| Summary: | guest kernel panic when cpu threads in same physical cpu are assigned to different numa node | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Product: | Red Hat Enterprise Linux 6 | Reporter: | Suqin Huang <shuang> | ||||||
| Component: | kernel | Assignee: | Lenny Szubowicz <lszubowi> | ||||||
| Status: | CLOSED CURRENTRELEASE | QA Contact: | Red Hat Kernel QE team <kernel-qe> | ||||||
| Severity: | medium | Docs Contact: | |||||||
| Priority: | medium | ||||||||
| Version: | 6.2 | CC: | areis, jshortt, juzhang, michen, mkenneth, qzhang, rbalakri, rpacheco, tburke, virt-maint | ||||||
| Target Milestone: | rc | ||||||||
| Target Release: | --- | ||||||||
| Hardware: | Unspecified | ||||||||
| OS: | Unspecified | ||||||||
| Whiteboard: | |||||||||
| Fixed In Version: | Doc Type: | Bug Fix | |||||||
| Doc Text: | Story Points: | --- | |||||||
| Clone Of: | |||||||||
| : | 1066286 (view as bug list) | Environment: | |||||||
| Last Closed: | 2015-10-08 18:13:50 UTC | Type: | --- | ||||||
| Regression: | --- | Mount Type: | --- | ||||||
| Documentation: | --- | CRM: | |||||||
| Verified Versions: | Category: | --- | |||||||
| oVirt Team: | --- | RHEL 7.3 requirements from Atomic Host: | |||||||
| Cloudforms Team: | --- | Target Upstream Version: | |||||||
| Bug Depends On: | |||||||||
| Bug Blocks: | 833130, 1066286, 1269638 | ||||||||
| Attachments: |
|
||||||||
|
Description
Suqin Huang
2011-11-25 03:25:53 UTC
Created attachment 536138 [details]
kernel panic
Can you please try to use an accurate numa topology setup (that the host have)? Apparently, with -node settings qemu provides numa settings that are wrong and the guest crashes (divides by zero) because of them. It's low priority but nice to fix. This request was not resolved in time for the current release. Red Hat invites you to ask your support representative to propose this request, if still desired, for consideration in the next release of Red Hat Enterprise Linux. This request was erroneously removed from consideration in Red Hat Enterprise Linux 6.4, which is currently under development. This request will be evaluated for inclusion in Red Hat Enterprise Linux 6.4. It's actually a guest bug, but probably low priority because it is triggered by an unusual configuration.
"-numa node -numa node" generates a round-robin VCPU configuration (VCPU 0 on node 0, VCPU 1 on node 1, VCPU 2 on node 0, VCPU 3 on node 1). I can reproduce it very easily if I use "-smp 4,threads=2 -numa node -numa node". I couldn't reproduce it using "-smp 4,cores=2 -numa node -numa node", even using the upstream QEMU.
Moving to kernel and attaching console output of a more recent kernel version. Reproduced using Fedora 17 qemu-kvm (qemu-kvm-1.0-18.fc17.x86_64). Probably it's low priority as having CPU threads on separate NUMA nodes is a very unusual situation.
The following command-line triggers a guest crash:
/usr/bin/qemu-kvm -enable-kvm -M pc -smp 4,cores=1,threads=2,sockets=2 -m 2048 -hda ~/system/vmachines/rhel63.img -net nic,model=rtl8139 -net user -numa node -numa node -serial file:/tmp/crash
The following boots properly:
/usr/bin/qemu-kvm -enable-kvm -M pc -smp 4,cores=2,threads=1,sockets=2 -m 2048 -hda ~/system/vmachines/rhel63.img -net nic,model=rtl8139 -net user -numa node -numa node
Guest is running kernel 2.6.32-287.el6. Crash is at:
(gdb) l *0xffffffff8105bcb1
0xffffffff8105bcb1 is in select_task_rq_fair (kernel/sched_fair.c:2173).
2168
2169 avg_load += load;
2170 }
2171
2172 /* Adjust by relative CPU power of the group */
2173 avg_load = (avg_load * SCHED_LOAD_SCALE) / group->cpu_power;
2174
2175 if (local_group) {
2176 this_load = avg_load;
2177 this = group;
(gdb)
Created attachment 602835 [details]
console output of 2.6.32-287.el6 crash
|