| Message ID | 20210930082032.1237812-1-lvivier@redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | failover: fix unplug pending detection | expand |
On Thu, 30 Sep 2021, Laurent Vivier wrote: > Failover needs to detect the end of the PCI unplug to start migration > after the VFIO card has been unplugged. > > To do that, a flag is set in pcie_cap_slot_unplug_request_cb() and reset in > pcie_unplug_device(). > > But since > 17858a169508 ("hw/acpi/ich9: Set ACPI PCI hot-plug as default on Q35") > we have switched to ACPI unplug and these functions are not called anymore > and the flag not set. So failover migration is not able to detect if card > is really unplugged and acts as it's done as soon as it's started. So it > doesn't wait the end of the unplug to start the migration. We don't see any > problem when we test that because ACPI unplug is faster than PCIe native > hotplug and when the migration really starts the unplug operation is > already done. > > See c000a9bd06ea ("pci: mark device having guest unplug request pending") > a99c4da9fc2a ("pci: mark devices partially unplugged") Ok so I have a basic question about partially_hotplugged flag in the device struct (there were no comments added in a99c4da9fc2a39847 explaining it). It seems we return early from pcie_unplug_device() when this flag is set from failover_unplug_primary() in virtio-net. What is the purpose of this flag? It seems we are not doing a full unplug of the primary device? > > Signed-off-by: Laurent Vivier <lvivier@redhat.com> > --- > hw/acpi/pcihp.c | 6 ++++++ > 1 file changed, 6 insertions(+) > > diff --git a/hw/acpi/pcihp.c b/hw/acpi/pcihp.c > index f610a25d2ef9..a2d27a3c4763 100644 > --- a/hw/acpi/pcihp.c > +++ b/hw/acpi/pcihp.c > @@ -366,6 +366,11 @@ void acpi_pcihp_device_unplug_cb(HotplugHandler *hotplug_dev, AcpiPciHpState *s, > trace_acpi_pci_unplug(PCI_SLOT(pdev->devfn), > acpi_pcihp_get_bsel(pci_get_bus(pdev))); > > + if (pdev->partially_hotplugged) { > + pdev->qdev.pending_deleted_event = false; > + return; > + } > + > /* > * clean up acpi-index so it could reused by another device > */ > @@ -396,6 +401,7 @@ void acpi_pcihp_device_unplug_request_cb(HotplugHandler *hotplug_dev, > return; > } > > + pdev->qdev.pending_deleted_event = true; > s->acpi_pcihp_pci_status[bsel].down |= (1U << slot); > acpi_send_event(DEVICE(hotplug_dev), ACPI_PCI_HOTPLUG_STATUS); > } > -- > 2.31.1 > >
On 30/09/2021 11:24, Ani Sinha wrote: > > > On Thu, 30 Sep 2021, Laurent Vivier wrote: > >> Failover needs to detect the end of the PCI unplug to start migration >> after the VFIO card has been unplugged. >> >> To do that, a flag is set in pcie_cap_slot_unplug_request_cb() and reset in >> pcie_unplug_device(). >> >> But since >> 17858a169508 ("hw/acpi/ich9: Set ACPI PCI hot-plug as default on Q35") >> we have switched to ACPI unplug and these functions are not called anymore >> and the flag not set. So failover migration is not able to detect if card >> is really unplugged and acts as it's done as soon as it's started. So it >> doesn't wait the end of the unplug to start the migration. We don't see any >> problem when we test that because ACPI unplug is faster than PCIe native >> hotplug and when the migration really starts the unplug operation is >> already done. >> >> See c000a9bd06ea ("pci: mark device having guest unplug request pending") >> a99c4da9fc2a ("pci: mark devices partially unplugged") > > Ok so I have a basic question about partially_hotplugged flag in the > device struct (there were no comments added in a99c4da9fc2a39847 > explaining it). It seems we return early from pcie_unplug_device() when > this flag is set from failover_unplug_primary() in virtio-net. What is the > purpose of this flag? It seems we are not doing a full unplug of the > primary device? Yes, to be able to plug it back in case of migration failure we must keep all the data structures. But reading the code again it seems this part should be in acpi_pcihp_eject_slot() rather than in acpi_pcihp_device_unplug_cb() to prevent the hotplug_handler_unplug()/object_unparent() rather than the qdev_unrealize() (to be like in pcie.c). Thanks, Laurent
On Thu, 30 Sep 2021, Laurent Vivier wrote: > On 30/09/2021 11:24, Ani Sinha wrote: > > > > > > On Thu, 30 Sep 2021, Laurent Vivier wrote: > > > > > Failover needs to detect the end of the PCI unplug to start migration > > > after the VFIO card has been unplugged. > > > > > > To do that, a flag is set in pcie_cap_slot_unplug_request_cb() and reset > > > in > > > pcie_unplug_device(). > > > > > > But since > > > 17858a169508 ("hw/acpi/ich9: Set ACPI PCI hot-plug as default on > > > Q35") > > > we have switched to ACPI unplug and these functions are not called anymore > > > and the flag not set. So failover migration is not able to detect if card > > > is really unplugged and acts as it's done as soon as it's started. So it > > > doesn't wait the end of the unplug to start the migration. We don't see > > > any > > > problem when we test that because ACPI unplug is faster than PCIe native > > > hotplug and when the migration really starts the unplug operation is > > > already done. > > > > > > See c000a9bd06ea ("pci: mark device having guest unplug request pending") > > > a99c4da9fc2a ("pci: mark devices partially unplugged") > > > > Ok so I have a basic question about partially_hotplugged flag in the > > device struct (there were no comments added in a99c4da9fc2a39847 > > explaining it). It seems we return early from pcie_unplug_device() when > > this flag is set from failover_unplug_primary() in virtio-net. What is the > > purpose of this flag? It seems we are not doing a full unplug of the > > primary device? > > Yes, to be able to plug it back in case of migration failure we must keep all > the data structures. Ok so two things here: (a) could you please add a comment to PCIDevice struct in pci.h to clarify what the flag actually means, why it is there and what it is supposed to do. (b) the naming of the variable could be something like do_partial_unplug or some such. This could be a separate patch. > > But reading the code again it seems this part should be in > acpi_pcihp_eject_slot() rather than in acpi_pcihp_device_unplug_cb() to > prevent the hotplug_handler_unplug()/object_unparent() rather than the > qdev_unrealize() (to be like in pcie.c). Correct. You need to place the check earlier so as to be equivalent to what the native hotplug code does.
On 01/10/2021 07:19, Ani Sinha wrote: > > > On Thu, 30 Sep 2021, Laurent Vivier wrote: > >> On 30/09/2021 11:24, Ani Sinha wrote: >>> >>> >>> On Thu, 30 Sep 2021, Laurent Vivier wrote: >>> >>>> Failover needs to detect the end of the PCI unplug to start migration >>>> after the VFIO card has been unplugged. >>>> >>>> To do that, a flag is set in pcie_cap_slot_unplug_request_cb() and reset >>>> in >>>> pcie_unplug_device(). >>>> >>>> But since >>>> 17858a169508 ("hw/acpi/ich9: Set ACPI PCI hot-plug as default on >>>> Q35") >>>> we have switched to ACPI unplug and these functions are not called anymore >>>> and the flag not set. So failover migration is not able to detect if card >>>> is really unplugged and acts as it's done as soon as it's started. So it >>>> doesn't wait the end of the unplug to start the migration. We don't see >>>> any >>>> problem when we test that because ACPI unplug is faster than PCIe native >>>> hotplug and when the migration really starts the unplug operation is >>>> already done. >>>> >>>> See c000a9bd06ea ("pci: mark device having guest unplug request pending") >>>> a99c4da9fc2a ("pci: mark devices partially unplugged") >>> >>> Ok so I have a basic question about partially_hotplugged flag in the >>> device struct (there were no comments added in a99c4da9fc2a39847 >>> explaining it). It seems we return early from pcie_unplug_device() when >>> this flag is set from failover_unplug_primary() in virtio-net. What is the >>> purpose of this flag? It seems we are not doing a full unplug of the >>> primary device? >> >> Yes, to be able to plug it back in case of migration failure we must keep all >> the data structures. > > Ok so two things here: > (a) could you please add a comment to PCIDevice struct in pci.h to clarify > what the flag actually means, why it is there and what it is supposed to > do. Will be in v3. > (b) the naming of the variable could be something like do_partial_unplug > or some such. This could be a separate patch. OK, I'll do that on a separate patch: I'm already working on a patch series moving most of the failover code to PCI files (hotplug is a PCI feature not a virtio one). https://patchew.org/QEMU/20210820142002.152994-1-lvivier@redhat.com/ > > >> But reading the code again it seems this part should be in >> acpi_pcihp_eject_slot() rather than in acpi_pcihp_device_unplug_cb() to >> prevent the hotplug_handler_unplug()/object_unparent() rather than the >> qdev_unrealize() (to be like in pcie.c). > > Correct. You need to place the check earlier so as to be equivalent to > what the native hotplug code does. > Thanks, Laurent
diff --git a/hw/acpi/pcihp.c b/hw/acpi/pcihp.c index f610a25d2ef9..a2d27a3c4763 100644 --- a/hw/acpi/pcihp.c +++ b/hw/acpi/pcihp.c @@ -366,6 +366,11 @@ void acpi_pcihp_device_unplug_cb(HotplugHandler *hotplug_dev, AcpiPciHpState *s, trace_acpi_pci_unplug(PCI_SLOT(pdev->devfn), acpi_pcihp_get_bsel(pci_get_bus(pdev))); + if (pdev->partially_hotplugged) { + pdev->qdev.pending_deleted_event = false; + return; + } + /* * clean up acpi-index so it could reused by another device */ @@ -396,6 +401,7 @@ void acpi_pcihp_device_unplug_request_cb(HotplugHandler *hotplug_dev, return; } + pdev->qdev.pending_deleted_event = true; s->acpi_pcihp_pci_status[bsel].down |= (1U << slot); acpi_send_event(DEVICE(hotplug_dev), ACPI_PCI_HOTPLUG_STATUS); }
Failover needs to detect the end of the PCI unplug to start migration after the VFIO card has been unplugged. To do that, a flag is set in pcie_cap_slot_unplug_request_cb() and reset in pcie_unplug_device(). But since 17858a169508 ("hw/acpi/ich9: Set ACPI PCI hot-plug as default on Q35") we have switched to ACPI unplug and these functions are not called anymore and the flag not set. So failover migration is not able to detect if card is really unplugged and acts as it's done as soon as it's started. So it doesn't wait the end of the unplug to start the migration. We don't see any problem when we test that because ACPI unplug is faster than PCIe native hotplug and when the migration really starts the unplug operation is already done. See c000a9bd06ea ("pci: mark device having guest unplug request pending") a99c4da9fc2a ("pci: mark devices partially unplugged") Signed-off-by: Laurent Vivier <lvivier@redhat.com> --- hw/acpi/pcihp.c | 6 ++++++ 1 file changed, 6 insertions(+)