Bug 513317
| Summary: | PCI passthrough with kvm guest cause libvirtd dead | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Product: | Red Hat Enterprise Linux 5 | Reporter: | zhanghaiyan <yoyzhang> | ||||||||||||
| Component: | libvirt | Assignee: | Daniel Berrangé <berrange> | ||||||||||||
| Status: | CLOSED ERRATA | QA Contact: | Virtualization Bugs <virt-bugs> | ||||||||||||
| Severity: | medium | Docs Contact: | |||||||||||||
| Priority: | low | ||||||||||||||
| Version: | 5.4 | CC: | berrange, llim, markmc, mshao, sghosh, veillard, virt-maint, xen-maint | ||||||||||||
| Target Milestone: | rc | ||||||||||||||
| Target Release: | --- | ||||||||||||||
| Hardware: | All | ||||||||||||||
| OS: | Linux | ||||||||||||||
| Whiteboard: | |||||||||||||||
| Fixed In Version: | Doc Type: | Bug Fix | |||||||||||||
| Doc Text: | Story Points: | --- | |||||||||||||
| Clone Of: | Environment: | ||||||||||||||
| Last Closed: | 2009-09-02 09:22:39 UTC | Type: | --- | ||||||||||||
| Regression: | --- | Mount Type: | --- | ||||||||||||
| Documentation: | --- | CRM: | |||||||||||||
| Verified Versions: | Category: | --- | |||||||||||||
| oVirt Team: | --- | RHEL 7.3 requirements from Atomic Host: | |||||||||||||
| Cloudforms Team: | --- | Target Upstream Version: | |||||||||||||
| Embargoed: | |||||||||||||||
| Attachments: |
|
||||||||||||||
I can't reproduce this. Can you attempt to capture a stack trace, - Install libvirt-debuginfo RPM - Run 'service libvirtd start' - Run 'ps -auxfw | grep libvirtd' to find the PID of the libvirtd process - Start 'gdb' - In the gdb console, type 'attach <PID-OF-LIBVIRTD>' and then 'cont' Now in another console attempt to run your test to make libvirtd crash. When it crashes, go back to the GDB console and type 'thread apply all backtrace' And then upload all the data from that as an attachment to this bug. Can you also provide the output of 'virsh nodedev-list --tree' And 'virsh nodedev-dumpxml pci_8086_10bd' And finally, the full XML config of the guest, and any /var/log/libvirt/qemu/demo.log that may exist Now, the test result is a little different
after step4 # virsh start demo
It hangs.
Attached gdb.log
nodedev-list
nodedev-dumpxml
kvm-test1.xml
test1.log
Created attachment 354972 [details]
gdb.log
Created attachment 354973 [details]
test1-kvm.xml
Created attachment 354974 [details]
test1.log
Created attachment 354975 [details]
nodedev-list
Created attachment 354976 [details]
nodedev-dumpxml
excerpt from stack trace:
#0 0x000000386747268e in free () from /lib64/libc.so.6
#1 0x0000003a66e1865c in virFree (ptrptr=<value optimized out>)
at memory.c:177
#2 0x0000003a66e1890a in pciReadDeviceID (dev=<value optimized out>,
---Type <return> to continue, or q <return> to quit---
id_name=<value optimized out>) at pci.c:839
#3 0x0000003a66e18a01 in pciGetDevice (conn=<value optimized out>,
domain=<value optimized out>, bus=<value optimized out>,
slot=<value optimized out>, function=<value optimized out>) at pci.c:875
#4 0x0000000000420d15 in qemudStartVMDaemon (conn=0x5e30f50,
driver=0x5d8f950, vm=0x5e31090, migrateFrom=0x0, stdin_fd=-1)
at qemu_driver.c:1251
Ah, that will probably be this upstream bug fix http://libvirt.org/git/?p=libvirt.git;a=commit;h=4a7acedd3c59a6a750576cb8680bc3f08fe0b52c IIRC it triggers if you configure a PCI device that does not actually exist on the host Yep, confirmed here
The device being attached has a slot '25' (decimal)
<device>
<name>pci_8086_10bd</name>
<parent>computer</parent>
<capability type='pci'>
<domain>0</domain>
<bus>0</bus>
<slot>25</slot>
<function>0</function>
<product id='0x10bd'>82566DM-2 Gigabit Network Connection</product>
<vendor id='0x8086'>Intel Corporation</vendor>
</capability>
</device>
The guest XML has been configure with slot 0x25 (hexidecimal).
<source>
<address bus='0x00' slot='0x25' function='0x00'/>
</source>
It is a shame the node-device XML prints decimal, but not hexidecimal when dumping XML, but that's life. Upon reading XML both domain & nodedevice XML accept any number base.
So changing the XML to
<source>
<address bus='00' slot='25' function='0'/>
</source>
should avoid the crash, but clearly we should still fix this.
I tried with XML
<source>
<address bus='00' slot='25' function='0'/>
</source>
YES, can passthrough PCI successfully.
libvirt-0.6.3-17.el5 has been built in dist-5E-qu-candidate with the fix Daniel Verified on libvirt-0.6.3-15.el5, cannot reproduce this bug Update comment #16. Verified on libvirt-0.6.3-17.el5, cannot reproduce this bug An advisory has been issued which should help the problem described in this bug report. This report is therefore being closed with a resolution of ERRATA. For more information on therefore solution and/or where to find the updated files, please follow the link below. You may reopen this bug report if the solution does not work for you. http://rhn.redhat.com/errata/RHEA-2009-1269.html |
Description of problem: PCI passthrough with kvm guest fail and can cause libvirtd dead Version-Release number of selected component (if applicable): - libvirt-0.6.3-15.el5 - xen-3.0.3-90.el5 - kvm-83-90.el5 - rhel-5.4 (2.6.18-158.el5) How reproducible: 100% Steps to Reproduce: 1.# virsh nodedev-dettach pci_8086_10bd Device pci_8086_10bd dettached 2.# virsh nodedev-reset pci_8086_10bd Device pci_8086_10bd reset 3.# virsh edit demo Domain demo XML configuration edited. <hostdev mode='subsystem' type='pci' managed='no'> <source> <address bus='0x00' slot='0x25' function='0x00'/> </source> </hostdev> 4. # virsh start demo error: Failed to start domain demo error: server closed connection 5.# virsh error: unable to connect to '/var/run/libvirt/libvirt-sock': Connection refused error: failed to connect to the hypervisor 6.# service libvirtd status libvirtd dead but pid file exists Actual results: PCI passthrough fail and cause libvirtd dead Expected results: PCI passthrough success Additional info: