Re: [Xen-devel] [PATCH 3 of 4 RFC] x86/nmi: Prevent reentrant execution of the C nmi handler

[Tim Deegan <tim@xxxxxxx>]

At 18:16 +0000 on 04 Dec (1354644970), Andrew Cooper wrote:
> +/* This function is NOT SAFE to call from C code in general.
> + * Use with extreme care! */
> +void do_nmi(struct cpu_user_regs *regs)
> +{
> +    bool_t * in_progress = &this_cpu(nmi_in_progress);
> +
> +    if ( is_panic_in_progress() )
> +    {
> +        /* A panic is already in progress.  It may have reenabled NMIs,
> +         * or we are simply unluckly to receive one right now.  Either
> +         * way, bail early, as Xen is about to die.
> +         *
> +         * TODO: Ideally we should exit without executing an iret, to
> +         * leave NMIs disabled, but that option is not currently
> +         * available to us.
> +         */
> +        return;
> +    }
> +
> +    if ( test_and_set_bool(*in_progress) )
> +    {
> +        /* Crash in an obvious mannor, as opposed to falling into
> +         * infinite loop because our exception frame corrupted the
> +         * exception frame of the previous NMI.
> +         *
> +         * TODO: This check does not cover all possible cases of corrupt
> +         * exception frames, but it is substantially better than
> +         * nothing.
> +         */
> +        console_force_unlock();
> +        show_execution_state(regs);
> +        panic("Reentrant NMI detected\n");
> +    }
> +
> +    _do_nmi(regs);

I think you need a barrier() above and below _do_nmi() in case the
compiler inlines _do_nmi() and then decides it can shuffle the writes to
in_progress around.

Otherwise, this looks fine, and you can add my ack if you like.

> +    *in_progress = 0;
> +}
> +
>  void set_nmi_callback(nmi_callback_t callback)
>  {
>      nmi_callback = callback;
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@xxxxxxxxxxxxx
> http://lists.xen.org/xen-devel

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel