2:12 p.m.
On Wed, Jun 16, 2021 at 8:30 AM Boris Fiuczynski <fiuczy(a)linux.ibm.com> wrote:
On 6/14/21 10:46 PM, Jonathon Jongsma wrote:
> On Mon, Jun 14, 2021 at 12:27 PM Boris Fiuczynski <fiuczy(a)linux.ibm.com>
wrote:
>>
>> On 6/3/21 10:11 PM, Jonathon Jongsma wrote:
>>> Implement these new API functions in the nodedev driver.
>>>
>>> Signed-off-by: Jonathon Jongsma <jjongsma(a)redhat.com>
>>> ---
>>> src/node_device/node_device_driver.c | 50 ++++++++++++++++++++++++++++
>>> src/node_device/node_device_driver.h | 6 ++++
>>> src/node_device/node_device_udev.c | 21 +++++++-----
>>> 3 files changed, 69 insertions(+), 8 deletions(-)
>>>
>>> diff --git a/src/node_device/node_device_driver.c
b/src/node_device/node_device_driver.c
>>> index 9ebe609aa4..75391f18b8 100644
>>> --- a/src/node_device/node_device_driver.c
>>> +++ b/src/node_device/node_device_driver.c
>>> @@ -1804,3 +1804,53 @@ nodeDeviceGetAutostart(virNodeDevice *device,
>>> virNodeDeviceObjEndAPI(&obj);
>>> return ret;
>>> }
>>> +
>>> +
>>> +int
>>> +nodeDeviceIsPersistent(virNodeDevice *device)
>>> +{
>>> + virNodeDeviceObj *obj = NULL;
>>> + virNodeDeviceDef *def = NULL;
>>> + int ret = -1;
>>> +
>>> + if (nodeDeviceInitWait() < 0)
>>> + return -1;
>>> +
>>> + if (!(obj = nodeDeviceObjFindByName(device->name)))
>>> + return -1;
>>> + def = virNodeDeviceObjGetDef(obj);
>>> +
>>> + if (virNodeDeviceIsPersistentEnsureACL(device->conn, def) < 0)
>>> + goto cleanup;
>>> +
>>> + ret = virNodeDeviceObjIsPersistent(obj);
>>> +
>>> + cleanup:
>>> + virNodeDeviceObjEndAPI(&obj);
>>> + return ret;
>>> +}
>>> +
>>> +
>>> +int
>>> +nodeDeviceIsActive(virNodeDevice *device)
>>> +{
>>> + virNodeDeviceObj *obj = NULL;
>>> + virNodeDeviceDef *def = NULL;
>>> + int ret = -1;
>>> +
>>> + if (nodeDeviceInitWait() < 0)
>>> + return -1;
>>> +
>>> + if (!(obj = nodeDeviceObjFindByName(device->name)))
>>> + return -1;
>>> + def = virNodeDeviceObjGetDef(obj);
>>> +
>>> + if (virNodeDeviceIsActiveEnsureACL(device->conn, def) < 0)
>>> + goto cleanup;
>>> +
>>> + ret = virNodeDeviceObjIsActive(obj);
>>> +
>>> + cleanup:
>>> + virNodeDeviceObjEndAPI(&obj);
>>> + return ret;
>>> +}
>>> diff --git a/src/node_device/node_device_driver.h
b/src/node_device/node_device_driver.h
>>> index d178a18180..744dd42632 100644
>>> --- a/src/node_device/node_device_driver.h
>>> +++ b/src/node_device/node_device_driver.h
>>> @@ -180,6 +180,12 @@ int
>>> nodeDeviceGetAutostart(virNodeDevice *dev,
>>> int *autostart);
>>>
>>> +int
>>> +nodeDeviceIsPersistent(virNodeDevice *dev);
>>> +
>>> +int
>>> +nodeDeviceIsActive(virNodeDevice *dev);
>>> +
>>> virCommand*
>>> nodeDeviceGetMdevctlSetAutostartCommand(virNodeDeviceDef *def,
>>> bool autostart,
>>> diff --git a/src/node_device/node_device_udev.c
b/src/node_device/node_device_udev.c
>>> index 21273083a6..eb15ccce7f 100644
>>> --- a/src/node_device/node_device_udev.c
>>> +++ b/src/node_device/node_device_udev.c
>>> @@ -1487,7 +1487,7 @@ udevAddOneDevice(struct udev_device *device)
>>> virObjectEvent *event = NULL;
>>> bool new_device = true;
>>> int ret = -1;
>>> - bool was_persistent = false;
>>> + bool persistent = true;
>>> bool autostart = true;
>>> bool is_mdev;
>>>
>>> @@ -1518,7 +1518,8 @@ udevAddOneDevice(struct udev_device *device)
>>>
>>> if (is_mdev)
>>> nodeDeviceDefCopyFromMdevctl(def, objdef);
>>> - was_persistent = virNodeDeviceObjIsPersistent(obj);
>>> +
>>> + persistent = virNodeDeviceObjIsPersistent(obj);
>>> autostart = virNodeDeviceObjIsAutostart(obj);
>>>
>>> /* If the device was defined by mdevctl and was never
instantiated, it
>>> @@ -1527,11 +1528,12 @@ udevAddOneDevice(struct udev_device *device)
>>>
>>> virNodeDeviceObjEndAPI(&obj);
>>> } else {
>>> - /* All non-mdev devices report themselves as autostart since they
>>> - * should still be present and active after a reboot unless the
device
>>> - * is removed from the host. Mediated devices can only be
persistent if
>>> - * they are in already in the device list from parsing the mdevctl
>>> - * output. */
>>> + /* All non-mdev devices report themselves as persistent and
autostart
>>> + * since they should still be present and active after a reboot
unless
>>> + * the device is removed from the host. Mediated devices can only
be
>>> + * persistent if they are in already in the device list from
parsing
>>> + * the mdevctl output. */
>>
>> The assumption for all non-mdev devices ends up very misleading.
