Re: [Qemu-devel] [PATCH 3/3] hw/pcie: better hotplug/hotunplug support

[Michael S. Tsirkin]

On Sun, Jun 22, 2014 at 01:54:06PM +0300, Marcel Apfelbaum wrote:
> On Thu, 2014-06-19 at 17:43 +0300, Michael S. Tsirkin wrote:
> > On Thu, Jun 19, 2014 at 04:52:21PM +0300, Marcel Apfelbaum wrote:
> > > Hotplug triggers both 'present detect change' and
> > > 'attention button pressed'.
> > > 
> > > Hotunplug starts by triggering 'attention button pressed',
> > > then waits for the OS to power off the device and only
> > > then detaches it.
> > > 
> > 
> > Pls not here that current code is broken: it does surprise removal which
> > crashes guests.
> I'll add a note, sure.
> 
> > 
> > > Signed-off-by: Marcel Apfelbaum <address@hidden>
> > > ---
> > >  hw/pci/pcie.c | 20 +++++++++++++++-----
> > >  1 file changed, 15 insertions(+), 5 deletions(-)
> > > 
> > > diff --git a/hw/pci/pcie.c b/hw/pci/pcie.c
> > > index f8bf515..9cfd93d 100644
> > > --- a/hw/pci/pcie.c
> > > +++ b/hw/pci/pcie.c
> > > @@ -258,7 +258,8 @@ void pcie_cap_slot_hotplug_cb(HotplugHandler 
> > > *hotplug_dev, DeviceState *dev,
> > >  
> > >      pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
> > >                                 PCI_EXP_SLTSTA_PDS);
> > > -    pcie_cap_slot_event(PCI_DEVICE(hotplug_dev), PCI_EXP_HP_EV_PDC);
> > > +    pcie_cap_slot_event(PCI_DEVICE(hotplug_dev),
> > > +                        PCI_EXP_HP_EV_PDC | PCI_EXP_HP_EV_ABP);
> > >  }
> > >  
> > >  void pcie_cap_slot_hot_unplug_cb(HotplugHandler *hotplug_dev, 
> > > DeviceState *dev,
> > > @@ -268,10 +269,7 @@ void pcie_cap_slot_hot_unplug_cb(HotplugHandler 
> > > *hotplug_dev, DeviceState *dev,
> > >  
> > >      pcie_cap_slot_hotplug_common(PCI_DEVICE(hotplug_dev), dev, &exp_cap, 
> > > errp);
> > >  
> > > -    object_unparent(OBJECT(dev));
> > > -    pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTSTA,
> > > -                                 PCI_EXP_SLTSTA_PDS);
> > > -    pcie_cap_slot_event(PCI_DEVICE(hotplug_dev), PCI_EXP_HP_EV_PDC);
> > > +    pcie_cap_slot_push_attention_button(PCI_DEVICE(hotplug_dev));
> > >  }
> > >  
> > >  /* pci express slot for pci express root/downstream port
> > > @@ -389,6 +387,18 @@ void pcie_cap_slot_write_config(PCIDevice *dev,
> > >                          sltsta);
> > >      }
> > >  
> > 
> > pls add code comments explaining the logic here.
> I thought is clear :(
> Basically: If the device is present, power indicator off and power
> controller off, it is safe to detach the device.

Sure, put this in the comment.

> > 
> > > +    if ((sltsta & PCI_EXP_SLTSTA_PDS) && (val & PCI_EXP_SLTCTL_PCC) &&
> > > +        ((val & PCI_EXP_SLTCTL_PIC_OFF) == PCI_EXP_SLTCTL_PIC_OFF)) {
> > > +        PCIDevice *slot_dev = 
> > > pci_bridge_get_sec_bus(PCI_BRIDGE(dev))->devices[0];
> > > +        if (slot_dev) {
> > 
> > Here you want to remove all devices behind the bridge?
> Yes, but since it is PCIe we only have one device,
> but we may have a multi-function device...

Exactly. Up to 256 functions.

> > You need to do this for all functions, not just function 0.
> So bus devices are actually functions... OK, I'll run a loop here.
> 
> > 
> > > +            object_unparent(OBJECT(slot_dev));
> > > +            pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTSTA,
> > > +                                         PCI_EXP_SLTSTA_PDS);
> > > +            pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
> > > +                                       PCI_EXP_SLTSTA_PDC);
> > 
> > These bits need to be cleared in any case?
> No, "PDS on" means the devices is present, so we clear it only after the
> OS  powers it off.

So why not clear PDS unconditionally?

> > 
> > > +        }
> > > +    }
> > > +
> > >      hotplug_event_notify(dev);
> > >  
> > >      /* 
> > > -- 
> > > 1.8.3.1
> 
>