[PATCH] arm64: signal: Update sigcontext reservations table

[Dave Martin]

On Tue, Sep 03, 2024 at 07:26:04PM +0100, Mark Brown wrote:
> On Tue, Sep 03, 2024 at 03:18:25PM +0100, Dave Martin wrote:
> > On Fri, Aug 23, 2024 at 12:46:05PM +0100, Will Deacon wrote:
> 
> > > I suppose we could, but I must confess that I find this comment a lot
> > > easier to digest that the fiddly maze of inconsistent macros we have
> > > for the different contexts. Then again, all that really matters, I
> > > suppose, is that we don't accidentally over-allocate the maximum size
> > > of the sigcontext. That ought to be enforceable.
> 
> ...
> 
> > I think my best approach for now would be to try to wire up something
> > that at least helps remind us that we need to review this table when
> > something new is added in the sigframe, unless you can think of a
> > better idea.
> 
> It be good to add a selftest that flags this, that way people might
> notice when adding things and if we miss something it'll probably turn
> up in one of the CIs at some point (possibly after it's too late but at
> least we'd know).  That'd give us some level of integration test with
> whatever libcs and other default software are actually doing, as opposed
> to what we think they'll do.

I suppose we could write a test that sets VL=64, SVL=32 and dirties the
SVE and SME regs before triggering a signal, then checks that
extra_context is not there.  This will only work if SVE and SME are
present and big enough.  If we can run this as a routine CI test on a
model, it might be useful though.

Cheers
---Dave