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

[malc]

On Mon, 26 Oct 2009, Stefan Weil wrote:

> malc schrieb:
> >
> >
> > On Sun, 11 Oct 2009, Stefan Weil wrote:
> >> Stuart Brady schrieb:
> >>> On Mon, Sep 28, 2009 at 06:50:21PM +0200, Stefan Weil wrote:
> >>>> Please send patches / enhancements based on my latest
> >>>> version from git://repo.or.cz/qemu/ar7.git.
> >>> Just bug reports for now, unfortunately...
> >
> > [..snip..]
> >
> >> git://repo.or.cz/qemu/ar7.git contains the latest changes.
> >>
> >> If the maintainers want to integrate tci in the official qemu,
> >> I can prepare the patches needed to add bytecode generator
> >> and interpreter and tcg host support for any host.
> >
> > I've looked at the code, and the problem with helpers is essentially
> > unsolved, if only things were as simple as casting things to
> > `helper_function' and expecting it will work, not that it's _that_
> > much more complex, but still..
> 
> Hello Malc,
> 
> I read your comments on the possible ABI problems.
> 
> Nevertheless, as far as I could see all existing TCG hosts simply
> use fixed registers when calling helper functions.

The problem, as i explained earlier, is that TCI doesn't take types
of the parameters into considerations and this will break SVR4 PPC ABI
for instance, since long long arguments are "aligned". Please read
this for details:
http://marc.info/?l=qemu-devel&m=125535217403861&w=2

> 
> So the same approach should work for TCI, too.
> 
> At least it is possible to run complete operating systems
> using this approach.
> 
> What do you think would be needed to get a first stage of
> TCI integrated in QEMU master?
> 

Fixing the aforementioned problem first, then convincing someone
to commit it. I'd guess that having qemu/tci run on something we
do not have TCG support for would imrpove the odds.

-- 
mailto:address@hidden