diff --git a/.github/scripts/install-bpftrace.sh b/.github/scripts/install-bpftrace.sh
index a225f25..e81474c 100755
--- a/.github/scripts/install-bpftrace.sh
+++ b/.github/scripts/install-bpftrace.sh
@@ -3,27 +3,19 @@
 set -x -euo pipefail
 
 BPFTRACE_VERSION=${BPFTRACE_VERSION:-0.22.1}
-BUILD_BPFTRACE=${BUILD_BPFTRACE:-}
-
 BIN_DIR=/usr/local/bin
 sudo mkdir -p $BIN_DIR
 
-if [[ -n "$BUILD_BPFTRACE" ]]; then
-    # see https://github.com/bpftrace/bpftrace/blob/master/docker/Dockerfile.ubuntu
-    sudo apt-get install -y \
-         asciidoctor binutils-dev bison build-essential clang cmake flex git libbpf-dev \
-         libbpfcc-dev libcereal-dev libelf-dev libiberty-dev libpcap-dev llvm-dev liblldb-dev \
-         libclang-dev systemtap-sdt-dev zlib1g-dev
-    git clone --depth 1 -b v${BPFTRACE_VERSION} https://github.com/bpftrace/bpftrace
-    cd bpftrace/src
-    cmake -B build -DBUILD_TESTING=OFF -DCMAKE_BUILD_TYPE=Release -DLLVM_DIR=$(llvm-config --prefix)/lib/cmake/llvm
-    make -C build -j$(nproc)
-    cd -
-    sudo mv bpftrace/src/build/bpftrace $BIN_DIR/bpftrace
-else # download AppImage
-    sudo curl -L -o $BIN_DIR/bpftrace https://github.com/bpftrace/bpftrace/releases/download/v${BPFTRACE_VERSION}/bpftrace
+if [ $(uname -m) == "x86_64" ]; then
+     URL=https://github.com/bpftrace/bpftrace/releases/download/v${BPFTRACE_VERSION}/bpftrace
+elif [ $(uname -m) == "aarch64" ]; then
+     URL=https://github.com/theihor/bpftrace/releases/download/v${BPFTRACE_VERSION}-arm64/bpftrace-arm64
+else
+    echo "Unexpected arch: $(uname -m)"
+    exit 1
 fi
 
+sudo curl -L -o $BIN_DIR/bpftrace $URL
 sudo chmod +x $BIN_DIR/bpftrace
 
 # mount tracefs to avoid warnings from bpftrace
diff --git a/.github/workflows/test.yml b/.github/workflows/test.yml
index 0aed5ee..fd5f59e 100644
--- a/.github/workflows/test.yml
+++ b/.github/workflows/test.yml
@@ -12,7 +12,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        runs-on: [ 'ubuntu-latest' ]
+        runs-on: [ 'ubuntu-latest', 'ubuntu-24.04-arm' ]
         shared: [ 0, 1 ]
         compiler:
           - { name: 'gcc', version: 11 }
@@ -25,10 +25,9 @@ jobs:
 
     runs-on: ${{ matrix.runs-on }}
 
-    name: ${{ endsWith(matrix.runs-on, '-arm') && 'arm64' || 'x86_64' }} SHARED=${{ matrix.shared }} ${{ matrix.compiler.name }}-${{ matrix.compiler.version }}
+    name: ${{ endsWith(matrix.runs-on, '-arm') && 'arm64' || 'x86_64' }} ${{ matrix.shared == 1 && 'shared' || 'static' }} ${{ matrix.compiler.name }}-${{ matrix.compiler.version }}
 
     env:
-      ARCH: ${{ endsWith(matrix.runs-on, '-arm') && 'arm64' || 'x86_64' }}
       COMPILER: ${{ matrix.compiler.name }}
       COMPILER_VERSION: ${{ matrix.compiler.version }}
       SHARED: ${{ matrix.shared }}
@@ -44,3 +43,11 @@ jobs:
 
     - name: Test SHARED=${{ matrix.shared }}
       run: make -C tests test
+
+    - uses: actions/upload-artifact@v4
+      if: always()
+      with:
+        name: tests-output-${{ endsWith(matrix.runs-on, '-arm') && 'arm64' || 'x86_64' }}-${{ matrix.shared == 1 && 'shared' || 'static' }}-${{ matrix.compiler.name }}-${{ matrix.compiler.version }}
+        if-no-files-found: warn
+        path: tests/.output
+        include-hidden-files: true
diff --git a/tests/arg_consts.c b/tests/arg_consts.c
index 99b9f85..bd50e71 100644
--- a/tests/arg_consts.c
+++ b/tests/arg_consts.c
@@ -21,6 +21,17 @@ int main(int argc, char **argv)
 	return 0;
 }
 
