Re: [PATCH] Clear *VAL in regcache_raw_read_unsigned

[Pedro Alves <palves@redhat.com>]

On 02/11/2016 03:15 PM, Yao Qi wrote:

> I think this is better than 'make gdbserver use extract_unsigned_integer',
> which looks overkill to me.

I think it's only going to bite us back in the future.

From the perspective of potentially making it easier to share more
code between gdb and gdbserver, I'd prefer that.  Would you object it?

> 
>>
>> Or maybe better, just byte the bullet and make gdbserver use
>> extract_unsigned_integer, like gdb.
>>
>> The problem with that is that gdbserver can't currently use 'enum bfd_endian',
>> which IIRC, was already an issue in the get-next-pcs stuff.  I've attached a
> 
> get-next-pcs stuff needs endianness in GDB side.  In GDBserver,
> endianness is not needed.

The get-next-pcs stuff does have endianness bits, but it works
around the lack of 'enum bfd_endian' by hacking it through an int instead:

void
arm_get_next_pcs_ctor (struct arm_get_next_pcs *self,
		       struct arm_get_next_pcs_ops *ops,
		       int byte_order,
		       int byte_order_for_code,
		       int has_thumb2_breakpoint,
		       struct regcache *regcache)
{

$ grep byte_order *

arm-get-next-pcs.c:                    int byte_order,
arm-get-next-pcs.c:                    int byte_order_for_code,
arm-get-next-pcs.c:  self->byte_order = byte_order;
arm-get-next-pcs.c:  self->byte_order_for_code = byte_order_for_code;
arm-get-next-pcs.c:  int byte_order_for_code = self->byte_order_for_code;
arm-get-next-pcs.c:  insn1 = self->ops->read_mem_uint (loc, 2, byte_order_for_code);
arm-get-next-pcs.c:  insn2 = self->ops->read_mem_uint (loc, 2, byte_order_for_code);
arm-get-next-pcs.c:      insn1 = self->ops->read_mem_uint (loc, 2,byte_order_for_code);
arm-get-next-pcs.c:       insn2 = self->ops->read_mem_uint (loc, 2, byte_order_for_code);
arm-get-next-pcs.c:  int byte_order_for_code = self->byte_order_for_code;
arm-get-next-pcs.c:  insn = self->ops->read_mem_uint (loc, 4, byte_order_for_code);
arm-get-next-pcs.c:      insn = self->ops->read_mem_uint (loc, 4, byte_order_for_code);
arm-get-next-pcs.c:  int byte_order = self->byte_order;
arm-get-next-pcs.c:  int byte_order_for_code = self->byte_order_for_code;
arm-get-next-pcs.c:  inst1 = self->ops->read_mem_uint (pc, 2, byte_order_for_code);
arm-get-next-pcs.c:           inst1 = self->ops->read_mem_uint (pc, 2,byte_order_for_code);
arm-get-next-pcs.c:               inst1 = self->ops->read_mem_uint (pc, 2, byte_order_for_code);
arm-get-next-pcs.c:               inst1 = self->ops->read_mem_uint (pc, 2, byte_order_for_code);
arm-get-next-pcs.c:      nextpc = self->ops->read_mem_uint (sp + offset, 4, byte_order);
arm-get-next-pcs.c:      inst2 = self->ops->read_mem_uint (pc + 2, 2, byte_order_for_code);
arm-get-next-pcs.c:           nextpc = self->ops->read_mem_uint (addr + offset, 4, byte_order);
arm-get-next-pcs.c:           = self->ops->read_mem_uint (base, 4, byte_order);
arm-get-next-pcs.c:       length = 2 * self->ops->read_mem_uint (table + offset, 1, byte_order);
arm-get-next-pcs.c:       length = 2 * self->ops->read_mem_uint (table + offset, 2, byte_order);
arm-get-next-pcs.c:  int byte_order = self->byte_order;
arm-get-next-pcs.c:  int byte_order_for_code = self->byte_order_for_code;
arm-get-next-pcs.c:  this_instr = self->ops->read_mem_uint (pc, 4, byte_order_for_code);
arm-get-next-pcs.c:                                                         4, byte_order);
arm-get-next-pcs.c:                                                              4, byte_order);
arm-get-next-pcs.h:  ULONGEST (*read_mem_uint) (CORE_ADDR memaddr, int len, int byte_order);
arm-get-next-pcs.h:  int byte_order;
arm-get-next-pcs.h:  int byte_order_for_code;
arm-get-next-pcs.h:                         int byte_order,
arm-get-next-pcs.h:                         int byte_order_for_code,


> 
>> patch series that handles that by moving bfd_endian to a separate header.
>> I've pushed it to the users/palves/gdbserver-extract-unsigned-integer branch
>> as well.
>>
>> If you agree with this, I'll run the bfd_endian patch by the binutils folks.
> 
> Since the endianness of GDBserver is always identical to the endianness
> of the program, I am not sure sharing extract_unsigned_integer between
> GDB and GDBserver is necessary.

Thanks,
Pedro Alves