Index: sys/vm/vm_fault.c
===================================================================
--- sys/vm/vm_fault.c
+++ sys/vm/vm_fault.c
@@ -81,6 +81,7 @@
 
 #include <sys/param.h>
 #include <sys/systm.h>
+#include <sys/counter.h>
 #include <sys/kernel.h>
 #include <sys/lock.h>
 #include <sys/mman.h>
@@ -162,12 +163,30 @@
     "Number of page allocation attempts in page fault handler before it "
     "triggers OOM handling");
 
+static SYSCTL_NODE(_vm_stats, OID_AUTO, fault, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
+    "VM fault statistics");
+
 static int vm_pfault_oom_wait = 10;
 SYSCTL_INT(_vm, OID_AUTO, pfault_oom_wait, CTLFLAG_RWTUN,
     &vm_pfault_oom_wait, 0,
     "Number of seconds to wait for free pages before retrying "
     "the page fault handler");
 
+static COUNTER_U64_DEFINE_EARLY(vmpfw_shared_read);
+SYSCTL_COUNTER_U64(_vm_stats_fault, OID_AUTO, vmpfw_shared_read,
+    CTLFLAG_RD, &vmpfw_shared_read,
+    "Number of waits on valid+sbusy pages caused by read faults");
+
+static COUNTER_U64_DEFINE_EARLY(vmpfw_shared_cow);
+SYSCTL_COUNTER_U64(_vm_stats_fault, OID_AUTO, vmpfw_shared_cow,
+    CTLFLAG_RD, &vmpfw_shared_cow,
+    "Number of waits on valid+sbusy pages caused by CoW faults");
+
+static COUNTER_U64_DEFINE_EARLY(vmpfw_exclusive);
+SYSCTL_COUNTER_U64(_vm_stats_fault, OID_AUTO, vmpfw_exclusive,
+    CTLFLAG_RD, &vmpfw_exclusive,
+    "Number of waits on invalid/xbusy pages");
+
 static inline void
 fault_page_release(vm_page_t *mp)
 {
@@ -1223,11 +1242,11 @@
  * and we could end up trying to pagein and pageout the same page
  * simultaneously.
  *
- * We can theoretically allow the busy case on a read fault if the page
- * is marked valid, but since such pages are typically already pmap'd,
- * putting that special case in might be more effort then it is worth.
- * We cannot under any circumstances mess around with a shared busied
- * page except, perhaps, to pmap it.
+ * We can theoretically allow the busy case on a read fault or a CoW
+ * fault if the page is marked valid, but since such pages are typically
+ * already pmap'd, putting that special case in might be more effort than
+ * it is worth.  We cannot under any circumstances mess around with a
+ * shared busied page except, perhaps, to pmap it.
  */
 static void
 vm_fault_busy_sleep(struct faultstate *fs)
@@ -1244,9 +1263,25 @@
 	}
 	vm_object_pip_wakeup(fs->object);
 	unlock_map(fs);
-	if (fs->m == vm_page_lookup(fs->object, fs->pindex))
+	if (fs->m == vm_page_lookup(fs->object, fs->pindex)) {
+		/*
+		 * Provide statistics on waits that we issue for shared
+		 * and valid pages that could potentially be avoided,
+		 * in comparison to the expected common case of waits
+		 * that cannot be avoided.
+		 */
+		if (vm_page_all_valid(fs->m) && vm_page_sbusied(fs->m)) {
+			if ((fs->fault_type &
+			    (VM_PROT_COPY | VM_PROT_WRITE)) == 0)
+				counter_u64_add(vmpfw_shared_read, 1);
+			else if (fs->object != fs->first_object)
+				counter_u64_add(vmpfw_shared_cow, 1);
+			else
+				counter_u64_add(vmpfw_exclusive, 1);
+		} else
+			counter_u64_add(vmpfw_exclusive, 1);
 		vm_page_busy_sleep(fs->m, "vmpfw", false);
-	else
+	} else
 		VM_OBJECT_WUNLOCK(fs->object);
 	VM_CNT_INC(v_intrans);
 	vm_object_deallocate(fs->first_object);