gmane.comp.emulators.kvm.devel

UDP problem with virtio...

Erik Brakkee — 2012-05-25T20:55:37

Hi,


I am running a guest that exposes a NFS share to be used by a TViX box. The TViX box is not that advanced and uses NFS over UDP instead of TCP.
What I am seeing is that when I use the virtio network driver, the TViX box cannot mount the filesystem. However, when I use rtl8139 instead, it can.

The problem is easily reproduced:
1. Create a guest that uses a virtio network interface (bridging setup)
2. Expose an NFS share on the guest
3. mount the NFS share using 'mount -o udp host:/share /localdir'
4. verify that the NFS share cannot be mounted.
5. Modify the guest to use rtl8139 device emulation and stop the guest and start it
6. mount the NFS share using 'mount -o udp host:/share /localdir'
7. Verify tha the NFS share can be mounted

Now the bug reporting guidelines say that I should always use the latest KVM version but compiling KVM from source and using that is not really an option. The server is important enough not to mess around with. Therefore, my question is whether this problem has been seen

[PATCH v2] KVM: Cleanup the kvm_print functions and introduce pr_XX wrappers

Christoffer Dall — 2012-05-25T16:22:59

Introduces a couple of print functions, which are essentially wrappers
around standard printk functions, with a KVM: prefix.

Functions introduced or modified are:
 - kvm_err(fmt, ...)
 - kvm_info(fmt, ...)
 - kvm_debug(fmt, ...)
 - kvm_pr_unimpl(fmt, ...)
 - pr_unimpl(vcpu, fmt, ...) -> vcpu_unimpl(vcpu, fmt, ...)

Applies to kvm-next

Changelog[2]:
 - Added PID to print functions
 - Renamed vcpu_pr_unimpl to vcpu_unimpl

 Signed-off-by: Christoffer Dall <c.dall< at >virtualopensystems.com>
---
 arch/x86/kvm/svm.c       |    6 +++--
 arch/x86/kvm/vmx.c       |    2 +-
 arch/x86/kvm/x86.c       |   54 +++++++++++++++++++++++-----------------------
 include/linux/kvm_host.h |   17 +++++++++-----
 4 files changed, 42 insertions(+), 37 deletions(-)

diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
index f75af40..7a41878 100644
--- a/arch/x86/kvm/svm.c
+++ b/arch/x86/kvm/svm.c
< at >< at > -3185,8 +3185,8 < at >< at > static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
 break;
 case MSR_IA32_DEBUGCTLMSR:
 if (

Re: Guest hangs after some stress tests

Marcelo Tosatti — 2012-05-25T15:37:57

If kvmclock is not enabled a number of problems arise (miscalibrated
loops per jiffy, sched_clock using tsc's out of sync on smp). Should
set the hypervisor cpuid bit (see 49ab56ac6e1b907b7da).

--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo< at >vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v4 13/16] net: Make "info network" output more readable info

Zhi Yong Wu — 2012-05-25T14:56:29

OK, done
No matter. :)

Re: [PATCH v3] uq/master: Expose CPUID leaf 7 only for -cpu host

Eduardo Habkost — 2012-05-25T14:56:07

Fixed and sent v4.

[PATCH v4] uq/master: Expose CPUID leaf 7 only for -cpu host

Eduardo Habkost — 2012-05-25T14:55:09

Changes v3 -> v4:
  - Wrap line at cpu_x86_fill_host() to make checkpatch.pl happy

Changes v2 -> v3;
  - Check for kvm_enabled() before setting cpuid_7_0_ebx_features

Changes v1 -> v2:
  - Use kvm_arch_get_supported_cpuid() instead of host_cpuid() on
    cpu_x86_fill_host().

  We should use GET_SUPPORTED_CPUID for all bits on "-cpu host"
  eventually, but I am not changing all the other CPUID leaves because
  we may not be able to test such an intrusive change in time for 1.1.

Description of the bug:

Since QEMU 0.15, the CPUID information on CPUID[EAX=7,ECX=0] is being
returned unfiltered to the guest, directly from the GET_SUPPORTED_CPUID
return value.

The problem is that this makes the resulting CPU feature flags
unpredictable and dependent on the host CPU and kernel version. This
breaks live-migration badly if migrating from a host CPU that supports
some features on that CPUID leaf (running a recent kernel) to a kernel
or host CPU that doesn't support it.

