x86/PoD: tie together P2M update and increment of entry count

When not holding the PoD lock across the entire region covering P2M
update and stats update, the entry count - if to be incorrect at all -
should indicate too large a value in preference to a too small one, to
avoid functions bailing early when they find the count is zero. However,
instead of moving the increment ahead (and adjust back upon failure),
extend the PoD-locked region.

Fixes: 99af3cd40b ("x86/mm: Rework locking in the PoD layer")
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: George Dunlap <george.dunlap@cloud.com>
master commit: cc950c49ae
master date: 2024-03-21 09:48:10 +0100

xen/arch/x86/mm/p2m-pod.c

@@ -1370,19 +1370,28 @@ mark_populate_on_demand(struct domain *d, unsigned long gfn_l,
         }
     }
 
+    /*
+     * P2M update and stats increment need to collectively be under PoD lock,
+     * to prevent code elsewhere observing PoD entry count being zero despite
+     * there actually still being PoD entries (created by the p2m_set_entry()
+     * invocation below).
+     */
+    pod_lock(p2m);
+
     /* Now, actually do the two-way mapping */
     rc = p2m_set_entry(p2m, gfn, INVALID_MFN, order,
                        p2m_populate_on_demand, p2m->default_access);
     if ( rc == 0 )
     {
-        pod_lock(p2m);
         p2m->pod.entry_count += 1UL << order;
         p2m->pod.entry_count -= pod_count;
         BUG_ON(p2m->pod.entry_count < 0);
-        pod_unlock(p2m);
-
-        ioreq_request_mapcache_invalidate(d);
     }
+
+    pod_unlock(p2m);
+
+    if ( rc == 0 )
+        ioreq_request_mapcache_invalidate(d);
     else if ( order )
     {
         /*