>> For example: The parent device of an mdev needs to be bound to a vfio
>> device driver. Without it the device ends up without the capability to
>> create mdevs.
>> If this driver binding is not persisted (e.g. with setting up driverctl)
>> but instead the device is just manually being rebound to a vfio device
>> driver than after reboot neither the mdev capability on the parent
>> device nor the mdev device as a child device will be existing.
>> A user calling nodedev-info before the reboot gets
>> on the parent device
>> Persistent: yes
>> Autostart: yes
>> and on the mdev device
>> Persistent: yes
>> Autostart: yes
>> After a reboot he ends up with with nodedev-info
>> on the parent device
>> Persistent: yes
>> Autostart: yes
>> and the mdev device does not exist.
>
> Before I get to the rest of your suggestion, I'd like to know more
> about this. If the mdev device is persistent (i.e. "defined" in
> mdevctl terminology), then it should still exist after a reboot, even
> if it's not started. If it doesn't, then it's a bug. An mdev can be
> defined even if its parent device is not present.
>
> Does this device show up if you run 'mdevctl list --defined'?
Yes, the mdev definition exists.
Here is the information before the reboot with a vfio-ccw device setup
by manually binding it to the vfio-ccw device driver. Manually means
that I did not use driverctl to persist the device driver binding.
# virsh nodedev-info mdev_e60cef97_3f6b_485e_ac46_0520f9f66ac2
Name: mdev_e60cef97_3f6b_485e_ac46_0520f9f66ac2
Parent: css_0_0_0033
Active: yes
Persistent: yes
Autostart: yes
# mdevctl list -d
e60cef97-3f6b-485e-ac46-0520f9f66ac2 0.0.0033 vfio_ccw-io auto (active)
# virsh nodedev-dumpxml mdev_e60cef97_3f6b_485e_ac46_0520f9f66ac2
<device>
<name>mdev_e60cef97_3f6b_485e_ac46_0520f9f66ac2</name>
<path>/sys/devices/css0/0.0.0033/e60cef97-3f6b-485e-ac46-0520f9f66ac2</path>
<parent>css_0_0_0033</parent>
<driver>
<name>vfio_mdev</name>
</driver>
<capability type='mdev'>
<type id='vfio_ccw-io'/>
<uuid>e60cef97-3f6b-485e-ac46-0520f9f66ac2</uuid>
<iommuGroup number='1'/>
</capability>
</device>
# virsh nodedev-info css_0_0_0033
Name: css_0_0_0033
Parent: computer
Active: yes
Persistent: yes
Autostart: yes
This is the state before rebooting.
After the reboot:
# virsh nodedev-list --tree
...
+- css_0_0_0033
| |
| +- ccw_0_0_c670
| |
| +- block_dasdb_IBM_750000000KMV11_c600_70
|
+- css_0_0_0034
| |
| +- ccw_0_0_c671
| |
| +- block_dasda_IBM_750000000KMV11_c600_71
...
# virsh nodedev-info css_0_0_0033
Name: css_0_0_0033
Parent: computer
Active: yes
Persistent: yes
Autostart: yes
# virsh nodedev-info ccw_0_0_c670
Name: ccw_0_0_c670
Parent: css_0_0_0033
Active: yes
Persistent: yes
Autostart: yes
# virsh nodedev-dumpxml css_0_0_0033
<device>
<name>css_0_0_0033</name>
<path>/sys/devices/css0/0.0.0033</path>
<parent>computer</parent>
<driver>
<name>io_subchannel</name>
</driver>
<capability type='css'>
<cssid>0x0</cssid>
<ssid>0x0</ssid>
<devno>0x0033</devno>
</capability>
</device>
# virsh nodedev-list --cap mdev --all
mdev_e60cef97_3f6b_485e_ac46_0520f9f66ac2
# virsh nodedev-info mdev_e60cef97_3f6b_485e_ac46_0520f9f66ac2
Name: mdev_e60cef97_3f6b_485e_ac46_0520f9f66ac2
Parent: 0.0.0033
Active: no
Persistent: yes
Autostart: yes
As you can see the default device driver is bound to the parent device
now and therefore the mdev does not get created although being defined
in mdevctl.
Also note that Persistent ond Autostart are misleading as they show on
the mdev definition as well as the ccw device being set both "yes".
On s390 there is also cio_ignore which allows devices to be ignored.
This can also cause that devices exist before the reboot afterwards due
to being ignored no longer exist on the system.
For these reasons I think that these two new attributes show something
that libvirt should not make assumptions on unless knowing about it.
This is an interesting case. The vast majority of these devices will
be "persistent" for all practical purposes, because they will still be
there on the next reboot, and they will still be using the same
drivers. On the other hand, it is true that there is no record of the
device that remains after the physical device is removed, so strictly
speaking they're not persistently defined. After reflecting, I've kind
of flip-flopped and decided that maybe the best choice would be to
mark them as non-persistent (and also non-autostart?). Would there be
any downside to such a designation? They don't exactly match the
behavior of other transient objects (domains, etc) because they
(mostly) will still be present after a reboot...I still don't really
like the idea of introducing a tristate return type. Opinions from
longtime libvirt developers particularly appreciated.
Jonathon