Fix stack memory, undefined behavior, etc.

libc/crt/crt.S

@@ -67,16 +67,11 @@ _start:
 	lea	16(%rsp,%rbx,8),%rdx		// envp
 	mov	%rsp,__oldstack(%rip)
 
-//	setup a stack frame
-//	align stack to GetStackSize() so GetStackAddr() is fast
-	.weak	ape_stack_memsz
-	mov	$ape_stack_memsz,%r9d
-	mov	$16,%r8d
-	test	%r9d,%r9d
-	cmovnz	%r9,%r8
-	neg	%r8
-	and	%r8,%rsp
+//	setup backtraces
 	xor	%ebp,%ebp
+
+//	make process stack (8mb) follow thread stack (256kb) alignment
+	and	$-(256*1024),%rsp
 //	bofram	9f
 
 #if SupportsWindows()
@@ -126,6 +121,10 @@ _start:
 //	this is the first argument to cosmo() below
 	mov	x0,sp
 
+//	make process stack (8mb) conform to thread stack (256kb) alignment
+	mov	x1,sp
+	and	sp,x1,-(256*1024)
+
 //	second arg shall be struct Syslib passed by ape-m1.c
 //	used to talk to apple's authoritarian libraries
 //	should be set to zero on other platforms

libc/integral/normalize.inc

@@ -74,11 +74,11 @@
 #define __BIGGEST_ALIGNMENT__ 16
 #endif
 
-#define APE_STACKSIZE 8388608 /* default 8mb stack */
-#define APE_PAGESIZE  0x10000 /* i386+ */
+#define APE_STACKSIZE 8388608
+#define APE_PAGESIZE  65536
 #ifdef _COSMO_SOURCE
-#define FRAMESIZE 0x10000
-#define _PAGESIZE 0x1000 /* i386+ */
+#define FRAMESIZE 65536
+#define _PAGESIZE 4096
 #endif
 
 #define BUFSIZ     0x1000 /* best stdio default */

libc/intrin/bsrl.c

@@ -49,7 +49,7 @@ int(_bsrl)(long x) {
   x |= x >> 8;
   x |= x >> 16;
   x |= x >> 32;
-  return kDebruijn[(x * 0x03f79d71b4cb0a89) >> 58];
+  return kDebruijn[(x * 0x03f79d71b4cb0a89ull) >> 58];
 }
 
 __weak_reference(_bsrl, _bsrll);

libc/intrin/describeframe.c

@@ -54,7 +54,7 @@ static const char *GetFrameName(int x) {
               x <= ((GetStaticStackAddr(0) + GetStackSize() +
                      sizeof(struct WinArgs) - 1) >>
                     16))) {
-    return "winargs";
+    return "mainstack";
   } else if ((int)((intptr_t)__executable_start >> 16) <= x &&
              x <= (int)(((intptr_t)_end - 1) >> 16)) {
     return "image";

libc/runtime/mapstack.c

@@ -45,7 +45,7 @@ void *NewCosmoStack(void) {
                 MAP_STACK | MAP_ANONYMOUS, -1, 0)) != MAP_FAILED) {
     if (IsAsan()) {
       __asan_poison(p + GetStackSize() - 16, 16, kAsanStackOverflow);
-      __asan_poison(p, 4096, kAsanStackOverflow);
+      __asan_poison(p, GetGuardSize(), kAsanStackOverflow);
     }
     return p;
   } else {

libc/runtime/mmap.c

@@ -372,13 +372,13 @@ dontasan inline void *__mmap_unlocked(void *addr, size_t size, int prot,
       if ((dm = sys_mmap(p + size - SIGSTKSZ, SIGSTKSZ, prot,
                          f | MAP_GROWSDOWN_linux, fd, off))
               .addr != MAP_FAILED) {
-        npassert(sys_mmap(p, page_size, PROT_NONE,
+        npassert(sys_mmap(p, GetGuardSize(), PROT_NONE,
                           MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0)
                      .addr == p);
         dm.addr = p;
         p = __finish_memory(p, size, prot, flags, fd, off, f, x, n, dm);
         if (IsAsan() && p != MAP_FAILED) {
-          __asan_poison(p, page_size, kAsanStackOverflow);
+          __asan_poison(p, GetGuardSize(), kAsanStackOverflow);
         }
         return p;
       } else if (errno == ENOTSUP) {

libc/runtime/stack.h

@@ -81,9 +81,18 @@ extern char ape_stack_memsz[] __attribute__((__weak__));
 extern char ape_stack_align[] __attribute__((__weak__));
 
 /**
- * Returns size of stack, which is always a two power.
+ * Returns preferred size and alignment of thread stack.
+ *
+ * This will always be equal to `PTHREAD_STACK_MIN`.
  */
-#define GetStackSize() ((uintptr_t)ape_stack_memsz)
+#define GetStackSize() 262144
+
+/**
+ * Returns preferred stack guard size.
+ *
+ * This is the max cpu page size of supported architectures.
+ */
+#define GetGuardSize() 16384
 
 /**
  * Returns address of bottom of stack.
@@ -124,9 +133,13 @@ extern char ape_stack_align[] __attribute__((__weak__));
 /**
  * Returns true if at least `n` bytes of stack are available.
  */
-#define HaveStackMemory(n) \
-  ((intptr_t)__builtin_frame_address(0) >= GetStackAddr() + 16384 + (n))
+#define HaveStackMemory(n)                 \
+  ((intptr_t)__builtin_frame_address(0) >= \
+   GetStackAddr() + GetGuardSize() + (n))
 
+/**
+ * Extends stack memory by poking large allocations.
+ */
 forceinline void CheckLargeStackAllocation(void *p, ssize_t n) {
   for (; n > 0; n -= 4096) {
     ((char *)p)[n - 1] = 0;

libc/thread/thread.h

@@ -2,7 +2,7 @@
 #define COSMOPOLITAN_LIBC_THREAD_THREAD_H_
 
 #define PTHREAD_KEYS_MAX              128
-#define PTHREAD_STACK_MIN             65536
+#define PTHREAD_STACK_MIN             262144
 #define PTHREAD_DESTRUCTOR_ITERATIONS 4
 
 #define PTHREAD_BARRIER_SERIAL_THREAD 31337