rte_pktmbuf_alloc( ) returns same mbuf address which is not freed

rte_pktmbuf_alloc( ) returns same mbuf address which is not freed

[Gokilavani A]

[-- Attachment #1: Type: text/plain, Size: 1032 bytes --]

Hi,



We are using *DPDK-19.11.11* for our application in which 4 construction
threads construct packets periodically, place it in a circular queue and

a transmission thread pushes that packet by reading that circular queue.


In the long run, We are facing a crash in *eth_igb_xmit_pkts() *call which
is called during transmission.


After an inspect with GDB, we came to know that the non-freed mbuf address
is returned by* rte_pktmbuf_alloc() *for another packet during
construction, even though that mbuf is still not transmitted by the
transmission thread, which makes the corresponding construction thread to
update the same memory again.


It is not clear why rte_pktmbuf_alloc() returns the mbuf pointer before
that pointer is not getting freed by rte_eth_tx_burst() call?.


We have created the mempool using the call,

*rte_pktmbuf_pool_create(txMemPoolName, 16384,RTE_MEMPOOL_CACHE_MAX_SIZE,0,
2048 +RTE_PKTMBUF_HEADROOM,  rte_eth_dev_socket_id(0));*


Please guide us in resolving this issue.



Thanks,

Gokilavani A

[-- Attachment #2: Type: text/html, Size: 8421 bytes --]

Re: rte_pktmbuf_alloc( ) returns same mbuf address which is not freed

[Gokilavani A]

[-- Attachment #1: Type: text/plain, Size: 1288 bytes --]

Hi,

We have been blocked by this issue.

Can anyone help us to resolve the issue?

Thanks
Gokilavani A

On Thu, Oct 13, 2022 at 4:39 PM Gokilavani A <gokilavanianbazhagan@gmail.com>
wrote:

> Hi,
>
>
>
> We are using *DPDK-19.11.11* for our application in which 4 construction
> threads construct packets periodically, place it in a circular queue and
>
> a transmission thread pushes that packet by reading that circular queue.
>
>
> In the long run, We are facing a crash in *eth_igb_xmit_pkts() *call
> which is called during transmission.
>
>
> After an inspect with GDB, we came to know that the non-freed mbuf address
> is returned by* rte_pktmbuf_alloc() *for another packet during
> construction, even though that mbuf is still not transmitted by the
> transmission thread, which makes the corresponding construction thread to
> update the same memory again.
>
>
> It is not clear why rte_pktmbuf_alloc() returns the mbuf pointer before
> that pointer is not getting freed by rte_eth_tx_burst() call?.
>
>
> We have created the mempool using the call,
>
> *rte_pktmbuf_pool_create(txMemPoolName,
> 16384,RTE_MEMPOOL_CACHE_MAX_SIZE,0, 2048
> +RTE_PKTMBUF_HEADROOM,  rte_eth_dev_socket_id(0));*
>
>
> Please guide us in resolving this issue.
>
>
>
> Thanks,
>
> Gokilavani A
>
>
>

[-- Attachment #2: Type: text/html, Size: 8995 bytes --]

Re: rte_pktmbuf_alloc( ) returns same mbuf address which is not freed

[Stephen Hemminger]

On Tue, 25 Oct 2022 14:11:26 +0530
Gokilavani A <gokilavanianbazhagan@gmail.com> wrote:

> Hi,
> 
> We have been blocked by this issue.
> 
> Can anyone help us to resolve the issue?
> 
> Thanks
> Gokilavani A
> 
> On Thu, Oct 13, 2022 at 4:39 PM Gokilavani A <gokilavanianbazhagan@gmail.com>
> wrote:
> 
> > Hi,
> >
> >
> >
> > We are using *DPDK-19.11.11* for our application in which 4 construction
> > threads construct packets periodically, place it in a circular queue and
> >
> > a transmission thread pushes that packet by reading that circular queue.
> >
> >
> > In the long run, We are facing a crash in *eth_igb_xmit_pkts() *call
> > which is called during transmission.
> >
> >
> > After an inspect with GDB, we came to know that the non-freed mbuf address
> > is returned by* rte_pktmbuf_alloc() *for another packet during
> > construction, even though that mbuf is still not transmitted by the
> > transmission thread, which makes the corresponding construction thread to
> > update the same memory again.
> >
> >
> > It is not clear why rte_pktmbuf_alloc() returns the mbuf pointer before
> > that pointer is not getting freed by rte_eth_tx_burst() call?.
> >
> >
> > We have created the mempool using the call,
> >
> > *rte_pktmbuf_pool_create(txMemPoolName,
> > 16384,RTE_MEMPOOL_CACHE_MAX_SIZE,0, 2048
> > +RTE_PKTMBUF_HEADROOM,  rte_eth_dev_socket_id(0));*
> >
> >
> > Please guide us in resolving this issue.
> >
> >
> >
> > Thanks,
> >
> > Gokilavani A
> >
> >
> >  

Looks like an application bug.
Did you try building with every possible debug setting enabled.
Things like MBUF_DEBUG and the address sanitizier.

Re: rte_pktmbuf_alloc( ) returns same mbuf address which is not freed

[Stephen Hemminger]

On Tue, 25 Oct 2022 14:11:26 +0530
Gokilavani A <gokilavanianbazhagan@gmail.com> wrote:

> Hi,
> 
> We have been blocked by this issue.
> 
> Can anyone help us to resolve the issue?
> 
> Thanks
> Gokilavani A
> 
> On Thu, Oct 13, 2022 at 4:39 PM Gokilavani A <gokilavanianbazhagan@gmail.com>
> wrote:
> 
> > Hi,
> >
> >
> >
> > We are using *DPDK-19.11.11* for our application in which 4 construction
> > threads construct packets periodically, place it in a circular queue and
> >
> > a transmission thread pushes that packet by reading that circular queue.
> >
> >
> > In the long run, We are facing a crash in *eth_igb_xmit_pkts() *call
> > which is called during transmission.
> >
> >
> > After an inspect with GDB, we came to know that the non-freed mbuf address
> > is returned by* rte_pktmbuf_alloc() *for another packet during
> > construction, even though that mbuf is still not transmitted by the
> > transmission thread, which makes the corresponding construction thread to
> > update the same memory again.
> >
> >
> > It is not clear why rte_pktmbuf_alloc() returns the mbuf pointer before
> > that pointer is not getting freed by rte_eth_tx_burst() call?.
> >

Off list discussion. This application was using m->next for private data.
Stop, this is not expected to work.