Re: [dpdk-dev] Issues running Ethtool on KNI interfaces

["Zhang, Helin" <helin.zhang@intel.com>]

Hi Desh

> -----Original Message-----
> From: Deshvanth Mirle Jayaprakash (dmirleja) [mailto:dmirleja@cisco.com]
> Sent: Friday, October 17, 2014 3:58 PM
> To: Zhang, Helin; dev@dpdk.org
> Subject: Re: Issues running Ethtool on KNI interfaces
> 
> Hi Helin,
> 
>   We have been using KNI application to create Virtual interfaces. We would
> like to use KNI interfaces for control plane application only. As intended, We are
> still planning to use UIO interfaces to pass Data plane traffic which would bump
> up traffic throughput. We have assigned 4 vCPUs for KNI application (CoreMask
> is 0x1E (Core 1,2,3, and 4)). When I run top while KNI process is running in the
> background, see there is 100% CPU utilization on all 4 Cores.
Yes, all user space threads are 100% utilized in DPDK, including KNI user space threads.
I guess somebody is thinking of a good solution for this issue, but for now, it is what we can expected.

> 
>  Is there a way to fix this or is this expected? In our case, the CPU utilization
> seems to cause lot of issues wrt to other processes running on Linux OS. Am I
> missing any EAL option in KNI.Also, I was wondering If there is a way to run KNI
> without Hugepages.
See my answer above, KNI user space app is actually a special DPDK application, it is based on all DPDK technologies, including hugepages.

> 
> The Command:
> 
> ./kni -c 0x1E -n 4 --socket-mem 512 -- -p 0x3 --config="(0,1,3),(1,2,4)" &
> 
> Thanks,
> Desh
> 
>  There is some issue with respect to Hugepage allocation that prevents other
> processes from On 10/13/14 12:46 AM, "Zhang, Helin"
> <helin.zhang@intel.com> wrote:
> 
> >Hi Desh
> >
> >Actually KNI provide a path to exchange info/actions between user space
> >and kernel space. You can read kni example application and KNI kernel
> >module and KNI library for more details. It already supports setting
> >MTU, link up/down the port, etc.
> >
> >Regards,
> >Helin
> >
> >> -----Original Message-----
> >> From: Deshvanth Mirle Jayaprakash (dmirleja)
> >> [mailto:dmirleja@cisco.com]
> >> Sent: Monday, October 13, 2014 3:39 PM
> >> To: Zhang, Helin; dev@dpdk.org
> >> Subject: Re: Issues running Ethtool on KNI interfaces
> >>
> >> Thanks Helin, have been trying this for some time, Is there any other
> >>way I can  pass IOCTLs to IGB-UIO interfaces. Can I use /dev/uio0 and
> >>/dev/uio1 ?
> >>Can I
> >> use IOCTLS on these references in Kernel to gather interface
> >>statistics, set  MTU or bring up/down UIO interfaces.
> >>
> >> /Desh
> >>
> >> On 10/13/14 12:34 AM, "Zhang, Helin" <helin.zhang@intel.com> wrote:
> >>
> >> >Hi Desh
> >> >
> >> >You tried to use ethtool for KNI interfaces in VM, right? I don't
> >> >think it is supported in VM.
> >> >Currently it just supports ethtool for KNI interfaces in host for
> >> >some igb and ixgbe NICs.
> >> >
> >> >Regards,
> >> >Helin
> >> >
> >> >> -----Original Message-----
> >> >> From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Deshvanth
> >> >> Mirle Jayaprakash (dmirleja)
> >> >> Sent: Monday, October 13, 2014 3:21 PM
> >> >> To: dev@dpdk.org
> >> >> Subject: [dpdk-dev] Issues running Ethtool on KNI interfaces
> >> >>
> >> >> Hi,
> >> >>
> >> >> Hypervisor: KVM
> >> >>
> >> >> VM: Linux OS with 2.6.32 Kernel
> >> >>
> >> >> VM Settings:
> >> >>
> >> >> 8 vCPUs, 8192 MB of memory, CPU Configuration: Copy Host CPU
> >> >> Config (SandyBridge), Manually set CPU topology: Sockets=2,
> >> >> Cores=4,
> >> >> Threads=1
> >> >>
> >> >> 10Gigi Passthrough Interfaces attached to VM: Intel X520
> >> >>
> >> >>
> >> >> DPDK settings:
> >> >>
> >> >>
> >> >> DPDK Version: 1.6R2
> >> >>
> >> >>
> >> >> echo 512 >
> /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages
> >> >>
> >> >> rmmod ixgbe
> >> >>
> >> >> insmod igb_uio.ko
> >> >>
> >> >> mkdir /mnt/hugepages
> >> >>
> >> >> mount -t hugetlbfs nodev /mnt/hugepages
> >> >>
> >> >> #./igb_uio_bind.py --status
> >> >>
> >> >> #./igb_uio_bind.py --bind=igb_uio 03:01.0
> >> >>
> >> >> #./igb_uio_bind.py --bind=igb_uio 03:02.0
> >> >>
> >> >> #./igb_uio_bind.py --status
> >> >>
> >> >> modprobe hwmon
> >> >>
> >> >> insmod rte_kni.ko
> >> >>
> >> >> ./kni -c 0x1E -n 2 --socket-mem 512 -- -p 0x3
> >> >>--config="(0,1,3),(1,2,4)" &
> >> >>
> >> >>
> >> >> Having KNI process running in the background, creates two KNI
> >> >>interfaces  called vEth0 and vEth1 which corresponds to IGB-UIO
> >> >>interfaces attached to  the Userspace. The issue is that the
> >> >>Ethtool doesn't work with KNI interfaces.
> >> >> Is there anything that I am missing, How can I get Ethtool to work
> >> >>with KNI  interfaces? Also, If I do not use --socket-mem OR -m EAL
> >> >>option, KNI application  consumes all the huge pages(It doesn't
> >> >>mater
> >> >>512/4096) assigned in the very  first step.Is that expected behavior?
> >> >>Can I get KNI to work without assigning  any Hugepages?
> >> >>
> >> >>
> >> >>
> >> >> Also, I have attached screen shots of KNI log file. Please take a
> >>look.
> >> >>
> >> >>
> >> >> Thanks,
> >> >>
> >> >> Desh
> >

Regards,
Helin