Migration also is incorrect (the virtual CP

Re: [PATCH v4 13/16] net: Make "info network" output more readable info

Jan Kiszka — 2012-05-25T14:50:50

Then non-hub needs adjustment to a single space as well. Just remove the
two spaces I added to indent the categories (VLAN vs. non-VLAN listing)
for _both_ peer lines.

(I love nit-picking discussions :) )

Jan

Re: [PATCH v4 13/16] net: Make "info network" output more readable info

Zhi Yong Wu — 2012-05-25T14:44:38

Yeah, but it is three spaces, not one. You can check the code in
do_info_network. For non-hub peers, it is also three spaces.
That is the result i got when i was running the code.

Re: [PATCH v3] uq/master: Expose CPUID leaf 7 only for -cpu host

Jan Kiszka — 2012-05-25T14:42:00

checkpatch.pl?

Jan

Re: [PATCH v4 13/16] net: Make "info network" output more readable info

Jan Kiszka — 2012-05-25T14:40:05

This line is for printing peers attached to a hub, isn't it? So it
should be exactly one space, like for the non-hub peers. You should see
it better than I when running the code.

Jan

Re: [PATCH v4 13/16] net: Make "info network" output more readable info

Zhi Yong Wu — 2012-05-25T14:25:56

I have noticed this, and split out them to other patch.
This will have the following layout:
(qemu) info network
hub 1
   \ dump.0: type=dump,dump to qemu-vlan1.pcap (len=65536)
   \ user.1: type=user,net=10.0.2.0,restrict=off
   \ virtio-net-pci.0: type=nic,model=virtio-net-pci,macaddr=52:54:00:12:34:58
hub 0
   \ user.0: type=user,net=10.0.2.0,restrict=off
   \ e1000.0: type=nic,model=e1000,macaddr=52:54:00:12:34:57
virtio-net-pci.1: type=nic,model=virtio-net-pci,macaddr=52:54:00:12:34:56
   \ ur: type=tap,ifname=tap0,script=/etc/qemu-ifup,downscript=/etc/qemu-ifdown
^^^
It is three space, not two.
Do you mean that it should be like "monitor_printf(mon, "\\ ");" but
this indention will not match with those peers in hub case.

Re: [PATCH] PCI: Mark INTx masking support of Chelsio T310 10GbE NIC as broken

Jan Kiszka — 2012-05-25T14:20:25

Sorry, it's documented here:
http://thread.gmane.org/gmane.comp.emulators.kvm.devel/91388. Will
repost with this information included once Alexey provided his tested-by.

Jan

Re: [PATCH v4 13/16] net: Make "info network" output more readable info

Jan Kiszka — 2012-05-25T14:17:39

Please don't put reviewed-by tags here before the reviewer had a chance
to look at the code.


This should be obsolete now.


Why introduce a different hub output format at all? Do it in the final
right from the start.


I bet those two prototypes are required by some other patch (or are even
redundant).


Two space too much of indention (you remove them above for the existing
peer \ peer outputs).


Jan

Re: [PATCH] PCI: Mark INTx masking support of Chelsio T310 10GbE NIC as broken

Bjorn Helgaas — 2012-05-25T14:11:53

Please include a reference to a bugzilla, mailing list discussion, or
other details about how this was found and debugged.  Thanks!

--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo< at >vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v4 13/16] net: Make "info network" output more readable info

2012-05-25T14:11:04

From: Zhi Yong Wu <wuzhy< at >linux.vnet.ibm.com>

Reviewed-by:   Jan Kiszka  <jan.kiszka< at >siemens.com>
Signed-off-by: Zhi Yong Wu <wuzhy< at >linux.vnet.ibm.com>
---
 net.c     |   18 ++++++++++++++----
 net.h     |    1 +
 net/hub.c |   23 +++++++++++++++++++++--
 net/hub.h |    1 +
 4 files changed, 37 insertions(+), 6 deletions(-)

diff --git a/net.c b/net.c
index 61dc28d..ae0deec 100644
--- a/net.c
+++ b/net.c
< at >< at > -887,6 +887,12 < at >< at > static const struct {
         },
     },
 #endif /* CONFIG_NET_BRIDGE */
+    [NET_CLIENT_TYPE_HUB] = {
+        .type = "hubport",
+        .desc = {
+            { /* end of list */ }
+        },
+    },
 };
 
 int net_client_init(Monitor *mon, QemuOpts *opts, int is_netdev)
< at >< at > -1068,7 +1074,7 < at >< at > int do_netdev_del(Monitor *mon, const QDict *qdict, QObject **ret_data)
     return 0;
 }
 
