Re: [Qemu-devel] [PATCH 13/13] tcg: rework TCG ops flags

[Aurelien Jarno]

On Thu, Sep 27, 2012 at 03:00:15PM -0700, Richard Henderson wrote:
> On 09/27/2012 01:37 PM, Aurelien Jarno wrote:
> >> > I don't agree with SIDE_EFFECTS implying exceptions.  How can "br" cause 
> >> > an
> >> > exception?  Or for that matter "st_i32", recalling that we're not storing
> >> > to guest memory.
> > That's exactly why SIDE_EFFECTS has been removed from this op in the
> > previous patch.
> 
> Well, you removed it from br, but not st.

Oh correct. st doesn't need one as it has zero output arguments, so it
won't be removed.

> > I think it implies exception, because I don't see why an op shouldn't be
> > removed otherwise (remember ops without outputs are never removed).
> 
> In which case, because the non-qemu store insns cannot raise exceptions,
> there ought to be exactly zero instances of TCG_OPF_SIDE_EFFECTS remaining.
> At which point we simply ought to remove it.
> 

I don't understand, the qemu/load store still need to keep this
TCG_OPF_SIDE_EFFECTS. Even a load to a dead output argument might 
trigger a TLB miss exception, and thus should be fixed.

-- 
Aurelien Jarno                          GPG: 1024D/F1BCDB73
address@hidden                 http://www.aurel32.net