Rewrite memory block allocator.

Use a mix of linear probing and pseudo-random probing.
Workaround for 1GB MAP_32BIT limit on Linux/x64. Now 2GB with !LJ_GC64.
Enforce 128TB LJ_GC64 limit for > 47 bit memory layouts (ARM64).

src/lj_alloc.c

@@ -72,13 +72,56 @@
 
 #define IS_DIRECT_BIT		(SIZE_T_ONE)
 
+
+/* Determine system-specific block allocation method. */
 #if LJ_TARGET_WINDOWS
 
 #define WIN32_LEAN_AND_MEAN
 #include <windows.h>
 
-#if LJ_64 && !LJ_GC64
+#define LJ_ALLOC_VIRTUALALLOC	1
 
+#if LJ_64 && !LJ_GC64
+#define LJ_ALLOC_NTAVM		1
+#endif
+
+#else
+
+#include <errno.h>
+/* If this include fails, then rebuild with: -DLUAJIT_USE_SYSMALLOC */
+#include <sys/mman.h>
+
+#define LJ_ALLOC_MMAP		1
+
+#if LJ_64
+
+#define LJ_ALLOC_MMAP_PROBE	1
+
+#if LJ_GC64
+#define LJ_ALLOC_MBITS		47	/* 128 TB in LJ_GC64 mode. */
+#elif LJ_TARGET_X64 && LJ_HASJIT
+/* Due to limitations in the x64 compiler backend. */
+#define LJ_ALLOC_MBITS		31	/* 2 GB on x64 with !LJ_GC64. */
+#else
+#define LJ_ALLOC_MBITS		32	/* 4 GB on other archs with !LJ_GC64. */
+#endif
+
+#endif
+
+#if LJ_64 && !LJ_GC64 && defined(MAP_32BIT)
+#define LJ_ALLOC_MMAP32		1
+#endif
+
+#if LJ_TARGET_LINUX
+#define LJ_ALLOC_MREMAP		1
+#endif
+
+#endif
+
+
+#if LJ_ALLOC_VIRTUALALLOC
+
+#if LJ_ALLOC_NTAVM
 /* Undocumented, but hey, that's what we all love so much about Windows. */
 typedef long (*PNTAVM)(HANDLE handle, void **addr, ULONG zbits,
 		       size_t *size, ULONG alloctype, ULONG prot);
@@ -89,14 +132,15 @@ static PNTAVM ntavm;
 */
 #define NTAVM_ZEROBITS		1
 
-static void INIT_MMAP(void)
+static void init_mmap(void)
 {
   ntavm = (PNTAVM)GetProcAddress(GetModuleHandleA("ntdll.dll"),
 				 "NtAllocateVirtualMemory");
 }
+#define INIT_MMAP()	init_mmap()
 
 /* Win64 32 bit MMAP via NtAllocateVirtualMemory. */
-static LJ_AINLINE void *CALL_MMAP(size_t size)
+static void *CALL_MMAP(size_t size)
 {
   DWORD olderr = GetLastError();
   void *ptr = NULL;
@@ -107,7 +151,7 @@ static LJ_AINLINE void *CALL_MMAP(size_t size)
 }
 
 /* For direct MMAP, use MEM_TOP_DOWN to minimize interference */
-static LJ_AINLINE void *DIRECT_MMAP(size_t size)
+static void *DIRECT_MMAP(size_t size)
 {
   DWORD olderr = GetLastError();
   void *ptr = NULL;
@@ -119,10 +163,8 @@ static LJ_AINLINE void *DIRECT_MMAP(size_t size)
 
 #else
 
-#define INIT_MMAP()		((void)0)
-
 /* Win32 MMAP via VirtualAlloc */
-static LJ_AINLINE void *CALL_MMAP(size_t size)
+static void *CALL_MMAP(size_t size)
 {
   DWORD olderr = GetLastError();
   void *ptr = VirtualAlloc(0, size, MEM_RESERVE|MEM_COMMIT, PAGE_READWRITE);
@@ -131,7 +173,7 @@ static LJ_AINLINE void *CALL_MMAP(size_t size)
 }
 
