Re: [Qemu-devel] [PATCH 2/3] Add pci_bus_reset() function.

[Gleb Natapov]

On Tue, Jun 09, 2009 at 06:07:00PM +0300, Blue Swirl wrote:
> On 6/9/09, Gleb Natapov <address@hidden> wrote:
> > On Tue, Jun 09, 2009 at 12:42:16AM +0300, Dor Laor wrote:
> >  > Gleb Natapov wrote:
> >  >> On Mon, Jun 08, 2009 at 05:35:11PM +0300, Avi Kivity wrote:
> >  >>
> >  >>> Gleb Natapov wrote:
> >  >>>
> >  >>>> To reset internal irq handling data structures.
> >  >>>>
> >  >>>> Signed-off-by: Gleb Natapov <address@hidden>
> >  >>>> Signed-off-by: Yaniv Kamay <address@hidden>
> >  >>>> ---
> >  >>>>  hw/pci.c |   16 ++++++++++++++++
> >  >>>>  1 files changed, 16 insertions(+), 0 deletions(-)
> >  >>>>
> >  >>>> diff --git a/hw/pci.c b/hw/pci.c
> >  >>>> index 02b335f..89fefdf 100644
> >  >>>> --- a/hw/pci.c
> >  >>>> +++ b/hw/pci.c
> >  >>>> @@ -88,6 +88,21 @@ static int  pcibus_load(QEMUFile *f, void *opaque, 
> > int version_id)
> >  >>>>      return 0;
> >  >>>>  }
> >  >>>>  +static void pci_bus_reset(void *opaque)
> >  >>>> +{
> >  >>>> +    PCIBus *bus = (PCIBus *)opaque;
> >  >>>> +    int i;
> >  >>>> +
> >  >>>> +    for (i = 0; i < bus->nirq; i++) {
> >  >>>> +        bus->irq_count[i] = 0;
> >  >>>> +    }
> >  >>>> +    for (i = 0; i < PCI_DEVICES_MAX; i++) {
> >  >>>> +        if (bus->devices[i])
> >  >>>> +            memset(bus->devices[i]->irq_state, 0,
> >  >>>> +                   sizeof(bus->devices[i]->irq_state));
> >  >>>> +    }
> >  >>>> +}
> >  >>>> +
> >  >>>>
> >  >>> Shouldn't each device's reset function bring its line low, thus
> >  >>> zeroing  the irq_state naturally?
> >  >>>
> >  >>> If not, we have a bug somewhere.  Note we have exactly the same issue
> >  >>>  with save/restore.
> >  >>>
> >  >>>
> >  >> They should, but I haven't found one that does.
> >  >>
> >  > virtio does and so do many others. Sad thing is that all should do it
> >  > since the line is shared.
> >  > e1000 and rtl8139 do not register a reset handler.
> >  >
> >
> > So somebody (not me) should go and fix all others.
> 
> Here's a 5 min patch to add reset to e1000 and rtl8139. Not too difficult?
> 
Very handy. I just discovered that e1000 continue to write to memory
after OS reboot with some OSes. Will check if this fixes it.


> >  > Maybe we should make it a required callback for pci_qdev_register? or 
> > better
> >  > have every (pci?) device register several callback together.
> >
> > May be.
> 
> If we go to this, I'd also make savevm/loadvm mandatory.

We need this for device hot unplug. We have to be able to reset only one
device at a time.

--
                        Gleb.