Re: [PATCH v3 0/10] add failover feature for assigned network devices
On Thu, Oct 17, 2019 at 06:51:54AM -0600, Alex Williamson wrote: On Thu, 17 Oct 2019 12:33:47 +0200 Jens Freimann wrote: On Tue, Oct 15, 2019 at 01:03:17PM -0600, Alex Williamson wrote: >On Fri, 11 Oct 2019 13:20:05 +0200 >Jens Freimann wrote: > >> This is implementing the host side of the net_failover concept >> (https://www.kernel.org/doc/html/latest/networking/net_failover.html) >> >> Changes since v2: >> * back out of creating failover pair when it is a non-networking >> vfio-pci device (Alex W) >> * handle migration state change from within the migration thread. I do a >> timed wait on a semaphore and then check if all unplugs were >> succesful. Added a new function to each device that checks the device >> if the unplug for it has happened. When all devices report the succesful >> unplug *or* the time/retries is up, continue with the migration or >> cancel. When not all devices could be unplugged I am cancelling at the >> moment. It is likely that we can't plug it back at the destination which >> would result in degraded network performance. >> * fix a few bugs regarding re-plug on migration source and target >> * run full set of tests including migration tests >> * add patch for libqos to tolerate new migration state >> * squashed patch 1 and 2, added patch 8 >> >> The general idea is that we have a pair of devices, a vfio-pci and a >> virtio-net device. Before migration the vfio device is unplugged and data >> flows to the virtio-net device, on the target side another vfio-pci device >> is plugged in to take over the data-path. In the guest the net_failover >> module will pair net devices with the same MAC address. >> >> * Patch 1 adds the infrastructure to hide the device for the qbus and qdev APIs >> >> * Patch 2 sets a new flag for PCIDevice 'partially_hotplugged' which we >> use to skip the unrealize code path when doing a unplug of the primary >> device >> >> * Patch 3 sets the pending_deleted_event before triggering the guest >> unplug request > >These only cover pcie hotplug, is this feature somehow dependent on >pcie? There's also ACPI-based PCI hotplug, SHPC hotplug, and it looks >like s390 has it's own version (of course) of PCI hotplug. IMO, we >either need to make an attempt to support this universally or the >option needs to fail if the hotplug controller doesn't support partial >removal. Thanks, It is possible to make it work with non-pcie hotplug but as the first step I want to enable it for pcie only. For that I would add a check into pci_qdev_realize(), where I also check if the device is an ethernet device, and fail if it is not a pcie device. Would that work for you? How would libvirt introspect what topologies are supported rather than trial and error? I think this solves my issue that I get bugs that the failover pair option doesn't work on vfio-pci depending on the topology, but it really just pushes the problem up the stack. Thanks, Good point Alex. Laine, any idea what would be required for this? How does libvirt introspect other properties of PCI devices if any? I only found checks for commandline options in libvirt, but I might miss something. regards, Jens
Re: [PATCH v3 0/10] add failover feature for assigned network devices
On Thu, 17 Oct 2019 12:33:47 +0200 Jens Freimann wrote: > On Tue, Oct 15, 2019 at 01:03:17PM -0600, Alex Williamson wrote: > >On Fri, 11 Oct 2019 13:20:05 +0200 > >Jens Freimann wrote: > > > >> This is implementing the host side of the net_failover concept > >> (https://www.kernel.org/doc/html/latest/networking/net_failover.html) > >> > >> Changes since v2: > >> * back out of creating failover pair when it is a non-networking > >> vfio-pci device (Alex W) > >> * handle migration state change from within the migration thread. I do a > >> timed wait on a semaphore and then check if all unplugs were > >> succesful. Added a new function to each device that checks the device > >> if the unplug for it has happened. When all devices report the succesful > >> unplug *or* the time/retries is up, continue with the migration or > >> cancel. When not all devices could be unplugged I am cancelling at the > >> moment. It is likely that we can't plug it back at the destination which > >> would result in degraded network performance. > >> * fix a few bugs