Get rid of MAP_FIXED that can cause memory corruption on Mac ARM, rem…

…oved unnecessary APPLE checks

.gitignore

@@ -34,3 +34,4 @@ Release/
 
 new_dynarec_debug.txt
 new_dynarec_disasm.txt
+*.zip

libretro/libretro.c

@@ -2085,10 +2085,6 @@ static void format_saved_memory(void)
 
 bool retro_load_game(const struct retro_game_info *game)
 {
-#if defined(__APPLE__) && defined(__arm64__)
-   extern void new_dynarec_create_mapping();
-   new_dynarec_create_mapping();
-#endif
    format_saved_memory();
 
    update_variables(true);

mupen64plus-core/src/memory/m64p_memory.c

@@ -51,7 +51,7 @@
 #include <stddef.h>
 #include <stdint.h>
 
-#if !defined(__APPLE__) || !defined(__arm64__)
+#if !defined(__arm64__)
 #if NEW_DYNAREC < NEW_DYNAREC_ARM
 // address : address of the read/write operation being done
 uint32_t address = 0;

mupen64plus-core/src/memory/memory.h

@@ -42,14 +42,14 @@
 #define write_hword_in_memory() writememh[mupencoreaddress >>16]()
 #define write_dword_in_memory() writememd[mupencoreaddress >>16]()
 
-#if !defined(__APPLE__) || !defined(__arm64__)
+#if !defined(__arm64__)
 extern uint32_t address, cpu_word;
 extern uint8_t cpu_byte;
 extern uint16_t cpu_hword;
 extern uint64_t cpu_dword;
 #define mupencoreaddress address
 #else
-#include "../r4300/new_dynarec/arm64/apple_memory_layout.h"
+#include "../r4300/new_dynarec/arm64/memory_layout_arm64.h"
 #define mupencoreaddress (RECOMPILER_MEMORY->rml_address)
 #define cpu_word         (RECOMPILER_MEMORY->rml_cpu_word)
 #define cpu_byte         (RECOMPILER_MEMORY->rml_cpu_byte)

mupen64plus-core/src/r4300/cp0.c

@@ -34,7 +34,7 @@
 #endif
 
 /* global variable */
-#if !defined(__APPLE__) || !defined(__arm64__)
+#if !defined(__arm64__)
 #if NEW_DYNAREC < NEW_DYNAREC_ARM
 /* ARM backend requires a different memory layout
  * and therefore manually allocate that variable */

mupen64plus-core/src/r4300/cp0_private.h

@@ -24,10 +24,10 @@
 
 #include "cp0.h"
 
-#if !defined(__APPLE__) || !defined(__arm64__)
+#if !defined(__arm64__)
 extern uint32_t g_cp0_regs[CP0_REGS_COUNT];
 #else
-#include "new_dynarec/arm64/apple_memory_layout.h"
+#include "new_dynarec/arm64/memory_layout_arm64.h"
 #define g_cp0_regs (RECOMPILER_MEMORY->rml_g_cp0_regs)
 #endif
 

mupen64plus-core/src/r4300/cp1.c

@@ -26,7 +26,7 @@
 
 #include "new_dynarec/new_dynarec.h"
 
-#if !defined(__APPLE__) || !defined(__arm64__)
+#if !defined(__arm64__)
 #if NEW_DYNAREC < NEW_DYNAREC_ARM
 float *reg_cop1_simple[32];
 double *reg_cop1_double[32];
@@ -35,7 +35,7 @@ uint32_t FCR0, FCR31;
 #else
 /* ARM backend requires a different memory layout
  * and therefore manually allocates these variables */
-#include "new_dynarec/arm64/apple_memory_layout.h"
+#include "new_dynarec/arm64/memory_layout_arm64.h"
 #define reg_cop1_simple (RECOMPILER_MEMORY->rml_reg_cop1_simple)
 #define reg_cop1_double (RECOMPILER_MEMORY->rml_reg_cop1_double)
 #define FCR0            (RECOMPILER_MEMORY->rml_FCR0)

mupen64plus-core/src/r4300/cp1_private.h

@@ -26,12 +26,12 @@
 
 #include "cp1.h"
 
-#if !defined(__APPLE__) || !defined(__arm64__)
+#if !defined(__arm64__)
 extern float *reg_cop1_simple[32];
 extern double *reg_cop1_double[32];
 extern uint32_t FCR0, FCR31;
 #else
-#include "new_dynarec/arm64/apple_memory_layout.h"
+#include "new_dynarec/arm64/memory_layout_arm64.h"
 #define reg_cop1_simple (RECOMPILER_MEMORY->rml_reg_cop1_simple)
 #define reg_cop1_double (RECOMPILER_MEMORY->rml_reg_cop1_double)
 #define FCR0            (RECOMPILER_MEMORY->rml_FCR0)

mupen64plus-core/src/r4300/new_dynarec/arm64/assem_arm64.c

@@ -313,14 +313,10 @@ static void set_jump_target(intptr_t addr,uintptr_t target)
 static void *dynamic_linker_impl(void * src, u_int vaddr);
 static void *dynamic_linker(void * src, u_int vaddr)
 {
-#if defined(__APPLE__) && defined(__arm64__)
   apple_jit_wx_unprotect_enter();
   void* r = dynamic_linker_impl(src, vaddr);
   apple_jit_wx_unprotect_exit();
   return r;
-#else
-  return dynamic_linker_impl(src, vaddr);
-#endif
 }
 
 static void *dynamic_linker_impl(void * src, u_int vaddr)
@@ -427,14 +423,10 @@ static void *dynamic_linker_impl(void * src, u_int vaddr)
 static void *dynamic_linker_ds_impl(void * src, u_int vaddr);
 static void *dynamic_linker_ds(void * src, u_int vaddr)
 {
-#if defined(__APPLE__) && defined(__arm64__)
   apple_jit_wx_unprotect_enter();
   void* r = dynamic_linker_ds_impl(src, vaddr);
   apple_jit_wx_unprotect_exit();
   return r;
-#else
-  return dynamic_linker_ds_impl(src, vaddr);
-#endif
 }
 
 static void *dynamic_linker_ds_impl(void * src, u_int vaddr)
@@ -6763,6 +6755,7 @@ static void arch_init(void) {
   jump_table_symbols[18] = (intptr_t) cached_interpreter_table.TLBP;
 #endif
 
+  trampoline_init(base_addr);
   apple_jit_wx_unprotect_enter();
   trampolines_reg_jump_t jumps = trampoline_alloc_reg_jump(&jump_vaddr);
   apple_jit_wx_unprotect_exit();

mupen64plus-core/src/r4300/new_dynarec/arm64/assem_arm64.h

@@ -97,7 +97,7 @@ void write_rdramd_new(void);
 void breakpoint(void);
 
 #ifdef __APPLE__
-#include "apple_memory_layout.h"
+#include "memory_layout_arm64.h"
 #define invc_ptr          (RECOMPILER_MEMORY->rml_invc_ptr)
 #define cycle_count       (RECOMPILER_MEMORY->rml_cycle_count)
 #define last_count        (RECOMPILER_MEMORY->rml_last_count)

