fix(benchmarks): bypass native_sim clock intercept via raw syscall

native_sim replaces libc's clock_gettime() to return simulated Zephyr time.
Using syscall(__NR_clock_gettime, ...) bypasses the intercept entirely and
reads CLOCK_MONOTONIC directly from the Linux kernel, giving real ns-precision
wall-clock timing even during CPU-bound loops with no timer interrupts.

Co-Authored-By: Oz <oz-agent@warp.dev>

Author: tbitcs

tests/benchmarks/kalman_benchmark/src/main.c

@@ -19,11 +19,17 @@
  * Zephyr kernel ticks which do NOT advance during CPU-bound loops.
  */
 #ifdef CONFIG_NATIVE_SIMULATOR
+/* native_sim replaces the C-library's clock_gettime() with one that returns
+ * simulated Zephyr time (stuck at 0 during CPU-bound loops). A raw Linux
+ * syscall bypasses the intercept and reads the real CLOCK_MONOTONIC directly
+ * from the kernel, giving true nanosecond-resolution wall-clock timing.
+ */
 #include <time.h>
+#include <sys/syscall.h>
 static inline uint64_t bench_ns(void)
 {
 	struct timespec ts;
-	clock_gettime(CLOCK_MONOTONIC, &ts);
+	syscall(__NR_clock_gettime, CLOCK_MONOTONIC, &ts);
 	return (uint64_t)ts.tv_sec * 1000000000ULL + (uint64_t)ts.tv_nsec;
 }
 #else

tests/benchmarks/pid_benchmark/src/main.c

@@ -23,10 +23,11 @@
 
 #ifdef CONFIG_NATIVE_SIMULATOR
 #include <time.h>
+#include <sys/syscall.h>
 static inline uint64_t bench_ns(void)
 {
 	struct timespec ts;
-	clock_gettime(CLOCK_MONOTONIC, &ts);
+	syscall(__NR_clock_gettime, CLOCK_MONOTONIC, &ts);
 	return (uint64_t)ts.tv_sec * 1000000000ULL + (uint64_t)ts.tv_nsec;
 }
 #else