Re: [PATCH 1/5] xen/spinlock: Fix locking path engaging too soon under PVHVM.

[Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>]

On Mon, Sep 09, 2013 at 11:31:23AM +0100, David Vrabel wrote:
> On 07/09/13 14:46, Konrad Rzeszutek Wilk wrote:
> > The xen_lock_spinning has a check for the kicker interrupts
> > and if it is not initialised it will spin normally (not enter
> > the slowpath).
> > 
> > But for PVHVM case we would initialise the kicker interrupt
> > before the CPU came online. This meant that if the booting
> > CPU used a spinlock and went in the slowpath - it would
> > enter the slowpath and block forever. The forever part b/c
> 
> b/c? Ewww.  Proper English please.
> 
> > during bootup the interrupts are disabled - so the CPU would
> > never get an IPI kick and would stay stuck in the slowpath
> > logic forever.
> 
> This description isn't right -- VCPUs blocked in SCHEDOP_poll can be
> unblocked on the event they're waiting for even if local irq delivery is
> disabled.
> 
> > Why would the booting CPU never get an IPI kick? B/c in both
> > PV and PVHVM we consult the cpu_online_mask to determine whether
> > the IPI should go to its CPU destination. Since the booting
> > CPU has not yet finished and set that flag, it meant that
> > if any spinlocks were taken before the booting CPU had gotten to:
> 
> I can't find where the online mask is being checked in
> xen_send_IPI_one().  Is this really the reason why it didn't work?

More details in fc78d343fa74514f6fd117b5ef4cd27e4ac30236
Author: Chuck Anderson <chuck.anderson@oracle.com>
Date:   Tue Aug 6 15:12:19 2013 -0700

    xen/smp: initialize IPI vectors before marking CPU online

I will add that part in.
> 
> This fix looks fine but both the description and the comment need to be
> fixed/clarified.

U r Right!

> 
> David