mupen64plus-core/src/r4300/new_dynarec/arm64/linkage_aarch64.S

@@ -289,11 +289,8 @@ GLOBAL_FUNCTION(_jump_eret):
     b      .E8
 
 GLOBAL_FUNCTION(_new_dyna_start):
-    adrp   x1, _base_addr@PAGE
-    add    x1, x1, _base_addr@PAGEOFF
-    ldr    x16, [x1] /* _extra_memory=base_addr */
-    mov    x17, #0x2000000
-    add    x16, x16, x17 /* _dynarec_local */
+    adrp   x16, _memory_layout@PAGE
+    add    x16, x16, _memory_layout@PAGEOFF
     add    x16, x16, #152
     mov    w0, #0xa4000000
     stp    x19,x20,[x16,#0]
@@ -303,6 +300,8 @@ GLOBAL_FUNCTION(_new_dyna_start):
     stp    x27,x28,[x16,#64]
     stp    x29,x30,[x16,#80]
     sub    x29, x16, #152
+    adrp   x1, _base_addr@PAGE
+    add    x1, x1, _base_addr@PAGEOFF
     ldr    x19, [x1]
     add    w0, w0, #0x40
     bl     _new_recompile_block

mupen64plus-core/src/r4300/new_dynarec/arm64/apple_memory_layout.h → mupen64plus-core/src/r4300/new_dynarec/arm64/memory_layout_arm64.h

@@ -1,13 +1,15 @@
-#pragma once
+#ifndef M64P_MEMORY_LAYOUT_ARM64_H
+#define M64P_MEMORY_LAYOUT_ARM64_H
 
-#ifdef __APPLE__
 #include <stdint.h>
+#ifdef __APPLE__
 #include <sys/types.h>
+#endif
+
 #include "../../recomp_types.h"
 
 typedef struct recompiler_memory_layout
 {
-    char rml_extra_memory[0x2000000]; // rwx memory
     char rml_dynarec_local[256];
     uint32_t rml_next_interrupt;
     int rml_cycle_count;
@@ -44,12 +46,12 @@ typedef struct recompiler_memory_layout
     uint64_t rml_memory_map[1048576];
 } recompiler_memory_layout_t;
 
-extern recompiler_memory_layout_t* base_addr;
-#define RECOMPILER_MEMORY (base_addr)
+extern recompiler_memory_layout_t memory_layout;
+#define RECOMPILER_MEMORY (&memory_layout)
 
 // Copied from linkage_aarch64.S declaration
 #define RML_SIZE_EXTRA_MEMORY 0
-#define RML_SIZE_DYNAREC_LOCAL     RML_SIZE_EXTRA_MEMORY      + 33554432
+#define RML_SIZE_DYNAREC_LOCAL     RML_SIZE_EXTRA_MEMORY      + 0
 #define RML_SIZE_NEXT_INTERRUPT    RML_SIZE_DYNAREC_LOCAL     + 256
 #define RML_SIZE_CYCLE_COUNT       RML_SIZE_NEXT_INTERRUPT    + 4
 #define RML_SIZE_LAST_COUNT        RML_SIZE_CYCLE_COUNT       + 4
@@ -82,6 +84,7 @@ extern recompiler_memory_layout_t* base_addr;
 #define RML_SIZE_LINK_REGISTER     RML_SIZE_INSTR_ADDR        + 8
 #define RML_SIZE_MEMORY_MAP        RML_SIZE_LINK_REGISTER     + 8
 
+#if __STDC_VERSION__ > 201112L
 #define RML_CHECK_SIZE(name, off) _Static_assert(offsetof(recompiler_memory_layout_t, name) == (off), "Recompiler layout for '" #name "' does not match asm")
 RML_CHECK_SIZE(rml_dynarec_local    , RML_SIZE_DYNAREC_LOCAL);
 RML_CHECK_SIZE(rml_next_interrupt   , RML_SIZE_NEXT_INTERRUPT);
@@ -116,5 +119,6 @@ RML_CHECK_SIZE(rml_instr_addr       , RML_SIZE_INSTR_ADDR);
 RML_CHECK_SIZE(rml_link_register    , RML_SIZE_LINK_REGISTER);
 RML_CHECK_SIZE(rml_memory_map       , RML_SIZE_MEMORY_MAP);
 #undef RML_CHECK_SIZE
+#endif
 
