[PATCH] net/ice: fix get link status timeout

[Yang, Qiming]

Hi, Minjin

> -----Original Message-----
> From: Ye, MingjinX <mingjinx.ye at intel.com>
> Sent: Monday, February 6, 2023 6:52 PM
> To: 韩爽 <hanshuang at qianxin.com>; Thomas Monjalon
> <thomas at monjalon.net>
> Cc: dev at dpdk.org; stable at dpdk.org; Zhou, YidingX
> <yidingx.zhou at intel.com>; Yang, Qiming <qiming.yang at intel.com>; Zhang,
> Qi Z <qi.z.zhang at intel.com>; david.marchand at redhat.com;
> ferruh.yigit at amd.com
> Subject: RE: [PATCH] net/ice: fix get link status timeout
> 
> 
> 
> > -----Original Message-----
> > From: 韩爽 <hanshuang at qianxin.com>
> > Sent: 2023年2月6日 15:57
> > To: Thomas Monjalon <thomas at monjalon.net>; Ye, MingjinX
> > <mingjinx.ye at intel.com>
> > Cc: dev at dpdk.org; stable at dpdk.org; Zhou, YidingX
> > <yidingx.zhou at intel.com>; Yang, Qiming <qiming.yang at intel.com>; Zhang,
> > Qi Z <qi.z.zhang at intel.com>; Ye, MingjinX <mingjinx.ye at intel.com>;
> > david.marchand at redhat.com; ferruh.yigit at amd.com
> > Subject: Re: [PATCH] net/ice: fix get link status timeout
> >
> >
> >
> > > -----Original Message-----
> > > From: Thomas Monjalon <thomas at monjalon.net>
> > > Sent: Monday, February 06, 2023 3:23 PM
> > > To: Mingjin Ye <mingjinx.ye at intel.com>
> > > Cc: dev at dpdk.org; stable at dpdk.org; yidingx.zhou at intel.com; Qiming
> > Yang
> > > <qiming.yang at intel.com>; Qi Zhang <qi.z.zhang at intel.com>; Mingjin Ye
> > > <mingjinx.ye at intel.com>; david.marchand at redhat.com;
> > > ferruh.yigit at amd.com
> > > Subject: Re: [PATCH] net/ice: fix get link status timeout
> > >
> > > 06/02/2023 07:22, Mingjin Ye:
> > > > When hw is just started, it will immediately obtain the link
> > > > status, and the longest attempt is 1 second. Some NICs are slow to
> > > > initialize, which make it fails to obtain the link status.
> > > >
> > > > The patch fixes this issue by modifying the longest attempt to 5 seconds.
> > >
> > > What is the consequence?
> > > In which case, DPDK application would be blocked during 5 seconds?
> > >
> >
> > I think ice_link_update with wait_to_complete needs to be optimized
> > when ice_link_update.
> > Our system integrates dozens or more interfaces(E810), even if wait 1
> > second, the system starts very slowly.
> 
> Your proposal is very interesting. We will discuss this in our internal meeting.
> Just verified some issues:
> ----------------------------------------------------------------------------------------------
> ice_dev_start
> 	ice_dev_set_link_up
> 		ice_link_update(dev, 1)
> e810 takes 700ms, e823 takes 1980ms.
> -----------------------------------------------------------------------------------------------
> ice_dev_start
> 	ice_dev_set_link_up
> 		sleep(5)
> 		ice_link_update(dev, 1)
>  e810 and e823 need about 5~8ms.
> -----------------------------------------------------------------------------------------------
> The execution timing of ice_link_update needs to be optimized. The root
> cause is that the hw  needs time to complete initialization. This involves
> initializing the hw, which can be a lot of work.
> This is not the same issue that the current patch solves.

I think it's a cycle, so even we set the max waiting time to 1s, it will run out when link status result get.
Will not cause system start slowly only when HW meet some errors and can't get link status.
But I also suggest you to use a reasonable time based on the test result.