DPDK usage discussions
 help / color / mirror / Atom feed
From: Kai Zhang <kay21s@gmail.com>
To: "Van Haaren, Harry" <harry.van.haaren@intel.com>
Cc: "Wiles, Keith" <keith.wiles@intel.com>,
	"users@dpdk.org" <users@dpdk.org>
Subject: Re: [dpdk-users] Issue with more Cores assigned: Cannot mmap device resource file
Date: Mon, 13 Mar 2017 18:59:24 +0800	[thread overview]
Message-ID: <CACKSPme71Gzjrp1_AdsbXo2fqw9yNjhwvPWgA55_r=zi7XO98w@mail.gmail.com> (raw)
In-Reply-To: <E923DB57A917B54B9182A2E928D00FA6129FB458@IRSMSX102.ger.corp.intel.com>

Thank you for your info, Harry.

Even if the ASLR is the root reason, I don't think DPDK should expect users
to disable it to use the primary/secondary model. Is it possible for the
DPDK team to check this issue and fix the bug?

Regards,
Kai

On Mon, Mar 13, 2017 at 5:58 PM, Van Haaren, Harry <
harry.van.haaren@intel.com> wrote:

> > From: users [mailto:users-bounces@dpdk.org] On Behalf Of Kai Zhang
> > Subject: Re: [dpdk-users] Issue with more Cores assigned: Cannot mmap
> device resource
> > file
> >
> > Yes, my application is somewhat special and should run with the
> > primary/secondary mode. I will search for the way to turn of the random
> > page mapping and try it.
>
>
> You're searching for ASLR, or Address Space Layout Randomization.
>
> Some useful links regarding ASLR, DPDK and Linux;
> http://dpdk.readthedocs.io/en/v16.04/prog_guide/multi_proc_
> support.html#multi-process-limitations
> http://askubuntu.com/questions/318315/how-can-i-temporarily-disable-aslr-
> address-space-layout-randomization
> http://dpdk.org/ml/archives/dev/2015-June/019364.html
>
> Please note that ASLR is a security feature of the OS, think twice before
> disabling it.
>
>
> Hope that helps, -Harry
>
>
> > Thanks for your help :)
> >
> > Regards,
> > Kai
> >
> > On Mon, Mar 13, 2017 at 3:24 AM, Wiles, Keith <keith.wiles@intel.com>
> wrote:
> >
> > >
> > > > On Mar 12, 2017, at 6:39 PM, Kai Zhang <kay21s@gmail.com> wrote:
> > > >
> > > >
> > > > Your application may be attaching to the same port for each core.
> > > Normally this means the each core could be allocating memory and the
> 4th
> > > core just goes over the amount of memory you have reserved.
> > > >
> > > > I don't think so. Because the error is in the rte_eal_init(), which
> is
> > > executed in the first line of the main() function. At the time, the
> other
> > > threads are not even launched.
> > > >
> > > > Is it possible to consider this as a bug in DPDK?
> > >
> > > One more thing, I run Pktgen as two processes all of the time. The big
> > > difference is I do not run in primary and secondary modes. I run two
> > > different instances of pktgen at the same time without seeing this type
> > > problem. If the failure is associated with primary/secondary
> application
> > > model, then it could be a bug in that code as a lot of syncing up
> between
> > > the two processes needs to be done because of memory/device sharing.
> One
> > > problem with P/S applications is memory needs to be mapped at the same
> > > address between the processes and Linux has the Random memory mapping
> > > builtin for security reasons. I forget the name of the mode in Linux to
> > > turn off the random page mapping and google is not work for me ATM.
> > >
> > > Does your application require running as a primary/secondary
> application?
> > >
> > > >
> > > > Regards,
> > > > Kai
> > > >
> > > >
> > > > >
> > > > > EAL: Cannot mmap device resource file /sys/bus/pci/devices/0000:02:
> 00.0/resource0
> > > to address: 0x7fff65bfc000
> > > > > EAL: Error - exiting with code: 1
> > > > >   Cause: Requested device 0000:02:00.0 cannot be used
> > > > >
> > > > > Regards,
