[dpdk-dev] rte_eth_from_rings

[dpdk-dev] rte_eth_from_rings

[Sridhar Pitchai]

Hi,

I am trying to write a data path for packets punted to CPU(slowpath) from vender silicon like broadcom. I am planing to use "rte_eth_from_rings" like model where I will be able to read and write to the ring for the packets punted from vendor chip.


the eth_dev abstraction provided by "rte_eth_from_rings" helps to build the dpdk data path. Can someone help me on how to read and write to the rings that is emulating the eth_dev.


Thanks,

Sridhar

Re: [dpdk-dev] rte_eth_from_rings

[Bruce Richardson]

On Fri, Jan 27, 2017 at 07:16:25PM +0000, Sridhar Pitchai wrote:
> Hi,
> 
> I am trying to write a data path for packets punted to CPU(slowpath) from vender silicon like broadcom. I am planing to use "rte_eth_from_rings" like model where I will be able to read and write to the ring for the packets punted from vendor chip.
> 
> 
> the eth_dev abstraction provided by "rte_eth_from_rings" helps to build the dpdk data path. Can someone help me on how to read and write to the rings that is emulating the eth_dev.
> 
>
To use the rings like an ethdev just use rte_eth_rx_burst and
rte_eth_tx_burst, passing in the port id of your newly created rings
ethdev. To access them directly as rings, just use the ring
enqueue/dequeue functions passing in the ring pointer as normal.

Regards,
/Bruce

Re: [dpdk-dev] rte_eth_from_rings

[Sridhar Pitchai]

Thanks Bruce.


I have created eth_dev from the rings as below.
rt = rte_eth_from_rings(port_p->name,
             (struct rte_ring * const*)port_p->rx_ring_p, 2,
                            (struct rte_ring * const*)port_p->tx_ring_p, 2,
                            rte_socket_id());

Lets say I have a call back something like

        pkt_rx(void * pkt_p, struct pkt_metadata_t *meta_p)

which is called when there is a pkt at the chip.
inside this function(pkt_rx) i will find the port and the corresponding ring_p and
enqueue the pkt into the queue. I am assuming this should work. Kindly correct me if i misunderstood you.


question 2:
Can use this eth_dev to create KNI interface, like below.
rte_eth_dev_info_get(fp_p->port_list[port].key.port_id, &dev_info);
;
;
;
ops.port_id = fp_p->port_list[port].key.port_id;  // rt from rte_eth_from_rings(...)
                ops.change_mtu = _kni_ifconfig_mtu; // static function
                ops.config_network_if = _kni_ifconfig; // static functions
PORT(fp_p, port).kni_p = rte_kni_alloc(PORT(fp_p, port).mbuf_p,
                                                                &conf, &ops);

I am assuming this should work as well. I find the netdevs created but when i try to configure
them i am facing the following error.

root@ VM snaproute/softpath (master) >ifconfig PORT_8
PORT_8    Link encap:Ethernet  HWaddr be:35:c3:0f:f8:2f
          BROADCAST MULTICAST  MTU:1500  Metric:1
          RX packets:0 errors:0 dropped:0 overruns:0 frame:0
          TX packets:0 errors:0 dropped:0 overruns:0 carrier:0
          collisions:0 txqueuelen:1000
          RX bytes:0 (0.0 B)  TX bytes:0 (0.0 B)

root@ VM snaproute/softpath (master) >ifconfig PORT_8 20.1.1.1/24 up
SIOCSIFFLAGS: Timer expired
SIOCSIFFLAGS: Timer expired
root@ VM snaproute/softpath (master) >

Thanks for the help.

Thanks,
Sridhar Pitchai


________________________________
From: Bruce Richardson <bruce.richardson@intel.com>
Sent: Friday, January 27, 2017 1:36 PM
To: Sridhar Pitchai
Cc: dev@dpdk.org
Subject: Re: [dpdk-dev] rte_eth_from_rings

On Fri, Jan 27, 2017 at 07:16:25PM +0000, Sridhar Pitchai wrote:
> Hi,
>
> I am trying to write a data path for packets punted to CPU(slowpath) from vender silicon like broadcom. I am planing to use "rte_eth_from_rings" like model where I will be able to read and write to the ring for the packets punted from vendor chip.
>
>
> the eth_dev abstraction provided by "rte_eth_from_rings" helps to build the dpdk data path. Can someone help me on how to read and write to the rings that is emulating the eth_dev.
>
>
To use the rings like an ethdev just use rte_eth_rx_burst and
rte_eth_tx_burst, passing in the port id of your newly created rings
ethdev. To access them directly as rings, just use the ring
enqueue/dequeue functions passing in the ring pointer as normal.