 #endif

mupen64plus-core/src/r4300/new_dynarec/new_dynarec.h

@@ -30,7 +30,7 @@
 #define NEW_DYNAREC_ARM 3
 #define NEW_DYNAREC_ARM64 4
 
-#if !defined(__APPLE__) || !defined(__arm64__)
+#if !defined(__arm64__)
 #ifdef __cplusplus
 extern "C" {
 #endif
@@ -40,7 +40,7 @@ extern int pending_exception;
 }
 #endif
 #else
-#include "arm64/apple_memory_layout.h"
+#include "arm64/memory_layout_arm64.h"
 #define pcaddr            (RECOMPILER_MEMORY->rml_pcaddr)
 #define pending_exception (RECOMPILER_MEMORY->rml_pending_exception)
 #endif

mupen64plus-core/src/r4300/new_dynarec/new_dynarec_64.c

@@ -63,7 +63,7 @@
 #include "arm/assem_arm.h"
 #elif NEW_DYNAREC == NEW_DYNAREC_ARM64
 #include "arm64/apple_jit_protect.h"
-#include "arm64/apple_memory_layout.h"
+#include "arm64/memory_layout_arm64.h"
 #include "arm64/assem_arm64.h"
 #else
 #error Unsupported dynarec architecture
@@ -107,11 +107,10 @@ struct ll_entry
   struct ll_entry *next;
 };
 
-#ifdef __APPLE__
-recompiler_memory_layout_t* base_addr = NULL;
-#else
-void *base_addr = NULL;
+#ifdef __arm64__
+recompiler_memory_layout_t memory_layout;
 #endif
+void *base_addr = NULL;
 
 u_char *out;
 ALIGN(16, uintptr_t hash_table[65536][4]);
@@ -1305,6 +1304,7 @@ void invalidate_block(u_int block)
   #ifdef USE_MINI_HT
   memset(mini_ht,-1,sizeof(mini_ht));
   #endif
+
   apple_jit_wx_unprotect_exit();
 }
 
@@ -7752,57 +7752,6 @@ static void disassemble_inst(int i)
 }
 #endif
 
-#if defined(__APPLE__) && defined(__arm64__)
-#include "arm64/trampoline_arm64.h"
-
-void new_dynarec_create_mapping(void)
-{
-  if (base_addr)
-    return;
-
-  // For Mac ARM (and it can be ported to other ARM64 platforms), we want a special layout
-  //  <-  0x2000000  -> |
-  // -------------------|-------------------|------------------- 
-  //        RW^X        ^         RW^X      ^     RW
-  //                    base_addr           dynarec_local
-
-  // pre base_addr is trampolines we need to have, growing like a stack downwards
-  // after base_addr is normal JIT stuff generated
-  // We also want addresses to be in 32-bit space because there are some casts to 32bit unsigned int
-
-  // On macOS we are not allowed to use MAP_FIXED with MAP_JIT so allocate first 2 segments together, then use MAP_FIXED for RW segment
-  // This is a stupid hack - we want part of mapping to be RWX and other part of map to be RW
-  // We hope that the 2nd map wont fail but really this should just do it couple hundred times
-
-#define ATTEMPTS_COUNT 1000
-
-  for (int i = 0; i < ATTEMPTS_COUNT; i++)
-  {
-    char* addr = (char*) mmap(NULL, 2 * (1<<TARGET_SIZE_2), PROT_READ | PROT_WRITE | PROT_EXEC, MAP_ANON | MAP_PRIVATE | MAP_JIT, -1, 0);
-    if (addr == MAP_FAILED)
-    {
-      DebugMessage(M64MSG_ERROR, "mmap() for JIT failed: %d[%s]", errno, strerror(errno));
-      return;
-    }
-
-    // now we need need to attach the RW buffer right after
-    void* expected_rw_addr = addr + 2 * (1<<TARGET_SIZE_2);
-    void* rw_addr = mmap(expected_rw_addr, 1 << TARGET_SIZE_2, PROT_READ | PROT_WRITE, MAP_FIXED | MAP_ANON | MAP_PRIVATE, -1, 0);
-    if (rw_addr == MAP_FAILED)
-    {
-      DebugMessage(M64MSG_WARNING, "Failed to append RW pages at %p after JIT pages to %p", expected_rw_addr, addr);
-      continue;
-    }
-
-    // We have everything setup now, can use base_addr
-    base_addr = (recompiler_memory_layout_t*) (addr + (1<<TARGET_SIZE_2));
-    trampoline_init(base_addr);
-    // TODO DK: debug without this break for large addresses
-    break;
-  }
-}
-#endif
-
 void new_dynarec_init(void)
 {
 #ifdef NEW_DYNAREC_DEBUG
@@ -7815,15 +7764,25 @@ void new_dynarec_init(void)
 #ifndef PROFILER
   DebugMessage(M64MSG_INFO, "Init new dynarec");
 
