Re: [Qemu-devel] [4406] fixed global variable handling with qemu load/ stores - initial global prologue/epilogue implementation

[Fabrice Bellard <fabrice@bellard.org>]

Stuart Brady wrote:
> On Sat, May 10, 2008 at 01:45:54PM +0100, Stuart Brady wrote:
>> On Sat, May 10, 2008 at 10:52:06AM +0000, Fabrice Bellard wrote:
>>
>>> -#ifdef HOST_HPPA
>>> -    func_addr = (tcg_target_long)__canonicalize_funcptr_for_compare((void *)func_addr);
>>> -#endif
>> How would you prefer this to be handled?
> 
> Ouch -- I've just noticed that the right place to handle this was
> tcg_out_call()... which in fact already does it!
> 
> I obviously meant to remove the __canonicalize_funcptr_for_compare from
> tcg.c before I even submitted the code.  Thanks for fixing it.
> 
> Another change I made was adding the 'func_addr' variable, and using it
> in place of ts->val.  I can submit a patch to revert that change if you
> would like.

An important note: the only "officially" supported hosts for TCG are x86
and x86_64. I prefer not to support other hosts at this point because
TCG is evolving. However I accept that code is commited for them
provided it does not make heavy change in the TCG core.

In order to test a TCG target support, it is important to use a fully
converted target such as SPARC. Testing it on non fully converted
targets such as x86 only tests a small fraction of the code.

Regards,

Fabrice.