regarding re-plug on migration source and target > >> * run full set of tests including migration tests > >> * add patch for libqos to tolerate new migration state > >> * squashed patch 1 and 2, added patch 8 > >> > >> The general idea is that we have a pair of devices, a vfio-pci and a > >> virtio-net device. Before migration the vfio device is unplugged and data > >> flows to the virtio-net device, on the target side another vfio-pci device > >> is plugged in to take over the data-path. In the guest the net_failover > >> module will pair net devices with the same MAC address. > >> > >> * Patch 1 adds the infrastructure to hide the device for the qbus and qdev > >> APIs > >> > >> * Patch 2 sets a new flag for PCIDevice 'partially_hotplugged' which we > >> use to skip the unrealize code path when doing a unplug of the primary > >> device > >> > >> * Patch 3 sets the pending_deleted_event before triggering the guest > >> unplug request > > > >These only cover pcie hotplug, is this feature somehow dependent on > >pcie? There's also ACPI-based PCI hotplug, SHPC hotplug, and it looks > >like s390 has it's own version (of course) of PCI hotplug. IMO, we > >either need to make an attempt to support this universally or the > >option needs to fail if the hotplug controller doesn't support partial > >removal. Thanks, > > It is possible to make it work with non-pcie hotplug but as the first > step I want to enable it for pcie only. For that I would add a check > into pci_qdev_realize(), where I also check if the device is an > ethernet device, and fail if it is not a pcie device. Would that work > for you? How would libvirt introspect what topologies are supported rather than trial and error? I think this solves my issue that I get bugs that the failover pair option doesn't work on vfio-pci depending on the topology, but it really just pushes the problem up the stack. Thanks, Alex
Re: [PATCH v3 0/10] add failover feature for assigned network devices
On Tue, Oct 15, 2019 at 01:03:17PM -0600, Alex Williamson wrote: On Fri, 11 Oct 2019 13:20:05 +0200 Jens Freimann wrote: This is implementing the host side of the net_failover concept (https://www.kernel.org/doc/html/latest/networking/net_failover.html) Changes since v2: * back out of creating failover pair when it is a non-networking vfio-pci device (Alex W) * handle migration state change from within the migration thread. I do a timed wait on a semaphore and then check if all unplugs were succesful. Added a new function to each device that checks the device if the unplug for it has happened. When all devices report the succesful unplug *or* the time/retries is up, continue with the migration or cancel. When not all devices could be unplugged I am cancelling at the moment. It is likely that we can't plug it back at the destination which would result in degraded network performance. * fix a few bugs regarding re-plug on migration source and target * run full set of tests including migration tests * add patch for libqos to tolerate new migration state * squashed patch 1 and 2, added patch 8 The general idea is that we have a pair of devices, a vfio-pci and a virtio-net device. Before migration the vfio device is unplugged and data flows to the virtio-net device, on the target side another vfio-pci device is plugged in to take over the data-path. In the guest the net_failover module will pair net devices with the same MAC address. * Patch 1 adds the infrastructure to hide the device for the qbus and qdev APIs * Patch 2 sets a new flag for PCIDevice 'partially_hotplugged' which we use to skip the unrealize code path when doing a unplug of the primary device * Patch 3 sets the pending_deleted_event before triggering the guest unplug request These only cover pcie hotplug, is this feature somehow dependent on pcie? There's also ACPI-based PCI hotplug, SHPC hotplug, and it looks like s390 has it's own version (of course) of PCI hotplug. IMO, we either need to make an attempt to support this universally or the option needs to fail if the hotplug controller doesn't support partial removal. Thanks, It is possible to make it work with non-pcie hotplug but as the first step I want to enable it for pcie only. For that I would add a check into pci_qdev_realize(), where I also check if the device is an ethernet device, and fail if it is not a pcie device. Would that work for you? regards, Jens
