[Users] ofed 4.8.1: opemsm infiniband-diags build time circular dependency?

daggs daggs at gmx.com
Fri Nov 10 08:12:44 PST 2017


Greetings Hal,
 

>On Fri, Nov 10, 2017 at 9:57 AM, daggs <daggs at gmx.com[mailto:daggs at gmx.com]> wrote:Greetings Hal,
>
>>Hi Dagg,
>> 
>>On Fri, Nov 10, 2017 at 3:44 AM, daggs <daggs at gmx.com[mailto:daggs at gmx.com][mailto:daggs at gmx.com[mailto:daggs at gmx.com]]> wrote:Greetings,
>>
>>I need to build a local version of opensm and infiniband-diags to bringup a mlx based fabric.
> 
>Why do you need a local version to do this ? Do you plan on making modifications for this fabric or just want to run these tools on that fabric where some machine is attached to >multiple disjoint fabrics/subnets ?
>>>>mainly because the distro I use doesn't have the 4.8.1 in its repo, in addition as I don't have root permissions, I'm to install it on to my home folder.
> 
> 
>By default, you're going to need to be root to run any of these tools unless the admin changes udev permissions for umad things so even if you get things built they will not run unless the admin has made umad >things accessible more widely than root.
> > 
>> 
>>Anyhow, there is --prefix option to configure where things are installed.
>my apologies, not sure how --prefix can help my issue.
> 
>Look at configure --help for things like:
>Installation directories:
>  --prefix=PREFIX         install architecture-independent files in PREFIX
>                          [/usr/local]
>  --exec-prefix=EPREFIX   install architecture-dependent files in EPREFIX
>                          [PREFIX]
> 
>Fine tuning of the installation directories:
>...
> 
>> 
>>Note also that before running OpenSM, you need to configure things in the options/conf file  so it does not step on various files that your primary OpenSM.
>I know, thing is, if I run configure on opensm it complains that libumand is missing.
>if I run configure on infiniband-diags, it complains opensm is missing.
> 
> 
>So first issue is to determine why it thinks libibumad is missing.
> 
> >>I'm not using rpm based distro so I need to compile the code myself.
>> 
>>Are you looking to build specific versions or from latest source ? There are tar balls for the specific versions and the latest sources are available in various git repos.
>I've searched ofed download on google, got to the ofed alliance web page, clicked download,and selected 4.8.1 from the folder (latest rc).
> 
>Oh, you're talking about OFED. I think OFED ships sources as well as rpms. So you should be able to build from the sources there. I haven't tried this with OFED. I know it has install >script and not sure what options would be needed or whether some additional procedure would be needed. The email list for that is >ewg at lists.openfabrics.org[mailto:ewg at lists.openfabrics.org] although there might be some people on that list who are lurking on this one. 
> 
>>>I've looked into the build dependencies of the two pkgs mentioned above, I both state that I need libibumad-devel, first glance on the tar file reveals there is no libibumad rpm.
>> 
>>libibumad is needed for both infiniband-diags and OpenSM. It must be built and installed first.
>> 
>>Recently, libibmad was incorporated into infiniband-diags so it depends on which version you want to use as to whether you need to build libibmad separately and priot to >>infiniband->diags, which should be built last.
>>>so I've looked into the other rpms and found it inside of infiniband-diags.
>>>looking on the build dependencies of infiniband-diags shows it needs opensm thus creating a circular dependency issue.
>> 
>>infiniband-diags depends on some things like node-name-map support and ib_types.h so OpenSM needs to be built next and finally libibmad/infiniband-diags last.
>that is my question, how can I build libibumad when it merged into infiniband-diags? is there a specific target?
> 
> 
>No; libibmad is merged into infiniband-diags. libibumad is different from libibmad. I believe OFED 4.8.1 ships libibumad and rdma-core (which also includes libibumad - a newer one) >but not sure.
> 
I think I've got everything mixed up, I'll verify and try again.

Dagg.



More information about the Users mailing list