Re: [rfc] Target-described register support for MIPS

["Maciej W. Rozycki" <macro at linux-mips dot org>]

On Mon, 21 May 2007, Daniel Jacobowitz wrote:

> Maciej, I copied you in case you see anything obviously wrong with my
> choices of required registers, et cetera.  You may not be familiar
> with the protocol extensions and so forth I've added for target
> descriptions - it's been my pet project for the last year or two - but
> this is something that I hope would be very useful for MIPS.  Could
> you look it over?  Additional information about the feature is already
> in the manual; this is just the MIPS-specific support.

 Sorry about the delay -- my time constraints are very tight these days.  
Here are some comments, but I have only had a brief look over your changes 
and certainly not even attempted to try them at the run time.

> +/* Aliases for o32 and most other ABIs.  */
> +const struct register_alias mips_o32_aliases[] = {
> +  { "ta0", 12 },
> +  { "ta1", 13 },
> +  { "ta2", 14 },
> +  { "ta3", 15 }
> +};

 Hmm, these look wrong -- no "ta" registers have been defined for old 
ABIs.

> +      /* FIXME drow/2007-05-17: The FPU should be optional.  The MIPS
> +	 backend is not prepared for that, though.  */

 I do certainly have some patches covering this area -- please check with 
me before commencing any related work.  There are quite a lot of FPU 
configurations to handle too, including MIPS32r2 processors with 64-bit 
FPU and MIPS16 code using hard float.

> Index: gdb/features/mips64-cp0.xml
> ===================================================================
> --- /dev/null	1970-01-01 00:00:00.000000000 +0000
> +++ gdb/features/mips64-cp0.xml	2007-05-18 12:01:07.000000000 -0400
> @@ -0,0 +1,13 @@
> +<?xml version="1.0"?>
> +<!-- Copyright (C) 2007 Free Software Foundation, Inc.
> +
> +     Copying and distribution of this file, with or without modification,
> +     are permitted in any medium without royalty provided the copyright
> +     notice and this notice are preserved.  -->
> +
> +<!DOCTYPE feature SYSTEM "gdb-target.dtd">
> +<feature name="org.gnu.gdb.mips.cp0">
> +  <reg name="status" bitsize="64" regnum="32"/>
> +  <reg name="badvaddr" bitsize="64" regnum="35"/>
> +  <reg name="cause" bitsize="64" regnum="36"/>
> +</feature>

 Hmm, "status" and "cause" are generally 32-bit -- for the MIPS64 
architecture dmfc0/dmtc0 on such registers are defined as yielding 
unpredictable results (and I think at least one implementer did take this 
seriously), so they should be accessed as 32-bit registers.  For legacy 
chips the results may vary too.  The cause register is 64-bit for the 
R8000 IIRC, but whether it matters should probably be verified by an IRIX 
user.

 These are just minor nits.  Overall I like the change, though to cover 
all the optional subsets of cp0 that MIPS32 and MIPS64 specs define quite 
a lot of DTDs will have to be created.

  Maciej