Zombie guests and the inability to launch guests afterwards

Zombie guests and the inability to launch guests afterwards

[Konrad Rzeszutek Wilk]

So I launched a guest (Windows7 + GPLV some version) and when it shut
itself off it hanged on to some grant entries:

(XEN) gnttab_usage_print_all [ key 'g' pressed
(XEN)       -------- active --------       -------- shared --------
(XEN) [ref] localdom mfn      pin          localdom gmfn     flags
(XEN) grant-table for remote domain:    0 ... no active grant table entries
(XEN)       -------- active --------       -------- shared --------
(XEN) [ref] localdom mfn      pin          localdom gmfn     flags
(XEN) grant-table for remote domain:   18 (v1)
(XEN) [16228]        0 0x2ba341 0x00000001          0 0x07f541 0x19
(XEN) [16339]        0 0x2ba339 0x00000001          0 0x07f539 0x19
(XEN) [16383]        0 0x2bb0eb 0x00000001          0 0x07e6eb 0x19
(XEN)       -------- active --------       -------- shared --------

[root@phenom ~]# xl list
Name                                        ID   Mem VCPUs      State   Time(s)
Domain-0                                     0  9522     7     r-----    3789.0
(null)                                      18     0     4     --ps-d    2021.5
Fedora                                      21  2048     2     -b----     499.7

Fine, I should be able to launch other domains, right?
No. 'xl' tells me:
failed to free memory for the domain

But I do have enough memory:
[root@phenom pciback]# xl info  | grep free
free_memory            : 4507

However, xen hypervisor tells me something is off:

(XEN) mm.c:2019:d0 Error pfn 2ba555: rd=ffff83042fe6c000, od=0000000000000000, caf=180000000000000, taf=7400000000000001
(XEN) mm.c:2019:d0 Error pfn 2ba51b: rd=ffff83042fe6c000, od=0000000000000000, caf=180000000000000, taf=7400000000000001
(XEN) mm.c:2019:d0 Error pfn 2ba580: rd=ffff83042fe6c000, od=0000000000000000, caf=180000000000000, taf=7400000000000001

I presume that those three PFNs are associated with those zombie grants.

Now I would think we should still be able to do still launch/etc even
with a zombie around - and I know for fact that 3.4 Xen could do it.
(The reason I know this is that at Virtual Iron where I worked we had
the support to rename guests that had become zombiefied to keep on working).

So, has anybody looked at this ?