diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 1983375..3034c7b 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -2,6 +2,9 @@ name: mt32-pi CI
 
 on: [push, pull_request]
 
+env:
+  toolchain-version: 9.2-2019.12
+
 jobs:
   build:
     runs-on: ubuntu-latest
@@ -9,7 +12,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        board: [pi2, pi3, pi4]
+        board: [pi2, pi3-64, pi4-64]
     env:
       toolchain-version: 9.2-2019.12
 
@@ -18,22 +21,29 @@ jobs:
       with:
         submodules: recursive
 
-    - name: Cache ARM toolchain
+    - name: Cache ARM toolchains
+      id: cache-toolchains
       uses: actions/cache@v2
       with:
-        path: ~/gcc-arm-${{ env.toolchain-version }}-x86_64-arm-none-eabi
-        key: cache-arm-toolchain-${{ env.toolchain-version }}
+        path: |
+          ~/gcc-arm-${{ env.toolchain-version }}-x86_64-arm-none-eabi
+          ~/gcc-arm-${{ env.toolchain-version }}-x86_64-aarch64-none-elf
+        key: cache-arm-toolchains-${{ env.toolchain-version }}
 
-    - name: Install toolchain if not retrieved from cache
+    - name: Install toolchains if not retrieved from cache
+      if: steps.cache-toolchains.outputs.cache-hit != 'true'
       run: |
-        if [ ! -d ~/gcc-arm-${{ env.toolchain-version }}-x86_64-arm-none-eabi ]; then
-          wget -q https://developer.arm.com/-/media/Files/downloads/gnu-a/${{ env.toolchain-version }}/binrel/gcc-arm-${{ env.toolchain-version }}-x86_64-arm-none-eabi.tar.xz
-          tar -xf gcc-arm-${{ env.toolchain-version }}-x86_64-arm-none-eabi.tar.xz -C ~
-          rm gcc-arm-${{ env.toolchain-version }}-x86_64-arm-none-eabi.tar.xz
-        fi
+        for triplet in arm-none-eabi aarch64-none-elf; do
+          wget -q https://developer.arm.com/-/media/Files/downloads/gnu-a/${{ env.toolchain-version }}/binrel/gcc-arm-${{ env.toolchain-version }}-x86_64-${triplet}.tar.xz
+          tar -xf gcc-arm-${{ env.toolchain-version }}-x86_64-${triplet}.tar.xz -C ~
+          rm gcc-arm-${{ env.toolchain-version }}-x86_64-${triplet}.tar.xz
+        done
 
     - name: Update PATH
-      run: echo "$HOME/gcc-arm-${{ env.toolchain-version }}-x86_64-arm-none-eabi/bin" >> $GITHUB_PATH
+      run: |
+        for triplet in arm-none-eabi aarch64-none-elf; do
+          echo "$HOME/gcc-arm-${{ env.toolchain-version }}-x86_64-${triplet}/bin" >> $GITHUB_PATH
+        done
 
     - name: Build
       run: make -j BOARD=${{ matrix.board }}
@@ -56,9 +66,11 @@ jobs:
         submodules: recursive
 
     - name: Cache boot files
+      id: cache-boot-files
       uses: actions/cache@v2
       with:
         path: |
+          sdcard/armstub8-rpi4.bin
           sdcard/bcm2711-rpi-4-b.dtb
           sdcard/bootcode.bin
           sdcard/COPYING.linux
@@ -69,20 +81,36 @@ jobs:
           sdcard/start4cd.elf
         key: cache-boot-files-${{ hashFiles(join(env.boot-home, '/Makefile')) }}
 
-    - name: Download boot files if not retrieved from cache
+    - name: Cache ARM toolchains
+      if: steps.cache-boot-files.outputs.cache-hit != 'true'
+      uses: actions/cache@v2
+      with:
+        path: |
+          ~/gcc-arm-${{ env.toolchain-version }}-x86_64-arm-none-eabi
+          ~/gcc-arm-${{ env.toolchain-version }}-x86_64-aarch64-none-elf
+        key: cache-arm-toolchains-${{ env.toolchain-version }}
+
+    - name: Update PATH
+      if: steps.cache-boot-files.outputs.cache-hit != 'true'
       run: |
-        if [ ! -f sdcard/bootcode.bin ]; then
-          make -C ${{ env.boot-home }}
-          cp  ${{ env.boot-home }}/bcm2711-rpi-4-b.dtb \
-              ${{ env.boot-home }}/bootcode.bin \
-              ${{ env.boot-home }}/COPYING.linux \
-              ${{ env.boot-home }}/fixup_cd.dat \
-              ${{ env.boot-home }}/fixup4cd.dat \
-              ${{ env.boot-home }}/LICENCE.broadcom \
-              ${{ env.boot-home }}/start_cd.elf \
-              ${{ env.boot-home }}/start4cd.elf \
-              sdcard
-        fi
+        for triplet in arm-none-eabi aarch64-none-elf; do
+          echo "$HOME/gcc-arm-${{ env.toolchain-version }}-x86_64-${triplet}/bin" >> $GITHUB_PATH
+        done
+
+    - name: Download boot files and build ARM stub if not retrieved from cache
+      if: steps.cache-boot-files.outputs.cache-hit != 'true'
+      run: |
+        make -C ${{ env.boot-home }} firmware armstub64
+        cp  ${{ env.boot-home }}/armstub8-rpi4.bin \
+            ${{ env.boot-home }}/bcm2711-rpi-4-b.dtb \
+            ${{ env.boot-home }}/bootcode.bin \
+            ${{ env.boot-home }}/COPYING.linux \
+            ${{ env.boot-home }}/fixup_cd.dat \
+            ${{ env.boot-home }}/fixup4cd.dat \
+            ${{ env.boot-home }}/LICENCE.broadcom \
+            ${{ env.boot-home }}/start_cd.elf \
+            ${{ env.boot-home }}/start4cd.elf \
+            sdcard
 
     - name: Retrieve built kernels
       uses: actions/download-artifact@v2
diff --git a/Config.mk b/Config.mk
index 79ce5b9..36febf3 100644
--- a/Config.mk
+++ b/Config.mk
@@ -2,8 +2,9 @@
 # Build configuration
 #
 
-# Path to ARM toolchain
+# Paths to ARM toolchains
 ARM_HOME?=$(HOME)/gcc-arm-9.2-2019.12-x86_64-arm-none-eabi
+AARCH64_HOME?=$(HOME)/gcc-arm-9.2-2019.12-x86_64-aarch64-none-elf
 
 # Valid options: pi0, pi2, pi3, pi4, pi4-64
 BOARD?=pi4
@@ -17,6 +18,42 @@ USERBAUD?=115200
 # Enable section garbage collection
 GC_SECTIONS?=0
 
+# Toolchain setup
+ifeq ($(BOARD), pi0)
+RASPBERRYPI=1
+BITS=32
+PREFIX=arm-none-eabi-
+else ifeq ($(BOARD), pi2)
+RASPBERRYPI=2
+BITS=32
+PREFIX=arm-none-eabi-
+else ifeq ($(BOARD), pi3)
+RASPBERRYPI=3
+BITS=32
+PREFIX=arm-none-eabi-
+else ifeq ($(BOARD), pi3-64)
+RASPBERRYPI=3
+BITS=64
+PREFIX=aarch64-none-elf-
+else ifeq ($(BOARD), pi4)
+RASPBERRYPI=4
+BITS=32
+PREFIX=arm-none-eabi-
+else ifeq ($(BOARD), pi4-64)
+RASPBERRYPI=4
+BITS=64
+PREFIX=aarch64-none-elf-
+else
+$(error Invalid board type "$(BOARD)"; please specify one of [ pi0 | pi2 | pi3 | pi3-64 | pi4 | pi4-64 ])
+endif
+
+ifeq ($(PREFIX), arm-none-eabi-)
+CMAKE_TOOLCHAIN_FLAGS=-DARM_HOME=$(ARM_HOME) -DCMAKE_TOOLCHAIN_FILE=../cmake/arm-none-eabi.cmake
+else
+CMAKE_TOOLCHAIN_FLAGS=-DAARCH64_HOME=$(AARCH64_HOME) -DCMAKE_TOOLCHAIN_FILE=../cmake/aarch64-none-elf.cmake
+endif
+
+# Paths
 CIRCLESTDLIBHOME=$(realpath external/circle-stdlib)
 CIRCLE_STDLIB_CONFIG=$(CIRCLESTDLIBHOME)/Config.mk
 
