[Lguest] My vector reserve patches etc. so Plan 9 boots.

[Rusty Russell]

On Fri, 2007-08-31 at 14:16 -0700, ron minnich wrote:
> These are against -rc4.
> 
> 0002 adds the syscall_vec struct member to lguest_data.

This is fine, but since we have no ABI and other things are going to
change anyway for 2.6.25, we don't need the "if it's zero" check.

> Both plan 9 and linux boot fine. There was a little confusion on
> everyone's part from the earlier discussion and the original patches:
> what we needed to do was reserve a vector, not an irq. The code in
> patch 0003 supports reservations of vectors so that the io_apic won't
> try to use it. Although a boot-time command parameter is supported, it
> is not needed for lguest to work. This code also removes the hard-code
> compare to SYSCALL_VECTOR, and replaces it with a hard-code to reserve
> the SYSCALL_VECTOR :-) vector. There are some unresolved questions
> here, which I note in the comment for patch 0003.

Unfortunately, this doesn't quite work....

> patch 0004 is a slight mod of Rusty's earlier work. The change is that
> we're not doing request_irq, since this is not an IRQ we want, but a
> vector; and hence we don't need the null handler.

If vector is already in use, this code will reserve it anyway.  I think
what we actually want is a "used_vectors" bitmap and associated lock.
It's clear and simple, and avoids the abstraction layer...

Rusty.