Re: [Qemu-devel] Shifts, ppc[64], xtensa

[Peter Maydell]

On 19 September 2012 19:01, Richard Henderson <address@hidden> wrote:
> On 09/19/2012 10:51 AM, Aurelien Jarno wrote:
>> That said it is not a valid reason to not keep the value during
>> re-translation, as it means the TB will exit instead of linking to
>> the next one. The consequences are only the performance.
>
> We still have the problem of when is the goto_tb link initialized the *first* 
> time?
> Where we expect the goto_tb to fall through to stuff+exit_tb?
>
> For i386 it's during translation, with no care for re-translation.
>
> For ARM?  I can't see that it is.

I think the answer to this is that the only caller of cpu_gen_code()
is tb_gen_code(), which always then calls tb_link_page()
which calls tb_reset_jump() which calls tb_set_jmp_target().

> For PPC, malc has already verified that it *never* happens.  If he
> puts "trap" insns there instead of "nop" insns, he'll see the trap.

...but on the other hand that ought to work for PPC too, so
presumably my analysis is wrong somewhere.

-- PMM