The following reply was made to PR lib/29248; it has been noted by GNATS.

From: jtc@acorntoolworks.com (J.T. Conklin)
To: Geoff Wing <gcw@pobox.com>
Cc: NetBSD GNATS <gnats-bugs@netbsd.org>
Subject: Re: lib/29248: memchr.S
Date: Sun, 06 Feb 2005 07:37:12 -0800

 Geoff Wing <gcw@pobox.com> writes:
 > OK, this fixes my problems.  Maybe memset.S strchr.S strrchr.S also
 > need similar clearing on %ecx.
 
 Thanks for tracking this down.  My str* and mem* regression test never
 checked for "character" arguments with values in their high bytes.
 
 Zero extending the value when it's read from the stack frame is better
 than clearing it later.
 
 Here is a patch for memchr.S, strchr.S, and strrchr.S.  Memset already
 used movzbl.
 
     --jtc
 
 --- memchr.S	2004/04/14 02:10:40	1.8
 +++ memchr.S	2005/02/06 15:26:57
 @@ -12,7 +12,7 @@
  ENTRY(memchr)
  	pushl	%esi
  	movl	 8(%esp),%eax
 -	movl	12(%esp),%ecx
 +	movzbl	12(%esp),%ecx
  	movl	16(%esp),%esi
  
  	/*
 @@ -29,6 +29,7 @@
  	incl	%eax
  	decl	%esi
  	jnz	.Lalign
 +	jmp	.Lzero
  	
  .Lword_aligned:
  	/* copy char to all bytes in word */
 --- strchr.S	2004/04/14 02:17:24	1.9
 +++ strchr.S	2005/02/06 15:30:52
 @@ -17,7 +17,7 @@
  	pushl	%esi
  	pushl	%ebx
  	movl	12(%esp),%eax
 -	movl	16(%esp),%ecx
 +	movzbl	16(%esp),%ecx
  
  	/*
  	 * Align to word boundary.
 --- strrchr.S	2004/04/16 16:41:10	1.7
 +++ strrchr.S	2005/02/06 15:31:04
 @@ -18,7 +18,7 @@
  	pushl	%edi
  	pushl	%ebx
  	movl	16(%esp),%edx
 -	movl	20(%esp),%ecx
 +	movzbl	20(%esp),%ecx
  
  	/* zero return value */
  	xorl	%eax,%eax
 
 -- 
 J.T. Conklin