Re: [PATCH v2 3/3] Add inject plugin and x86_64 target for the inject plugin

[Pierrick Bouvier]

On 12/6/24 17:02, Rowan Hart wrote:
> > > +++ b/tests/tcg/plugins/inject.c
> >
> > Could we find a better name?
>
> For sure, maybe "hypercalls.c" since that's really what it's mostly about.

Sounds good.

> > > @@ -0,0 +1,206 @@
> > > +/*
> > > + * Copyright (C) 2024, Rowan Hart <rowanbhart@gmail.com>
> > > + *
> > > + * License: GNU GPL, version 2 or later.
> > > + *   See the COPYING file in the top-level directory.
> > > + */
> >
> > We can add a comment here about what the plugin is doing.
>
> Will do!
>
> > One challenge with picking a random value, is how to ensure this pattern has no 
> > other meaning for all architectures? I'm not sure we can find a single pattern 
> > of bytes that works for all arch, even though that would be definitely stylish 
> > :).
> >
> > In more, it seems that we are reinventing the syscall interface, while we 
> > already have it. But as the current instrumentation only works for user-mode, 
> > having a specific hypercall interface might be worth it for plugins, so system 
> > mode could benefit from it too.
> >
> > The work done here could serve later to define a proper interface.
>
>
> I'll see what I can do about this. SIMICS supports many architectures and has a
> "magic instruction" interface[0] (basically hypercalls) and has these
> instructions defined per-architecture in a way that at minimum there are 12
> values available which work on every architecture the simulator supports. QEMU
> supports more architectures than SIMICS but I think we could start there and
> follow a similar approach.
>
> [0]:
> https://intel.github.io/tsffs/simics/simics-user-guide/breakpoints.html#Magic-Breakpoints

Looks like a good model to reuse if we want to implement something similar.

> -Rowan