Re: [Qemu-devel] [PATCH] tcg, tci: Add TCG and interpreter for bytecode (virtual machine)

[malc]

yOn Mon, 12 Oct 2009, Stuart Brady wrote:

> On Sun, Oct 11, 2009 at 06:33:57PM +0200, Stefan Weil wrote:
> > Stuart Brady schrieb:
> > > exec-all.h defines 'USE_DIRECT_JUMP' based on the host architecture,
> > > which is not relevant under TCI. Also, a tb_set_jmp_target1()
> > > implementation is needed for TCI.
> 
> > Done. Tested with mips / mipsel host.
> 
> I think the #ifdef CONFIG_TCG_INTERPRETER sections belong before
> the arch-specific sections. :-)
> 
> > Alignment is still to be done (my test platforms don't need it).
> 
> I will do this at some point if nobody beats me to it. :-)
> 
> BTW, I'm wondering about certain places that require host-specific
> code... for example:
> 
>  * qemu-lock.h, which is used by cpu_exec() for the tb lock and by
>    cpu_unlink_tb() for non-NPTL hosts.
> 
>  * The generic cpu_get_real_ticks() doesn't seem great... I suppose
>    the MIPS version using clock() might be more appropriate?
> 
> Apart from this, I suppose user mode emulation simply has to depend
> on host-specific code for signal handling... :-(
> 
> I gather INDEX_op_call handling could be improved?  Presumably that's
> really just a matter of casting to the right type of function pointer
> when making each call?

No it's not, given that there are ABIs with argument alignment and
the fact that information about the types doesn't exist at the point
of op_call things just aren't that simple.

-- 
mailto:address@hidden