 /* For direct MMAP, use MEM_TOP_DOWN to minimize interference */
-static LJ_AINLINE void *DIRECT_MMAP(size_t size)
+static void *DIRECT_MMAP(size_t size)
 {
   DWORD olderr = GetLastError();
   void *ptr = VirtualAlloc(0, size, MEM_RESERVE|MEM_COMMIT|MEM_TOP_DOWN,
@@ -143,7 +185,7 @@ static LJ_AINLINE void *DIRECT_MMAP(size_t size)
 #endif
 
 /* This function supports releasing coalesed segments */
-static LJ_AINLINE int CALL_MUNMAP(void *ptr, size_t size)
+static int CALL_MUNMAP(void *ptr, size_t size)
 {
   DWORD olderr = GetLastError();
   MEMORY_BASIC_INFORMATION minfo;
@@ -163,10 +205,7 @@ static LJ_AINLINE int CALL_MUNMAP(void *ptr, size_t size)
   return 0;
 }
 
-#else
+#elif LJ_ALLOC_MMAP
-
-#include <errno.h>
-#include <sys/mman.h>
 
 #define MMAP_PROT		(PROT_READ|PROT_WRITE)
 #if !defined(MAP_ANONYMOUS) && defined(MAP_ANON)
@@ -174,107 +213,145 @@ static LJ_AINLINE int CALL_MUNMAP(void *ptr, size_t size)
 #endif
 #define MMAP_FLAGS		(MAP_PRIVATE|MAP_ANONYMOUS)
 
-#if LJ_64 && !LJ_GC64
+#if LJ_ALLOC_MMAP_PROBE
-/* 64 bit mode with 32 bit pointers needs special support for allocating
+
-** memory in the lower 2GB.
+#define LJ_ALLOC_MMAP_PROBE_MAX		30
+#define LJ_ALLOC_MMAP_PROBE_LINEAR	5
+
+#define LJ_ALLOC_MMAP_PROBE_LOWER	((uintptr_t)0x4000)
+
+/* No point in a giant ifdef mess. Just try to open /dev/urandom.
+** It doesn't really matter if this fails, since we get some ASLR bits from
+** every unsuitable allocation, too. And we prefer linear allocation, anyway.
 */
+#include <fcntl.h>
+#include <unistd.h>
 
-#if defined(MAP_32BIT)
+static uintptr_t mmap_probe_seed(void)
-
-#if defined(__sun__)
-#define MMAP_REGION_START	((uintptr_t)0x1000)
-#else
-/* Actually this only gives us max. 1GB in current Linux kernels. */
-#define MMAP_REGION_START	((uintptr_t)0)
-#endif
-
-static LJ_AINLINE void *CALL_MMAP(size_t size)
 {
-  int olderr = errno;
+  uintptr_t val;
-  void *ptr = mmap((void *)MMAP_REGION_START, size, MMAP_PROT, MAP_32BIT|MMAP_FLAGS, -1, 0);
+  int fd = open("/dev/urandom", O_RDONLY);
-  errno = olderr;
+  if (fd != -1) {
-  return ptr;
+    int ok = ((size_t)read(fd, &val, sizeof(val)) == sizeof(val));
+    (void)close(fd);
+    if (ok) return val;
+  }
+  return 1;  /* Punt. */
 }
 
