On 11.08.2014 11:16, Daniel P. Berrange wrote:
On Tue, Aug 05, 2014 at 11:16:53AM -0600, Jim Fehlig wrote:
> Jim Fehlig wrote:
>> Until recently, the qemu driver has always succeeded in loading even on
>> a system booted to Xen. Commit be0782e1 broke this, hence libvirtd no
>> longer loads
>>
>> xendom0 # /usr/sbin/libvirtd -l
>> 2014-08-04 22:27:31.586+0000: 35859: info : libvirt version: 1.2.8
>> 2014-08-04 22:27:31.586+0000: 35859: error :
>> virFileGetDefaultHugepageSize:2958 : internal error: Unable to parse
>> /proc/meminfo
>> 2014-08-04 22:27:31.586+0000: 35859: error : virStateInitialize:749 :
>> Initialization of QEMU state driver failed: internal error: Unable to
>> parse /proc/meminfo
>> 2014-08-04 22:27:31.586+0000: 35859: error : daemonRunStateInit:922 :
>> Driver state initialization failed
>>
>> Problem is the Xen kernel does not provide any huge page info in
>> /proc/meminfo
>>
>> xendom0 # cat /proc/meminfo | grep -i huge
>> xendom0 #
>>
>
> BTW, I think this would be the case for any kernel with
> CONFIG_HUGETLB_PAGE turned off.
Yes, we should be robust to this situation.
And as of d26e810838fad we are fault tolerant.
>> Michal, short of always building with '--without-qemu' on a Xen system,
>> is it possible to make detection of huge pages less fatal wrt loading
>> the qemu driver?
>>
>
> E.g. assume zero/default for entries not present?
If the dirs / files don't exist, we should just not report any huge
page info at all.
Regards,
Daniel
Michal