+#ifdef __aarch64__
+const char *USDT_SPECS =
+"test:chars base=BASE1 sema=0 argn=2 args=1@1 -1@-1.\n"
+"test:shorts base=BASE1 sema=0 argn=2 args=2@2 -2@-2.\n"
+"test:ints base=BASE1 sema=0 argn=2 args=4@3 -4@-3.\n"
+"test:longs base=BASE1 sema=0 argn=2 args=*@4 -*@-4.\n"
+"test:longlongs base=BASE1 sema=0 argn=2 args=8@5 -8@-5.\n"
+"test:size_ts base=BASE1 sema=0 argn=2 args=*@6 -*@-6.\n"
+"test:ptrs base=BASE1 sema=0 argn=2 args=*@0 *@7.\n"
+;
+#else
 const char *USDT_SPECS =
 "test:chars base=BASE1 sema=0 argn=2 args=1@$1 -1@$-1.\n"
 "test:shorts base=BASE1 sema=0 argn=2 args=2@$2 -2@$-2.\n"
@@ -30,7 +41,7 @@ const char *USDT_SPECS =
 "test:size_ts base=BASE1 sema=0 argn=2 args=*@$6 -*@$-6.\n"
 "test:ptrs base=BASE1 sema=0 argn=2 args=*@$0 *@$7.\n"
 ;
-
+#endif
 const char *BPFTRACE_SCRIPT =
 "test:chars { arg0=%hhu arg1=%hhd -> arg0, arg1 }\n"
 "test:shorts { arg0=%hu arg1=%hd -> arg0, arg1 }\n"
diff --git a/tests/arg_nums.c b/tests/arg_nums.c
index b1e87a4..20b0840 100644
--- a/tests/arg_nums.c
+++ b/tests/arg_nums.c
@@ -25,6 +25,23 @@ int main(int argc, char **argv)
 	return 0;
 }
 
+#ifdef __aarch64__
+const char *USDT_SPECS =
+"test:name0 base=BASE1 sema=0 argn=0 args=.\n"
+"test:name1 base=BASE1 sema=0 argn=1 args=-4@1.\n"
+"test:name2 base=BASE1 sema=0 argn=2 args=-4@1 -4@2.\n"
+"test:name3 base=BASE1 sema=0 argn=3 args=-4@1 -4@2 -4@3.\n"
+"test:name4 base=BASE1 sema=0 argn=4 args=-4@1 -4@2 -4@3 -4@4.\n"
+"test:name5 base=BASE1 sema=0 argn=5 args=-4@1 -4@2 -4@3 -4@4 -4@5.\n"
+"test:name6 base=BASE1 sema=0 argn=6 args=-4@1 -4@2 -4@3 -4@4 -4@5 -4@6.\n"
+"test:name7 base=BASE1 sema=0 argn=7 args=-4@1 -4@2 -4@3 -4@4 -4@5 -4@6 -4@7.\n"
+"test:name8 base=BASE1 sema=0 argn=8 args=-4@1 -4@2 -4@3 -4@4 -4@5 -4@6 -4@7 -4@8.\n"
+"test:name9 base=BASE1 sema=0 argn=9 args=-4@1 -4@2 -4@3 -4@4 -4@5 -4@6 -4@7 -4@8 -4@9.\n"
+"test:name10 base=BASE1 sema=0 argn=10 args=-4@1 -4@2 -4@3 -4@4 -4@5 -4@6 -4@7 -4@8 -4@9 -4@10.\n"
+"test:name11 base=BASE1 sema=0 argn=11 args=-4@1 -4@2 -4@3 -4@4 -4@5 -4@6 -4@7 -4@8 -4@9 -4@10 -4@11.\n"
+"test:name12 base=BASE1 sema=0 argn=12 args=-4@1 -4@2 -4@3 -4@4 -4@5 -4@6 -4@7 -4@8 -4@9 -4@10 -4@11 -4@12.\n"
+;
+#else
 const char *USDT_SPECS =
 "test:name0 base=BASE1 sema=0 argn=0 args=.\n"
 "test:name1 base=BASE1 sema=0 argn=1 args=-4@$1.\n"
@@ -40,6 +57,7 @@ const char *USDT_SPECS =
 "test:name11 base=BASE1 sema=0 argn=11 args=-4@$1 -4@$2 -4@$3 -4@$4 -4@$5 -4@$6 -4@$7 -4@$8 -4@$9 -4@$10 -4@$11.\n"
 "test:name12 base=BASE1 sema=0 argn=12 args=-4@$1 -4@$2 -4@$3 -4@$4 -4@$5 -4@$6 -4@$7 -4@$8 -4@$9 -4@$10 -4@$11 -4@$12.\n"
 ;
+#endif
 
