[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [PATCH v6 04/10] device_core: use drain_call_rcu in in qmp_device_ad
From: |
Paolo Bonzini |
Subject: |
Re: [PATCH v6 04/10] device_core: use drain_call_rcu in in qmp_device_add |
Date: |
Fri, 2 Oct 2020 23:22:52 +0200 |
User-agent: |
Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101 Thunderbird/68.11.0 |
On 02/10/20 19:35, Paolo Bonzini wrote:
> From: Maxim Levitsky <mlevitsk@redhat.com>
>
> Soon, a device removal might only happen on RCU callback execution.
> This is okay for device-del which provides a DEVICE_DELETED event,
> but not for the failure case of device-add. To avoid changing
> monitor semantics, just drain all pending RCU callbacks on error.
Nope, this is not enough... qemu-iotests test 067 still breaks; if I
leave the drain_call_rcu() in qmp_device_del, on the other hand, the
hotplug test in qtest-ppc64:qos-test hangs because DEVICE_DELETED comes
before the {'return'}.
One way to handle this is:
1) handle events properly in libqtest. For example, rename
qtest_qmp_receive to qtest_qmp_receive_dict (most callers can keep on
calling qtest_qmp_receive, though some have to switch if they look for
events); a new qtest_qmp_receive that calls qtest_qmp_receive_dict until
it doesn't get an event, and in the meanwhile stashes the events in a
GList. Then qtest_qmp_eventwait_ref can look for events in the GList
prior to calling qtest_qmp_receive_any if it doesn't find any.
2) with (1) in place we can probably keep the drain_call_rcu(), but
perhaps we can drop it if we remove test 067 and add equivalent tests to
tests/qtest/drive_del.c.
The reason I'm not fond of the drain_call_rcu() in device_del is that
there is already a DEVICE_DELETED callback, and device-del is (at least
for some buses) an advisory operation that the guest needs to attend to.
So there's no guarantee that you get the DEVICE_DELETED event before
drain_call_rcu().
I can look at this next week.
Paolo
- [PATCH v6 00/10] Fix scsi devices plug/unplug races w.r.t virtio-scsi iothread, Paolo Bonzini, 2020/10/02
- [PATCH v6 01/10] qdev: add "check if address free" callback for buses, Paolo Bonzini, 2020/10/02
- [PATCH v6 02/10] scsi: switch to bus->check_address, Paolo Bonzini, 2020/10/02
- [PATCH v6 04/10] device_core: use drain_call_rcu in in qmp_device_add, Paolo Bonzini, 2020/10/02
- Re: [PATCH v6 04/10] device_core: use drain_call_rcu in in qmp_device_add,
Paolo Bonzini <=
- [PATCH v6 03/10] scsi/scsi_bus: switch search direction in scsi_device_find, Paolo Bonzini, 2020/10/02
- [PATCH v6 05/10] device-core: use RCU for list of children of a bus, Paolo Bonzini, 2020/10/02
- [PATCH v6 06/10] device-core: use atomic_set on .realized property, Paolo Bonzini, 2020/10/02
- [PATCH v6 07/10] scsi/scsi-bus: scsi_device_find: don't return unrealized devices, Paolo Bonzini, 2020/10/02
- [PATCH v6 10/10] scsi/scsi_bus: fix races in REPORT LUNS, Paolo Bonzini, 2020/10/02
- [PATCH v6 08/10] scsi/scsi_bus: Add scsi_device_get, Paolo Bonzini, 2020/10/02
- [PATCH v6 09/10] virtio-scsi: use scsi_device_get, Paolo Bonzini, 2020/10/02
- Re: [PATCH v6 00/10] Fix scsi devices plug/unplug races w.r.t virtio-scsi iothread, no-reply, 2020/10/03