Re: [PATCH v3 0/10] add failover feature for assigned network devices
On Tue, Oct 15, 2019 at 01:03:17PM -0600, Alex Williamson wrote: > On Fri, 11 Oct 2019 13:20:05 +0200 > Jens Freimann wrote: > > > This is implementing the host side of the net_failover concept > > (https://www.kernel.org/doc/html/latest/networking/net_failover.html) > > > > Changes since v2: > > * back out of creating failover pair when it is a non-networking > > vfio-pci device (Alex W) > > * handle migration state change from within the migration thread. I do a > > timed wait on a semaphore and then check if all unplugs were > > succesful. Added a new function to each device that checks the device > > if the unplug for it has happened. When all devices report the succesful > > unplug *or* the time/retries is up, continue with the migration or > > cancel. When not all devices could be unplugged I am cancelling at the > > moment. It is likely that we can't plug it back at the destination which > > would result in degraded network performance. > > * fix a few bugs regarding re-plug on migration source and target > > * run full set of tests including migration tests > > * add patch for libqos to tolerate new migration state > > * squashed patch 1 and 2, added patch 8 > > > > The general idea is that we have a pair of devices, a vfio-pci and a > > virtio-net device. Before migration the vfio device is unplugged and data > > flows to the virtio-net device, on the target side another vfio-pci device > > is plugged in to take over the data-path. In the guest the net_failover > > module will pair net devices with the same MAC address. > > > > * Patch 1 adds the infrastructure to hide the device for the qbus and qdev > > APIs > > > > * Patch 2 sets a new flag for PCIDevice 'partially_hotplugged' which we > > use to skip the unrealize code path when doing a unplug of the primary > > device > > > > * Patch 3 sets the pending_deleted_event before triggering the guest > > unplug request > > These only cover pcie hotplug, is this feature somehow dependent on > pcie? There's also ACPI-based PCI hotplug, SHPC hotplug, and it looks > like s390 has it's own version (of course) of PCI hotplug. IMO, we > either need to make an attempt to support this universally or the > option needs to fail if the hotplug controller doesn't support partial > removal. Thanks, > > Alex Alex, could you please comment a bit more on vfio patches? Besides what you point out here, any other issues? -- MST
Re: [PATCH v3 0/10] add failover feature for assigned network devices
On Fri, 11 Oct 2019 13:20:05 +0200 Jens Freimann wrote: > This is implementing the host side of the net_failover concept > (https://www.kernel.org/doc/html/latest/networking/net_failover.html) > > Changes since v2: > * back out of creating failover pair when it is a non-networking > vfio-pci device (Alex W) > * handle migration state change from within the migration thread. I do a > timed wait on a semaphore and then check if all unplugs were > succesful. Added a new function to each device that checks the device > if the unplug for it has happened. When all devices report the succesful > unplug *or* the time/retries is up, continue with the migration or > cancel. When not all devices could be unplugged I am cancelling at the > moment. It is likely that we can't plug it back at the destination which > would result in degraded network performance. > * fix a few bugs regarding re-plug on migration source and target > * run full set of tests including migration tests > * add patch for libqos to tolerate new migration state > * squashed patch 1 and 2, added patch 8 > > The general idea is that we have a pair of devices, a vfio-pci and a > virtio-net device. Before migration the vfio device is unplugged and data > flows to the virtio-net device, on the target side another vfio-pci device > is plugged in to take over the data-path. In the guest the net_failover > module will pair net devices with the same MAC address. > > * Patch 1 adds the infrastructure to hide the device for the qbus and qdev > APIs > > * Patch 2 sets a new flag for PCIDevice 'partially_hotplugged' which we > use to skip the unrealize code path when doing a unplug of the primary > device > > * Patch 3 sets the pending_deleted_event before triggering the guest > unplug request These only cover pcie hotplug, is this feature somehow dependent on pcie? There's also ACPI-based PCI hotplug, SHPC hotplug, and it looks like s390 has it's own version (of course) of PCI hotplug. IMO, we either need to make an attempt to support this universally or the option needs to fail if the hotplug controller doesn't support partial removal. Thanks, Alex