-#elif LJ_TARGET_OSX || LJ_TARGET_PS4 || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) || defined(__NetBSD__) || defined(__OpenBSD__) || defined(__DragonFly__) || defined(__sun__) || LJ_TARGET_CYGWIN
+static void *mmap_probe(size_t size)
-
-/* OSX and FreeBSD mmap() use a naive first-fit linear search.
-** That's perfect for us. Except that -pagezero_size must be set for OSX,
-** otherwise the lower 4GB are blocked. And the 32GB RLIMIT_DATA needs
-** to be reduced to 250MB on FreeBSD.
-*/
-#if LJ_TARGET_OSX || defined(__DragonFly__)
-#define MMAP_REGION_START	((uintptr_t)0x10000)
-#elif LJ_TARGET_PS4
-#define MMAP_REGION_START	((uintptr_t)0x4000)
-#else
-#define MMAP_REGION_START	((uintptr_t)0x10000000)
-#endif
-#define MMAP_REGION_END		((uintptr_t)0x80000000)
-
-#if (defined(__FreeBSD__) || defined(__FreeBSD_kernel__)) && !LJ_TARGET_PS4
-#include <sys/resource.h>
-#endif
-
-static LJ_AINLINE void *CALL_MMAP(size_t size)
 {
-  int olderr = errno;
   /* Hint for next allocation. Doesn't need to be thread-safe. */
-  static uintptr_t alloc_hint = MMAP_REGION_START;
+  static uintptr_t hint_addr = 0;
-  int retry = 0;
+  static uintptr_t hint_prng = 0;
-#if (defined(__FreeBSD__) || defined(__FreeBSD_kernel__)) && !LJ_TARGET_PS4
+  int olderr = errno;
-  static int rlimit_modified = 0;
+  int retry;
-  if (LJ_UNLIKELY(rlimit_modified == 0)) {
+  for (retry = 0; retry < LJ_ALLOC_MMAP_PROBE_MAX; retry++) {
-    struct rlimit rlim;
+    void *p = mmap((void *)hint_addr, size, MMAP_PROT, MMAP_FLAGS, -1, 0);
-    rlim.rlim_cur = rlim.rlim_max = MMAP_REGION_START;
+    uintptr_t addr = (uintptr_t)p;
-    setrlimit(RLIMIT_DATA, &rlim);  /* Ignore result. May fail below. */
+    if ((addr >> LJ_ALLOC_MBITS) == 0 && addr >= LJ_ALLOC_MMAP_PROBE_LOWER) {
-    rlimit_modified = 1;
+      /* We got a suitable address. Bump the hint address. */
-  }
+      hint_addr = addr + size;
-#endif
-  for (;;) {
-    void *p = mmap((void *)alloc_hint, size, MMAP_PROT, MMAP_FLAGS, -1, 0);
-    if ((uintptr_t)p >= MMAP_REGION_START &&
-	(uintptr_t)p + size < MMAP_REGION_END) {
-      alloc_hint = (uintptr_t)p + size;
       errno = olderr;
       return p;
     }
-    if (p != CMFAIL) munmap(p, size);
+    if (p != MFAIL) {
-#if defined(__sun__) || defined(__DragonFly__)
+      munmap(p, size);
-    alloc_hint += 0x1000000;  /* Need near-exhaustive linear scan. */
+    } else if (errno == ENOMEM) {
-    if (alloc_hint + size < MMAP_REGION_END) continue;
+      return MFAIL;
-#endif
+    }
-    if (retry) break;
+    if (hint_addr) {
-    retry = 1;
+      /* First, try linear probing. */
-    alloc_hint = MMAP_REGION_START;
+      if (retry < LJ_ALLOC_MMAP_PROBE_LINEAR) {
+	hint_addr += 0x1000000;
+	if (((hint_addr + size) >> LJ_ALLOC_MBITS) != 0)
+	  hint_addr = 0;
+	continue;
+      } else if (retry == LJ_ALLOC_MMAP_PROBE_LINEAR) {
+	/* Next, try a no-hint probe to get back an ASLR address. */
+	hint_addr = 0;
+	continue;
+      }
+    }
+    /* Finally, try pseudo-random probing. */
+    if (LJ_UNLIKELY(hint_prng == 0)) {
+      hint_prng = mmap_probe_seed();
+    }
+    /* The unsuitable address we got has some ASLR PRNG bits. */
+    hint_addr ^= addr & ~((uintptr_t)(LJ_PAGESIZE-1));
+    do {  /* The PRNG itself is very weak, but see above. */
+      hint_prng = hint_prng * 1103515245 + 12345;
+      hint_addr ^= hint_prng * (uintptr_t)LJ_PAGESIZE;
+      hint_addr &= (((uintptr_t)1 << LJ_ALLOC_MBITS)-1);
+    } while (hint_addr < LJ_ALLOC_MMAP_PROBE_LOWER);
   }
   errno = olderr;
-  return CMFAIL;
+  return MFAIL;
 }
 
