[dpdk-dev] [PATCH v7 1/2] eal: add uevent monitor for hot plug

Guo, Jia jia.guo at intel.com
Tue Jan 9 12:39:37 CET 2018


So, how can separate the patch into more small patch, use stake or null implement in function. I think we should consider if it is a economic way now, if I could explain more detail in code for you all not very familiar the background? I have sent v8, please check, thanks all. 

Best regards,
Jeff Guo

-----Original Message-----
From: Thomas Monjalon [mailto:thomas at monjalon.net] 
Sent: Tuesday, January 9, 2018 6:48 PM
To: Guo, Jia <jia.guo at intel.com>
Cc: Mordechay Haimovsky <motih at mellanox.com>; dev at dpdk.org; stephen at networkplumber.org; Richardson, Bruce <bruce.richardson at intel.com>; Yigit, Ferruh <ferruh.yigit at intel.com>; gaetan.rivet at 6wind.com; Ananyev, Konstantin <konstantin.ananyev at intel.com>; shreyansh.jain at nxp.com; Wu, Jingjing <jingjing.wu at intel.com>; Zhang, Helin <helin.zhang at intel.com>; Van Haaren, Harry <harry.van.haaren at intel.com>
Subject: Re: [dpdk-dev] [PATCH v7 1/2] eal: add uevent monitor for hot plug

09/01/2018 11:31, Mordechay Haimovsky:
> From: Guo, Jia [mailto:jia.guo at intel.com]
> > On 1/9/2018 8:39 AM, Thomas Monjalon wrote:
> > > At last there is the kernel binding effort - this one will 
> > > probably be ignored for 18.02, because it is another huge topic.
> > > Without bothering with kernel binding, we can at least remove a 
> > > device, get a notification, and eventually re-add it. It is a good first step.
> > > Anyway your testpmd patch tests exactly this scenario (totally new 
> > > devices are not seen).
> > 
> > i will  separate it for you all to benefit  for review.  for kernel 
> > binding, i just let it automatically compare with the first time 
> > manually binding, and it is the part of he hot plug flow. so i 
> > suggest to review more about that if it is not side effect and workable, beg for keep on.
> 
> This will not work for Mellanox which uses several drivers and 
> services in order to map the device and device queues to user space. 
> For example, the mlx4 PMD (PMD for ConnectX-3 devices) requires that 
> mlx4_core mlx4_en and mlx4_ib drivers to be loaded, and for RDM -core 
> user-space libraries and daemons to be loaded.

Yes automatic binding is a feature which requires more work.
It cannot be ready for 18.02.


More information about the dev mailing list