Re: [PATCH v3 0/10] add failover feature for assigned network devices
Patchew URL: https://patchew.org/QEMU/2019102015.11785-1-jfreim...@redhat.com/ Hi, This series failed the docker-quick@centos7 build test. Please find the testing commands and their output below. If you have Docker installed, you can probably reproduce it locally. === TEST SCRIPT BEGIN === #!/bin/bash make docker-image-centos7 V=1 NETWORK=1 time make docker-test-quick@centos7 SHOW_ENV=1 J=14 NETWORK=1 === TEST SCRIPT END === TESTiotest-qcow2: 025 TESTiotest-qcow2: 027 ** ERROR:/tmp/qemu-test/src/tests/migration-test.c:903:wait_for_migration_fail: assertion failed: (!strcmp(status, "setup") || !strcmp(status, "failed") || (allow_active && !strcmp(status, "active"))) ERROR - Bail out! ERROR:/tmp/qemu-test/src/tests/migration-test.c:903:wait_for_migration_fail: assertion failed: (!strcmp(status, "setup") || !strcmp(status, "failed") || (allow_active && !strcmp(status, "active"))) make: *** [check-qtest-aarch64] Error 1 make: *** Waiting for unfinished jobs TESTiotest-qcow2: 029 TESTcheck-unit: tests/test-qdist --- TESTiotest-qcow2: 172 TESTiotest-qcow2: 174 ** ERROR:/tmp/qemu-test/src/tests/migration-test.c:903:wait_for_migration_fail: assertion failed: (!strcmp(status, "setup") || !strcmp(status, "failed") || (allow_active && !strcmp(status, "active"))) ERROR - Bail out! ERROR:/tmp/qemu-test/src/tests/migration-test.c:903:wait_for_migration_fail: assertion failed: (!strcmp(status, "setup") || !strcmp(status, "failed") || (allow_active && !strcmp(status, "active"))) make: *** [check-qtest-x86_64] Error 1 TESTiotest-qcow2: 176 TESTiotest-qcow2: 177 TESTiotest-qcow2: 179 --- raise CalledProcessError(retcode, cmd) subprocess.CalledProcessError: Command '['sudo', '-n', 'docker', 'run', '--label', 'com.qemu.instance.uuid=8a50cfce678f45bfb554befdb722504e', '-u', '1001', '--security-opt', 'seccomp=unconfined', '--rm', '-e', 'TARGET_LIST=', '-e', 'EXTRA_CONFIGURE_OPTS=', '-e', 'V=', '-e', 'J=14', '-e', 'DEBUG=', '-e', 'SHOW_ENV=1', '-e', 'CCACHE_DIR=/var/tmp/ccache', '-v', '/home/patchew/.cache/qemu-docker-ccache:/var/tmp/ccache:z', '-v', '/var/tmp/patchew-tester-tmp-lrrx9ztr/src/docker-src.2019-10-11-11.53.24.30391:/var/tmp/qemu:z,ro', 'qemu:centos7', '/var/tmp/qemu/run', 'test-quick']' returned non-zero exit status 2. filter=--filter=label=com.qemu.instance.uuid=8a50cfce678f45bfb554befdb722504e make[1]: *** [docker-run] Error 1 make[1]: Leaving directory `/var/tmp/patchew-tester-tmp-lrrx9ztr/src' make: *** [docker-run-test-quick@centos7] Error 2 real10m49.386s user0m8.737s The full log is available at http://patchew.org/logs/2019102015.11785-1-jfreim...@redhat.com/testing.docker-quick@centos7/?type=message. --- Email generated automatically by Patchew [https://patchew.org/]. Please send your feedback to patchew-de...@redhat.com
Re: [PATCH v3 0/10] add failover feature for assigned network devices
On Fri, Oct 11, 2019 at 01:20:05PM +0200, Jens Freimann wrote: > This is implementing the host side of the net_failover concept > (https://www.kernel.org/doc/html/latest/networking/net_failover.html) I posted a comment about the migration changes. Besides that this looks good to me. I did not look at VFIO parts at all yet. Alex, could you pls review/ack the VFIO patch? > Changes since v2: > * back out of creating failover pair when it is a non-networking > vfio-pci device (Alex W) > * handle migration state change from within the migration thread. I do a > timed wait on a semaphore and then check if all unplugs were > succesful. Added a new function to each device that checks the device > if the unplug for it has happened. When all devices report the succesful > unplug *or* the time/retries is up, continue with the migration or > cancel. When not all devices could be unplugged I am cancelling at the > moment. It is likely that we can't plug it back at the destination which > would result in degraded network performance. > * fix