Re: RFA: handle missing fpregs

[Kevin Buettner <kevinb@redhat.com>]

On 06 May 2004 17:24:11 -0500
Jim Blandy <jimb@redhat.com> wrote:

> 2004-05-06  Jim Blandy  <jimb@redhat.com>
> 
> 	* ppc-tdep.h (struct gdbarch_tdep): Change definition of
> 	ppc_fp0_regnum and ppc_fpscr_regnum: if they are -1, then this
> 	processor variant lacks those registers.
> 	(ppc_floating_point_unit_p): Change description to make it clear
> 	that this returns info about the ISA, not the ABI.
> 	* rs6000-tdep.c (ppc_floating_point_unit_p): Decide whether to
> 	return true or false by checking tdep->ppc_fp0_regnum and
> 	tdep->ppc_fpscr_regnum.  The original code replicated the BFD
> 	arch/mach switching done in rs6000_gdbarch_init; it's better to
> 	keep that logic there, and just check the results here.
> 	(rs6000_gdbarch_init): On the E500, set tdep->ppc_fp0_regnum and
> 	tdep->ppc_fpscr_regnum to -1 to indicate that we have no
> 	floating-point registers.
> 	(ppc_supply_fpregset, ppc_collect_fpregset,
> 	rs6000_push_dummy_call, rs6000_extract_return_value,
> 	rs6000_store_return_value): Assert that we have floating-point
> 	registers.
> 	(rs6000_dwarf2_stab_reg_to_regnum): Add FIXME.
> 	(rs6000_frame_cache): Don't note the locations at which
> 	floating-point registers were saved if we have no fprs.
> 	* aix-thread.c (supply_fprs, fill_fprs): Assert that we have FP
> 	registers.
> 	(fetch_regs_user_thread, fetch_regs_kernel_thread,
> 	store_regs_user_thread, store_regs_kernel_thread): Only call
> 	supply_fprs / fill_fprs if we actually have floating-point
> 	registers.
> 	(special_register_p): Check ppc_fpscr_regnum before matching
> 	against it.
> 	(supply_sprs64, supply_sprs32, fill_sprs64, fill_sprs32): Don't
> 	supply / collect fpscr if we don't have it.
> 	* ppc-bdm.c: #include "gdb_assert.h".
> 	(bdm_ppc_fetch_registers, bdm_ppc_store_registers): Assert that we
> 	have floating-point registers, since I can't test this code on
> 	FP-free systems to adapt it.
> 	* ppc-linux-nat.c (ppc_register_u_addr): Don't match against the
> 	fpscr and floating point register numbers if they don't exist.
> 	(fetch_register): Assert that we have
> 	floating-point registers before we reach the code that handles
> 	them.
> 	(store_register): Same.  And use tdep instead of calling
> 	gdbarch_tdep again.
> 	(fill_fpregset): Don't try to collect FP registers and fpscr if we
> 	don't have them.
> 	(ppc_linux_sigtramp_cache): Don't record the saved locations of
> 	fprs and fpscr if we don't have them.
> 	(ppc_linux_supply_fpregset): Don't supply fp regs and fpscr if we
> 	don't have them.
> 	* ppcnbsd-nat.c: #include "gdb_assert.h".
> 	(getfpregs_supplies): Assert that we have floating-point registers.
> 	* ppcnbsd-tdep.c (ppcnbsd_supply_fpreg, ppcnbsd_fill_fpreg): Same.
> 	* ppcobsd-tdep.c: #include "gdb_assert.h".
> 	(ppcobsd_supply_gregset, ppcobsd_collect_gregset): Assert that we
> 	have floating-point registers.
> 	* rs6000-nat.c (regmap): Don't match against the
> 	fpscr and floating point register numbers if they don't exist.
> 	(fetch_inferior_registers, store_inferior_registers,
> 	fetch_core_registers): Only fetch / store / supply the
> 	floating-point registers and the fpscr if we have them.
> 	* Makefile.in (ppc-bdm.o, ppc-linux-nat.o, ppcnbsd-nat.o,
> 	ppcobsd-tdep.o): Update dependencies.

Okay except for a few nits.

This portion of the above ChangeLog entry...

> 	(ppc_supply_fpregset, ppc_collect_fpregset,
> 	rs6000_push_dummy_call, rs6000_extract_return_value,
> 	rs6000_store_return_value): Assert that we have floating-point

...should be formated differently:

 	(ppc_supply_fpregset, ppc_collect_fpregset)
 	(rs6000_push_dummy_call, rs6000_extract_return_value)
 	(rs6000_store_return_value): Assert that we have floating-point

Personally, I prefer the way you did it, but I've been told that emacs
likes the other form better.

Regarding:
> *** gdb/ppc-tdep.h	2004-05-06 14:18:00.000000000 -0500
> --- gdb/ppc-tdep.h	2004-05-06 14:35:11.000000000 -0500
[...]
> *************** struct gdbarch_tdep
> *** 150,158 ****
> --- 150,162 ----
>       int ppc_lr_regnum;		/* Link register */
>       int ppc_ctr_regnum;		/* Count register */
>       int ppc_xer_regnum;		/* Integer exception register */
> + 
> +     /* On RS6000 variants that have no floating-point registers, the
> +        next two members will be -1.  */

I'm not comfortable with the term "RS6000 variants" here.  I'd be happier
with "PPC variants", though that's probably not strictly correct either.
I suppose you could just say "On cores that have no floating-point
registers...".

Regarding:

> *** gdb/rs6000-nat.c	2004-05-06 14:18:00.000000000 -0500
> --- gdb/rs6000-nat.c	2004-05-06 14:35:11.000000000 -0500
[...]
> *************** fetch_core_registers (char *core_reg_sec
> *** 583,591 ****
>         for (regi = 0; regi < 32; regi++)
>           supply_register (regi, (char *) &regs->r64.gpr[regi]);
>   
> !       for (regi = 0; regi < 32; regi++)
> ! 	supply_register (tdep->ppc_fp0_regnum + regi,
> !                          (char *) &regs->r64.fpr[regi]);
>   
>         supply_register (PC_REGNUM, (char *) &regs->r64.iar);
>         supply_register (tdep->ppc_ps_regnum, (char *) &regs->r64.msr);
> --- 589,598 ----
>         for (regi = 0; regi < 32; regi++)
>           supply_register (regi, (char *) &regs->r64.gpr[regi]);
>   
> !       if (tdep->ppc_fp0_regnum >= 0)
> !         for (regi = 0; regi < 32; regi++)
> !           supply_register (tdep->ppc_fp0_regnum + regi,
> !                            (char *) &regs->r64.fpr[regi]);
>   
>         supply_register (PC_REGNUM, (char *) &regs->r64.iar);
>         supply_register (tdep->ppc_ps_regnum, (char *) &regs->r64.msr);

I know it's not really related to this patch, but I happened to notice
that constant 32 in the above (and elsewhere too).  If you get a
chance, could you change these to use either ppc_num_fprs or
ppc_num_gprs?  (I mention this because I noticed that you had made
this kind of change at several points elsewhere in the current patch. 
A separate patch which addressed these remaining occurrences would
certainly be welcome.)

Kevin