return ERR_PTR(ret);
 }
 
+static int nvme_global_check_duplicate_ids(struct nvme_subsystem *this,
+               struct nvme_ns_ids *ids)
+{
+       struct nvme_subsystem *s;
+       int ret = 0;
+
+       /*
+        * Note that this check is racy as we try to avoid holding the global
+        * lock over the whole ns_head creation.  But it is only intended as
+        * a sanity check anyway.
+        */
+       mutex_lock(&nvme_subsystems_lock);
+       list_for_each_entry(s, &nvme_subsystems, entry) {
+               if (s == this)
+                       continue;
+               mutex_lock(&s->lock);
+               ret = nvme_subsys_check_duplicate_ids(s, ids);
+               mutex_unlock(&s->lock);
+               if (ret)
+                       break;
+       }
+       mutex_unlock(&nvme_subsystems_lock);
+
+       return ret;
+}
+
 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
                struct nvme_ns_ids *ids, bool is_shared)
 {
        struct nvme_ctrl *ctrl = ns->ctrl;
        struct nvme_ns_head *head = NULL;
-       int ret = 0;
+       int ret;
+
+       ret = nvme_global_check_duplicate_ids(ctrl->subsys, ids);
+       if (ret) {
+               dev_err(ctrl->device,
+                       "globally duplicate IDs for nsid %d\n", nsid);
+               return ret;
+       }
 
        mutex_lock(&ctrl->subsys->lock);
        head = nvme_find_ns_head(ctrl->subsys, nsid);
                ret = nvme_subsys_check_duplicate_ids(ctrl->subsys, ids);
                if (ret) {
                        dev_err(ctrl->device,
-                               "duplicate IDs for nsid %d\n", nsid);
+                               "duplicate IDs in subsystem for nsid %d\n",
+                               nsid);
                        goto out_unlock;
                }
                head = nvme_alloc_ns_head(ctrl, nsid, ids);