 const char *BPFTRACE_SCRIPT =
 "test:name0 { triggered }\n"
diff --git a/tests/arg_types.c b/tests/arg_types.c
index dfc5de7..375da20 100644
--- a/tests/arg_types.c
+++ b/tests/arg_types.c
@@ -122,7 +122,7 @@ const char *BPFTRACE_SCRIPT =
 	str(arg0),								\
 	*(int32 *)(arg1 + 0), *(int32 *)(arg1 + 4), *(int32 *)(arg1 + 8) }\n"
 "test:struct_by_val_reg { arg0=%hhu arg1=%u -> arg0, arg1 }\n"
-#ifndef __clang__ /* bpftrace generates a bad BPF program in this case */
+#if ALLOW_STRUCT_BY_VALUE_TEST
 "test:struct_by_val_reg_pair { s.x=%llx -> arg0 }\n" /* captures first half of a struct */
 #endif
 /* bpftrace can't handle 24-byte struct-by-value case in struct_by_val_stack */
@@ -140,7 +140,7 @@ const char *BPFTRACE_OUTPUT =
 "test:ptrs: arg0=0x* arg1='some literal' arg2=&42\n"
 "test:arrs: arg0='STRING' arg1=(-100,-200,-300)\n"
 "test:struct_by_val_reg: arg0=1 arg1=2\n"
-#ifndef __clang__
+#if ALLOW_STRUCT_BY_VALUE_TEST
 "test:struct_by_val_reg_pair: s.x=3\n"
 #endif
 "test:structs_by_ref: a=(1) b=(2) c=(3,4) d=(4,5,6)\n"
diff --git a/tests/common.h b/tests/common.h
index 094dd10..11339f3 100644
--- a/tests/common.h
+++ b/tests/common.h
@@ -8,6 +8,15 @@
 #define __always_inline inline __attribute__((always_inline))
 #endif
 
+/* Struct-by-value USDT argument currently only works on x86_64 with gcc
+ * See: https://github.com/bpftrace/bpftrace/issues/3798
+ */
+#if defined(__clang__) || defined(__aarch64__)
+  #define ALLOW_STRUCT_BY_VALUE_TEST 0
+#else
+  #define ALLOW_STRUCT_BY_VALUE_TEST 1
+#endif
+
 #ifdef __cplusplus
 extern "C" {
 #endif
diff --git a/tests/cxx_arg_consts.cpp b/tests/cxx_arg_consts.cpp
index 62eca84..90ca910 100644
--- a/tests/cxx_arg_consts.cpp
+++ b/tests/cxx_arg_consts.cpp
@@ -30,6 +30,17 @@ int main(int argc, char **argv)
 	return 0;
 }
 
+#ifdef __aarch64__
+const char *USDT_SPECS =
+"test:chars base=BASE1 sema=0 argn=2 args=1@1 -1@-1.\n"
+"test:shorts base=BASE1 sema=0 argn=2 args=2@2 -2@-2.\n"
+"test:ints base=BASE1 sema=0 argn=2 args=4@3 -4@-3.\n"
+"test:longs base=BASE1 sema=0 argn=2 args=*@4 -*@-4.\n"
+"test:longlongs base=BASE1 sema=0 argn=2 args=8@5 -8@-5.\n"
+"test:size_ts base=BASE1 sema=0 argn=2 args=*@6 -*@-6.\n"
+"test:ptrs base=BASE1 sema=0 argn=2 args=*@0 *@7.\n"
+;
+#else
 const char *USDT_SPECS =
 "test:chars base=BASE1 sema=0 argn=2 args=1@$1 -1@$-1.\n"
 "test:shorts base=BASE1 sema=0 argn=2 args=2@$2 -2@$-2.\n"
@@ -39,6 +50,7 @@ const char *USDT_SPECS =
 "test:size_ts base=BASE1 sema=0 argn=2 args=*@$6 -*@$-6.\n"
 "test:ptrs base=BASE1 sema=0 argn=2 args=*@$0 *@$7.\n"
 ;
+#endif
 