-static void print_net_client(Monitor *mon, NetClientState *vc)
+void print_net_client(Monitor *mon, NetClientState *vc)
 {
     monitor_printf(mon, "%s: type=%s,%s\n", vc->name,
                    ne

[PATCH v3] uq/master: Expose CPUID leaf 7 only for -cpu host

Eduardo Habkost — 2012-05-25T14:12:05

[re-submitting to get this in through uq/master]

Changes v2 -> v3;
  - Check for kvm_enabled() before setting cpuid_7_0_ebx_features

Changes v1 -> v2:
  - Use kvm_arch_get_supported_cpuid() instead of host_cpuid() on
    cpu_x86_fill_host().

  We should use GET_SUPPORTED_CPUID for all bits on "-cpu host"
  eventually, but I am not changing all the other CPUID leaves because
  we may not be able to test such an intrusive change in time for 1.1.

Description of the bug:

Since QEMU 0.15, the CPUID information on CPUID[EAX=7,ECX=0] is being
returned unfiltered to the guest, directly from the GET_SUPPORTED_CPUID
return value.

The problem is that this makes the resulting CPU feature flags
unpredictable and dependent on the host CPU and kernel version. This
breaks live-migration badly if migrating from a host CPU that supports
some features on that CPUID leaf (running a recent kernel) to a kernel
or host CPU that doesn't support it.

Migration also is incorrect (the virtual CPU changes under the guest's
feet)

[PATCH v4 13/16] net: Make "info network" output more readable info

2012-05-25T14:02:09

From: Zhi Yong Wu <wuzhy< at >linux.vnet.ibm.com>

Reviewed-by:   Jan Kiszka  <jan.kiszka< at >siemens.com>
Signed-off-by: Zhi Yong Wu <wuzhy< at >linux.vnet.ibm.com>
---
 net.c     |   18 ++++++++++++++----
 net.h     |   12 ++++++++++++
 net/hub.c |   23 +++++++++++++++++++++--
 net/hub.h |    1 +
 4 files changed, 48 insertions(+), 6 deletions(-)

diff --git a/net.c b/net.c
index 61dc28d..ae0deec 100644
--- a/net.c
+++ b/net.c
< at >< at > -887,6 +887,12 < at >< at > static const struct {
         },
     },
 #endif /* CONFIG_NET_BRIDGE */
+    [NET_CLIENT_TYPE_HUB] = {
+        .type = "hubport",
+        .desc = {
+            { /* end of list */ }
+        },
+    },
 };
 
 int net_client_init(Monitor *mon, QemuOpts *opts, int is_netdev)
< at >< at > -1068,7 +1074,7 < at >< at > int do_netdev_del(Monitor *mon, const QDict *qdict, QObject **ret_data)
     return 0;
 }
 
-static void print_net_client(Monitor *mon, NetClientState *vc)
+void print_net_client(Monitor *mon, NetClientState *vc)
 {
     monitor_printf(mon, "%s: type=%s,%s\n", vc->name,

Re: [PATCH v3 13/16] net: Make the monitor output more reasonable hub info

Zhi Yong Wu — 2012-05-25T13:58:41

OK. sent out v4 for this patch.

Re: [Qemu-devel] [PATCH v3 00/16] net: hub-based networking

Luiz Capitulino — 2012-05-25T13:56:28

On Fri, 25 May 2012 15:47:28 +0200
Paolo Bonzini <pbonzini< at >redhat.com> wrote:


vlan is and the cleanest solution is to drop it.

But that's just my opnion, I won't (and possibly can't) nack this.
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo< at >vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v3 13/16] net: Make the monitor output more reasonable hub info

Jan Kiszka — 2012-05-25T13:49:42

Yes, this is what I was propose. You can just save the peer indentions
in the hub case.

Thanks!
Jan

Re: [Qemu-devel] [PATCH v3 00/16] net: hub-based networking

Luiz Capitulino — 2012-05-25T13:43:22

On Fri, 25 May 2012 15:37:15 +0200
Paolo Bonzini <pbonzini< at >redhat.com> wrote:


VDE allows this too :)


Let's start with what is hurting us.


I'm not sure. But note that openvswitch is a better alternative for linux.
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo< at >vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html