Merge all libs into libu

libu/amd64/.gitignore

@@ -0,0 +1 @@
+*.o

libu/amd64/_start.S

@@ -0,0 +1,7 @@
+.global _start
+_start:
+  xorq %rbp, %rbp
+  movq %rsp, %rbp
+  andq $-16, %rsp
+
+  callq __premain

libu/amd64/clone_tramp.S

@@ -0,0 +1,7 @@
+.global _clone_tramp
+_clone_tramp:
+  xorq %rbp, %rbp
+  movq %rsp, %rbp
+  andq $-16, %rsp
+
+  callq _clone_tramp1

libu/amd64/clone_tramp1.c

@@ -0,0 +1,12 @@
+#include <malloc.h>
+#include <process.h>
+#include <system.h>
+
+void _clone_tramp1 (void) {
+  struct process_data* pdata = argument_ptr ();
+
+  pdata->fn (pdata->arg_ptr);
+
+  free (pdata);
+  quit ();
+}

libu/amd64/src.mk

@@ -0,0 +1,12 @@
+c += amd64/syscall.c \
+		 amd64/stall.c \
+		 amd64/clone_tramp1.c
+
+S += amd64/_start.S \
+		 amd64/clone_tramp.S
+
+o += amd64/_start.o \
+		 amd64/syscall.o \
+		 amd64/stall.o \
+		 amd64/clone_tramp.o \
+		 amd64/clone_tramp1.o

libu/amd64/stall.c

@@ -0,0 +1,30 @@
+#include <stall.h>
+#include <stdint.h>
+
+static uint64_t stall_read_tsc (void) {
+  uint32_t lo, hi;
+  __asm__ volatile ("rdtsc" : "=a"(lo), "=d"(hi));
+  return ((uint64_t)hi << 32) | lo;
+}
+
+static uint64_t stall_get_tsc_freq_hz (void) {
+  uint32_t eax, ebx, ecx, edx;
+  __asm__ volatile ("cpuid" : "=a"(eax), "=b"(ebx), "=c"(ecx), "=d"(edx) : "a"(0x15));
+
+  if (eax == 0 || ebx == 0 || ecx == 0)
+    return 2500000000ULL;
+
+  return (uint64_t)ecx * ebx / eax;
+}
+
+void stall_ms (uint64_t ms) {
+  uint64_t freq_hz = stall_get_tsc_freq_hz ();
+
+  uint64_t cycles = freq_hz / 1000;
+  uint64_t wait_cycles = ms * cycles;
+
+  uint64_t now = stall_read_tsc ();
+
+  while ((stall_read_tsc () - now) < wait_cycles)
+    __asm__ volatile ("pause" ::: "memory");
+}

libu/amd64/syscall.c

@@ -0,0 +1,17 @@
+#include <amd64/syscall.h>
+#include <stddef.h>
+#include <stdint.h>
+
+uintptr_t amd64_syscall (int syscall_num, uintptr_t a1, uintptr_t a2, uintptr_t a3, uintptr_t a4,
+                         uintptr_t a5, uintptr_t a6) {
+  uint64_t result;
+  __asm__ volatile ("movq %[a4], %%r10\n"
+                    "movq %[a5], %%r8\n"
+                    "movq %[a6], %%r9\n"
+                    "syscall\n"
+                    : "=a"(result)
+                    : "a"(syscall_num), "D"(a1), "S"(a2),
+                      "d"(a3), [a4] "r"(a4), [a5] "r"(a5), [a6] "r"(a6)
+                    : "r10", "r8", "r9", "r11", "rcx", "cc", "memory");
+  return result;
+}

libu/amd64/syscall.h

@@ -0,0 +1,10 @@
+#ifndef _LIBMSL_AMD64_SYSCALL_H
+#define _LIBMSL_AMD64_SYSCALL_H
+
+#include <stdint.h>
+
+/* Performs a syscall for the AMD64  */
+uintptr_t amd64_syscall (int syscall_num, uintptr_t a1, uintptr_t a2, uintptr_t a3, uintptr_t a4,
+                         uintptr_t a5, uintptr_t a6);
+
+#endif // _LIBMSL_AMD64_SYSCALL_H