Bug 1007698

Can reproduce on RHEL7:
# rpm -qa libvirt qemu-kvm
libvirt-1.1.1-4.el7.x86_64
qemu-kvm-1.5.3-3.el7.x86_64

Fixed upstream with v1.2.0-76-g231656b -- v1.2.0-79-gea130e3:

commit ea130e3bf666397a05a674ffcf15b9ab170b2255
Author: Martin Kletzander <mkletzan>
Date:   Mon Dec 9 11:32:48 2013 +0100

    conf: don't format memtune with unlimited values
    
commit 8d7c668e64b5bcd2d08aa5057c9aff43d1f73dfd
Author: Martin Kletzander <mkletzan>
Date:   Wed Dec 4 18:59:52 2013 +0100

    qemu: Fix minor inconsistency in error message

commit 0c2fdd7b14cbfc6ced77ed2a24f01f07a8a2f657
Author: Martin Kletzander <mkletzan>
Date:   Wed Dec 4 18:56:02 2013 +0100

    qemu: Report VIR_DOMAIN_MEMORY_PARAM_UNLIMITED properly
    
commit 231656bbeb9e4d3bedc44362784c35eee21cf0f4
Author: Martin Kletzander <mkletzan>
Date:   Wed Dec 4 16:54:29 2013 +0100

    cgroups: Redefine what "unlimited" means wrt memory limits

1> for patches in comment 3 , it seems like that is unrelated to this bug  .
and i found the following patch , but i am not sure if it is the correct one :
https://www.redhat.com/archives/libvir-list/2014-March/msg00921.html

2> for this bug , according to the description verify it via following steps:

version:

kernel-3.10.0-203.el7.x86_64
qemu-kvm-rhev-2.1.2-8.el7.x86_64
libvirt-1.2.8-6.el7.x86_64

steps to verify :

1. negative value

# virsh schedinfo r7
Scheduler      : posix
cpu_shares     : 1024
vcpu_period    : 100000
vcpu_quota     : -1
emulator_period: 100000
emulator_quota : -1

1.1 
set value as -1 :
# virsh schedinfo r7 --set cpu_shares=-1 --live
Scheduler      : posix
cpu_shares     : 262144
vcpu_period    : 100000
vcpu_quota     : -1
emulator_period: 100000
emulator_quota : -1

# virsh dumpxml r7 | grep cputune -A 5
  <cputune>
    <shares>262144</shares>    <=== the value is 262114 , same as return of  schedinfo
  </cputune>

1.2 
set value as -100 :
# virsh schedinfo r7 --set cpu_shares=-100 --live
Scheduler      : posix
cpu_shares     : 262144
vcpu_period    : 100000
vcpu_quota     : -1
emulator_period: 100000
emulator_quota : -1

# virsh dumpxml r7 | grep cputune -A 5
  <cputune>
    <shares>262144</shares>   <=== the value is 262114 , same as return of  schedinfo
  </cputune>
  <resource>

1.3 
set large value ,  larger values are capped at the maximum.

# virsh schedinfo r7 --set cpu_shares=18446744073709551615 --live
Scheduler      : posix
cpu_shares     : 262144
vcpu_period    : 100000
vcpu_quota     : -1
emulator_period: 100000
emulator_quota : -1

# virsh dumpxml r7 | grep cputune -A 3
  <cputune>
    <shares>262144</shares>
  </cputune>


2.
for set value 0 and 1, the minimal value of 2 will be added to domian xml.

# virsh schedinfo r7 --set cpu_shares=0 --live
Scheduler      : posix
cpu_shares     : 2
vcpu_period    : 100000
vcpu_quota     : -1
emulator_period: 100000
emulator_quota : -1

# virsh dumpxml r7 | grep cputune -A 3
  <cputune>
    <shares>2</shares>    <=====the minimal value of 2 was added , and same as return schedinfo
  </cputune>
  
# virsh schedinfo r7 --set cpu_shares=1 --live
Scheduler      : posix
cpu_shares     : 2
vcpu_period    : 100000
vcpu_quota     : -1
emulator_period: 100000
emulator_quota : -1

# virsh dumpxml r7 | grep cputune -A 3
  <cputune>
    <shares>2</shares>    <=====the minimal value of 2 was added , and same as return schedinfo
  </cputune>

3.
boundary value , larger values are capped at the maximum.:

