How to make JPA use a single join to get columns with conditions for both sides

My problem is the following,

There are two entity classes, let's call them Entity1 and Entity2 with One-to-Many relationship in between, i.e. one Entity1 contains multiple Entity2s, and Entity2 may have only one Entity1:

@Entity
@Table(name = "entity1")
public class Entity1 {

    int x;
    int y;
    ...
    @LazyCollection(LazyCollectionOption.TRUE)
    @OneToMany(mappedBy = "e1", cascade = CascadeType.ALL, orphanRemoval = true)
    private Set<Entity2> entity2s = new HashSet<>();

}


@Entity
@Table(name = "entity2")
public class Entity2 {

    int a;
    int b;  
    ...
    @ManyToOne
    @JoinColumn(name = "entity1_id")
    @JsonBackReference
    private Entity1 e1;

}

Now I would like to issue a query for retrieving Entity2s with WHERE conditions for both Entity2 and its corresponding Entity1:

@Query("SELECT " +
       " e2 " +
       "FROM " +
       " Entity2 e2 " +
       "WHERE " +
       " e2.a = '<val1>' AND e2.b = '<val2>' AND e2.e1.x = '<val3>' AND e2.e1.y ='<val4>'")
List<Entity2> findMyEntity2s(
        @Param...,
        @Param...,
);

So the problem with this approach is that, it indeed gets desired Entity2s by cross joining entity1 and entity2 tables with specified WHERE conditions BUT it fetches e1s for each of those Entity2s in the result with a separate query.
So for example if the result of join is 5 Entity2s, there will be 5 additional queries to entity1 table.

I tried to set @ManyToOne in Entity2 as @ManyToOne(fetch = FetchType.LAZY) but it didn't help. I guess that's expected because LAZY would simply postpone the retrieval of e1s but wouldn't eliminate it completely.

Next, I read about @EntityGraph, and added it to Entity2:

@Entity
@Table(name = "entity2")
@NamedEntityGraph(name = "graph.entity2.entity1",
    attributeNodes = { @NamedAttributeNode("e1") })
public class Entity2 {

    int a;
    int b;  
    ...
    @ManyToOne
    @JoinColumn(name = "entity1_id")
    @JsonBackReference
    private Entity1 e1;

}

and in the repository, I added it as:

@EntityGraph(value = "graph.entity2.entity1")
@Query("SELECT " +
       " e2 " +
       "FROM " +
       " Entity2 e2 " +
       "WHERE " +
       " e2.a = '<val1>' AND e2.b = '<val2>' AND e2.e1.x = '<val3>' AND e2.e1.y ='<val4>'")
List<Entity2> findMyEntity2s(
        @Param...,
        @Param...,
);

In this case, the separate SQL queries disappear, EntityGraph does left join and its result contains columns from both entity1 and entity2, BUT because the conditions for e2.e1 are still in WHERE clause, it adds ONE MORE unnecessary cross join with entity1 table (e2.e1 conditions are checked in that cross join).

I couldn't find a way to get rid of that extra cross join, so now I'm using the following query:

@EntityGraph(value = "graph.entity2.entity1")
@Query("SELECT " +
       " e2 " +
       "FROM " +
       " Entity2 e2 " +
       "WHERE " +
       " e2.a = '<val1>' AND e2.b = '<val2>'")
List<Entity2> findMyEntity2s(
        @Param...,
        @Param...,
);

So basically I get Entity2s and in the application I filter out based on conditions of Entity1 (e2.e1.x = '<val3>' AND e2.e1.y ='<val4>').

Is there a way to make it work with a single join only, for both entity's conditions, not only Entity2 conditions? The way I'm doing it now, does not seem correct and efficient to me, and I feel there's a way to do that using repository method only, without involving the app. Would appreciate any help on this

UPD. Read about nativeQuery option (nativeQuery = true) for @Query annotation, which allows specifying a raw query and thus bypassing entity-based query, but the query still fetches many-to-one e1 field, using entity1_id (entity graph was disabled). I tried to enable entity graph but it dropped exception stating that entity graph cannot be used with native query, which is expected