diff --git a/.github/cases/blastoise/default.json b/.github/cases/blastoise/default.json
index 3a46b2e41..14a4b6fd1 100644
--- a/.github/cases/blastoise/default.json
+++ b/.github/cases/blastoise/default.json
@@ -5,9 +5,9 @@
   "mlir.stripmining": 26736,
   "asm.mmm": 92573,
   "asm.smoke": 4867,
-  "intrinsic.conv2d_less_m2": 2612,
+  "intrinsic.conv2d_less_m2": 2594,
   "intrinsic.linear_normalization": 3280,
-  "intrinsic.softmax": 8348,
+  "intrinsic.softmax": 8347,
   "codegen.vaadd_vv": 91859,
   "codegen.vaadd_vx": 253663,
   "codegen.vaaddu_vv": 91859,
@@ -499,14 +499,14 @@
   "codegen.vfredusum_vs": 122275,
   "codegen.vfredmax_vs": 122275,
   "codegen.vfredmin_vs": 122275,
-  "rvv_bench.ascii_to_utf16": 1583921,
-  "rvv_bench.ascii_to_utf32": 704111,
-  "rvv_bench.byteswap": 3353234,
+  "rvv_bench.ascii_to_utf16": 1583663,
+  "rvv_bench.ascii_to_utf32": 703954,
+  "rvv_bench.byteswap": 3353148,
   "rvv_bench.chacha20": 2,
-  "rvv_bench.mandelbrot": 4056018,
-  "rvv_bench.memcpy": 2152069,
-  "rvv_bench.memset": 439171,
-  "rvv_bench.mergelines": 3338064,
+  "rvv_bench.mandelbrot": 4055961,
+  "rvv_bench.memcpy": 2131904,
+  "rvv_bench.memset": 438545,
+  "rvv_bench.mergelines": 3337870,
   "rvv_bench.poly1305": 2,
   "rvv_bench.strlen": 877539,
   "rvv_bench.utf8_count": 6340756
diff --git a/.github/cases/machamp/default.json b/.github/cases/machamp/default.json
index 000e0c79f..d7cde8503 100644
--- a/.github/cases/machamp/default.json
+++ b/.github/cases/machamp/default.json
@@ -5,7 +5,7 @@
   "mlir.stripmining": 13425,
   "asm.mmm": 91428,
   "asm.smoke": 5005,
-  "intrinsic.conv2d_less_m2": 2612,
+  "intrinsic.conv2d_less_m2": 2585,
   "codegen.vaadd_vv": 90595,
   "codegen.vaadd_vx": 250851,
   "codegen.vaaddu_vv": 90595,
@@ -435,13 +435,13 @@
   "codegen.vxor_vx": 63090,
   "codegen.vzext_vf2": 39398,
   "codegen.vzext_vf4": 6420,
-  "rvv_bench.ascii_to_utf16": 1460381,
-  "rvv_bench.ascii_to_utf32": 631370,
-  "rvv_bench.byteswap": 3259113,
+  "rvv_bench.ascii_to_utf16": 1460078,
+  "rvv_bench.ascii_to_utf32": 631187,
+  "rvv_bench.byteswap": 3259002,
   "rvv_bench.chacha20": 2,
-  "rvv_bench.memcpy": 1905444,
-  "rvv_bench.memset": 244925,
-  "rvv_bench.mergelines": 3137885,
+  "rvv_bench.memcpy": 1875713,
+  "rvv_bench.memset": 244319,
+  "rvv_bench.mergelines": 3137759,
   "rvv_bench.poly1305": 2,
   "rvv_bench.strlen": 710421,
   "rvv_bench.utf8_count": 5729721
diff --git a/.github/cases/sandslash/default.json b/.github/cases/sandslash/default.json
index 6fa9e360a..32e69fe3a 100644
--- a/.github/cases/sandslash/default.json
+++ b/.github/cases/sandslash/default.json
@@ -5,7 +5,7 @@
   "mlir.stripmining": 3577,
   "asm.mmm": 91437,
   "asm.smoke": 3558,
-  "intrinsic.conv2d_less_m2": 2612,
+  "intrinsic.conv2d_less_m2": 2576,
   "codegen.vaadd_vv": 119061,
   "codegen.vaadd_vx": 336401,
   "codegen.vaaddu_vv": 119061,
@@ -435,13 +435,13 @@
   "codegen.vxor_vx": 84837,
   "codegen.vzext_vf2": 134873,
   "codegen.vzext_vf4": 19800,
-  "rvv_bench.ascii_to_utf16": 1371886,
-  "rvv_bench.ascii_to_utf32": 583519,
-  "rvv_bench.byteswap": 3556432,
+  "rvv_bench.ascii_to_utf16": 1371550,
+  "rvv_bench.ascii_to_utf32": 583318,
+  "rvv_bench.byteswap": 3556315,
   "rvv_bench.chacha20": 2,
-  "rvv_bench.memcpy": 1739436,
-  "rvv_bench.memset": 131052,
-  "rvv_bench.mergelines": 3038652,
+  "rvv_bench.memcpy": 1704869,
+  "rvv_bench.memset": 130483,
+  "rvv_bench.mergelines": 3038600,
   "rvv_bench.poly1305": 2,
   "rvv_bench.strlen": 715272,
   "rvv_bench.utf8_count": 4797732
diff --git a/difftest/default.nix b/difftest/default.nix
index 41ed87e64..efcd77d71 100644
--- a/difftest/default.nix
+++ b/difftest/default.nix
@@ -1,4 +1,5 @@
 { lib
+, libspike
 , callPackage
 , elaborateConfig
 
