[dpdk-dev] [PATCH v3] i40e: configure MTU

Ananyev, Konstantin konstantin.ananyev at intel.com
Mon Jun 20 14:15:12 CEST 2016


Hi Beilei,

> -----Original Message-----
> From: Xing, Beilei
> Sent: Monday, June 20, 2016 1:05 PM
> To: Ananyev, Konstantin; Yong Wang; Olivier Matz; Wu, Jingjing
> Cc: dev at dpdk.org; Julien Meunier; Thomas Monjalon
> Subject: RE: [dpdk-dev] [PATCH v3] i40e: configure MTU
> 
> 
> > -----Original Message-----
> > From: Ananyev, Konstantin
> > Sent: Monday, June 20, 2016 4:05 PM
> > To: Xing, Beilei <beilei.xing at intel.com>; Yong Wang
> > <yongwang at vmware.com>; Olivier Matz <olivier.matz at 6wind.com>; Wu,
> > Jingjing <jingjing.wu at intel.com>
> > Cc: dev at dpdk.org; Julien Meunier <julien.meunier at 6wind.com>; Thomas
> > Monjalon <thomas.monjalon at 6wind.com>
> > Subject: RE: [dpdk-dev] [PATCH v3] i40e: configure MTU
> >
> >
> >
> > > -----Original Message-----
> > > From: dev [mailto:dev-bounces at dpdk.org] On Behalf Of Xing, Beilei
> > > Sent: Monday, June 20, 2016 5:50 AM
> > > To: Yong Wang; Olivier Matz; Wu, Jingjing
> > > Cc: dev at dpdk.org; Julien Meunier; Thomas Monjalon
> > > Subject: Re: [dpdk-dev] [PATCH v3] i40e: configure MTU
> > >
> > >
> > > > -----Original Message-----
> > > > From: Yong Wang [mailto:yongwang at vmware.com]
> > > > Sent: Friday, June 17, 2016 1:40 AM
> > > > To: Olivier Matz <olivier.matz at 6wind.com>; Xing, Beilei
> > > > <beilei.xing at intel.com>; Wu, Jingjing <jingjing.wu at intel.com>
> > > > Cc: dev at dpdk.org; Julien Meunier <julien.meunier at 6wind.com>; Thomas
> > > > Monjalon <thomas.monjalon at 6wind.com>
> > > > Subject: Re: [dpdk-dev] [PATCH v3] i40e: configure MTU
> > > >
> > > > On 5/16/16, 5:27 AM, "dev on behalf of Olivier Matz"
> > > > <dev-bounces at dpdk.org on behalf of olivier.matz at 6wind.com> wrote:
> > > >
> > > > >Hi Beilei,
> > > > >
> > > > >On 05/13/2016 10:15 AM, Beilei Xing wrote:
> > > > >> This patch enables configuring MTU for i40e.
> > > > >> Since changing MTU needs to reconfigure queue, stop port first
> > > > >> before configuring MTU.
> > > > >>
> > > > >> Signed-off-by: Beilei Xing <beilei.xing at intel.com>
> > > > >> ---
> > > > >> v3 changes:
> > > > >>  Add frame size with extra I40E_VLAN_TAG_SIZE.
> > > > >>  Delete i40e_dev_rx_init(pf) cause it will be called when port starts.
> > > > >>
> > > > >> v2 changes:
> > > > >>  If mtu is not within the allowed range, return -EINVAL instead of
> > -EBUSY.
> > > > >>  Delete rxq reconfigure cause rxq reconfigure will be finished in
> > > > >> i40e_dev_rx_init.
> > > > >>
> > > > >>  drivers/net/i40e/i40e_ethdev.c | 34
> > > > >> ++++++++++++++++++++++++++++++++++
> > > > >>  1 file changed, 34 insertions(+)
> > > > >>
> > > > >> [...]
> > > > >> +static int
> > > > >> +i40e_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu) {
> > > > >> +	struct i40e_pf *pf =
> > I40E_DEV_PRIVATE_TO_PF(dev->data->dev_private);
> > > > >> +	struct rte_eth_dev_data *dev_data = pf->dev_data;
> > > > >> +	uint32_t frame_size = mtu + ETHER_HDR_LEN
> > > > >> +			      + ETHER_CRC_LEN + I40E_VLAN_TAG_SIZE;
> > > > >> +	int ret = 0;
> > > > >> +
> > > > >> +	/* check if mtu is within the allowed range */
> > > > >> +	if ((mtu < ETHER_MIN_MTU) || (frame_size >
> > I40E_FRAME_SIZE_MAX))
> > > > >> +		return -EINVAL;
> > > > >> +
