于 2013年03月19日 16:58, Tomas Hozza 写道:
----- Original Message -----
-----Original Message-----
From: Tomas Hozza [mailto:thozza@xxxxxxxxxx]
Sent: Monday, March 18, 2013 11:39 AM
To: KY Srinivasan
Cc: Jason Wang
Subject: Re: Can HyperV KVP daemon detect if running under Windows
Server
----- Original Message -----
-----Original Message-----
From: Tomas Hozza [mailto:thozza@xxxxxxxxxx]
Sent: Monday, March 18, 2013 11:15 AM
To: KY Srinivasan
Cc: Jason Wang
Subject: Re: Can HyperV KVP daemon detect if running under
Windows
Server
I tested the hyperv KVP daemon on RHEL-6.4 running on libvirt
KVM virtual machine. None of hv_* kmod-s was loaded. The daemon
is running just fine and netlink_send() completes successfully
even if proper drivers are not loaded.
[root@localhost ~]# modprobe hv_vmbus
FATAL: Error inserting hv_vmbus (/lib/modules/2.6.32-
358.el6.x86_64/kernel/drivers/hv/hv_vmbus.ko): No such device
[root@localhost ~]# modprobe hv_utils
FATAL: Error inserting hv_utils (/lib/modules/2.6.32-
358.el6.x86_64/kernel/drivers/hv/hv_utils.ko): No such device
[root@localhost ~]# lsmod | grep hv
[root@localhost ~]# service hypervkvpd start
Starting Hyper-V KVP daemon [
OK ]
[root@localhost ~]# service hypervkvpd status
hv_kvp_daemon (pid 1509) is running...
[root@localhost ~]# grep KVP /var/log/messages
Mar 18 16:00:50 localhost KVP: KVP starting; pid is:1509
[root@localhost ~]# uname -a
Linux localhost.localdomain 2.6.32-358.el6.x86_64 #1 SMP Tue
Jan 29
11:47:41 EST
2013 x86_64 x86_64 x86_64 GNU/Linux
Maybe there could be some timeout for driver registration.
After it the daemon would exit if it got no reply from
kernel driver.
In that case, as you suggest maybe we can have a timeout on the
first
receive from the kernel. In the current code, once the daemon
registers, the kernel component sends the LIS version number and
this message is guaranteed to come soon after the daemon
registers.
If this does not happen, we can conclude that we are not on
Hyper-V.
After having a look on the kernel driver I know this can not
be done before some issues in the kernel driver are fixed.
The problem is that the kernel driver at the moment handles
daemon registration only for the first time. On the other
hand the daemon wants to register with the driver every time
it is started. This can be real problem for example when
the daemon is restarted (stopped and started). You can read
my longer and more explanatory comment in Red Hat Bugzilla [1].
Why would you want to stop/restart the daemon. Not having the
daemon,
can only create more problems since the kernel component holds
onto
the host transaction until the daemon registers. We should view
the
daemon as simply an extension of the kernel driver.
If you update the hypervkvpd package the daemon is restarted. In
the
current daemon code the registration part is written as:
...
if ((in_hand_shake) && (op == KVP_OP_REGISTER1)) {
/* handle registration */
}
...
so it is possible that you update the daemon, it is restarted and
wants
to register with the driver. It sends the registration message to
driver.
The driver ignores this message since it has no transaction active
from
the Host so it is not expecting any messages from Guest. The daemon
ends up in a state when it is "in_hand_shake" and waiting for the
registration
to complete, BUT (because how the registration part is written) it
will
handle requests from the driver (Host). I think this is not very
fortunate
and it is kind of "design issue". User should NOT have to restart
(or unload
kernel modules and load them again) after he updated the daemon.
I agree that in_hand_shake snippet like this is not good enough to
guarantee a registration is mandatory and can't be skipped.
I thought code like below:
if (in_hand_shake) {
if (op == KVP_OP_REGISTER1) {
...
in_hand_shake = 0;
}
if (in_hand_shake) {
continue;
}
}
is more reasonable and fault tolerable.
I suppose I've already encountered an issue like this -- kvp receviced
messages sent by libcgroup, registration phase is skipped and fault
packets are regarded as kvp packets.
to K. Y:
remind you on
http://driverdev.linuxdriverproject.org/pipermail/devel/2013-March/036306.html.
I updated it and asked Evgeniy Polyakov on how to subscribe group using
connector, waiting his feedback.
Regards,
Vaughan
True; but as I said the daemon is to be viewed as simply an extension
of the kernel component. If this is done, then you would not be
updating the daemon independently of the kernel component. I guess
the problem you are running into is because you are viewing the
daemon as an independent component that can be updated independently
of the kernel piece. Perhaps, this is a packaging issue.
Regards,
K. Y
I see your point. It is true that we ship hyperv daemon as a separate
package. I understand the daemon is just an extension to kernel driver
and has no purpose without it.
But I don't like the way it is now anyway. Even if you look at the daemon
as just an extension to the kernel part user should not need to restart
whole Guest to get it to an "initial" state of module and the daemon.
There might be some bug in the daemon, but the driver can be just fine.
In that case I would expect that the daemon can be updated without any
problems and without restarting the system (or unloading and loading driver).
I would like to know what other people think about this. CC-ing
Haiyang Zhang and devel@xxxxxxxxxxxxxxxxxxxxxx.
Regards,
Tomas Hozza
Regards,
Tomas Hozza
Regards,
Tomas Hozza
----- Original Message -----
Tomas,
Sorry for the delayed response. The kernel driver only loads
when
Linux is running on Hyper-V. We could leverage this to not
have
the
daemon running when not hosted on Hyper-V. The current code
as it
stands, the daemon exits if the netlink_send() fails. When
not
running on Hyper-V, does the initial send that registers the
daemon
succeed.
Regards,
K. Y
-----Original Message-----
From: Tomas Hozza [mailto:thozza@xxxxxxxxxx]
Sent: Monday, March 18, 2013 7:13 AM
To: Jason Wang
Cc: KY Srinivasan
Subject: Fwd: Can HyperV KVP daemon detect if running under
Windows
Server
Importance: High
Hi.
I got no response to my email I sent some time ago
so I'm sending it again. Maybe you just overlooked
it. Thank you in advance for the response.
Regards,
Tomas Hozza
----- Forwarded Message -----
Hi.
I'm wondering if there is any way for the hyperv kvp
daemon
detect if it is running as a guest on Windows Server host
or
not. And furthermore run only if it is a guest under
Windows
Server. Now hyperv kvp daemon is a default package in a
Virtualization group of packages on RHEL 6.4. Some people
are
not happy that it is installed and running by default
when
the Virtualization group is selected.
Do you have any suggestions? I don't know how the kernel
driver
is working (one that hyperv kvp daemon registers with).
Would
it
be possible to get some useful information from it?
Thank you.
Regards,
Tomas Hozza
_______________________________________________
devel mailing list
devel@xxxxxxxxxxxxxxxxxxxxxx
http://driverdev.linuxdriverproject.org/mailman/listinfo/devel
_______________________________________________
devel mailing list
devel@xxxxxxxxxxxxxxxxxxxxxx
http://driverdev.linuxdriverproject.org/mailman/listinfo/devel
