Correctly track index of first free irq map entry

Any time an IRQ map entry was removed, irq_map_first_free_idx was being
set to the index of the removed entry; this caused problems when
entries were removed in random order since irq_map_first_free_idx was
set to a larger value than the index of the first free map entry, and
in 9beb195fd9 ("Continue searching for an irq map from the start")
the IRQ map allocation code was adjusted to use irq_map_first_free_idx
as a starting point but ultimately scan the entire map if necessary,
including values less than irq_map_first_free_idx.

Remove that workaround and instead make irq_map_first_free_idx do what
the name suggests -- tracking the index of the first free map entry --
by only setting to the index of a newly-freed entry if that index is
lower than the existing irq_map_first_free_idx value.

Reviewed by:	imp, mhorne
Sponsored by:	Amazon
Differential Revision:	https://reviews.freebsd.org/D49543

sys/kern/subr_intr.c

@@ -1733,14 +1733,6 @@ intr_map_irq(device_t dev, intptr_t xref, struct intr_map_data *data)
 			return (i);
 		}
 	}
-	for (i = 0; i < irq_map_first_free_idx; i++) {
-		if (irq_map[i] == NULL) {
-			irq_map[i] = entry;
-			irq_map_first_free_idx = i + 1;
-			mtx_unlock(&irq_map_lock);
-			return (i);
-		}
-	}
 	mtx_unlock(&irq_map_lock);
 
 	/* XXX Expand irq_map table */
@@ -1760,7 +1752,8 @@ intr_unmap_irq(u_int res_id)
 		panic("Attempt to unmap invalid resource id: %u\n", res_id);
 	entry = irq_map[res_id];
 	irq_map[res_id] = NULL;
-	irq_map_first_free_idx = res_id;
+	if (res_id < irq_map_first_free_idx)
+		irq_map_first_free_idx = res_id;
 	mtx_unlock(&irq_map_lock);
 	intr_free_intr_map_data(entry->map_data);
 	free(entry, M_INTRNG);