> > > > > Kai
> > > > >
> > > > > On Sun, Mar 12, 2017 at 11:21 AM, Kai Zhang <kay21s@gmail.com>
> wrote:
> > > > >
> > > > > Command line:
> > > > > primary:      sudo ./primary -l 0,1,2,3 -n 4 --proc-type=primary
> > > > > secondary: sudo ./secondary -l 4,5,6,7,8 -n 4 --proc-type=secondary
> > > > >
> > > > > The configurations are as follows:
> > > > > A) 1 x Intel E5-2650 v4, 12 cores [UMA],     XL710 40GbE, bind
> > > 02:00.0,    2048 x 4k huge page
> > > > > 02:00.0 Ethernet controller: Intel Corporation Ethernet Controller
> > > XL710 for 40GbE QSFP+ (rev 02)   [<<- Only bind this one]
> > > > > 02:00.1 Ethernet controller: Intel Corporation Ethernet Controller
> > > XL710 for 40GbE QSFP+ (rev 02)
> > > > > 05:00.0 Ethernet controller: Intel Corporation I210 Gigabit Network
> > > Connection (rev 03)
> > > > > 06:00.0 Ethernet controller: Intel Corporation I210 Gigabit Network
> > > Connection (rev 03)
> > > > >         Socket 0
> > > > > --------
> > > > > Core 0  [0, 12]
> > > > > Core 1  [1, 13]
> > > > > Core 2  [2, 14]
> > > > > Core 3  [3, 15]
> > > > > Core 4  [4, 16]
> > > > > Core 5  [5, 17]
> > > > > Core 8  [6, 18]
> > > > > Core 9  [7, 19]
> > > > > Core 10 [8, 20]
> > > > > Core 11 [9, 21]
> > > > > Core 12 [10, 22]
> > > > > Core 13 [11, 23]
> > > > >
> > > > > B) 2 x Intel E5-2640 v4, 10 cores [NUMA],      No Port Bind,
> 2048 x
> > > 4k huge page
> > > > > 05:00.0 Ethernet controller: Intel Corporation I210 Gigabit Network
> > > Connection (rev 03)
> > > > > 06:00.0 Ethernet controller: Intel Corporation I210 Gigabit Network
> > > Connection (rev 03)
> > > > >         Socket 0        Socket 1
> > > > >         --------        --------
> > > > > Core 0  [0, 20]         [10, 30]
> > > > > Core 1  [1, 21]         [11, 31]
> > > > > Core 2  [2, 22]         [12, 32]
> > > > > Core 3  [3, 23]         [13, 33]
> > > > > Core 4  [4, 24]         [14, 34]
> > > > > Core 8  [5, 25]         [15, 35]
> > > > > Core 9  [6, 26]         [16, 36]
> > > > > Core 10 [7, 27]         [17, 37]
> > > > > Core 11 [8, 28]         [18, 38]
> > > > > Core 12 [9, 29]         [19, 39]
> > > > >
> > > > > Ah, as machine B does not have a 40GbE, I did not bind any NIC and
> run
> > > my program with locally generated packets. But I am using other DPDK
> > > features, such as memory sharing and message passing. Maybe that is the
> > > reason it works correctly? I can only access machine B remotely, so I
> am
> > > unable to install a NIC on it. I have another PC that is used as a
> client
> > > that only has four cores, which also cannot be used for verification...
> > > > >
> > > > > Regards,
> > > > > Kai
> > > > >
> > > > >
> > > > > On Sun, Mar 12, 2017 at 2:59 AM, Wiles, Keith <
> keith.wiles@intel.com>
> > > wrote:
> > > > >
> > > > > > On Mar 11, 2017, at 9:45 AM, Kai Zhang <kay21s@gmail.com> wrote:
> > > > > >
> > > > > > Hi Keith,
> > > > > >
> > > > > > Thank you for your reply.
> > > > > >
> > > > > > I have tested my program on two machines
> > > > > > A) 1 x Intel E5-2650 v4, 12 cores [UMA]
> > > > > > B) 2 x Intel E5-2640 v4, 10 cores [NUMA]
> > > > > >
> > > > > > I am very sure that the primary process uses different cores with
> > > the secondary process. The strange thing is that my program works
> correctly
> > > on machine B. But on machine A, the above issue happens with more than
> 4
> > > cores assigned to the secondary process.
> > > > > >
> > > > > > I have tried to assign cores 1-5  to the secondary process and
> also
> > > tried other core assignment policies, but the error still happens
> > > rte_eal_init() with more than 4 cores.
> > > > >
> > > > > It would be nice to see both command lines. I am not sure I can