@@ -48,6 +49,8 @@ let
     env = {
       VERILATED_INC_DIR = "${verilated}/include";
       VERILATED_LIB_DIR = "${verilated}/lib";
+      SPIKE_LIB_DIR = "${libspike}/lib";
+      SPIKE_INTERFACES_LIB_DIR = "${spike_interfaces}/lib";
       DESIGN_VLEN = elaborateConfig.parameter.vLen;
       DESIGN_DLEN = elaborateConfig.parameter.dLen;
     };
diff --git a/difftest/spike_interfaces/CMakeLists.txt b/difftest/spike_interfaces/CMakeLists.txt
index 6da21334c..fe5272891 100644
--- a/difftest/spike_interfaces/CMakeLists.txt
+++ b/difftest/spike_interfaces/CMakeLists.txt
@@ -4,7 +4,7 @@ set(CMAKE_CXX_STANDARD 17)
 
 find_package(libspike REQUIRED)
 
-add_library(${CMAKE_PROJECT_NAME} SHARED spike_interfaces.cc)
+add_library(${CMAKE_PROJECT_NAME} STATIC spike_interfaces.cc)
 
 target_link_libraries(${CMAKE_PROJECT_NAME} PUBLIC libspike)
 
diff --git a/difftest/spike_rs/build.rs b/difftest/spike_rs/build.rs
new file mode 100644
index 000000000..9399fdaf0
--- /dev/null
+++ b/difftest/spike_rs/build.rs
@@ -0,0 +1,18 @@
+use std::env;
+
+fn main() {
+  println!("cargo::rustc-link-search=native={}", env::var("SPIKE_LIB_DIR").expect("SPIKE_LIB_DIR should be set"));
+  println!("cargo::rustc-link-lib=static=riscv");
+  println!("cargo::rustc-link-lib=static=softfloat");
+  println!("cargo::rustc-link-lib=static=disasm");
+  println!("cargo::rustc-link-lib=static=fesvr");
+  println!("cargo::rustc-link-lib=static=fdt");
+
+  println!("cargo::rustc-link-search=native={}", env::var("SPIKE_INTERFACES_LIB_DIR").expect("SPIKE_INTERFACES_LIB_DIR should be set"));
+  println!("cargo::rustc-link-lib=static=spike_interfaces");
+
+  println!("cargo::rerun-if-env-changed=SPIKE_LIB_DIR");
+  println!("cargo::rerun-if-env-changed=SPIKE_INTERFACES_LIB_DIR");
+
+  println!("cargo::rustc-link-lib=stdc++");
+}
diff --git a/nix/overlay.nix b/nix/overlay.nix
index 662ae18ef..c6027ba31 100644
--- a/nix/overlay.nix
+++ b/nix/overlay.nix
@@ -18,7 +18,6 @@ rec {
   espresso = final.callPackage ./pkgs/espresso.nix { };
   dramsim3 = final.callPackage ./pkgs/dramsim3.nix { };
   libspike = final.callPackage ./pkgs/libspike.nix { };
-  libspike_interfaces = final.callPackage ../difftest/libspike_interfaces { };
   buddy-mlir = final.callPackage ./pkgs/buddy-mlir.nix { };
   fetchMillDeps = final.callPackage ./pkgs/mill-builder.nix { };
   circt-full = final.callPackage ./pkgs/circt-full.nix { };
diff --git a/nix/pkgs/libspike.nix b/nix/pkgs/libspike.nix
index dc3a28fdc..1ecb1c317 100644
--- a/nix/pkgs/libspike.nix
+++ b/nix/pkgs/libspike.nix
@@ -7,7 +7,7 @@ stdenv.mkDerivation {
   env.cmakeConfig = ''
     add_library(libspike STATIC IMPORTED GLOBAL)
     set_target_properties(libspike PROPERTIES
-      IMPORTED_LOCATION "${placeholder "out"}/lib/libriscv.so")
+      IMPORTED_LOCATION "${placeholder "out"}/lib/libriscv.a")
     target_include_directories(libspike AFTER INTERFACE
       "${placeholder "out"}/include"
       "${placeholder "out"}/include/riscv"
@@ -37,6 +37,7 @@ stdenv.mkDerivation {
     cp softfloat/*.h $out/include/softfloat
     cp config.h $out/include
     cp *.so $out/lib
+    cp *.a $out/lib
     echo "$cmakeConfig" > $out/lib/cmake/libspike/libspike-config.cmake
     runHook postInstall
   '';
diff --git a/t1/src/lsu/StoreUnit.scala b/t1/src/lsu/StoreUnit.scala
index fb5d9ee4a..69f28a6a0 100644
--- a/t1/src/lsu/StoreUnit.scala
+++ b/t1/src/lsu/StoreUnit.scala
@@ -58,7 +58,6 @@ class StoreUnit(param: MSHRParam) extends StrideBase(param) with LSUPublic {
   val lastDataGroupReg: UInt = RegEnable(lastDataGroupForInstruction, 0.U, lsuRequest.valid)
   val nextDataGroup: UInt = Mux(lsuRequest.valid, 0.U, dataGroup + 1.U)
   val isLastRead: Bool = dataGroup === lastDataGroupReg
-  val lastGroupAndNeedAlign: Bool = initOffset.orR && isLastRead
 
