Re: [PATCH 4/7] range stepping: gdb

[Yao Qi <yao at codesourcery dot com>]

On 04/11/2013 03:36 PM, Yao Qi wrote:
> It is incorrect to copy address to buffer via phex_nz here and
> somewhere else in remote.c.  We should use remote_address_masked first
> to mask the CORE_ADDR and then call hexnumstr to copy address to
> buffer.  Here is an updated one.

After a further thought, I decide to withdraw the updated patch 4/7, and 
keep using the original V2 patch 4/7.  The comments of variable 
'remote_address_size' say:

/* This variable sets the number of bits in an address that are to be
   sent in a memory ("M" or "m") packet.  Normally, after stripping
   leading zeros, the entire address would be sent.  This variable
   restricts the address to REMOTE_ADDRESS_SIZE bits.  HISTORY: The
   initial implementation of remote.c restricted the address sent in
   memory packets to ``host::sizeof long'' bytes - (typically 32
   bits).  Consequently, for 64 bit targets, the upper 32 bits of an
   address was never sent.  Since fixing this bug may cause a break in
   some remote targets this variable is principly provided to
   facilitate backward compatibility.  */

static unsigned int remote_address_size;

variable 'remote_address_size' has only effect on memory packet and 
breakpoint packet (it is not documented, but actually used).  Address in 
these packets should be masked by remote_address_masked, and address in 
new packets should not be, because the stub supports these new packets 
should be able to parse address correctly.

--
Yao (éå)