a few bugs regarding re-plug on migration source and target > * run full set of tests including migration tests > * add patch for libqos to tolerate new migration state > * squashed patch 1 and 2, added patch 8 > > The general idea is that we have a pair of devices, a vfio-pci and a > virtio-net device. Before migration the vfio device is unplugged and data > flows to the virtio-net device, on the target side another vfio-pci device > is plugged in to take over the data-path. In the guest the net_failover > module will pair net devices with the same MAC address. > > * Patch 1 adds the infrastructure to hide the device for the qbus and qdev > APIs > > * Patch 2 sets a new flag for PCIDevice 'partially_hotplugged' which we > use to skip the unrealize code path when doing a unplug of the primary > device > > * Patch 3 sets the pending_deleted_event before triggering the guest > unplug request > > * Patch 4 and 5 add new qmp events, one sends the device id of a device > that was just requested to be unplugged from the guest and another one > to let libvirt know if VIRTIO_NET_F_STANDBY was negotiated > > * Patch 6 make sure that we can unplug the vfio-device before > migration starts > > * Patch 7 adds a new migration state that is entered while we wait for > devices to be unplugged by guest OS > > * Patch 8 just adds the new migration state to a check in libqos code > > * Patch 9 In the second patch the virtio-net uses the API to defer adding the > vfio > device until the VIRTIO_NET_F_STANDBY feature is acked. It also > implements the migration handler to unplug the device from the guest and > re-plug in case of migration failure > > * Patch 10 allows migration for failover vfio-pci devices, but only when it is > a networking device > > Previous discussion: > RFC v1 https://patchwork.ozlabs.org/cover/989098/ > RFC v2 https://www.mail-archive.com/qemu-devel@nongnu.org/msg606906.html > v1: https://lists.gnu.org/archive/html/qemu-devel/2019-05/msg03968.html > v2: https://www.mail-archive.com/qemu-devel@nongnu.org/msg635214.html > > To summarize concerns/feedback from previous discussion: > 1.- guest OS can reject or worse _delay_ unplug by any amount of time. > Migration might get stuck for unpredictable time with unclear reason. > This approach combines two tricky things, hot/unplug and migration. > -> We need to let libvirt know what's happening. Add new qmp events > and a new migration state. When a primary device is (partially) > unplugged (only from guest) we send a qmp event with the device id. When > it is unplugged from the guest the DEVICE_DELETED event is sent. > Migration will enter the wait-unplug state while waiting for the guest > os to unplug all primary devices and then move on with migration. > 2. PCI devices are a precious ressource. The primary device should never > be added to QEMU if it won't be used by guest instead of hiding it in > QEMU. > -> We only hotplug the device when the standby feature bit was > negotiated. We save the device cmdline options until we need it for > qdev_device_add() > Hiding a device can be a useful concept to model. For example a > pci device in a powered-off slot could be marked as hidden until the > slot is > powered on (mst). > 3. Management layer software should handle this. Open Stack already has > components/code to handle unplug/replug VFIO devices and metadata to > provide to the guest for detecting which devices should be paired. > -> An approach that includes all software from firmware to > higher-level management software wasn't tried in the last years. This is > an attempt to keep it simple and contained in QEMU as much as possible. > One of the problems that stopped management software and libvirt from > implementing this idea is that it can't be sure that it's possible to
[PATCH v3 0/10] add failover feature for assigned network devices