-#if !defined(__APPLE__) || !defined(__arm64__)
-#if NEW_DYNAREC >= NEW_DYNAREC_ARM
-  if ((base_addr = mmap ((u_char *)BASE_ADDR, 1<<TARGET_SIZE_2,
+#if defined(__arm64__)
+  // trampolines are put behind the base_addr so allocate a bit extra memory
+  // TODO: Don't allocate 32MB, it is definitely too much 
+  if ((base_addr = mmap (NULL, 2 * (1<<TARGET_SIZE_2),
             PROT_READ | PROT_WRITE | PROT_EXEC,
-            MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS
+            MAP_PRIVATE | MAP_ANONYMOUS
 #ifdef __APPLE__
             | MAP_JIT
 #endif
-            ,
+             ,
+            -1, 0)) <= 0) {DebugMessage(M64MSG_ERROR, "mmap() failed");}
+
+  DebugMessage(M64MSG_INFO, "MAP_JIT pages are allocated at %p", base_addr);
+  base_addr = ((char*)base_addr) + (1<<TARGET_SIZE_2);
+#else
+#if NEW_DYNAREC >= NEW_DYNAREC_ARM
+  if ((base_addr = mmap ((u_char *)BASE_ADDR, 1<<TARGET_SIZE_2,
+            PROT_READ | PROT_WRITE | PROT_EXEC,
+            MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS,
             -1, 0)) <= 0) {DebugMessage(M64MSG_ERROR, "mmap() failed");}
 #else
 #if defined(WIN32)
@@ -7909,8 +7868,8 @@ void new_dynarec_cleanup(void)
 #endif
   int n;
 
-#if defined(__APPLE__) && defined(__arm64__)
-  // We are not doing anything, too unsafe
+#if defined(__arm64__)
+  if (munmap (((char*)base_addr) - (1<<TARGET_SIZE_2), 2 * (1<<TARGET_SIZE_2)) < 0) {DebugMessage(M64MSG_ERROR, "munmap() failed");}
 #else
 #if defined(WIN32)
   VirtualFree(base_addr, 0, MEM_RELEASE);
@@ -7933,14 +7892,10 @@ void new_dynarec_cleanup(void)
 static int new_recompile_block_impl(int addr);
 int new_recompile_block(int addr)
 {
-#if defined(__APPLE__) && defined(__arm64__)
   apple_jit_wx_unprotect_enter();
   int r = new_recompile_block_impl(addr);
   apple_jit_wx_unprotect_exit();
   return r;
-#else
-  return new_recompile_block_impl(addr);
-#endif
 }
 
 static int new_recompile_block_impl(int addr)
@@ -7983,7 +7938,6 @@ static int new_recompile_block_impl(int addr)
     else {
       assem_debug("Compile at unmapped memory address: %x ", (int)addr);
       assem_debug("start: %x next: %x",memory_map[start>>12],memory_map[(start+4096)>>12]);
-      __builtin_debugtrap();
       return 1; // Caller will invoke exception handler
     }
     //DebugMessage(M64MSG_VERBOSE, "source= %x",(intptr_t)source);