diff --git a/src/lsu/lsu.sv b/src/lsu/lsu.sv
index e10183a9e..12ab9930e 100644
--- a/src/lsu/lsu.sv
+++ b/src/lsu/lsu.sv
@@ -426,7 +426,7 @@ module lsu import cvw::*;  #(parameter cvw_t P) (
   if(MISALIGN_SUPPORT) begin
     subwordreaddouble #(P.LLEN) subwordread(.ReadDataWordMuxM(LittleEndianReadDataWordM), .PAdrM(PAdrM[2:0]), .BigEndianM,
       .FpLoadStoreM, .Funct3M(LSUFunct3M), .ReadDataM);
-    subwordwritedouble #(P.LLEN) subwordwrite(.LSUFunct3M, .PAdrM(PAdrM[2:0]), .FpLoadStoreM, .BigEndianM, .IMAFWriteDataM, .LittleEndianWriteDataM);
+    subwordwritedouble #(P.LLEN) subwordwrite(.LSUFunct3M, .PAdrM(PAdrM[2:0]), .FpLoadStoreM, .BigEndianM, .CacheableM, .IMAFWriteDataM, .LittleEndianWriteDataM);
   end else begin
     subwordread #(P.LLEN) subwordread(.ReadDataWordMuxM(LittleEndianReadDataWordM), .PAdrM(PAdrM[2:0]), .BigEndianM,
       .FpLoadStoreM, .Funct3M(LSUFunct3M), .ReadDataM);
diff --git a/src/lsu/subwordwritedouble.sv b/src/lsu/subwordwritedouble.sv
index 728a4f4aa..599d71984 100644
--- a/src/lsu/subwordwritedouble.sv
+++ b/src/lsu/subwordwritedouble.sv
@@ -33,6 +33,7 @@ module subwordwritedouble #(parameter LLEN) (
   input logic [2:0]         PAdrM,
   input logic               FpLoadStoreM, 
   input logic               BigEndianM, 
+  input logic               CacheableM,
   input logic [LLEN-1:0]    IMAFWriteDataM,
   output logic [LLEN*2-1:0] LittleEndianWriteDataM
 );
@@ -43,7 +44,13 @@ module subwordwritedouble #(parameter LLEN) (
   logic [4:0]               LengthM;
   // Funct3M[2] is the unsigned bit. mask upper bits.
   // Funct3M[1:0] is the size of the memory access.
-  assign PAdrSwap = BigEndianM ? BigEndianPAdr : {2'b0, PAdrM};
+  // cacheable, BigEndian
+  // 10: PAdrM[2:0]
+  // 11: BigEndianPAdr
+  // 00: 00000
+  // 01: 00111
+  mux4 #(5) OffsetMux(5'b0, 5'b11111, {2'b0, PAdrM}, BigEndianPAdr, {CacheableM, BigEndianM}, PAdrSwap);
+  //assign PAdrSwap = BigEndianM ? BigEndianPAdr : {2'b0, PAdrM};
   /* verilator lint_off WIDTHEXPAND */
   /* verilator lint_off WIDTHTRUNC */
   assign BigEndianPAdr = (LLEN/4) - PAdrM - LengthM;