 const char *BPFTRACE_SCRIPT =
 "test:chars { arg0=%hhu arg1=%hhd -> arg0, arg1 }\n"
diff --git a/tests/cxx_arg_nums.cpp b/tests/cxx_arg_nums.cpp
index 940b1fa..ae68117 100644
--- a/tests/cxx_arg_nums.cpp
+++ b/tests/cxx_arg_nums.cpp
@@ -35,6 +35,23 @@ int main(int argc, char **argv)
 	return 0;
 }
 
+#ifdef __aarch64__
+const char *USDT_SPECS =
+"test:name0 base=BASE1 sema=0 argn=0 args=.\n"
+"test:name1 base=BASE1 sema=0 argn=1 args=-4@1.\n"
+"test:name2 base=BASE1 sema=0 argn=2 args=-4@1 -4@2.\n"
+"test:name3 base=BASE1 sema=0 argn=3 args=-4@1 -4@2 -4@3.\n"
+"test:name4 base=BASE1 sema=0 argn=4 args=-4@1 -4@2 -4@3 -4@4.\n"
+"test:name5 base=BASE1 sema=0 argn=5 args=-4@1 -4@2 -4@3 -4@4 -4@5.\n"
+"test:name6 base=BASE1 sema=0 argn=6 args=-4@1 -4@2 -4@3 -4@4 -4@5 -4@6.\n"
+"test:name7 base=BASE1 sema=0 argn=7 args=-4@1 -4@2 -4@3 -4@4 -4@5 -4@6 -4@7.\n"
+"test:name8 base=BASE1 sema=0 argn=8 args=-4@1 -4@2 -4@3 -4@4 -4@5 -4@6 -4@7 -4@8.\n"
+"test:name9 base=BASE1 sema=0 argn=9 args=-4@1 -4@2 -4@3 -4@4 -4@5 -4@6 -4@7 -4@8 -4@9.\n"
+"test:name10 base=BASE1 sema=0 argn=10 args=-4@1 -4@2 -4@3 -4@4 -4@5 -4@6 -4@7 -4@8 -4@9 -4@10.\n"
+"test:name11 base=BASE1 sema=0 argn=11 args=-4@1 -4@2 -4@3 -4@4 -4@5 -4@6 -4@7 -4@8 -4@9 -4@10 -4@11.\n"
+"test:name12 base=BASE1 sema=0 argn=12 args=-4@1 -4@2 -4@3 -4@4 -4@5 -4@6 -4@7 -4@8 -4@9 -4@10 -4@11 -4@12.\n"
+;
+#else
 const char *USDT_SPECS =
 "test:name0 base=BASE1 sema=0 argn=0 args=.\n"
 "test:name1 base=BASE1 sema=0 argn=1 args=-4@$1.\n"
@@ -50,6 +67,7 @@ const char *USDT_SPECS =
 "test:name11 base=BASE1 sema=0 argn=11 args=-4@$1 -4@$2 -4@$3 -4@$4 -4@$5 -4@$6 -4@$7 -4@$8 -4@$9 -4@$10 -4@$11.\n"
 "test:name12 base=BASE1 sema=0 argn=12 args=-4@$1 -4@$2 -4@$3 -4@$4 -4@$5 -4@$6 -4@$7 -4@$8 -4@$9 -4@$10 -4@$11 -4@$12.\n"
 ;
+#endif
 