This is implementing the host side of the net_failover concept (https://www.kernel.org/doc/html/latest/networking/net_failover.html) Changes since v2: * back out of creating failover pair when it is a non-networking vfio-pci device (Alex W) * handle migration state change from within the migration thread. I do a timed wait on a semaphore and then check if all unplugs were succesful. Added a new function to each device that checks the device if the unplug for it has happened. When all devices report the succesful unplug *or* the time/retries is up, continue with the migration or cancel. When not all devices could be unplugged I am cancelling at the moment. It is likely that we can't plug it back at the destination which would result in degraded network performance. * fix a few bugs regarding re-plug on migration source and target * run full set of tests including migration tests * add patch for libqos to tolerate new migration state * squashed patch 1 and 2, added patch 8 The general idea is that we have a pair of devices, a vfio-pci and a virtio-net device. Before migration the vfio device is unplugged and data flows to the virtio-net device, on the target side another vfio-pci device is plugged in to take over the data-path. In the guest the net_failover module will pair net devices with the same MAC address. * Patch 1 adds the infrastructure to hide the device for the qbus and qdev APIs * Patch 2 sets a new flag for PCIDevice 'partially_hotplugged' which we use to skip the unrealize code path when doing a unplug of the primary device * Patch 3 sets the pending_deleted_event before triggering the guest unplug request * Patch 4 and 5 add new qmp events, one sends the device id of a device that was just requested to be unplugged from the guest and another one to let libvirt know if VIRTIO_NET_F_STANDBY was negotiated * Patch 6 make sure that we can unplug the vfio-device before migration starts * Patch 7 adds a new migration state that is entered while we wait for devices to be unplugged by guest OS * Patch 8 just adds the new migration state to a check in libqos code * Patch 9 In the second patch the virtio-net uses the API to defer adding the vfio device until the VIRTIO_NET_F_STANDBY feature is acked. It also implements the migration handler to unplug the device from the guest and re-plug in case of migration failure * Patch 10 allows migration for failover vfio-pci devices, but only when it is a networking device Previous discussion: RFC v1 https://patchwork.ozlabs.org/cover/989098/ RFC v2 https://www.mail-archive.com/qemu-devel@nongnu.org/msg606906.html v1: https://lists.gnu.org/archive/html/qemu-devel/2019-05/msg03968.html v2: https://www.mail-archive.com/qemu-devel@nongnu.org/msg635214.html To summarize concerns/feedback from previous discussion: 1.- guest OS can reject or worse _delay_ unplug by any amount of time. Migration might get stuck for unpredictable time with unclear reason. This approach combines two tricky things, hot/unplug and migration. -> We need to let libvirt know what's happening. Add new qmp events and a new migration state. When a primary device is (partially) unplugged (only from guest) we send a qmp event with the device id. When it is unplugged from the guest the DEVICE_DELETED event is sent. Migration will enter the wait-unplug state while waiting for the guest os to unplug all primary devices and then move on with migration. 2. PCI devices are a precious ressource. The primary device should never be added to QEMU if it won't be used by guest instead of hiding it in QEMU. -> We only hotplug the device when the standby feature bit was negotiated. We save the device cmdline options until we need it for qdev_device_add() Hiding a device can be a useful concept to model. For example a pci device in a powered-off slot could be marked as hidden until the slot is powered on (mst). 3. Management layer software should handle this. Open Stack already has components/code to handle unplug/replug VFIO devices and metadata to provide to the guest for detecting which devices should be paired. -> An approach that includes all software from firmware to higher-level management software wasn't tried in the last years. This is an attempt to keep it simple and contained in QEMU as much as possible. One of the problems that stopped management software and libvirt from implementing this idea is that it can't be sure that it's possible to re-plug the primary device. By not freeing the devices resources in QEMU and only asking the guest OS to unplug it is possible to re-plug the device in case of a migration failure. 4. Hotplugging a device and then making it part of a failover setup is not possible -> addressed by extending qdev hotplug functions to check for hidden attribute, so e.g. device_add can be used to plug a
