Re: RFA: New port: ia64-hp-openvms - the stub

[Pedro Alves <palves@redhat.com>]

On 02/24/2012 11:12 AM, Tristan Gingold wrote:
> 
> On Feb 24, 2012, at 11:46 AM, Pedro Alves wrote:
> 
>>>>>>> +#if 1
>>>>>>> +      /* What a mess.  Gdb and linux expects bsp to point after the current
>>>>>>> +         register frame.  Adjust.  */
>>>>>>
>>>>>> What does this mean?  Are we committing to a hack that will make our
>>>>>> lives hard when we want to fix it?
>>>>>
>>>>> I have updated the comment.  Currently, ia64-tdep.c follows the Linux convention, and we don't want to change that.  So the BSP has to be adjusted.
>>>>
>>>> Why don't we want to change it?
>>>
>>> I think it's too late.  That would be a change in gdbserver and other stubs, so it would break compatibility.  
>>
>> We have ways to make it happen without breaking other stubs.  The question is what is the
>> right thing to do.
> 
> I am not sure that modifying the convention of BSP is the right thing to do.
> 
>> Is this really Linux specific?  We have ia64 hpux support as well.
> 
> AFAIU, hpux also follows the Linux convention.
> 
>> I take it libunwind must follow the same convention as well, and I hope a libunwind
>> built on a Linux host doesn't behave different from a libunwind built on a
>> different host.
> 
> That's correct.  libunwind convention matches the VMS one, and therefore ia64-tdep.c has to translate the convention.
> See for example the case of UNW_IA64_AR_BSP in ia64-tdep.c:ia64_access_reg.

Damn, what a mess...  If we want to fix this, getting rid of the unnecessary conversions
all around, we can do it with a xml target description with a new standard feature
(org.gnu.gdb.ia64.something) so that gdb knows the registers are sent with the
libunwind/vms convention.  Without that, yes, we should feed the raw register set
as gdb has been expecting it.

> 
>> The adjustment in the VMS stubs isn't that complex.
>>
>> Hopefully it'll still be simple with infcalls in the game.  :-)  afaics, you're
>> not undoing the adjustment on writes.
> 
> Correct. Writing this register is used only to implement inferior call, which is not yet implemented.  The user will never modify it
> directly, as this value cannot be changed without modifying other registers.  HPUX even don't allow to modify it (for not so bad reasons).
> Welcome in ia64 world!
> 
> Tristan.
> 


-- 
Pedro Alves