rxq->sw_ring[i].mbuf)) {
> rte_pktmbuf_free_seg(rxq->sw_ring[i].mbuf);
>
>>
>>
>>> ? 2015?8?3????1:16?Lu, Wenzhuo <wenzhuo.lu at intel.com> ?
>> ??
>>>
>>> Hi Peng,
>>> Would you like to tell me more details about the panic?
>>> I saw there's refcnt check in rte_mbuf_sanity_check. I'm not sure what
>>> sanity
>> check you want to add.
>>> Thanks.
>>>
>>>> -----Original Message-----
>>>> From: HePeng [mailto:xnhp0320 at icloud.com]
>>>> Sent: Monday, August 3, 2015 10:54 AM
>>>> To: Lu, Wenzhuo
>>>> Subject: Re: [dpdk-dev] [PATCH] [new]ixgbe:set txep.mbuf to NULL when
>> calling
>>>> ixgbe_tx_free_bufs
>>>>
>>>> Hi, Wenzhuo
>>>> It will cause panic if you enable *CONFIG_RTE_MBUF_DEBUG* in you
>>>> config file. So if it is a designed behavior, some code fix may need for
>>>> *mbuf_sanity_check*.
>>>> Thanks.
>>>>
>>>>
>>>>> ? 2015?8?3????10:46?Lu, Wenzhuo <wenzhuo.lu at intel.com>
>> ?
>>>> ??
>>>>>
>>>>> Hi Peng,
>>>>>
>>>>>> -----Original Message-----
>>>>>> From: dev [mailto:dev-bounces at dpdk.org] On Behalf Of hepeng
>>>>>> Sent: Saturday, August 1, 2015 9:27 AM
>>>>>> To: dev at dpdk.org
>>>>>> Subject: [dpdk-dev] [PATCH] [new]ixgbe:set txep.mbuf to NULL when
>> calling
>>>>>> ixgbe_tx_free_bufs
>>>>>>
>>>>>> In *ixgbe_tx_free_bufs*, after recycling some tx entries, one should set
>> their
>>>>>> mbuf pointers to NULL.
>>>>>>
>>>>>> The first path is not correct, the txep->mbuf should be set to NULL no
>> matter
>>>> if
>>>>>> it is recycled into mempool
>>>>>> Signed-off-by: hepeng <xnhp0320 at icloud.com>
>>>>>> ---
>>>>>> drivers/net/ixgbe/ixgbe_rxtx_vec.c | 13 ++++++++++++-
>>>>>> 1 file changed, 12 insertions(+), 1 deletion(-)
>>>>>>
>>>>>> diff --git a/drivers/net/ixgbe/ixgbe_rxtx_vec.c
>>>>>> b/drivers/net/ixgbe/ixgbe_rxtx_vec.c
>>>>>> index 1c16dec..e7ce740 100644
>>>>>> --- a/drivers/net/ixgbe/ixgbe_rxtx_vec.c
>>>>>> +++ b/drivers/net/ixgbe/ixgbe_rxtx_vec.c
>>>>>> @@ -612,6 +612,7 @@ ixgbe_tx_free_bufs(struct ixgbe_tx_queue *txq)
>>>>>> */
>>>>>> txep = &txq->sw_ring_v[txq->tx_next_dd - (n - 1)];
>>>>>> m = __rte_pktmbuf_prefree_seg(txep[0].mbuf);
>>>>>> + txep[0].mbuf = NULL;
>>>>>> if (likely(m != NULL)) {
>>>>>> free[0] = m;
>>>>>> nb_free = 1;
>>>>>> @@ -632,11 +633,21 @@ ixgbe_tx_free_bufs(struct ixgbe_tx_queue *txq)
>>>>>> } else {
>>>>>> for (i = 1; i < n; i++) {
>>>>>> m = __rte_pktmbuf_prefree_seg(txep[i].mbuf);
>>>>>> - if (m != NULL)
>>>>>> + if (m != NULL) {
>>>>>> rte_mempool_put(m->pool, m);
>>>>>> + }
>>>>>> }
>>>>>> }
>>>>>>
>>>>>> + /*
>>>>>> + * No matter the mbufs have been put back to mempool or not,
>>>>>> + * we should set the txep[i].mbuf to NULL
>>>>>> + */
>>>>>> +
>>>>>> + for( i = 1; i < n; i++) {
>>>>>> + txep[i].mbuf = NULL;
>>>>>> + }
>>>>>> +
>>>>>> /* buffers were freed, update counters */
>>>>>> txq->nb_tx_free = (uint16_t)(txq->nb_tx_free + txq->tx_rs_thresh);
>>>>>> txq->tx_next_dd = (uint16_t)(txq->tx_next_dd + txq->tx_rs_thresh);
>>>>>> --
>>>>>> 1.9.1
>>>>>
>>>>> NACK.
>>>>> Thanks for looking into this code. But it's designed behavior, not an
>>>>> issue.
>>>>> BTW, if you want to send a new version, the tittle should be like this
>>>>> [PATCH
>> v2]
>>>> ixgbe: ..., and add "--in-reply-to your original mail" when sending the
>>>> mail,
>> and
>>>> add a v2 comments. You can reference the other's v2 patches for detail.