Regards,
/Bruce

Re: [dpdk-dev] rte_eth_from_rings

[Bruce Richardson]

On Fri, Jan 27, 2017 at 10:31:52PM +0000, Sridhar Pitchai wrote:
> Thanks Bruce.
> 
> 
> I have created eth_dev from the rings as below.
> rt = rte_eth_from_rings(port_p->name,
>              (struct rte_ring * const*)port_p->rx_ring_p, 2,
>                             (struct rte_ring * const*)port_p->tx_ring_p, 2,
>                             rte_socket_id());
> 
> Lets say I have a call back something like
> 
>         pkt_rx(void * pkt_p, struct pkt_metadata_t *meta_p)
> 
> which is called when there is a pkt at the chip.
> inside this function(pkt_rx) i will find the port and the corresponding ring_p and
> enqueue the pkt into the queue. I am assuming this should work. Kindly correct me if i misunderstood you.
> 

Yes, it should work.

> 
> question 2:
> Can use this eth_dev to create KNI interface, like below.
> rte_eth_dev_info_get(fp_p->port_list[port].key.port_id, &dev_info);
> ;
> ;
> ;
> ops.port_id = fp_p->port_list[port].key.port_id;  // rt from rte_eth_from_rings(...)
>                 ops.change_mtu = _kni_ifconfig_mtu; // static function
>                 ops.config_network_if = _kni_ifconfig; // static functions
> PORT(fp_p, port).kni_p = rte_kni_alloc(PORT(fp_p, port).mbuf_p,
>                                                                 &conf, &ops);
> 
> I am assuming this should work as well. I find the netdevs created but when i try to configure
> them i am facing the following error.
> 
> root@ VM snaproute/softpath (master) >ifconfig PORT_8
> PORT_8    Link encap:Ethernet  HWaddr be:35:c3:0f:f8:2f
>           BROADCAST MULTICAST  MTU:1500  Metric:1
>           RX packets:0 errors:0 dropped:0 overruns:0 frame:0
>           TX packets:0 errors:0 dropped:0 overruns:0 carrier:0
>           collisions:0 txqueuelen:1000
>           RX bytes:0 (0.0 B)  TX bytes:0 (0.0 B)
> 
> root@ VM snaproute/softpath (master) >ifconfig PORT_8 20.1.1.1/24 up
> SIOCSIFFLAGS: Timer expired
> SIOCSIFFLAGS: Timer expired
> root@ VM snaproute/softpath (master) >
> 

+Ferruh as KNI maintainer, to see if he can help out with this question.

/Bruce

> Thanks for the help.
> 
> Thanks,
> Sridhar Pitchai
> 
> 
> ________________________________
> From: Bruce Richardson <bruce.richardson@intel.com>
> Sent: Friday, January 27, 2017 1:36 PM
> To: Sridhar Pitchai
> Cc: dev@dpdk.org
> Subject: Re: [dpdk-dev] rte_eth_from_rings
> 
> On Fri, Jan 27, 2017 at 07:16:25PM +0000, Sridhar Pitchai wrote:
> > Hi,
> >
> > I am trying to write a data path for packets punted to CPU(slowpath) from vender silicon like broadcom. I am planing to use "rte_eth_from_rings" like model where I will be able to read and write to the ring for the packets punted from vendor chip.
> >
> >
> > the eth_dev abstraction provided by "rte_eth_from_rings" helps to build the dpdk data path. Can someone help me on how to read and write to the rings that is emulating the eth_dev.
> >
> >
> To use the rings like an ethdev just use rte_eth_rx_burst and
> rte_eth_tx_burst, passing in the port id of your newly created rings
> ethdev. To access them directly as rings, just use the ring
> enqueue/dequeue functions passing in the ring pointer as normal.
> 
> Regards,
> /Bruce

Re: [dpdk-dev] rte_eth_from_rings

[Sridhar Pitchai]

Thanks Bruce,


Hi Ferruh,

I experimented the following to get the KNI interface up, but not able to get it up yet.


rte_kni_alloc(..) // with rte_eth_dev_info = NULL

;

;

rte_kni_register_handlers // with rte_eth_dev_info set with right call_back for port_up_down and if_mtu


;

;

;

I also added


rte_kni_handle_request(kni_p)



with these i am not able to bring the port UP. I also tried to do ioctl in interface,


  ifr.ifr_flags |= (IFF_UP | IFF_RUNNING);
    ioctl(fp_p->port_ioctl, SIOCSIFFLAGS, &ifr);

this call fails.

