[libvirt] [PATCH] libxl: always enable pae for x86_64 HVM

[Jim Fehlig]

On 01/12/2017 07:50 AM, Martin Kletzander wrote:
> On Thu, Jan 12, 2017 at 07:31:29AM -0700, Jim Fehlig wrote:
>> On 01/12/2017 02:06 AM, Martin Kletzander wrote:
>>> On Wed, Jan 11, 2017 at 05:45:27PM -0700, Jim Fehlig wrote:
>>>> For HVM domains, pae is only set in libxl_domain_build_info when
>>>> explicitly specified in the hypervisor <features> config. This is
>>>> fine for i686 machines, but is incorrect behavior for x86_64 machines
>>>> where pae must always be enabled. See the following discussion for
>>>> additional details
>>>>
>>>> https://www.redhat.com/archives/libvir-list/2017-January/msg00254.html
>>>> Signed-off-by: Jim Fehlig <jfehlig at suse.com>
>>>> ---
>>>> src/libxl/libxl_domain.c | 6 ++++++
>>>> 1 file changed, 6 insertions(+)
>>>>
>>>> diff --git a/src/libxl/libxl_domain.c b/src/libxl/libxl_domain.c
>>>> index fbe7ee5..a5314b0 100644
>>>> --- a/src/libxl/libxl_domain.c
>>>> +++ b/src/libxl/libxl_domain.c
>>>> @@ -410,6 +410,12 @@ libxlDomainDefPostParse(virDomainDefPtr def,
>>>>     if (xenDomainDefAddImplicitInputDevice(def) < 0)
>>>>         return -1;
>>>>
>>>> +    /* For x86_64 HVM, always enable pae */
>>>> +    if (def->os.type == VIR_DOMAIN_OSTYPE_HVM &&
>>>> +        def->os.arch == VIR_ARCH_X86_64) {
>>>> +        def->features[VIR_DOMAIN_FEATURE_PAE] = VIR_TRISTATE_SWITCH_ON;
>>>> +    }
>>>> +
>>>
>>> This will even rewrite explicit <pae enabled='no/>
>>
>> According to docs/schemas/domaincommon.rng, pae is not a tristate :-). One of
>> the options for solving this problem is to make it one
>>
>> https://www.redhat.com/archives/libvir-list/2017-January/msg00254.html
>>
>
> I saw that, but I somehow thought it changed since, I haven't realized
> it's less than a week ago.
>
>>>
>>> I would check that it's not VIR_TRISTATE_SWITCH_OFF, then enable it and
>>> during startup just fail if it is VIR_TRISTATE_SWITCH_ON.
>>
>> I suppose you mean fail if it is *not* VIR_TRISTATE_SWITCH_ON?
>>
>
> No, we add features by default normally, I just thought it's a
> tristate.  Since it doesn't, this patch makes sense as it is.
>
> ACK, safe for freeze if you want to.

Thanks. Since I think the bug fix is noteworthy, I squashed in the below news 
entry and pushed the patch.

Regards,
Jim

--- a/docs/news.xml
+++ b/docs/news.xml
@@ -214,6 +214,17 @@
            default to the general working scenario.
          </description>
        </change>
+      <change>
+        <summary>
+          libxl: always enable pae for x86_64 HVM
+        </summary>
+        <description>
+          By default pae is disabled in libxl. Without an explicit <pae/>
+          setting in the domain <features> configuration, an x86_64 HVM
+          domain would be get an i686 environment. pae should always be enabled
+          for x86_64 HVM domains.
+        </description>
+      </change>