arc4random (again)

It's only stirred after first handling out 256 zeroes
- seeing the offending code right there in my previous message probably 
makes it the worst bugreport ever :)

Initializing n adds some bytes to arc4random.o, which go away inverting 
its meaning (ie, n = "# of uint32 available before next stir")

"n-=1" because I almost sent a patch with "n--"


Index: lib/arc4random.c
===================================================================
RCS file: /cvs/dietlibc/lib/arc4random.c,v
retrieving revision 1.2
diff -u -r1.2 arc4random.c
--- lib/arc4random.c	26 May 2015 02:55:23 -0000	1.2
+++ lib/arc4random.c	29 May 2015 15:46:39 -0000
@@ -7,8 +7,8 @@

  /* These come from OpenBSD: */
  uint32_t arc4random(void) {
-  if (n>=sizeof(buf)/sizeof(buf[0])) arc4random_stir();
-  return buf[n++];
+  if (n == 0) arc4random_stir();
+  return buf[n-=1];
  }

  void arc4random_buf(void* Buf, size_t N) {
@@ -41,7 +41,7 @@
  }

  void arc4random_stir(void) {
-  n=0;
+  n=sizeof(buf)/sizeof(buf[0]);
    if ((size_t)getrandom(buf,sizeof(buf),GRND_NONBLOCK) != sizeof(buf))
      abort();
  }