> > > > >> +	/* mtu setting is forbidden if port is start */
> > > > >> +	if (dev_data->dev_started) {
> > > > >> +		PMD_DRV_LOG(ERR,
> > > > >> +			    "port %d must be stopped before configuration\n",
> > > > >> +			    dev_data->port_id);
> > > > >> +		return -ENOTSUP;
> > > > >> +	}
> > > > >
> > > > >I'm not convinced that ENOTSUP is the proper return value here.
> > > > >It is usually returned when a function is not implemented, which is
> > > > >not the case here: the function is implemented but is forbidden
> > > > >because the port is running.
> > > > >
> > > > >I saw that Julien commented on your v1 that the return value should
> > > > >be one of:
> > > > > - (0) if successful.
> > > > > - (-ENOTSUP) if operation is not supported.
> > > > > - (-ENODEV) if *port_id* invalid.
> > > > > - (-EINVAL) if *mtu* invalid.
> > > > >
> > > > >But I think your initial value (-EBUSY) was fine. Maybe it should
> > > > >be added in the API instead, with the following description:
> > > > >  (-EBUSY) if the operation is not allowed when the port is running
> > > >
> > > > AFAICT, the same check is not done for other drivers that implement
> > > > the mac_set op. Wouldn’t it make more sense to have the driver
> > > > disable the port, reconfigure and re-enable it in this case, instead
> > > > of returning error code?  If the consensus in DPDK is to have the
> > > > application disable the port first, we need to enforce this policy across all
> > devices and clearly document this behavior.
> > > >
> > > Hi,
> > > For ixgbe/igb, there's register for MTU during runtime, but for i40e,
> > > setting MTU is finished by firmware during configuration. There'll be some risk
> > when disable the port, reconfigure and re-enable the port, so return error code
> > in this case currently.
> >
> > Ok, but then what is the point to introduce mtu_set() for i40e at all?
> > Let's just leave it unimplemented, as it is right now.
> > Then users would know that for that device they have to do
> > dev_stop/dev_configure().
> 
> Hi Konstantin,
> The original intention is to reduce ops gap in different PMDs, but unfortunately, setting MTU in i40e couldn’t be finished as in ixgbe,

Yes, I understand that from your previous explanation.

> so in the function, we change the configuration and note users that if want to configure MTU, stop ports first.

Well, you can do that now with dev_stop(); dev_configure(), right?
As I remember, the original intention of mtu_set() was to give user an ability
to change mtu (whenever possible) without stopping the port.
If the HW (i40e) doesn't support it - that's fine, we just leave mtu_set() for i40e unimplemented. 

> The sequence of > configuring MTU for i40e is: port stop->set MTU->port start.
> We will add according instruction in document.

Sorry, but I don’t agree here.
I think, in that case it is better just to leave mtu_set() unimplemented for i40e.
Konstantin

> Beilei
> 
> > Konstantin
> >
> > > Thanks for your comments, we'll improve the document later.
> > >
> > > > >This would allow the application to take its dispositions to stop
> > > > >the port and restart it with the proper jumbo_frame argument.
> > > > >
> > > > >+CC Thomas which maintains ethdev API.
> > > > >
> > > > >
> > > > >Regards,
> > > > >Olivier



More information about the dev mailing list