What am i missing ? can you kindly point out my mistake here. the eth_dev the KNI bound to is the eth_dev created from ring buffer. (my dev environment is ubuntu 16.04 VM but eventual the code will be running on ONL. do i need to have any kernel flags. i am running rte_kni.ko without any args)

Thanks,
Sridhar


**put some syslog into ret_kni and find
rte_kni_handle_request is failling at

ret = kni_fifo_get(kni->req_q, (void **)&req, 1);



________________________________
From: Bruce Richardson <bruce.richardson@intel.com>
Sent: Monday, January 30, 2017 2:33 AM
To: Sridhar Pitchai
Cc: dev@dpdk.org; ferruh.yigit@intel.com
Subject: Re: [dpdk-dev] rte_eth_from_rings

On Fri, Jan 27, 2017 at 10:31:52PM +0000, Sridhar Pitchai wrote:
> Thanks Bruce.
>
>
> I have created eth_dev from the rings as below.
> rt = rte_eth_from_rings(port_p->name,
>              (struct rte_ring * const*)port_p->rx_ring_p, 2,
>                             (struct rte_ring * const*)port_p->tx_ring_p, 2,
>                             rte_socket_id());
>
> Lets say I have a call back something like
>
>         pkt_rx(void * pkt_p, struct pkt_metadata_t *meta_p)
>
> which is called when there is a pkt at the chip.
> inside this function(pkt_rx) i will find the port and the corresponding ring_p and
> enqueue the pkt into the queue. I am assuming this should work. Kindly correct me if i misunderstood you.
>

Yes, it should work.

>
> question 2:
> Can use this eth_dev to create KNI interface, like below.
> rte_eth_dev_info_get(fp_p->port_list[port].key.port_id, &dev_info);
> ;
> ;
> ;
> ops.port_id = fp_p->port_list[port].key.port_id;  // rt from rte_eth_from_rings(...)
>                 ops.change_mtu = _kni_ifconfig_mtu; // static function
>                 ops.config_network_if = _kni_ifconfig; // static functions
> PORT(fp_p, port).kni_p = rte_kni_alloc(PORT(fp_p, port).mbuf_p,
>                                                                 &conf, &ops);
>
> I am assuming this should work as well. I find the netdevs created but when i try to configure
> them i am facing the following error.
>
> root@ VM snaproute/softpath (master) >ifconfig PORT_8
> PORT_8    Link encap:Ethernet  HWaddr be:35:c3:0f:f8:2f
>           BROADCAST MULTICAST  MTU:1500  Metric:1
>           RX packets:0 errors:0 dropped:0 overruns:0 frame:0
>           TX packets:0 errors:0 dropped:0 overruns:0 carrier:0
>           collisions:0 txqueuelen:1000
>           RX bytes:0 (0.0 B)  TX bytes:0 (0.0 B)
>
> root@ VM snaproute/softpath (master) >ifconfig PORT_8 20.1.1.1/24 up
> SIOCSIFFLAGS: Timer expired
> SIOCSIFFLAGS: Timer expired
> root@ VM snaproute/softpath (master) >
>

+Ferruh as KNI maintainer, to see if he can help out with this question.

/Bruce

> Thanks for the help.
>
> Thanks,
> Sridhar Pitchai
>
>
> ________________________________
> From: Bruce Richardson <bruce.richardson@intel.com>
> Sent: Friday, January 27, 2017 1:36 PM
> To: Sridhar Pitchai
> Cc: dev@dpdk.org
> Subject: Re: [dpdk-dev] rte_eth_from_rings
>
> On Fri, Jan 27, 2017 at 07:16:25PM +0000, Sridhar Pitchai wrote:
> > Hi,
> >
> > I am trying to write a data path for packets punted to CPU(slowpath) from vender silicon like broadcom. I am planing to use "rte_eth_from_rings" like model where I will be able to read and write to the ring for the packets punted from vendor chip.
> >
> >
> > the eth_dev abstraction provided by "rte_eth_from_rings" helps to build the dpdk data path. Can someone help me on how to read and write to the rings that is emulating the eth_dev.
> >
> >
> To use the rings like an ethdev just use rte_eth_rx_burst and
> rte_eth_tx_burst, passing in the port id of your newly created rings
> ethdev. To access them directly as rings, just use the ring
> enqueue/dequeue functions passing in the ring pointer as normal.
>
> Regards,
> /Bruce