[ewg] [Bug 2630] OFED-4.8-rc3 fails building QIB on 4.8.17 kernel + rhel6.6 distro

[Davis, Arlin R]

Any suggestions for this QIB build issue on 4.8? This is a blocker for RC4.

> -----Original Message-----
> From: bugzilla at openfabrics.org [mailto:bugzilla at openfabrics.org]
> Sent: Tuesday, May 30, 2017 6:45 AM
> To: Davis, Arlin R <arlin.r.davis at intel.com>
> Subject: [Bug 2630] OFED-4.8-rc3 fails building QIB on 4.8.17 kernel + rhel6.6
> distro
> 
> http://bugs.openfabrics.org/show_bug.cgi?id=2630
> 
> --- Comment #6 from lukasz.rostkowski at intel.com 2017-05-30 06:45:28 PDT
> --- Sorry guys we have a bit of a differece of opinion of how to approach this
> bug, so we would like to ask Vlad for help.
> 
> We have found, that the problem is caused by a macro introduced in the
> mm.h
> (include/linux/mm.h) file:
> 
> #ifdef HAVE_GET_USER_PAGES_6_PARAMS
> #define get_user_pages(p1, p2, p3, p4, p5, p6, p7, p8) \
>         get_user_pages(p3, p4, p5, p6, p7, p8) #endif
> 
> The qib is prepared to be used in both get_user_pages() flavours, becouse it
> has a backport for it including the ifdefs for
> GET_USER_PARAMS_HAS_6_PARAMS.
> Unfortunatelly the macro from the mm.h replaces proper function invocation
> with a faulty one.
> 
> We suggest, that it should be resovled by removing the mm.h declaration and
> let the drivers/modules handle this on their own instad of some strange
> generic replacement introduced by the mm.h file.
> 
> But if we do that at this point, there are few modules that would be affected
> (mthca, usnic, umem), which aren't yet prepared for using get_user_pages in
> two flavors and would be needed to be updated.
> 
> --
> Configure bugmail: http://bugs.openfabrics.org/userprefs.cgi?tab=email
> ------- You are receiving this mail because: ------- You are on the CC list for the
> bug.