Re: [RFA] Sparc/Linux fixes part 1

[Michael Snyder <msnyder at redhat dot com>]

"David S. Miller" wrote:
> 
> Ok, let's try to split up these changes even more in hopes
> someone might be able to start to look at least at the obvious
> bits.
> 
> 32-bit Sparc under Linux does not have a 16-byte long double,
> this causes testsuite failures.  Fix this by providing a
> SPARC_TARGET_LONG_DOUBLE_BYTES the OS specific target headers
> can override.
> 
> Tested on sparc-linux-gnu, no regressions, a few new successes
> :-)

The idea is sound, but the implementation is not very 
multi-arch-ish.  Is there a runtime way to do this, 
rather than by using a macro?  The whole purpose of
multi-arch is that these decisions get made at runtime.

> 
> 2002-04-19  David S. Miller  <davem@redhat.com>
> 
>         * config/sparc/tm-sparc.h (SPARC_TARGET_LONG_DOUBLE_BYTES): New.
>         (TARGET_LONG_DOUBLE_BIT): Define in non-multi-arch case.
>         * config/sparc/tm-linux.h (SPARC_TARGET_LONG_DOUBLE_BYTES):
>         Override.
>         * config/sparc/tm-sp64linux.h (SPARC_TARGET_LONG_DOUBLE_BYTES):
>         Override.
>         * sparc-tdep.c (sparc_gdbarch_init): Use
>         SPARC_TARGET_LONG_DOUBLE_BYTES in set_gdbarch_long_double_bit
>         call.  Do it after GDB_TARGET_IS_SPARC64 has stabilized.
> 
> --- ./config/sparc/tm-sparc.h.~1~       Mon Jan 28 20:42:44 2002
> +++ ./config/sparc/tm-sparc.h   Fri Apr 19 13:55:48 2002
> @@ -137,6 +137,8 @@ extern int sparc_intreg_size (void);
>  #endif
>  #endif
> 
> +#define SPARC_TARGET_LONG_DOUBLE_BYTES 16
> +
>  #if !defined (GDB_MULTI_ARCH) || (GDB_MULTI_ARCH == 0)
>  /*
>   * The following defines must go away for MULTI_ARCH
> @@ -175,6 +177,11 @@ extern int sparc_intreg_size (void);
>     real way to know how big a register is.  */
> 
>  #define REGISTER_SIZE 4
> +
> +/* Say how large a 'long double' is.  */
> +
> +#define TARGET_LONG_DOUBLE_BIT \
> +        (SPARC_TARGET_LONG_DOUBLE_BYTES * TARGET_CHAR_BIT)
> 
>  /* Number of machine registers */
> 
> --- ./config/sparc/tm-linux.h.~1~       Sun Feb 24 14:56:07 2002
> +++ ./config/sparc/tm-linux.h   Fri Apr 19 13:58:11 2002
> @@ -25,6 +25,9 @@
> 
>  #include "sparc/tm-sparc.h"
> 
> +#undef SPARC_TARGET_LONG_DOUBLE_BYTES
> +#define SPARC_TARGET_LONG_DOUBLE_BYTES 8
> +
>  #define SIGCONTEXT_PC_OFFSET 12
> 
>  #include "tm-linux.h"
> --- ./config/sparc/tm-sp64linux.h.~1~   Sun Feb 24 14:56:07 2002
> +++ ./config/sparc/tm-sp64linux.h       Fri Apr 19 13:58:03 2002
> @@ -25,6 +25,10 @@ Foundation, Inc., 59 Temple Place - Suit
> 
>  #include "sparc/tm-sp64.h"
> 
> +#undef SPARC_TARGET_LONG_DOUBLE_BYTES
> +#define SPARC_TARGET_LONG_DOUBLE_BYTES \
> +       (GDB_TARGET_IS_SPARC64 ? 16 : 8)
> +
>  #define SIGCONTEXT_PC_OFFSET 16  /* See asm-sparc64/sigcontext.h */
> 
>  /* We always want full V9 + Ultra VIS stuff... */
> --- ./sparc-tdep.c.~1~  Fri Apr 12 11:18:57 2002
> +++ ./sparc-tdep.c      Fri Apr 19 13:57:40 2002
> @@ -2970,7 +2970,6 @@ sparc_gdbarch_init (struct gdbarch_info
>    set_gdbarch_init_extra_frame_info (gdbarch, sparc_init_extra_frame_info);
>    set_gdbarch_inner_than (gdbarch, core_addr_lessthan);
>    set_gdbarch_int_bit (gdbarch, 4 * TARGET_CHAR_BIT);
> -  set_gdbarch_long_double_bit (gdbarch, 16 * TARGET_CHAR_BIT);
>    set_gdbarch_long_long_bit (gdbarch, 8 * TARGET_CHAR_BIT);
>    set_gdbarch_max_register_raw_size (gdbarch, 8);
>    set_gdbarch_max_register_virtual_size (gdbarch, 8);
> @@ -3114,6 +3113,11 @@ sparc_gdbarch_init (struct gdbarch_info
>        tdep->call_dummy_call_offset = 148 + 4 * 5;
>        break;
>      }
> +
> +  /* Set this after GDB_TARGET_IS_SPARC64 has stabilized.  */
> +  set_gdbarch_long_double_bit (gdbarch,
> +                              SPARC_TARGET_LONG_DOUBLE_BYTES
> +                              * TARGET_CHAR_BIT);
> 
>    /*
>     * Settings that vary per-architecture: