Bug 557 - Xen: OpenVZ in DomU, RHEL5 Xen kernel
: Xen: OpenVZ in DomU, RHEL5 Xen kernel
Status: RESOLVED FIXED
Product: OpenVZ
kernel
: rhel5-2.6.18_028stabXXX
: Other Other
: P2 normal
Assigned To: Evgeny Kravtsunov
: http://forum.openvz.org/index.php?t=p...
:
:
  Show dependency treegraph
 
Reported: 2007-04-26 08:17 EDT by Kirill Korotaev
Modified: 2012-01-22 16:00 EST (History)
5 users (show)

See Also:


Attachments

Note You need to log in before you can comment on or make changes to this bug.
Description Kirill Korotaev 2007-04-26 08:17:48 EDT
RHEL5 kernel has Xen inside. Need to fix its compilation, check that it works.

Thanks to seyko@ for some info here:
http://forum.openvz.org/index.php?t=post&reply_to=12374&SQ=244d57dc8c322ce78bb5cc7b62ebe6e7&
Comment 1 Kirill Korotaev 2007-04-26 08:18:44 EDT
as a final result: RHEL5 Xen kernel should be buildable from src.rpm

Some patches from seyko@ will be 100% applied. Some need rework.
Comment 2 Kirill Korotaev 2007-04-26 08:21:40 EDT
commited in 028stab031:
 openvz-Kconfig.patch
 openvz-tux.patch
 openvzelx-utsname.patch
 openvzelx_netback.patch
Comment 3 Kirill Korotaev 2007-04-26 11:36:43 EDT
ok. rest patches will be reviewed after 028stab031.

Sergey, I need your help in order to understand some of your other patches.
  openvzelx-tux.patch - what has it to do with tux? imho fixes compilation only
due to nsproxy...
  openvz-vsched.patch - why? what does it fix?
  openvz-utsname2.patch - the same... why?
Comment 4 Sergey Ya. Korshunoff 2007-04-26 17:53:38 EDT
(In reply to comment #3)
>   openvzelx-tux.patch - what has it to do with tux? imho fixes compilation only
> due to nsproxy...
Yes, patch name is wrong (i think nsproxy is a tux part)

>   openvz-vsched.patch - why? what does it fix?
vcpu_online() is not used anywhere in the kernel (there is no one
implementation of this function)

>   openvz-utsname2.patch - the same... why?
the same, it is not used anywhere
Comment 5 Vasily Tarasov 2007-04-27 04:26:31 EDT
*** Bug 544 has been marked as a duplicate of this bug. ***
Comment 6 Kirill Korotaev 2007-04-27 05:21:13 EDT
Evgeniy,

please fix compiltion of RHEL5-OVZ Xen kernel from .spec file.
check that OVZ works fine in Dom0 and in DomU.
Comment 7 Kirill Korotaev 2007-05-28 14:11:03 EDT
fixed in 028stab034
Comment 8 Sergey Ya. Korshunoff 2007-05-29 03:20:38 EDT
(In reply to comment #7)
> fixed in 028stab034

Where it is currently? Is there a inotify-reverse.patch appliend? (because main
2.6.18 do not have a modified inotify_add_watch() function...

There is a stange situation with openvz-el kernel: it do not have a repository,
patch is allways a commulative one, there is no simple way to find a changes in
versions for this kernel. Currenly I have a problem with compiling openvz-el
028.031 for XEN using a commulative patch from you (complains about ldt_pages).
And I succefully compiled a kernel with 028stab027-combined + 027-to-030.patch
from repositiory.

Conslusion: openvz-el need a more verbose info (like changelist with a
timeorder).
Comment 9 Sergey Ya. Korshunoff 2007-05-29 03:27:10 EDT
(In reply to comment #7)
> fixed in 028stab034

Some info abount XEN: xen-3.1.0 have a memory problems which do not allow to
use it in real life

Currently I going to test openvz-el-xen for a memory problems (dom0 w/o any
domU).

Test description:
   open in MidnightCommander 1.4Gb tar.bzip2 archive (press ENTER on archive
name). Compiling a kernel is a plus. After some time linux-xen-3.1.0 give you a
kernel dump with message abount XEN list-memory problem
Comment 10 Sergey Ya. Korshunoff 2007-05-29 03:29:41 EDT
(In reply to comment #9)
> (In reply to comment #7)
> Test description:
>    open in MidnightCommander 1.4Gb tar.bzip2 archive (press ENTER on archive
> name). Compiling a kernel is a plus. After some time linux-xen-3.1.0 give you a
> kernel dump with message abount XEN list-memory problem

Abount hardware: Intel Celeron 1.7 Ghz with 512Mb of RAM (and the same kernel
with the same kernel config but w/o XEN-patch-monitor do not have such problems
with memory)
Comment 11 Kirill Korotaev 2007-05-29 03:49:20 EDT
> Where it is currently? Is there a inotify-reverse.patch appliend? (because main
> 2.6.18 do not have a modified inotify_add_watch() function...

it is currently in CVS. In a couple of days I will build a kernel based on this
patchset.

> There is a stange situation with openvz-el kernel: it do not have a repository,
> patch is allways a commulative one, there is no simple way to find a changes in
> versions for this kernel.

Yes, you are right, there is such problem. This is due to the fact that git is
not suitable for such developement. The process looks like the following:
- we have a bunch of RHEL patches and spec's where we simply add one OVZ core
patch and driver related patches (since they are not OVZ modifications related)
- OVZ patch is constructed from the list of patches with similar to quilt
process.

> Currenly I have a problem with compiling openvz-el
> 028.031 for XEN using a commulative patch from you (complains about ldt_pages).
> And I succefully compiled a kernel with 028stab027-combined + 027-to-030.patch
> from repositiory.

This is fixed by Evgeniy in 028.034. The reason is simple - in 028.028 4GB
split feature was added. It is a huge patch actually which increases normal
zone size up to 3gb on i686. RHEL5 dropped this feature, while we need it quite
much. So I ported it.

> Conslusion: openvz-el need a more verbose info (like changelist with a
> timeorder).

Will do one day. Will broken-out patch list suit you?
Meanwhile you can simply use `interdiff -p1 patch-028stab027-core
patch-028stab031-core` to see all the changes.
Comment 12 Kirill Korotaev 2007-05-29 03:54:43 EDT
> Where it is currently? Is there a inotify-reverse.patch appliend? (because main
> 2.6.18 do not have a modified inotify_add_watch() function...

ohh, forget to answer on this.
no, inotify-reverse is not applied, since this feature is required.
otherwise checkpointing sometimes don't work on RHEL5 VEs since they use
inotify interfaces in udev code :/

Intead, I suggest to rename functions in appropriate way - i.e. keep the
original functions with the same argument set, while add new functions with
additional argument which we need.

we have a separate bug #552 for this issue.