Re: [PATCH 2/2] i387: split up <asm/i387.h> into exported and internal interfaces

[Avi Kivity <avi@redhat.com>]

On 02/28/2012 09:26 PM, Linus Torvalds wrote:
> On Tue, Feb 28, 2012 at 11:06 AM, Avi Kivity <avi@redhat.com> wrote:
> >
> > No, the scheduler saves the state into task_struct.  I need it saved
> > into the vcpu structure.  We have two fpu states, the user state, and
> > the guest state.  APIs that take a task_struct as a parameter, or
> > reference current implicitly, aren't going to work.
>
> As far as I can tell, you should do the saving into the vcpu structure
> when you actually switch the thing around.
>
> In fact, you can do it these days by just playing around with the
> "tsk->thread.fpu.state" pointer, I guess.

Good idea.  I can't say I like poking into struct fpu's internals, but
we can treat it as an opaque structure and copy it around.

We can also do this in kernel_fpu_begin(), and allow it to be preemptible.

> But it all boils down to the fact that your code is not just ugly,
> it's *buggy*. If you play around with setting TS, you *will* be hit by
> interrupts etc that will start to use the FP code that you "don't
> use".
>
> And there is no excuse for you touching the host TS. The kernel does
> that for you, and does it better. And caches the end result in
> TS_USEDFPU (old) or in some variable that you shouldn't look at but
> can access with the user_has_fpu() helpers.

Again, I can't avoid touching it.  I can try to get the hardware to
always preserve its value, but that comes with a cost.

btw, I think the current code is safe wrt kvm.  If the guest fpu has
been loaded, then we know that that TS_USEDFPU is set, since we will
have saved the user fpu earlier.  Yes it's "accidental" and needs to be
improved, but I don't think it's a data corruptor waiting to happen.

-- 
error compiling committee.c: too many arguments to function