Re: [PATCH v13 3/3] trace,x86: Add irq vector tracepoints

[Steven Rostedt]

On Tue, 2013-06-04 at 11:15 -0700, H. Peter Anvin wrote:
> On 06/03/2013 04:53 PM, Steven Rostedt wrote:
> > 
> > This way we wont be opening up any easy root holes where if a process
> > finds a way to modify some arbitrary kernel memory, we can prevent it
> > from modifying the current_idt_descr_ptr and have a nice way to exploit
> > the IDT. Sure, one can argue that if they can modify arbitrary kernel
> > memory, we may already be lost, but lets not make it easier for them
> > than need be.
> > 
> 
> I don't like current_idt_descr_ptr if we can avoid it.  It is a direct
> proxy for reading and writing the original IDT, in other words, it
> really hasn't really addressed the issue.
> 
> What I'm thinking we really should have is a function that returns the
> IDT that we currently should be using, based on the current state.  If
> that state is, say, tracing on/off and NMI on/off, then that can be
> indicated by bits in a state vector.

The NMI on/off may be a bit trickier, as it is also a debug state as
well. When we go into a nested debug or NMI state we use the same IDT.

> 
> The point is that the IDT address itself should not be mutable state if
> it can be at all avoided.

Hmm, maybe we can do it. Have two counters, a debug_idt_ctr and a
trace_idt_ctr, then have a function that basically does this:

	if (this_cpu_read(debug_idt_ctr))
		load_idt(&nmi_idt_descr); /* probably should rename to debug_idt_descr) */
	else if (trace_idt_ctr)
		load_idt(&trace_idt_descr);
	else
		load_idt(&idt_descr);

Then all modifications of the idt would call this function.

-- Steve



--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/