diff --git a/src/libc/crt0/makefile b/src/libc/crt0/makefile
index 34abdf385..029b073a9 100644
--- a/src/libc/crt0/makefile
+++ b/src/libc/crt0/makefile
@@ -12,7 +12,8 @@ SRC += c1args.c
 SRC += c1loadef.c
 SRC += c1pglob.c
 SRC += crt1.c
-SRC += mcount.c
+SRC += mcount_i.c
+SRC += mcount.S
 SRC += memhandl.c
 SRC += rfinfo.c
 SRC += dfinfo.c
@@ -40,5 +41,5 @@ $(LIB)/gcrt0.o : gcrt0.S crt0.S exit16.ah sbrk16.ah
 	$(MISC) rm gcrt0.o
 	sed 's@gcrt0.o@$(LIB)/gcrt0.o@' gcrt0.d > gcrt02.d
 
-mcount.o : mcount.c
+mcount_i.o : mcount_i.c
 	$(XNOPGGCC) -c $<
diff --git a/src/libc/crt0/mcount.S b/src/libc/crt0/mcount.S
new file mode 100644
index 000000000..823580d5b
--- /dev/null
+++ b/src/libc/crt0/mcount.S
@@ -0,0 +1,25 @@
+	.intel_syntax noprefix
+	.text
+	.globl	_mcount
+_mcount:
+	cmp	esp, offset _etext
+	jb	Lstack_overflow
+	push	ebp
+	push	ecx
+	push	eax
+	mov	ecx, [ebp+4]
+	mov	ebp, esp
+	and	esp, -16		# Keep stack aligned
+	mov	eax, [ebp+12]
+	sub	esp, 4
+	push	edx			# Pointer to cached MTAB entry
+	push	eax			# Our return address (callee)
+	push	ecx			# Callee's return address (caller)
+	call	___mcount_internal
+	mov	esp, ebp
+	pop	eax
+	pop	ecx
+	pop	ebp
+	ret
+Lstack_overflow:
+	ud2
diff --git a/src/libc/crt0/mcount.c b/src/libc/crt0/mcount_i.c
similarity index 91%
rename from src/libc/crt0/mcount.c
rename to src/libc/crt0/mcount_i.c
index a0479b1bd..7b16ac062 100644
--- a/src/libc/crt0/mcount.c
+++ b/src/libc/crt0/mcount_i.c
@@ -38,7 +38,7 @@ typedef struct MTAB {
 
 static header h;
 static short *histogram;
-static int mcount_skip = 1;
+static volatile unsigned char mcount_skip = 1;
 static int histlen;
 static MTAB *mtab=0;
 
@@ -48,35 +48,24 @@ extern int etext __asm__("etext");
 
 static int profiling_p;
 
-/* called by functions.  Use the pointer it provides to cache
+/* called by mcount.  Use the pointer it provides to cache
 ** the last used MTABE, so that repeated calls to/from the same
 ** pair works quickly - no lookup.
 */
-void mcount(int _to);
-void mcount(int _to)
+void __mcount_internal(unsigned long from, unsigned long to, MTABE **cache);
+void __mcount_internal(unsigned long from, unsigned long to, MTABE **cache)
 {
   MTAB *m;
   int i;
-  unsigned int to;
-  int ebp;
-  unsigned int from;
   int mtabi;
-  MTABE **cache;
-
-  /* obtain the cached pointer */
-  __asm__ __volatile__ ("movl %%edx,%0" : "=g" (cache));
 
   mcount_skip = 1;
   /* Do nothing if profiling is disabled.  */
   if (!profiling_p)
     return;
 
-  if (&_to < &etext)
-    *(int *)(-1) = 0; /* fault! */
+  to -= 12;
 
-  to = *((&_to)-1) - 12;
-  ebp = *((&_to)-2); /* glean the caller's return address from the stack */
-  from = ((int *)ebp)[1];
   /* Do nothing if the FROM address is outside the sampling range.  */
   if (from < h.low || from >= h.high)
     return;