Re: (FWD) Re: ** potential patch for 2.10 ** DEC Alpha bug relatedto AT register

Re: (FWD) Re: ** potential patch for 2.10 ** DEC Alpha bug relatedto AT register

[Nick Clifton]

Hi Doug,

: David has asked me to comment on this minor patch to binutils to try to
: get it included in the 2.10 release. I have been working with the alpha
: architecture for several years and I did the original FreeBSD/alpha kernel
: port to this architecture.

I assume, although you do not mention this specifically, that you
would also like to see this patch go into the mainline sources as
well? 

The patch itself is fine.  If you can provide a ChangeLog entry then I
will be happy to check it into the repository.

Cheers
	Nick

Re: (FWD) Re: ** potential patch for 2.10 ** DEC Alpha bug related to AT register

[David O'Brien]

On Thu, Jun 08, 2000 at 10:07:13AM -0700, Nick Clifton wrote:
> : David has asked me to comment on this minor patch to binutils to try to
> : get it included in the 2.10 release. I have been working with the alpha
> : architecture for several years and I did the original FreeBSD/alpha kernel
> : port to this architecture.
> 
> I assume, although you do not mention this specifically, that you
> would also like to see this patch go into the mainline sources as
> well? 
>
> The patch itself is fine.  If you can provide a ChangeLog entry then I
> will be happy to check it into the repository.

I've already got a ChangeLog for it, it actually isn't Doug's patch.
If this is permission, I will commit it to HEAD and 2.10 branches.
If I am only cleared to fix it in HEAD please mention that too. :-)

-- 
-- David    (obrien@NUXI.com)

Re: (FWD) Re: ** potential patch for 2.10 ** DEC Alpha bug relatedto AT register

[Matthew Jacob]

> I've already got a ChangeLog for it, it actually isn't Doug's patch.

It's jointly an O'Brien, Rabson, Jacob production.



> If this is permission, I will commit it to HEAD and 2.10 branches.
> If I am only cleared to fix it in HEAD please mention that too. :-)
> 
> -- 
> -- David    (obrien@NUXI.com)
> 

Re: (FWD) Re: ** potential patch for 2.10 ** DEC Alpha bug relatedto AT register

[Doug Rabson]

On Thu, 8 Jun 2000, Nick Clifton wrote:

> Hi Doug,
> 
> : David has asked me to comment on this minor patch to binutils to try to
> : get it included in the 2.10 release. I have been working with the alpha
> : architecture for several years and I did the original FreeBSD/alpha kernel
> : port to this architecture.
> 
> I assume, although you do not mention this specifically, that you
> would also like to see this patch go into the mainline sources as
> well? 

Yes please.

> 
> The patch itself is fine.  If you can provide a ChangeLog entry then I
> will be happy to check it into the repository.

The ChangeLog would be something along the lines of:

	Fix md_undefined_symbol() so that it does not confuse $f28
	with $at and produce an irrelavent warning.


-- 
Doug Rabson				Mail:  dfr@nlsystems.com
Nonlinear Systems Ltd.			Phone: +44 20 8442 9037

Re: (FWD) Re: ** potential patch for 2.10 ** DEC Alpha bug relatedto AT register

[Matthew Jacob]

> 	Fix md_undefined_symbol() so that it does not confuse $f28
> 	with $at and produce an irrelavent warning.
                                irrelevant

Re: (FWD) Re: ** potential patch for 2.10 ** DEC Alpha bug relatedto AT register

[Matthew Jacob]

Exactly. The original usage of is_float is as an additive value to begin with.

