Re: [Qemu-devel] [PATCH v3 10/19] pc: register created initial and hotpluged CPUs in one place pc_cpu_plug()

[Eduardo Habkost]

On Thu, Jul 14, 2016 at 11:18:36AM +0200, Igor Mammedov wrote:
> On Wed, 13 Jul 2016 20:37:13 -0300
> Eduardo Habkost <address@hidden> wrote:
> 
> > On Wed, Jul 13, 2016 at 06:59:21PM -0400, Bandan Das wrote:
> > > Eduardo Habkost <address@hidden> writes:
> > >   
> > > > On Wed, Jul 13, 2016 at 06:32:27PM -0400, Bandan Das wrote:  
> > > >> Igor Mammedov <address@hidden> writes:
> > > >>   
> > > >> > consolidate possible_cpus array management in pc_cpu_plug()
> > > >> > for smp_cpus, coldplugged with -device and hotplugged with
> > > >> > device_add.  
> > > >> 
> > > >> So, this takes care of the hotplug case and 09/19 took care of the
> > > >> coldplug case, right ? If yes, we should probably modify this commit
> > > >> message a little.  
> > > >
> > > > This makes pc_cpu_plug() take care of both: now it will handle
> > > > the coldplug case also (in addition to the hotplug case, that it
> > > > already handled). I don't understand what you mean.
> > > >
> > > > Are you talking about the rtc_set_memory() call, only? This seems
> > > > to be the only hotplug-specific code that still remains, in this  
> > > 
> > > Right, I thought the rtc_set_memory() call in 09/19 takes care of the
> > > case when the user boots with "-device" and this takes care of the case
> > > when user invokes device_add, no ?  
> > 
> > Yes for rtc_set_memory(). But the main purpose of this patch is
> > to reuse all the rest of pc_cpu_plug() (i.e. everything except
> > for the rtc_set_memory() call) for the coldplug CPUs too.
> > 
> > Note that I didn't review the patch completely, yet. The
> > replacement for the possible_cpus code looks OK, but I didn't
> > check if it is safe to call
> > HOTPLUG_HANDLER_GET_CLASS(pcms->acpi_dev)->plug() for coldplugged
> > CPUs too.
> for cpus created at  pc_cpus_init() time, that path is unreachable
> since pcms->acpi_dev == NULL,

Sounds fragile, as it requires specific initialization ordering
in the pc_cpus_init() caller. But:

> 
> even if in the future that path becomes reachable
> (when we could specify all CPUs including BSP with -device, it doesn't work 
> for BSP yet)
>  i.e. pcms->acpi_dev != NULL
> then acpi_device will have be initialized see cpu_hotplug_hw_init()
> and following plug handler will be called acpi_cpu_plug_cb() doing what
> it was supposed to do.

For reference for others, acpi_cpu_plug_cb() is:

void acpi_cpu_plug_cb(HotplugHandler *hotplug_dev,
                      CPUHotplugState *cpu_st, DeviceState *dev, Error **errp)
{
    AcpiCpuStatus *cdev;

    cdev = get_cpu_status(cpu_st, dev);
    if (!cdev) {
        return;
    }

    cdev->cpu = CPU(dev);
    if (dev->hotplugged) {
        cdev->is_inserting = true;
        acpi_send_event(DEVICE(hotplug_dev), ACPI_CPU_HOTPLUG_STATUS);
    }
}

get_cpu_status() should return non-NULL for all possible CPUs. So the only code
for coldplug case is cdev->cpu = CPU(dev), which is really something we want to
run.

So, that means it works properly on both cases: if pcms->acpi_dev==NULL and if
pcms->acpi_dev != NULL.

Reviewed-by: Eduardo Habkost <address@hidden>

Related question: why does CPUHotplugState needs to duplicate the list of
(CPUState *cpu, uint64_t arch_id) pairs, if PCMachineState already maintains a
list with exactly the same information?

> 
> > 
> > >   
> > > > patch.
> > > >  
> > > >> 
> > > >> Bandan  
> > > >> > Signed-off-by: Igor Mammedov <address@hidden>
> > > >> > ---
> > > >> >  hw/i386/pc.c | 25 +++++++++----------------
> > > >> >  1 file changed, 9 insertions(+), 16 deletions(-)
> > > >> >
> > > >> > diff --git a/hw/i386/pc.c b/hw/i386/pc.c
> > > >> > index 3206572..0f85b56 100644
> > > >> > --- a/hw/i386/pc.c
> > > >> > +++ b/hw/i386/pc.c
> > > >> > @@ -1142,7 +1142,6 @@ void pc_cpus_init(PCMachineState *pcms)
> > > >> >          if (i < smp_cpus) {
> > > >> >              cpu = pc_new_cpu(typename, 
> > > >> > x86_cpu_apic_id_from_index(i),
> > > >> >                               &error_fatal);
> > > >> > -            pcms->possible_cpus->cpus[i].cpu = CPU(cpu);
> > > >> >              object_unref(OBJECT(cpu));
> > > >> >          }
> > > >> >      }
> > > >> > @@ -1697,25 +1696,19 @@ static void pc_cpu_plug(HotplugHandler 
> > > >> > *hotplug_dev,
> > > >> >      Error *local_err = NULL;
> > > >> >      PCMachineState *pcms = PC_MACHINE(hotplug_dev);
> > > >> >  
> > > >> > -    if (!dev->hotplugged) {
> > > >> > -        goto out;
> > > >> > -    }
> > > >> > -
> > > >> > -    if (!pcms->acpi_dev) {
> > > >> > -        error_setg(&local_err,
> > > >> > -                   "cpu hotplug is not enabled: missing acpi 
> > > >> > device");
> > > >> > -        goto out;
> > > >> > +    if (pcms->acpi_dev) {
> > > >> > +        hhc = HOTPLUG_HANDLER_GET_CLASS(pcms->acpi_dev);
> > > >> > +        hhc->plug(HOTPLUG_HANDLER(pcms->acpi_dev), dev, &local_err);
> > > >> > +        if (local_err) {
> > > >> > +            goto out;
> > > >> > +        }
> > > >> >      }
> > > >> >  
> > > >> > -    hhc = HOTPLUG_HANDLER_GET_CLASS(pcms->acpi_dev);
> > > >> > -    hhc->plug(HOTPLUG_HANDLER(pcms->acpi_dev), dev, &local_err);
> > > >> > -    if (local_err) {
> > > >> > -        goto out;
> > > >> > +    if (dev->hotplugged) {
> > > >> > +        /* increment the number of CPUs */
> > > >> > +        rtc_set_memory(pcms->rtc, 0x5f, rtc_get_memory(pcms->rtc, 
> > > >> > 0x5f) + 1);
> > > >> >      }
> > > >> >  
> > > >> > -    /* increment the number of CPUs */
> > > >> > -    rtc_set_memory(pcms->rtc, 0x5f, rtc_get_memory(pcms->rtc, 0x5f) 
> > > >> > + 1);
> > > >> > -
> > > >> >      found_cpu = pc_find_cpu_slot(pcms, CPU(dev), NULL);
> > > >> >      found_cpu->cpu = CPU(dev);
> > > >> >  out:  
> > 
> 

-- 
Eduardo