Re: HVM/PVH Balloon crash

[Jan Beulich <jbeulich@suse.com>]

On 27.09.2021 00:53, Elliott Mitchell wrote:
> On Wed, Sep 15, 2021 at 08:05:05AM +0200, Jan Beulich wrote:
>> On 15.09.2021 04:40, Elliott Mitchell wrote:
>>> On Tue, Sep 07, 2021 at 05:57:10PM +0200, Jan Beulich wrote:
>>>> On 07.09.2021 17:03, Elliott Mitchell wrote:
>>>>>  Could be this system is in an
>>>>> intergenerational hole, and some spot in the PVH/HVM code makes an
>>>>> assumption of the presence of NPT guarantees presence of an operational
>>>>> IOMMU.  Otherwise if there was some copy and paste while writing IOMMU
>>>>> code, some portion of the IOMMU code might be checking for presence of
>>>>> NPT instead of presence of IOMMU.
>>>>
>>>> This is all very speculative; I consider what you suspect not very likely,
>>>> but also not entirely impossible. This is not the least because for a
>>>> long time we've been running without shared page tables on AMD.
>>>>
>>>> I'm afraid without technical data and without knowing how to repro, I
>>>> don't see a way forward here.
>>>
>>> Downtimes are very expensive even for lower-end servers.  Plus there is
>>> the issue the system wasn't meant for development and thus never had
>>> appropriate setup done.
>>>
>>> Experimentation with a system of similar age suggested another candidate.
>>> System has a conventional BIOS.  Might some dependancies on the presence
>>> of UEFI snuck into the NPT code?
>>
>> I can't think of any such, but as all of this is very nebulous I can't
>> really rule out anything.
> 
> Getting everything right to recreate is rather inexact.  Having an
> equivalent of `sysctl` to turn on the serial console while running might
> be handy...
> 
> Luckily get things together and...
> 
> (XEN) mm locking order violation: 48 > 16
> (XEN) Xen BUG at mm-locks.h:82

Would you give the patch below a try? While against current staging it
looks to apply fine to 4.14.3.

Jan

x86/PoD: defer nested P2M flushes

With NPT or shadow in use, the p2m_set_entry() -> p2m_pt_set_entry() ->
write_p2m_entry() -> p2m_flush_nestedp2m() call sequence triggers a lock
order violation when the PoD lock is held around it. Hence such flushing
needs to be deferred. Steal the approach from p2m_change_type_range().

Reported-by: Elliott Mitchell <ehem+xen@m5p.com>
Signed-off-by: Jan Beulich <jbeulich@suse.com>

--- a/xen/arch/x86/mm/p2m-pod.c
+++ b/xen/arch/x86/mm/p2m-pod.c
@@ -24,6 +24,7 @@
 #include <xen/mm.h>
 #include <xen/sched.h>
 #include <xen/trace.h>
+#include <asm/hvm/nestedhvm.h>
 #include <asm/page.h>
 #include <asm/paging.h>
 #include <asm/p2m.h>
@@ -494,6 +495,13 @@ p2m_pod_offline_or_broken_replace(struct
 static int
 p2m_pod_zero_check_superpage(struct p2m_domain *p2m, gfn_t gfn);
 
+static void pod_unlock_and_flush(struct p2m_domain *p2m)
+{
+    pod_unlock(p2m);
+    p2m->defer_nested_flush = false;
+    if ( nestedhvm_enabled(p2m->domain) )
+        p2m_flush_nestedp2m(p2m->domain);
+}
 
 /*
  * This function is needed for two reasons:
@@ -514,6 +522,7 @@ p2m_pod_decrease_reservation(struct doma
 
     gfn_lock(p2m, gfn, order);
     pod_lock(p2m);
+    p2m->defer_nested_flush = true;
 
     /*
      * If we don't have any outstanding PoD entries, let things take their
@@ -665,7 +674,7 @@ out_entry_check:
     }
 
 out_unlock:
-    pod_unlock(p2m);
+    pod_unlock_and_flush(p2m);
     gfn_unlock(p2m, gfn, order);
     return ret;
 }
@@ -1144,8 +1153,10 @@ p2m_pod_demand_populate(struct p2m_domai
      * won't start until we're done.
      */
     if ( unlikely(d->is_dying) )
-        goto out_fail;
-
+    {
+        pod_unlock(p2m);
+        return false;
+    }
 
     /*
      * Because PoD does not have cache list for 1GB pages, it has to remap
@@ -1167,6 +1178,8 @@ p2m_pod_demand_populate(struct p2m_domai
                               p2m_populate_on_demand, p2m->default_access);
     }
 
+    p2m->defer_nested_flush = true;
+
     /* Only reclaim if we're in actual need of more cache. */
     if ( p2m->pod.entry_count > p2m->pod.count )
         pod_eager_reclaim(p2m);
@@ -1229,8 +1242,9 @@ p2m_pod_demand_populate(struct p2m_domai
         __trace_var(TRC_MEM_POD_POPULATE, 0, sizeof(t), &t);
     }
 
-    pod_unlock(p2m);
+    pod_unlock_and_flush(p2m);
     return true;
+
 out_of_memory:
     pod_unlock(p2m);
 
@@ -1239,12 +1253,14 @@ out_of_memory:
            p2m->pod.entry_count, current->domain->domain_id);
     domain_crash(d);
     return false;
+
 out_fail:
-    pod_unlock(p2m);
+    pod_unlock_and_flush(p2m);
     return false;
+
 remap_and_retry:
     BUG_ON(order != PAGE_ORDER_2M);
-    pod_unlock(p2m);
+    pod_unlock_and_flush(p2m);
 
     /*
      * Remap this 2-meg region in singleton chunks. See the comment on the