A20 delay

[ian at zembu.com (Ian Lance Taylor)]

   Date: Wed, 15 Dec 1999 23:40:01 -0500 (EST)
   From: Quality Quorum <qqi at world.std.com>

   >    BTW, gas still does not like 'jnz $1b' .
   > 
   > I'd be happy to look at a test case.

   Apply patch, replace `more:' with `1:' and `jnz more' with `jnz $1b'

First, I made a mistake talking about $1b.  Temporary dollar labels
use a trailing '$'.  The expression $1b refers to the label 1b as an
immediate value.

It helps a lot to talk about a specific test case.  You gave me
instructions, but I might be starting from the wrong file, or I might
follow the instructions incorrectly, or I might misinterpret the
result.

So, to be precise:

This test case:

	movl    $0x200, %eax
1:	dec     %eax
	jnz     $1b
	ret

fails, as I expect it to.  jnz does not take an immediate operand.  I
get this error message:

/home/ian/foo.s:3: Error: suffix or operands invalid for `jnz'


This test case:

	movl    $0x200, eax
1:	decl    eax
	jnz     1b
	ret

appears to succeed.

   > Note that gas supports two different conventions for local variables:
   > the dollar sign convention and the numeric f/b convention.  If you are
   > literally using `jnz $1b' then you are mixing them, which won't work.

   I am really puzzled, because there a few places  in pc386 where various 
   forms of 

	 jz $1f
   1:

   are both assembled without a hitch and seems working ok.

Again, a specific test case helps.  I could not find any code in pc386
matching your description.  It may be there even though I was unable
to find it.

I did find this case:

	ljmpl	$0x08, $1f
1:

This is not a branch instruction.  In gas syntax, the two operand ljmp
instruction takes an immediate operand.  $1f is an immediate operand
referring to the label 1f.

To see this, this test case fails:

	ljmpl	$0x08, foo
foo:	

while this test case succeeds:

	ljmpl	$0x08, $foo
foo:	

If you really have a test case which uses
	 jz $1f
   1:
and which succeeds, please send it to me.

Ian