Xen devel - Dec 2006 - [PATCH] Fix e820 mapping limit

The changeset ''12803:df5fa63490f4da7b65c56087a68783dbcb7944f8''
added a new
hypercall XENMEM_set_memory_map for specifying the e820 mapping. There was
a small bug in the userspace side of this change though - the e820 mapping
was specified based on the ''memory_static_min'' domain info
parameter which
means the memory map size is clamped to the value ''memory''
config option.

What we actually want is to setup the e820 map limit based on
''maxmem''
config option, so that we can balloon up the guest to its max limit.

 eg
  If we boot with ''memory=400'' and
''maxmem=800'', the e820 map should be
  set to 800 MB, and the guest should allocate  only 400 MB at startup.
  It should then be possible to use ''xm mem-set'' after boot to
assgin
  upto this 800 MB.

Thus, the attached patch fixed the image.py class to setup the e820 mapping
based on ''memory_static_max'' which is in turn based on
''maxmem''. Booting
with this patch I can verify that the guest kernel sees a 800 MB e820 map,
so this fixes the initial problem.

There appears to be a second issue somewhere in the stack though, because
even if do ''xm mem-set <domid> 600'', while the ballon
driver in the guest
will see the 600 MB target, it will still never try to allocate its 
increased target.

eg

  # grep mem /etc/xen/demo 
  memory = 410
  maxmem = 800
  # xm create demo

In the guest  ''dmesg'' shows:

  BIOS-provided physical RAM map:
   Xen: 0000000000000000 - 0000000032800000 (usable)

Back on the host:

  # xm list --long demo | grep mem
    (memory 410)
    (maxmem 800)
    (shadow_memory 0)
    (memory_dynamic_min 410)
    (memory_dynamic_max 410)

  # xm mem-set demo 600

  # xm list --long demo | grep mem
    (memory 600)
    (maxmem 800)
    (shadow_memory 0)
    (memory_dynamic_min 410)
    (memory_dynamic_max 410)

Looking at balloon driver in guest:

  $ cat /tmp/balloon 
  Current allocation:   419840 kB
  Requested target:     614400 kB
  Low-mem balloon:      407552 kB
  High-mem balloon:          0 kB
  Driver pages:              0 kB
  Xen hard limit:       419840 kB

And the memory target spceified in xenstore:

  # xenstore-ls | grep target
    target = "614400"


So, for some reason even though the guest e820 map is correct, and the memory
target got increased to 600 MB, the balloon driver is not actually ever trying
to increase its allocation.   Do we need to also fixup the
memory_dynamic_min/max
parameters in XenD too ?  xm mem-set only affects memory_static_min

Anyway, attaching the patch to fix e820 map limits.

  Signed-off-by: Daniel Berrange <berrange@redhat.com

Regards,
Dan.
-- 
|=- Red Hat, Engineering, Emerging Technologies, Boston.  +1 978 392 2496 -=|
|=-           Perl modules: http://search.cpan.org/~danberr/              -=|
|=-               Projects: http://freshmeat.net/~danielpb/               -=|
|=-  GnuPG: 7D3B9505   F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505  -=| 


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On Tue, Dec 12, 2006 at 06:25:25PM +0000, Daniel P. Berrange
wrote:
> The changeset
''12803:df5fa63490f4da7b65c56087a68783dbcb7944f8'' added a new
> hypercall XENMEM_set_memory_map for specifying the e820 mapping. There was
> a small bug in the userspace side of this change though - the e820 mapping
> was specified based on the ''memory_static_min'' domain
info parameter which
> means the memory map size is clamped to the value
''memory'' config option.
> 
> What we actually want is to setup the e820 map limit based on
''maxmem''
> config option, so that we can balloon up the guest to its max limit.
> 
>  eg
>   If we boot with ''memory=400'' and
''maxmem=800'', the e820 map should be
>   set to 800 MB, and the guest should allocate  only 400 MB at startup.
>   It should then be possible to use ''xm mem-set'' after
boot to assgin
>   upto this 800 MB.
> 
> Thus, the attached patch fixed the image.py class to setup the e820 mapping
> based on ''memory_static_max'' which is in turn based on
''maxmem''. Booting
> with this patch I can verify that the guest kernel sees a 800 MB e820 map,
> so this fixes the initial problem.
> 
> There appears to be a second issue somewhere in the stack though, because
> even if do ''xm mem-set <domid> 600'', while the
ballon driver in the guest
> will see the 600 MB target, it will still never try to allocate its 
> increased target.
Turns out this was a bug introduced in the refactoring of memory config
parameters from Xen API work. The memory & memmax parameters were getting
initialized the wrong way around. With this fixed & the e820 map fix, 
memory ballooning works perfectly:

  # xm create demo
  Using config file "/etc/xen/demo".
  Going to boot Fedora Core (2.6.18-1.2769.2.2.el5xen)
    kernel: /vmlinuz-2.6.18-1.2769.2.2.el5xen
    initrd: /initrd-2.6.18-1.2769.2.2.el5xen.img
  Started domain demo
  # xm list demo
  Name                                      ID   Mem VCPUs      State   Time(s)
  demo                                      16   410     2     -b----      0.4
  # xm mem-set demo 600
  # xm list demo
  Name                                      ID   Mem VCPUs      State   Time(s)
  demo                                      16   600     2     -b----      4.2


Attaching an updated patch which fixes both the memory ballooning issues
in one go

  Signed-off-by: Daniel P. Berrange <berrange@redhat.com>

Regards,
Dan.
-- 
|=- Red Hat, Engineering, Emerging Technologies, Boston.  +1 978 392 2496 -=|
|=-           Perl modules: http://search.cpan.org/~danberr/              -=|
|=-               Projects: http://freshmeat.net/~danielpb/               -=|
|=-  GnuPG: 7D3B9505   F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505  -=| 


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel