[Lustre-discuss] Patchless client modules on fedora14

Andreas Dilger adilger at whamcloud.com
Fri Jun 17 13:36:45 PDT 2011


On 2011-06-17, at 1:15 PM, Christos Triantafyllidis wrote:
>   i managed to build the patchless modules for fedora14 kernel (2.6.35.13-92.fc14) using the attached patch against the 1.8.5.56 tag from git.whamcloud.com [1].
> 
>   Most of the work is for supporting kernels like 2.6.35 and not fedora specific. If one wants to build the patchless modules for fedora14, he will need the complete fedora kernel headers [2] installed.
> 
>   I hope to receive comments on the patch, especially if i broke something :).

Thanks for sending this in.

The patch looks reasonable at first glance, but as it stands right now
it is not suitable for landing, since it will break building for all of
the earlier kernels.

What you need to do is:
- file a bug for tracking this work: http://jira.whamcloud.com/
- Make configure checks for each change - libcfs/autoconf/lustre-libcfs.m4
  and lustre/autoconf/lustre-core.m4 for many examples.  Each check should
  have a short description, the kernel version in which the change appeared
  (so it can be removed later when those kernels are obsolete), and be
  located in numerical order of the kernel version, to make that easier
- Make compatibility functions to handle the API changes as needed -
  libcfs/include/libcfs/linux/portals_compat25.h for functions affecting
  everything, and lustre/include/lustre/linux/lustre_compat25.h for VFS
  functions that only affect the lustre/ code
- These checks+fixes should be split into small patches (one API change
  per patch) so that they can be inspected, tested, and landed separately.
  Several times in the past someone submits a huge patch with dozens of
  different changes in it, but it can't be landed as-is for one reason or
  another, and it gets delayed and has to be refreshed repeatedly.  Making
  small patches means the obvious ones can be landed quickly, the tricky
  ones can be inspected/tested more thoroughly and they do not block the
  simple ones from landing.

Cheers, Andreas
--
Andreas Dilger 
Principal Engineer
Whamcloud, Inc.






More information about the lustre-discuss mailing list