add sched_getcpu vDSO support

This brings the call to an actually usable speed.
Quick unscientific benchmark: 14ns : 102ns :: vDSO : syscall

arch/x86_64/syscall_arch.h

@@ -64,3 +64,5 @@ static __inline long __syscall6(long n, long a1, long a2, long a3, long a4, long
 #define VDSO_USEFUL
 #define VDSO_CGT_SYM "__vdso_clock_gettime"
 #define VDSO_CGT_VER "LINUX_2.6"
+#define VDSO_GETCPU_SYM "__vdso_getcpu"
+#define VDSO_GETCPU_VER "LINUX_2.6"

src/sched/sched_getcpu.c

@@ -1,12 +1,43 @@
 #define _GNU_SOURCE
+#include <errno.h>
 #include <sched.h>
 #include "syscall.h"
+#include "atomic.h"
+
+#ifdef VDSO_GETCPU_SYM
+
+void *__vdsosym(const char *, const char *);
+
+static void *volatile vdso_func;
+
+typedef long (*getcpu_f)(unsigned *, unsigned *, void *);
+
+static long getcpu_init(unsigned *cpu, unsigned *node, void *unused)
+{
+	void *p = __vdsosym(VDSO_GETCPU_VER, VDSO_GETCPU_SYM);
+	getcpu_f f = (getcpu_f)p;
+	a_cas_p(&vdso_func, (void *)getcpu_init, p);
+	return f ? f(cpu, node, unused) : -ENOSYS;
+}
+
+static void *volatile vdso_func = (void *)getcpu_init;
+
+#endif
 
 int sched_getcpu(void)
 {
 	int r;
 	unsigned cpu;
 
+#ifdef VDSO_GETCPU_SYM
+	getcpu_f f = (getcpu_f)vdso_func;
+	if (f) {
+		r = f(&cpu, 0, 0);
+		if (!r) return cpu;
+		if (r != -ENOSYS) return __syscall_ret(r);
+	}
+#endif
+
 	r = __syscall(SYS_getcpu, &cpu, 0, 0);
 	if (!r) return cpu;
 	return __syscall_ret(r);