Re: [PATCH 3/4] Add FreeBSD/arm architecture.

[John Baldwin <jhb at freebsd dot org>]

On Wednesday, October 11, 2017 12:05:35 PM Pedro Alves wrote:
> On 09/14/2017 04:18 PM, John Baldwin wrote:
> > Support for collecting and supplying general purpose and floating
> > point registers is provided along with signal frame unwinding.  While
> > FreeBSD/arm kernels do populate NT_FPREGSET notes, they are always
> > zero-filled, so this implementation ignores them.  Recent FreeBSD/arm
> > kernels generate NT_ARM_VFP notes which are used to supply
> > floating-point registers.  As with Linux, the AT_HWCAP feature flags
> > are used to determine the correct target description.
> > 
> 
> Hi John.  FWIW, this looks good to me.  I'm comfortable with
> you self-approving this as FreeBSD maintainer after a
> period, BTW.  A few minor nits below.
> 
> > diff --git a/gdb/NEWS b/gdb/NEWS
> > index 2e6d48c016..f33b7ac49f 100644
> > --- a/gdb/NEWS
> > +++ b/gdb/NEWS
> > @@ -105,6 +105,7 @@ FreeBSD/aarch64			aarch64*-*-freebsd*
> >  * New targets
> >  
> >  FreeBSD/aarch64			aarch64*-*-freebsd*
> > +FreeBSD/arm			arm*-*-freebsd*
> 
> It'd be nice to update the hosts table at:
>  https://sourceware.org/gdb/wiki/Systems
> 
> (I've added a notes column now, thinking that we'd start
> saying something like "since GDB 8.1".  We could rename
> the column too.)
> 
> > +#define ARM_MCONTEXT_REG_SIZE		4
> > +#define ARM_MCONTEXT_VFP_REG_SIZE	8
> > +#define ARM_SIGFRAME_UCONTEXT_OFFSET	64
> > +#define ARM_UCONTEXT_MCONTEXT_OFFSET	16
> > +#define	ARM_MCONTEXT_VFP_PTR_OFFSET	72
> 
> Space vs tab after #define in the last line above.

Fixed.

> > +
> > +/* Implement the "init" method of struct tramp_frame.  */
> > +
> > +static void
> > +arm_fbsd_sigframe_init (const struct tramp_frame *self,
> > +			struct frame_info *this_frame,
> > +			struct trad_frame_cache *this_cache,
> > +			CORE_ADDR func)
> > +{
> > +  struct gdbarch *gdbarch = get_frame_arch (this_frame);
> > +  enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);
> > +  CORE_ADDR sp = get_frame_register_unsigned (this_frame, ARM_SP_REGNUM);
> > +  CORE_ADDR mcontext_addr =
> > +    sp
> > +    + ARM_SIGFRAME_UCONTEXT_OFFSET
> > +    + ARM_UCONTEXT_MCONTEXT_OFFSET;
> 
> = goes on next line.  Then per GNU standards (because that's
> what Emacs likes), to line up the multiple lines, wrap in parens:
> 
>   CORE_ADDR mcontext_addr 
>     = (sp
>        + ARM_SIGFRAME_UCONTEXT_OFFSET
>        + ARM_UCONTEXT_MCONTEXT_OFFSET);
> 
> Thought this would fit too, and is shorter:
> 
>   CORE_ADDR mcontext_addr = (sp
>                              + ARM_SIGFRAME_UCONTEXT_OFFSET
>                              + ARM_UCONTEXT_MCONTEXT_OFFSET);
 
Ok, the second looks better to me.

> > +  CORE_ADDR mcontext_vfp_addr;
> > +  gdb_byte buf[4];
> > +  int i;
> > +
> > +  for (i = 0; i < 16; i++)
> 
> Suggest writing:
> 
>  for (int i = 0; i < 16; i++)

Done.

> > +    {
> > +      trad_frame_set_reg_addr (this_cache,
> > +			       ARM_A1_REGNUM + i,
> > +			       mcontext_addr + i * ARM_MCONTEXT_REG_SIZE);
> > +    }
> > +  trad_frame_set_reg_addr (this_cache, ARM_PS_REGNUM,
> > +			   mcontext_addr + 16 * ARM_MCONTEXT_REG_SIZE);
> > +
> > +  mcontext_vfp_addr = 0;
> > +  if (target_read_memory (mcontext_addr + ARM_MCONTEXT_VFP_PTR_OFFSET, buf,
> > +			  4) == 0)
> > +    mcontext_vfp_addr = extract_unsigned_integer (buf, 4, byte_order);
> 
> I mildly wonder whether this be:
> 
>   if (safe_read_memory_unsigned_integer (mcontext_addr + ARM_MCONTEXT_VFP_PTR_OFFSET, 4,
>                                          byte_order, &mcontext_vfp_addr)
>     {
>       for (i = 0; i < 32; i++)
>        ....
> 
> I'd convey intention and avoid the "= 0" initialization + "!= 0" check
> (unless you need it anyway).

I do need the != 0 check, but safe_read_memory_unsigned_integer does indeed
look nicer.

BTW, while working on the arm and aarch64 FreeBSD backends I've really liked
having the arrays of 'struct regcache_map_entry' objects.  I think I'd like
to add a similar type of 'map' array for work with trad_frame used for
signal frames as then one can just define a structure to describe the layout
and perhaps give a starting CORE_ADDR and the 'map' array as args to a
trad_frame function that would handle calling trad_frame_set_reg_addr().

-- 
John Baldwin