Re: [patch] Get rid of current_gdbarch in xtensa

[Maxim Grigoriev <maxim@tensilica.com>]

Hello Ulrich and Markus,

Thanks for taking care of Xtensa.

I think xtensa_read_register() should be eliminated.
If you don't mind I can take the original Markus's patch
from here and finish and test it.

Thanks,
-- Maxim


Ulrich Weigand wrote:
> Markus Deuling wrote:
>
>   
>> @@ -2034,9 +2034,9 @@ call0_track_op (xtensa_c0reg_t dst[], xt
>>        /* 2 operands: dst, literal offset.  */
>>        gdb_assert (nods == 2);
>>        /* litbase = xtensa_get_litbase (pc); can be also used.  */
>> -      litbase = (gdbarch_tdep (current_gdbarch)->litbase_regnum == -1)
>> +      litbase = (gdbarch_tdep (gdbarch)->litbase_regnum == -1)
>>  	? 0 : xtensa_read_register
>> -		(gdbarch_tdep (current_gdbarch)->litbase_regnum);
>> +		(gdbarch_tdep (gdbarch)->litbase_regnum);
>>     
>
> There's really a different problem here: xtensa_read_register simply
> uses get_current_regcache () to retrieve that "litbase" register.
>
> This seems quite broken, in particular when call0_track_op is called
> out of the xtensa_skip_prologue code path, where the current regcache
> may not really be related at all to the function that is being
> queried ...
>
> In those cases where we have a proper frame to read from, the question
> of how to retrieve the litbase register number should resolve itself
> automatically.
>
> Maxim, do you have any suggestions how to handle this?  Is there a
> way to retrieve the "litbase" in the xtensa_skip_prologue case?
>
> Bye,
> Ulrich
>
>