> help
> > > more all I can do is suggest some ideas to look at.
> > > > >
> > > > > Does machine B have the same number and type of NICs? Use ‘lspci |
> > > grep Ethernet’ to get a list of all Ethernet devices on both machines.
> > > > >
> > > > > What is the number of hugepages you have allocated for both
> machines.
> > > > >
> > > > > Also look at the cpu_layout.py script to see why adding the 5th
> core
> > > would be different on the two machines and try to make them the same.
> > > > >
> > > > > >
> > > > > > Regards,
> > > > > > Kai
> > > > > >
> > > > > > On Sat, Mar 11, 2017 at 10:52 PM, Wiles, Keith <
> > > keith.wiles@intel.com> wrote:
> > > > > >
> > > > > > > On Mar 10, 2017, at 9:35 PM, Kai Zhang <kay21s@gmail.com>
> wrote:
> > > > > > >
> > > > > > > Hi, there
> > > > > > >
> > > > > > > I am using DPDK-16.11 on XL710 40GbE NIC. OS: CentOS 7.3.1611
> with
> > > Linux
> > > > > > > kernel version 3.8.0-30.
> > > > > > >
> > > > > > > I have a master process and a secondary process. When I run the
> > > secondary
> > > > > > > process with less than or equal to 4 cores, it works correctly.
> > > Such as:
> > > > > > > sudo ./program -l 4,5,6,7 -n 4 --proc-type=secondary
> > > > > > > sudo ./program -c 0x0f -n 4 --proc-type=secondary
> > > > > > >
> > > > > > > However, there will be error in the rte_eal_init if I assign
> more
> > > than 4
> > > > > > > cores.
> > > > > > > sudo ./program -l 0,1,2,3,4 -n 4 --proc-type=secondary
> > > > > > > sudo ./program -c 0x1f -n 4 --proc-type=secondary
> > > > > > >
> > > > > > > EAL: Cannot mmap device resource file
> > > > > > > /sys/bus/pci/devices/0000:02:00.0/resource0 to address:
> > > 0x7fff65bfc000
> > > > > > > EAL: Error - exiting with code: 1
> > > > > > >  Cause: Requested device 0000:02:00.0 cannot be used
> > > > > >
> > > > > > I assume you have at least 8 cores. Have you tried -l 1-5 on the
> > > secondary process.
> > > > > >
> > > > > > You did not show the primary process command line, but the if you
> > > use 1-5 then you can only give primary process -l 6-7 or two cores. It
> is
> > > always a reasonable thing is to leave core zero for linux to use.
> > > > > >
> > > > > > Also it could be you ran out of memory or hugepages you
> allocated to
> > > the system.
> > > > > >
> > > > > > >
> > > > > > > Anyone knows why this happens?
> > > > > > >
> > > > > > > Thanks a lot,
> > > > > > > Kai Zhang
> > > > > >
> > > > > > Regards,
> > > > > > Keith
> > > > > >
> > > > > >
> > > > >
> > > > > Regards,
> > > > > Keith
> > > > >
> > > > >
> > > > >
> > > >
> > > > Regards,
> > > > Keith
> > >
> > > Regards,
> > > Keith
> > >
> > >
>

  reply	other threads:[~2017-03-13 10:59 UTC|newest]

Thread overview: 16+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2017-03-11  3:35 Kai Zhang
2017-03-11 14:52 ` Wiles, Keith
2017-03-11 15:45   ` Kai Zhang
2017-03-11 18:59     ` Wiles, Keith
2017-03-12  3:21       ` Kai Zhang
2017-03-12  3:29         ` Kai Zhang
2017-03-12 10:32           ` Wiles, Keith
2017-03-12 10:39             ` Kai Zhang
2017-03-12 18:55               ` Wiles, Keith
2017-03-12 19:24               ` Wiles, Keith
2017-03-12 23:44                 ` Kai Zhang
2017-03-13  9:58                   ` Van Haaren, Harry
2017-03-13 10:59                     ` Kai Zhang [this message]
2017-03-15 14:56 David Coen
2017-03-15 15:48 David Coen
2017-03-15 17:02 ` Kai Zhang

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to='CACKSPme71Gzjrp1_AdsbXo2fqw9yNjhwvPWgA55_r=zi7XO98w@mail.gmail.com' \
    --to=kay21s@gmail.com \
    --cc=harry.van.haaren@intel.com \
    --cc=keith.wiles@intel.com \
    --cc=users@dpdk.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).