-matt


On Wed, 7 Jun 2000, Doug Rabson wrote:

> On 7 Jun 2000, Ian Lance Taylor wrote:
> 
> >    Date: Wed, 7 Jun 2000 09:18:07 +0100 (BST)
> >    From: Doug Rabson <dfr@nlsystems.com>
> > 
> >    -	  if (!alpha_noat_on && num == AXP_REG_AT)
> >    +	  if (!alpha_noat_on && (num + is_float) == AXP_REG_AT)
> > 
> > is_float is a terrible name for this variable, since the name implies
> > a boolean value.  I believe the patch is correct.  If the variable
> > were named something like regnum_offset or float_regnum_offset it
> > would be clearly correct.
> 
> I agree about the naming but the intention with the patch was to minimise
> changes.
> 
> -- 
> Doug Rabson				Mail:  dfr@nlsystems.com
> Nonlinear Systems Ltd.			Phone: +44 20 8442 9037
> 
> 

Re: (FWD) Re: ** potential patch for 2.10 ** DEC Alpha bug relatedto AT register

[Doug Rabson]

On 7 Jun 2000, Ian Lance Taylor wrote:

>    Date: Wed, 7 Jun 2000 09:18:07 +0100 (BST)
>    From: Doug Rabson <dfr@nlsystems.com>
> 
>    -	  if (!alpha_noat_on && num == AXP_REG_AT)
>    +	  if (!alpha_noat_on && (num + is_float) == AXP_REG_AT)
> 
> is_float is a terrible name for this variable, since the name implies
> a boolean value.  I believe the patch is correct.  If the variable
> were named something like regnum_offset or float_regnum_offset it
> would be clearly correct.

I agree about the naming but the intention with the patch was to minimise
changes.

-- 
Doug Rabson				Mail:  dfr@nlsystems.com
Nonlinear Systems Ltd.			Phone: +44 20 8442 9037

Re: (FWD) Re: ** potential patch for 2.10 ** DEC Alpha bug relatedto AT register

[Ian Lance Taylor]

   Date: Wed, 7 Jun 2000 09:18:07 +0100 (BST)
   From: Doug Rabson <dfr@nlsystems.com>

   -	  if (!alpha_noat_on && num == AXP_REG_AT)
   +	  if (!alpha_noat_on && (num + is_float) == AXP_REG_AT)

is_float is a terrible name for this variable, since the name implies
a boolean value.  I believe the patch is correct.  If the variable
were named something like regnum_offset or float_regnum_offset it
would be clearly correct.

Ian

Re: (FWD) Re: ** potential patch for 2.10 ** DEC Alpha bug relatedto AT register

[Doug Rabson]

David has asked me to comment on this minor patch to binutils to try to
get it included in the 2.10 release. I have been working with the alpha
architecture for several years and I did the original FreeBSD/alpha kernel
port to this architecture.

The background to this fix is that the alpha assembler includes a number
of builtin 'macros' to make life easier for the assembly language
programmer. These macros perform such chores as loading large constants,
performing byte/word sized loads and stores, loading addresses etc.

Since some of these macros require the use of a scratch register, normally
integer register 28 (often called $at) is reserved by the assembler. This
reservation can be controlled with the ".set noat" and ".set
at" directives. The assembler warns about uses of $at when ".set at" is in
effect.

The bug is simple. The code attempts to check for dangerous uses of
register 28 but inadvertantly the test used is also true for $f28 (which
is not used by any macro). This means that any code which uses $f28 will
generate a bogus warning. The patch adjusts the test so that it is only
true for $at.


Index: tc-alpha.c
===================================================================
RCS file: /home/ncvs/src/contrib/binutils/gas/config/tc-alpha.c,v
retrieving revision 1.1.1.3
diff -u -r1.1.1.3 tc-alpha.c
--- tc-alpha.c	2000/05/12 23:14:56	1.1.1.3
+++ tc-alpha.c	2000/06/07 08:08:33
@@ -1372,7 +1372,7 @@
 	  else
 	    break;
 
-	  if (!alpha_noat_on && num == AXP_REG_AT)
+	  if (!alpha_noat_on && (num + is_float) == AXP_REG_AT)
 	    as_warn(_("Used $at without \".set noat\""));
 	  return alpha_register_table[num + is_float];
 


-- 
Doug Rabson				Mail:  dfr@nlsystems.com
Nonlinear Systems Ltd.			Phone: +44 20 8442 9037