# virsh schedinfo r7 --set cpu_shares=262145 --live
Scheduler      : posix
cpu_shares     : 262144
vcpu_period    : 100000
vcpu_quota     : -1
emulator_period: 100000
emulator_quota : -1

# virsh dumpxml r7 | grep cputune -A 3
  <cputune>
    <shares>262144</shares>  <=== the value is 262114 , same as return of  schedinfo
  </cputune>


4.
set value as -1 with --config
# virsh schedinfo r7 --set cpu_shares=-1 --config
Scheduler      : posix
cpu_shares     : 18446744073709551615
vcpu_period    : 0
vcpu_quota     : 0
emulator_period: 0
emulator_quota : 0

check in domain xml , the value will not be changed.
# virsh dumpxml r7 | grep cputune -A 3
  <cputune>
    <shares>2</shares>
  </cputune>

restart domain , check the value which will be capped at the maximum.
# virsh destroy r7 ; virsh start r7
Domain r7 destroyed

Domain r7 started

# virsh schedinfo r7
Scheduler      : posix
cpu_shares     : 262144   
vcpu_period    : 100000
vcpu_quota     : -1
emulator_period: 100000
emulator_quota : -1

# virsh dumpxml r7 | grep cputune -A 5
  <cputune>
    <shares>262144</shares>    <=====the value was capped as maximum
  </cputune>


5.for a shut off domain:

# virsh list --all
 Id    Name                           State
----------------------------------------------------
 -     r7                             shut off

5.1
set value -1 with --config 
# virsh schedinfo r7 --set cpu_shares=-1 --config
Scheduler      : posix
cpu_shares     : 18446744073709551615
vcpu_period    : 0
vcpu_quota     : 0
emulator_period: 0
emulator_quota : 0
# virsh dumpxml r7 | grep cputune -A 5
  <cputune>
    <shares>18446744073709551615</shares>
  </cputune>

start the domain , the value is capped at the maximum :

# virsh start r7
Domain r7 started

# virsh dumpxml r7 | grep cputune -A 5
  <cputune>
    <shares>262144</shares>
  </cputune>

# virsh schedinfo r7
Scheduler      : posix
cpu_shares     : 262144
vcpu_period    : 100000
vcpu_quota     : -1
emulator_period: 100000
emulator_quota : -1


5.2 
for a shut off domain , set value 0 
# virsh schedinfo r7 --set cpu_shares=0 --config
Scheduler      : posix
cpu_shares     : 0
vcpu_period    : 0
vcpu_quota     : 0
emulator_period: 0
emulator_quota : 0

start to check , the minimal value of 2 will be added to domain.
# virsh start r7
Domain r7 started

# virsh schedinfo r7
Scheduler      : posix
cpu_shares     : 2    
vcpu_period    : 100000
vcpu_quota     : -1
emulator_period: 100000
emulator_quota : -1

# virsh dumpxml r7 | grep cputune -A 3
  <cputune>
    <shares>2</shares>
  </cputune>

According to bug description , now the cpu_shares value of domain xml are consistent with return value of schedinfo .

3> note : for --config option , a small question i am not sure if it is expected result .

# virsh schedinfo r7 --set cpu_shares=-1 --config
Scheduler      : posix
cpu_shares     : 18446744073709551615   <==== return a large value 
vcpu_period    : 0
vcpu_quota     : 0
emulator_period: 0
emulator_quota : 0

but it has no effect on domain , the value  is capped at the maximum (262144) after restart domain . details like above steps 4 and step 5 .

That is actually the desired output.  For values 0 and 1 the output is 2 because that's what you'll get from cgroups when you set 0 or 1 there.  This was intentionally done so that the output of virsh matches the data kernel operates with.  Different kernels might have different boundaries, so we cannot say/enforce any real limits.

However, lower or higher values are kept in the XML as specified.  So if you get the info from running domain (dumpxml without '--inactive' or schedinfo, while running) you'll get the capped value according to kernel.  If, however you get the info when the domain is not running (or use '--inactive' for dumpxml or'--config' for schedinfo), the value is taken right from the XML and it shows the exact data you specified there.

And -1 wraps to maximum we can use.  All this is in the manual already, and makes sense for me.  If there is a bit you want to adjust, I can have a look at it, but it makes sense from my point of view.

Thanks for the info . 
According to comment 7 and comment 8 , move the bug to verified.

Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory, and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

https://rhn.redhat.com/errata/RHSA-2015-0323.html