8685 uts and mdb: do not build 32bit kernel

Review Request #630 — Created July 19, 2017 and submitted

tsoome
illumos-gate
8685
629
0d5143f...
general

8685 uts and mdb: do not build 32bit kernel



  • 0
  • 0
  • 2
  • 0
  • 2
Description From Last Updated
tsoome
yuripv
  1. I've actually adopted the work you did here for my needs, and it looks good to me.

  2. 
      
hans
  1. Ship It!
  2. 
      
danmcd
  1. If this is indeed is a full 32-bit kernel elimination, add 32-bit libike, in.iked to the exception list.

    1. system-library.mf:file path=usr/lib/libike.so.1
      SUNWcs.mf:file path=usr/lib/inet/$(ARCH32)/in.iked mode=0555

  2. 
      
danmcd
  1. 
      
  2. usr/src/pkg/manifests/SUNWcs.mf (Diff revision 1)
     
     

    Nuke this one.

  3. usr/src/pkg/manifests/SUNWcs.mf (Diff revision 1)
     
     

    And likely these as well. They are entirely extraneous in a 64-bit-kernel world.

    There are likely more as well. And for those not closed-source, check their build routines as well.

    1. I am sure there are still leftovers etc, but those two cases do make sense, thanks.

  4. 
      
tsoome
tsoome
danmcd
  1. Shoot. I just realized 32-bit libike need to stay (ikecert(1M)), but why does 32-bit in.iked need to remain?

    1. ou, its not about stay, just removing via not extracting from tarball. but ikecert means this patch needs another round. will check...

    2. Oh, I see what happened.

      You will need to re-insert 32-bit libike, however, for the reasons I specified above (ikecert(1M) programs need to link to 32-bit libike.so.1).

  2. 
      
tsoome
andy_js
  1. FYI the debug version of the 32-bit kernel doesn't boot anymore. We hit an assert in vn_create_cache() because the size of vnode_t grew and VNODE_ALIGN_LOG2/VNODE_ALIGN need adjusting. If you fix this you end up hitting another assert in trap() and I have no idea where to begin with that. I think the sooner we can cut this stuff out the better.

  2. 
      
jjelinek
  1. I am in the process of pulling this patch into our smartos tree to build/test it. I haven't reviewed all of the files, but we had a conflict in usr/src/uts/intel/Makefile.intel and while resolving it I noticed quite a lot of 32-bit related entries and comments still in this file that have not been cleaned up. This is just an observation. I am not clear what the intention of the current change is supposed to be and if this stuff was missed or intentionally left.

    1. Oh, thanks. If you can provide the list, I can make update - I am sure there are many bits I have missed, so the more eyes the better:)

      One concern is that while we do have 32-bit userland, we want to be sure we do not break it - hence we need to provide 32-bit mdb for example. But the idea is to disable build of 32-bit kernel and modules, and clean up what we have spotted. The obvious issue is that once we stop building 32-bit kernel, we can only find the 32-bit related entries by visual inspection and therefore the cleanup will take more updates:)

  2. 
      
yuripv
  1. Ship It!
  2. 
      
ptribble
  1. Tested by building and installing a Tribblix ISO from a copy of the gate with this patch applied, which worked without any errors that I've been able to detect so far.

  2. 
      
hans
  1. Ship It!
  2. 
      
andy_js
  1. No issues running with these changes.

    1. Thanks, I did file RTI yesterday evening, but I'll post an note for advocates:)

  2. 
      
jjelinek
  1. I'm happy with the current changes, since we know there is a lot of cleanup left to do.

  2. 
      
tsoome
Review request changed

Status: Closed (submitted)

Loading...