Re: [PATCH -v3 3/6] x86, NMI, Rewrite NMI handler

[Huang Ying <ying.huang@intel.com>]

On Tue, 2010-10-12 at 14:31 +0800, Peter Zijlstra wrote:
> On Tue, 2010-10-12 at 14:14 +0800, Huang Ying wrote:
> > On Tue, 2010-10-12 at 14:04 +0800, Peter Zijlstra wrote:
> > > On Tue, 2010-10-12 at 08:50 +0800, Huang Ying wrote:
> > > > On Tue, 2010-10-12 at 00:13 +0800, Peter Zijlstra wrote:
> > > > > On Sat, 2010-10-09 at 14:49 +0800, Huang Ying wrote:
> > > > > > notify_die(DIE_NMI_IPI);
> > > > > > notify_die(DIE_NMI);
> > > > > > /* process io port 0x61 */
> > > > > > nmi_watchdog_touch();
> > > > > > unknown_nmi(); 
> > > > > 
> > > > > Why keep NMI_IPI? What the heck is it for?
> > > > 
> > > > DIE_NMI_IPI is used for CPU-specific or CPU-local NMIs, such as perf
> > > > NMI. While DIE_NMI is used for non-CPU-specific or global NMIs, such as
> > > > NMI notification from source bridge.
> > > > 
> > > > The order between these two is important. So we use two die value to
> > > > enforce the order.
> > > 
> > > But you can't know about that, there is no reason field to distinguish
> > > between these cases, so you might as well fold it into a single notifier
> > > chain and be done with it.
> > 
> > NMI users know that. Such as perf uses CPU-specific NMI, while APEI GHES
> > uses non-CPU-specific NMI. Different users expect different die values,
> > such as perf expects DIE_NMI_IPI, while APEI GHES expects DIE_NMI, so
> > that perf can be checked before APEI GHES.
> 
> GAAHHH, so why can't they live on a single notifier list? Its got
> priorities to deal with that?

It is possible to use priority to deal with the order. I just think two
die_value make order more explicit.

Best Regards,
Huang Ying