diff --git a/Makefile b/Makefile
index c0ed70c..8beae7c 100644
--- a/Makefile
+++ b/Makefile
@@ -12,24 +12,8 @@ include Config.mk
 #
 .ONESHELL:
 $(CIRCLE_STDLIB_CONFIG) $(CIRCLE_CONFIG)&:
-ifeq ($(BOARD), pi0)
-	@echo -n "Configuring for Raspberry Pi 0/1"
-	$(CIRCLESTDLIBHOME)/configure --raspberrypi=1
-else ifeq ($(BOARD), pi2)
-	@echo -n "Configuring for Raspberry Pi 2"
-	$(CIRCLESTDLIBHOME)/configure --raspberrypi=2
-else ifeq ($(BOARD), pi3)
-	@echo -n "Configuring for Raspberry Pi 3"
-	$(CIRCLESTDLIBHOME)/configure --raspberrypi=3
-else ifeq ($(BOARD), pi4)
-	@echo -n "Configuring for Raspberry Pi 4"
-	$(CIRCLESTDLIBHOME)/configure --raspberrypi=4
-else ifeq ($(BOARD), pi4-64)
-	@echo -n "Configuring for Raspberry Pi 4 (64 bit)"
-	$(CIRCLESTDLIBHOME)/configure --raspberrypi=4 --prefix=aarch64-none-elf
-else
-	$(error Invalid board type "$(BOARD)"; please specify one of [ pi0 | pi2 | pi3 | pi4 | pi4-64 ])
-endif
+	@echo "Configuring for Raspberry Pi $(RASPBERRYPI) ($(BITS) bit)"
+	$(CIRCLESTDLIBHOME)/configure --raspberrypi=$(RASPBERRYPI) --prefix=$(PREFIX)
 
 	# Enable multi-core
 	echo "DEFINE += -DARM_ALLOW_MULTI_CORE" >> $(CIRCLE_CONFIG)
@@ -48,6 +32,7 @@ circle-stdlib: $(CIRCLESTDLIBHOME)/.done
 $(CIRCLESTDLIBHOME)/.done: $(CIRCLE_STDLIB_CONFIG)
 	@$(MAKE) -C $(CIRCLESTDLIBHOME)
 	touch $@
+
 #
 # Build mt32emu
 #
@@ -56,9 +41,8 @@ mt32emu: $(MT32EMUBUILDDIR)/.done
 $(MT32EMUBUILDDIR)/.done: $(CIRCLESTDLIBHOME)/.done
 	@export CFLAGS="$(CFLAGS_FOR_TARGET)"
 	@export CXXFLAGS="$(CFLAGS_FOR_TARGET)"
-	@cmake   -B $(MT32EMUBUILDDIR) \
-			-DARM_HOME=$(ARM_HOME) \
-			-DCMAKE_TOOLCHAIN_FILE=../cmake/arm-none-eabi.cmake \
+	@cmake  -B $(MT32EMUBUILDDIR) \
+			$(CMAKE_TOOLCHAIN_FLAGS) \
 			-DCMAKE_CXX_FLAGS_RELEASE="-Ofast" \
 			-DCMAKE_BUILD_TYPE=Release \
 			-Dlibmt32emu_C_INTERFACE=FALSE \
diff --git a/cmake/aarch64-none-elf.cmake b/cmake/aarch64-none-elf.cmake
new file mode 100644
index 0000000..4edebac
--- /dev/null
+++ b/cmake/aarch64-none-elf.cmake
@@ -0,0 +1,7 @@
+set(CMAKE_SYSTEM_NAME Generic)
+set(CMAKE_SYSTEM_PROCESSOR aarch64)
+
+set(CMAKE_TRY_COMPILE_TARGET_TYPE STATIC_LIBRARY)
+
+set(CMAKE_C_COMPILER ${AARCH64_HOME}/bin/aarch64-none-elf-gcc)
+set(CMAKE_CXX_COMPILER ${AARCH64_HOME}/bin/aarch64-none-elf-g++)
diff --git a/sdcard/config.txt b/sdcard/config.txt
index 6df75eb..975c2f8 100644
--- a/sdcard/config.txt
+++ b/sdcard/config.txt
@@ -4,6 +4,12 @@ disable_splash=1
 force_eeprom_read=0
 gpu_mem=16
 
-# Ensure that the Pi 4 doesn't try to boot the Pi 3-optimized kernel (kernel8-32.img)
+# Enable 64-bit mode (AArch64) for Pi 3 and 4
+[pi3]
+arm_64bit=1
+kernel=kernel8.img
+
 [pi4]
-kernel=kernel7l.img
+arm_64bit=1
+armstub=armstub8-rpi4.bin
+kernel=kernel8-rpi4.img