Re: [PATCH 2/2 v5] KVM: PPC: booke: Add watchdog emulation

[Marcelo Tosatti <mtosatti@redhat.com>]

On Wed, Jul 25, 2012 at 06:55:16PM -0500, Scott Wood wrote:
> On 07/25/2012 03:37 PM, Marcelo Tosatti wrote:
> > On Fri, Jul 20, 2012 at 04:29:08PM -0500, Scott Wood wrote:
> >> On 07/20/2012 12:00 AM, Bharat Bhushan wrote:
> >>> This patch adds the watchdog emulation in KVM. The watchdog
> >>> emulation is enabled by KVM_ENABLE_CAP(KVM_CAP_PPC_WDT) ioctl.
> >>> The kernel timer are used for watchdog emulation and emulates
> >>> h/w watchdog state machine. On watchdog timer expiry, it exit to QEMU
> >>> if TCR.WRC is non ZERO. QEMU can reset/shutdown etc depending upon how
> >>> it is configured.
> > 
> > Why can't you do this in QEMU, given the kernel exits to userspace on
> > timer configuration?
> >
> > The QEMU timer handler used to emulate the watchdog then would have to
> > only read the register value from the vCPU.
> 
> Do what specifically in QEMU?  The whole watchdog mechanism?
>
> It's only on final (third) expiry, when a reset is needed, that it exits
> to QEMU.  The first expiry sets a bit in TSR that software can clear to
> clear the watchdog.  If software doesn't clear it, the second expiry
> sets a different bit in TSR that triggers a critical interrupt to the
> guest.  If software doesn't clear either of those bits by the time it
> expires again, it's reset time (if a reset action is enabled).
> 
> The TSR/TCR registers are shared with other timers that are implemented
> in the kernel, and it would be more complicated to implement just the
> watchdog in QEMU.  Plus, we currently have no interface for exiting to
> QEMU for emulation of an SPR.
> 
> Currently QEMU on PPC leaves all timekeeping stuff to the kernel, and
> I'd like to keep it that way.
> 
> -Scott
> 

OK.