   // stage1, 读vrf
   // todo: need hazardCheck?
@@ -159,12 +158,12 @@ class StoreUnit(param: MSHRParam) extends StrideBase(param) with LSUPublic {
   // 存每条cache 的mask, 也许能优化, 暂时先这样
   val maskForBufferData: Vec[UInt] = RegInit(VecInit(Seq.fill(8)(0.U(param.lsuTransposeSize.W))))
   val maskForBufferDequeue: UInt = maskForBufferData(cacheLineIndexInBuffer)
-  val tailLeft2: Bool = RegInit(false.B)
+  val lastDataGroupInDataBuffer: Bool = RegInit(false.B)
   val alignedDequeueFire: Bool = memRequest.fire
   // cache 不对齐的时候的上一条残留
   val cacheLineTemp: UInt = RegEnable(dataBuffer.head, 0.U((param.lsuTransposeSize * 8).W), alignedDequeueFire)
   val maskTemp: UInt = RegInit(0.U(param.lsuTransposeSize.W))
-  val tailValid: Bool = RegInit(false.B)
+  val canSendTail: Bool = RegInit(false.B)
   val isLastCacheLineInBuffer: Bool = cacheLineIndexInBuffer === lsuRequestReg.instructionInformation.nf
   val bufferWillClear: Bool = alignedDequeueFire && isLastCacheLineInBuffer
   accessBufferDequeueReady := !bufferValid || (memRequest.ready && isLastCacheLineInBuffer)
@@ -176,7 +175,7 @@ class StoreUnit(param: MSHRParam) extends StrideBase(param) with LSUPublic {
   // 把数据regroup, 然后放去 [[dataBuffer]]
   when(accessBufferDequeueFire) {
     maskForBufferData := cutUInt(fillBySeg, param.lsuTransposeSize)
-    tailLeft2 := lastGroupAndNeedAlign
+    lastDataGroupInDataBuffer := isLastRead
     // todo: 只是因为参数恰好是一个方形的, 需要写一个反的
     dataBuffer := Mux1H(dataEEWOH, Seq.tabulate(3) { sewSize =>
       // 每个数据块 2 ** sew byte
@@ -238,11 +237,12 @@ class StoreUnit(param: MSHRParam) extends StrideBase(param) with LSUPublic {
 
   when(lsuRequest.valid || alignedDequeueFire) {
     maskTemp := Mux(lsuRequest.valid, 0.U, maskForBufferDequeue)
-    tailValid := Mux(lsuRequest.valid, false.B, bufferValid && tailLeft2 && isLastCacheLineInBuffer)
+    canSendTail := !lsuRequest.valid && bufferValid && isLastCacheLineInBuffer && lastDataGroupInDataBuffer
   }
 
   // 连接 alignedDequeue
-  memRequest.valid := bufferValid || tailValid
+  val needSendTail: Bool = bufferBaseCacheLineIndex === cacheLineNumberReg
+  memRequest.valid := bufferValid || (canSendTail && needSendTail)
   // aligned
   memRequest.bits.data :=
     multiShifter(right = false, multiSize = 8)(dataBuffer.head ## cacheLineTemp, initOffset) >> cacheLineTemp.getWidth