Re: [PATCH 3/3] s390x/pci: drive ISM reset from subsystem reset

[Matthew Rosato <mjrosato@linux.ibm.com>]

On 1/17/24 10:19 AM, Matthew Rosato wrote:
> On 1/17/24 6:01 AM, Cédric Le Goater wrote:
>> Adding Alex,
>>
>> On 1/16/24 23:31, Matthew Rosato wrote:
>>> ISM devices are sensitive to manipulation of the IOMMU, so the ISM device
>>> needs to be reset before the vfio-pci device is reset (triggering a full
>>> UNMAP).  In order to ensure this occurs, trigger ISM device resets from
>>> subsystem_reset before triggering the PCI bus reset (which will also
>>> trigger vfio-pci reset).  This only needs to be done for ISM devices
>>> which were enabled for use by the guest.
>>> Further, ensure that AIF is disabled as part of the reset event.
>>>
>>> Fixes: ef1535901a ("s390x: do a subsystem reset before the unprotect on reboot")
>>> Fixes: 03451953c7 ("s390x/pci: reset ISM passthrough devices on shutdown and system reset")
>>> Reported-by: Cédric Le Goater <clg@redhat.com>
>>> Signed-off-by: Matthew Rosato <mjrosato@linux.ibm.com>
>>> ---
>>>   hw/s390x/s390-pci-bus.c         | 26 +++++++++++++++++---------
>>>   hw/s390x/s390-virtio-ccw.c      |  2 ++
>>>   include/hw/s390x/s390-pci-bus.h |  1 +
>>>   3 files changed, 20 insertions(+), 9 deletions(-)
>>>
>>> diff --git a/hw/s390x/s390-pci-bus.c b/hw/s390x/s390-pci-bus.c
>>> index 347580ebac..3e57d5faca 100644
>>> --- a/hw/s390x/s390-pci-bus.c
>>> +++ b/hw/s390x/s390-pci-bus.c
>>> @@ -151,20 +151,12 @@ static void s390_pci_shutdown_notifier(Notifier *n, void *opaque)
>>>       pci_device_reset(pbdev->pdev);
>>>   }
>>>   -static void s390_pci_reset_cb(void *opaque)
>>> -{
>>> -    S390PCIBusDevice *pbdev = opaque;
>>> -
>>> -    pci_device_reset(pbdev->pdev);
>>> -}
>>> -
>>>   static void s390_pci_perform_unplug(S390PCIBusDevice *pbdev)
>>>   {
>>>       HotplugHandler *hotplug_ctrl;
>>>         if (pbdev->pft == ZPCI_PFT_ISM) {
>>>           notifier_remove(&pbdev->shutdown_notifier);
>>> -        qemu_unregister_reset(s390_pci_reset_cb, pbdev);
>>>       }
>>>         /* Unplug the PCI device */
>>> @@ -1132,7 +1124,6 @@ static void s390_pcihost_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
>>>               if (pbdev->pft == ZPCI_PFT_ISM) {
>>>                   pbdev->shutdown_notifier.notify = s390_pci_shutdown_notifier;
>>>                   qemu_register_shutdown_notifier(&pbdev->shutdown_notifier);
>>> -                qemu_register_reset(s390_pci_reset_cb, pbdev);
>>>               }
>>>           } else {
>>>               pbdev->fh |= FH_SHM_EMUL;
>>> @@ -1279,6 +1270,23 @@ static void s390_pci_enumerate_bridge(PCIBus *bus, PCIDevice *pdev,
>>>       pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, s->bus_no, 1);
>>>   }
>>>   +void s390_pci_ism_reset(void)
>>> +{
>>> +    S390pciState *s = s390_get_phb();
>>> +
>>> +    S390PCIBusDevice *pbdev, *next;
>>> +
>>> +    /* Trigger reset event for each passthrough ISM device currently in-use */
>>> +    QTAILQ_FOREACH_SAFE(pbdev, &s->zpci_devs, link, next) {
>>> +        if (pbdev->interp && pbdev->pft == ZPCI_PFT_ISM &&
>>> +            pbdev->fh & FH_MASK_ENABLE) {
>>> +            s390_pci_kvm_aif_disable(pbdev);
>>> +
>>> +            pci_device_reset(pbdev->pdev);
>>> +        }
>>> +    }
>>> +}
>>
>>
>> Could we instead define a VFIOPCIDevice::resetfn handler for these
>> ISM devices (1014:04ed) ? This would be cleaner if possible.
>>
>> If so, as a prerequisite, we would need to introduce in a little VFIO
>> helper to define custom reset handlers.
>>
>> Thanks,
>>
>> C.
>>
> 
> Oh interesting, I had not noticed that.  This may well work -- resetfn is currently setup via vfio_setup_resetfn_quirk but it would probably be easier to have a helper that takes the vdev and a function pointer so that we can provide a platform-specific reset handler (rather than having hw/vfio/pci-quirks.c worry about CONFIG_S390 etc).  I'll have to play around with this.
>  
> 

Hmm, it was a good idea but I don't think this will work.  I tried to hack something together today but I'm definitely seeing paths where the vfio_listener_region_del happens before the call to vfio_pci_reset (which would ultimately trigger the new custom resetfn).

Perhaps we should stick with the call from subsystem_reset -- it will ensure that the ISM cleanup happens after guest CPUs are stopped but before vfio does its cleanup.