-#else
-
-#error "NYI: need an equivalent of MAP_32BIT for this 64 bit OS"
-
 #endif
 
-#else
+#if LJ_ALLOC_MMAP32
 
-/* 32 bit mode and GC64 mode is easy. */
+#if defined(__sun__)
-static LJ_AINLINE void *CALL_MMAP(size_t size)
+#define LJ_ALLOC_MMAP32_START	((uintptr_t)0x1000)
+#else
+#define LJ_ALLOC_MMAP32_START	((uintptr_t)0)
+#endif
+
+static void *mmap_map32(size_t size)
+{
+#if LJ_ALLOC_MMAP_PROBE
+  static int fallback = 0;
+  if (fallback)
+    return mmap_probe(size);
+#endif
+  {
+    int olderr = errno;
+    void *ptr = mmap((void *)LJ_ALLOC_MMAP32_START, size, MMAP_PROT, MAP_32BIT|MMAP_FLAGS, -1, 0);
+    errno = olderr;
+    /* This only allows 1GB on Linux. So fallback to probing to get 2GB. */
+#if LJ_ALLOC_MMAP_PROBE
+    if (ptr == MFAIL) {
+      fallback = 1;
+      return mmap_probe(size);
+    }
+#endif
+    return ptr;
+  }
+}
+
+#endif
+
+#if LJ_ALLOC_MMAP32
+#define CALL_MMAP(size)		mmap_map32(size)
+#elif LJ_ALLOC_MMAP_PROBE
+#define CALL_MMAP(size)		mmap_probe(size)
+#else
+static void *CALL_MMAP(size_t size)
 {
   int olderr = errno;
   void *ptr = mmap(NULL, size, MMAP_PROT, MMAP_FLAGS, -1, 0);
   errno = olderr;
   return ptr;
 }
+#endif
+
+#if (defined(__FreeBSD__) || defined(__FreeBSD_kernel__)) && !LJ_TARGET_PS4
+
+#include <sys/resource.h>
+
+static void init_mmap(void)
+{
+  struct rlimit rlim;
+  rlim.rlim_cur = rlim.rlim_max = 0x10000;
+  setrlimit(RLIMIT_DATA, &rlim);  /* Ignore result. May fail later. */
+}
+#define INIT_MMAP()	init_mmap()
 
 #endif
 
-#define INIT_MMAP()		((void)0)
+static int CALL_MUNMAP(void *ptr, size_t size)
-#define DIRECT_MMAP(s)		CALL_MMAP(s)
-
-static LJ_AINLINE int CALL_MUNMAP(void *ptr, size_t size)
 {
   int olderr = errno;
   int ret = munmap(ptr, size);
@@ -282,10 +359,9 @@ static LJ_AINLINE int CALL_MUNMAP(void *ptr, size_t size)
   return ret;
 }
 
-#if LJ_TARGET_LINUX
+#if LJ_ALLOC_MREMAP
 /* Need to define _GNU_SOURCE to get the mremap prototype. */
-static LJ_AINLINE void *CALL_MREMAP_(void *ptr, size_t osz, size_t nsz,
+static void *CALL_MREMAP_(void *ptr, size_t osz, size_t nsz, int flags)
-				     int flags)
 {
   int olderr = errno;
   ptr = mremap(ptr, osz, nsz, flags);
@@ -305,6 +381,15 @@ static LJ_AINLINE void *CALL_MREMAP_(void *ptr, size_t osz, size_t nsz,
 
 #endif
 
+
+#ifndef INIT_MMAP
+#define INIT_MMAP()		((void)0)
+#endif
+
+#ifndef DIRECT_MMAP
+#define DIRECT_MMAP(s)		CALL_MMAP(s)
+#endif
+
 #ifndef CALL_MREMAP
 #define CALL_MREMAP(addr, osz, nsz, mv) ((void)osz, MFAIL)
 #endif