userland/arch/aarch64/freestanding/linux/gem5_exit.S to benchmark immediate gem5 exit

ldxr_stxr does an ldxr on thread 0 to fall in the ISA case where event is certain.

Also add ldxr_str which tests yet another case where event is certain: non conditional
write.

userland/arch/aarch64/freestanding/linux/gem5_exit.S

@@ -0,0 +1,10 @@
+/* https://cirosantilli.com/linux-kernel-module-cheat#benchmark-emulators-on-userland-executables */
+
+#define LKMC_M5OPS_ENABLE 1
+#include "lkmc/m5ops.h"
+
+.text
+.global _start
+_start:
+asm_main_after_prologue:
+    LKMC_M5OPS_EXIT_ASM

userland/arch/aarch64/inline_asm/wfe_ldxr_str.cpp

@@ -0,0 +1,35 @@
+// https://cirosantilli.com/linux-kernel-module-cheat#arm-wfe-global-monitor-events
+
+#ifndef _GNU_SOURCE
+#define _GNU_SOURCE
+#endif
+#include <atomic>
+#include <iostream>
+#include <mutex>
+#include <thread>
+
+#include <lkmc/futex.h>
+
+std::atomic_ulong done;
+int futex = 1;
+
+void myfunc() {
+    __asm__ __volatile__ ("ldxr x0, [%0];wfe;wfe" : : "r" (&futex) : "x0");
+    done.store(futex);
+}
+
+int main(int argc, char **argv) {
+    bool do_sev = true;
+    if (argc > 1) {
+        do_sev = (argv[1][0] != '0');
+    }
+    done.store(0);
+    std::thread thread;
+    thread = std::thread(myfunc);
+    while (!done.load()) {
+        if (do_sev) {
+            __asm__ __volatile__ ("mov x0, 1;str x0, [%0]" : : "r" (&futex) : "x0", "x1");
+        }
+    }
+    thread.join();
+}

userland/arch/aarch64/inline_asm/wfe_ldxr_stxr.cpp

@@ -28,7 +28,7 @@ int main(int argc, char **argv) {
     thread = std::thread(myfunc);
     while (!done.load()) {
         if (do_sev) {
-            __asm__ __volatile__ ("mov x0, 1;stxr w1, x0, [%0]" : : "r" (&futex) : "x0", "x1");
+            __asm__ __volatile__ ("mov x0, 1;ldxr x0, [%0];stxr w1, x0, [%0]" : : "r" (&futex) : "x0", "x1");
         }
     }
     thread.join();