diff --git a/src/main/scala/coupledL2/CoupledL2.scala b/src/main/scala/coupledL2/CoupledL2.scala
index 41b09d29..0093bbd1 100644
--- a/src/main/scala/coupledL2/CoupledL2.scala
+++ b/src/main/scala/coupledL2/CoupledL2.scala
@@ -245,7 +245,8 @@ class CoupledL2(implicit p: Parameters) extends LazyModule with HasCoupledL2Para
     node.edges.in.headOption.foreach { n =>
       n.client.clients.zipWithIndex.foreach {
         case (c, i) =>
-          println(s"\t${i} <= ${c.name}")
+          println(s"\t${i} <= ${c.name};" +
+            s"\tsourceRange: ${c.sourceId.start}~${c.sourceId.end}")
       }
     }
 
diff --git a/src/main/scala/coupledL2/GrantBuffer.scala b/src/main/scala/coupledL2/GrantBuffer.scala
index e1d94d07..b868a82e 100644
--- a/src/main/scala/coupledL2/GrantBuffer.scala
+++ b/src/main/scala/coupledL2/GrantBuffer.scala
@@ -296,6 +296,6 @@ class GrantBuffer(implicit p: Parameters) extends L2Module {
     }
     // pftRespQueue is about to be full, and using back pressure to block All MainPipe Entrance
     // which can SERIOUSLY affect performance, should consider less drastic prefetch policy
-    XSPerfAccumulate(cacheParams, "WARNING_pftRespQueue_about_to_full", noSpaceForSinkPft.getOrElse(false.B))
+    XSPerfAccumulate(cacheParams, "WARNING_pftRespQueue_about_to_full", noSpaceForMSHRPft.getOrElse(false.B))
   }
 }
diff --git a/src/main/scala/coupledL2/SourceB.scala b/src/main/scala/coupledL2/SourceB.scala
index fa90dee7..731aeacc 100644
--- a/src/main/scala/coupledL2/SourceB.scala
+++ b/src/main/scala/coupledL2/SourceB.scala
@@ -47,12 +47,16 @@ class SourceB(implicit p: Parameters) extends L2Module {
     val grantStatus = Input(Vec(grantBufInflightSize, new GrantStatus))
   })
 
+  val dcacheSourceIdStart = edgeIn.client.clients
+    .filter(_.supports.probe)
+    .map(c => c.sourceId.start.U).head
+
   def toTLBundleB(task: SourceBReq) = {
     val b = Wire(new TLBundleB(edgeIn.bundle))
     b.opcode  := task.opcode
     b.param   := task.param
     b.size    := offsetBits.U
-    b.source  := 0.U // make sure there are only 1 client
+    b.source  := dcacheSourceIdStart
     b.address := Cat(task.tag, task.set, 0.U(offsetBits.W))
     b.mask    := Fill(beatBytes, 1.U(1.W))
     b.data    := Cat(task.alias.getOrElse(0.U), 0.U(1.W)) // this is the same as HuanCun