 const char *BPFTRACE_SCRIPT =
 "test:name0 { triggered }\n"
diff --git a/tests/cxx_arg_types.cpp b/tests/cxx_arg_types.cpp
index 90e3619..7135fda 100644
--- a/tests/cxx_arg_types.cpp
+++ b/tests/cxx_arg_types.cpp
@@ -122,7 +122,7 @@ const char *BPFTRACE_SCRIPT =
 	str(arg0),								\
 	*(int32 *)(arg1 + 0), *(int32 *)(arg1 + 4), *(int32 *)(arg1 + 8) }\n"
 "test:struct_by_val_reg { arg0=%hhu arg1=%u -> arg0, arg1 }\n"
-#ifndef __clang__ /* bpftrace generates a bad BPF program in this case */
+#if ALLOW_STRUCT_BY_VALUE_TEST
 "test:struct_by_val_reg_pair { s.x=%llx -> arg0 }\n" /* captures first half of a struct */
 #endif
 /* bpftrace can't handle 24-byte struct-by-value case in struct_by_val_stack */
@@ -140,7 +140,7 @@ const char *BPFTRACE_OUTPUT =
 "test:ptrs: arg0=0x* arg1='some literal' arg2=&42\n"
 "test:arrs: arg0='STRING' arg1=(-100,-200,-300)\n"
 "test:struct_by_val_reg: arg0=1 arg1=2\n"
-#ifndef __clang__
+#if ALLOW_STRUCT_BY_VALUE_TEST
 "test:struct_by_val_reg_pair: s.x=3\n"
 #endif
 "test:structs_by_ref: a=(1) b=(2) c=(3,4) d=(4,5,6)\n"
diff --git a/tests/fetch-usdts.awk b/tests/fetch-usdts.awk
index 0cfadd9..6f690dd 100644
--- a/tests/fetch-usdts.awk
+++ b/tests/fetch-usdts.awk
@@ -13,20 +13,26 @@ function reset_entry()
 
 function print_entry()
 {
-	if (base != 0 && !(base in basemap)) {
-		basemap[base] = ++base_cnt;
+	base_key = filename ":" base;
+	if (base != 0 && !(base_key in basemap)) {
+		basemap[base_key] = ++base_cnt;
 	}
-	if (sema != 0 && !(sema in semamap)) {
-		semamap[sema] = ++sema_cnt;
+
+	sema_key = filename ":" sema;
+	if (sema != 0 && !(sema_key in semamap)) {
+		semamap[sema_key] = ++sema_cnt;
 	}
-	base_stub = (base == 0) ? "0" : sprintf("BASE%d", basemap[base]);
-	sema_stub = (sema == 0) ? "0" : sprintf("SEMA%d", semamap[sema]);
+
+	base_stub = (base == 0) ? "0" : sprintf("BASE%d", basemap[base_key]);
+	sema_stub = (sema == 0) ? "0" : sprintf("SEMA%d", semamap[sema_key]);
+
 	printf "%s:%s base=%s sema=%s argn=%d args=%s.\n",
 	       grp, name, base_stub, sema_stub, argn, args;
 }
 
 BEGIN {
 	reset_entry();
+	filename = "";
 }
 
 #  stapsdt              0x0000003b       NT_STAPSDT (SystemTap probe descriptors)
@@ -35,6 +41,14 @@ BEGIN {
 #    Location: 0x0000000000401198, Base: 0x0000000000402043, Semaphore: 0x0000000000000000
 #    Arguments: -4@$1 -4@$2 -4@$3 -4@$4
 
+/File:/ {
+	if (entry != "") {
+		print_entry();
+		reset_entry();
+	}
+	filename = $2;
+}
+
 /\sstapsdt\s/ {
 	if (entry != "") {
 		print_entry();
@@ -47,7 +61,12 @@ BEGIN {
 /Provider:/ { grp = $2; }
 /Name:/ { name = $2; }
 /Location:/ { base = strtonum($4); sema = strtonum($6); }
-/Arguments:/ { argn = NF - 1; args = (argn > 0) ? substr($0, index($0,$2)) : ""; }
+/Arguments:/ {
+	arg_str = substr($0, index($0, "Arguments: ") + 11);
+	# Count arguments by looking for patterns like "-4@" that start each argument
+	argn = gsub(/-?[0-9]+@/, "&", arg_str);
+	args = (argn > 0) ? arg_str : "";
+}
 
 END {
 	if (entry != "")
diff --git a/tests/run_test.sh b/tests/run_test.sh
index 7caaafc..540c54b 100755
--- a/tests/run_test.sh
+++ b/tests/run_test.sh
@@ -65,7 +65,7 @@ $TEST_BIN -B > $TEST_BTOUT_SPEC
 
 if [ -s "$TEST_BTSCRIPT" ]; then
 	# start attaching bpftrace
-	setsid sudo $bpftrace ${V:+-v} -v -B none "$TEST_BTSCRIPT" >"$TEST_BTOUT_RAW" 2>&1 &
+	setsid sudo $bpftrace ${V:+-v} -B none "$TEST_BTSCRIPT" >"$TEST_BTOUT_RAW" 2>&1 &
 	bt_pid=$!
 	bt_pgid="$(ps -opgid= "$bt_pid" | tr -d ' ')"