Bug 1238574
| Summary: | qemu-kvm does not expose expected cpu topology to guest when wrong cpu topology is defined. | ||
|---|---|---|---|
| Product: | Red Hat Enterprise Linux 7 | Reporter: | Dan Zheng <dzheng> |
| Component: | qemu-kvm-rhev | Assignee: | Thomas Huth <thuth> |
| Status: | CLOSED ERRATA | QA Contact: | Virtualization Bugs <virt-bugs> |
| Severity: | medium | Docs Contact: | |
| Priority: | medium | ||
| Version: | 7.2 | CC: | dgibson, dyuan, dzheng, gsun, hannsj_uhl, jen, juzhang, knoel, lhuang, lvivier, michen, mrezanin, mzhan, ngu, qzhang, Robert.Hu, thuth, virt-maint, xuma |
| Target Milestone: | rc | ||
| Target Release: | --- | ||
| Hardware: | ppc64le | ||
| OS: | Linux | ||
| Whiteboard: | |||
| Fixed In Version: | qemu-kvm-rhev-2.5.0-1.el7 | Doc Type: | Bug Fix |
| Doc Text: | Story Points: | --- | |
| Clone Of: | Environment: | ||
| Last Closed: | 2016-11-07 20:26:20 UTC | Type: | Bug |
| Regression: | --- | Mount Type: | --- |
| Documentation: | --- | CRM: | |
| Verified Versions: | Category: | --- | |
| oVirt Team: | --- | RHEL 7.3 requirements from Atomic Host: | |
| Cloudforms Team: | --- | Target Upstream Version: | |
| Embargoed: | |||
| Bug Depends On: | |||
| Bug Blocks: | 1359843 | ||
Dan, Is this problem specific to ppc64le? Or, does it also happen on X86_64? Thanks. Your understanding the -smp options is not correct. -smp n,sockets=s,cores=c,threads=t n = total number of threads in the whole system s = total number of sockets in the system c = number of cores per socket t = number of threads per core If n != s * c * t, then n takes precedence over s. So, qemu will allocate all the threads in the first core, then all the threads in the next core until the total number of threads is reached. So, the expected topology is 1/2/2 which is what you see. In scenario B, qemu will allocate all 4 threads in core 0, then 2 of the 4 threads in core 1 before running out of total threads. Because the guest expects cores to have an equal number of threads, this is represented as 2 cores of 4 threads each, but the last two threads of core1 are offline as shown in the lscpu output. I think "3 threads per core" value shown by lscpu is just lscpu being confused by the strange situation of having cores with a different number of online threads each. Re-test above two scenarios on x86. The two results are as expected. Here is the bug tracking similar matter for x86. https://bugzilla.redhat.com/show_bug.cgi?id=689665 Version-Release number of selected component (if applicable): libvirt-1.2.17-1.el7.x86_64 qemu-kvm-rhev-2.3.0-5.el7.x86_64 kernel-3.10.0-221.el7.x86_64 Scenario A: In guest: [root@localhost ~]# lscpu Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Byte Order: Little Endian CPU(s): 4 On-line CPU(s) list: 0-3 Thread(s) per core: 1 Core(s) per socket: 4 Socket(s): 1 NUMA node(s): 1 Vendor ID: GenuineIntel CPU family: 6 Scenario B: In guest: [root@localhost ~]# lscpu Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Byte Order: Little Endian CPU(s): 6 On-line CPU(s) list: 0-5 Thread(s) per core: 2 Core(s) per socket: 3 Socket(s): 1 NUMA node(s): 1 Vendor ID: GenuineIntel CPU family: 6 (In reply to Karen Noel from comment #2) > Dan, Is this problem specific to ppc64le? Or, does it also happen on X86_64? > > Thanks. This problem is specific to ppc64le now. Similar problem on X86_64 has been fixed before. See https://bugzilla.redhat.com/show_bug.cgi?id=689665 I just had a look at this bug ticket, and I wonder whether there is a "right" topology in this case at all ... I mean, if the user provides wrong input, there can hardly be a real right output, can it? IMHO QEMU should simply exit with an error message when this misconfiguration occured. Would that be an acceptable solution for you, too, Dan Zheng? Then we could try to provide an according patch (which rejects wrong configurations) upstream first and then backport it if it is accepted. (In reply to Thomas Huth from comment #6) > I just had a look at this bug ticket, and I wonder whether there is a > "right" topology in this case at all ... I mean, if the user provides wrong > input, there can hardly be a real right output, can it? > > IMHO QEMU should simply exit with an error message when this > misconfiguration occured. Would that be an acceptable solution for you, too, > Dan Zheng? Then we could try to provide an according patch (which rejects > wrong configurations) upstream first and then backport it if it is accepted. It sounds reasonable. It is a good idea to reject this kind of wrong configurations. Thank you. Ok, I'll try to get a check for such wrong configurations into upstream... I've now suggested a patch upstream: http://lists.nongnu.org/archive/html/qemu-devel/2015-07/msg04549.html Thomas' patch is upstream in time for qemu-2.4, so we should get it it RHEL 7.3 with the first rebase. (In reply to David Gibson from comment #11) > Thomas' patch is upstream in time for qemu-2.4, so we should get it it RHEL > 7.3 with the first rebase. As far as I can see, it will be in QEMU 2.5 ... QEMU 2.4 does not contain the patch yet. Ah, yes, sorry. Off-by-one error in my thinking. Reproduced the issue on old version: Version-Release number of selected component (if applicable): Guest kernel: 3.10.0-327.el7.ppc64le qemu-kvm-rhev:qemu-kvm-rhev-2.3.0-31.el7_2.8.ppc64le Host kernel:3.10.0-418.el7.ppc64le Steps to Reproduce: 1. start a guest with command: /usr/libexec/qemu-kvm \ -name xuma-vm \ -smp 4,sockets=1,cores=4,threads=2 \ -m 8192 \ -realtime mlock=off \ -monitor stdio \ -rtc base=localtime,clock=host \ -boot strict=on \ -vnc 0:59 \ -qmp tcp:0:9999,server,nowait \ \ -device virtio-scsi-pci,bus=pci.0,addr=0x5 \ \ -device scsi-hd,id=scsi-hd0,drive=scsi-hd-dr0,bootindex=0 \ -drive file=RHEL-7.2.qcow2,if=none,id=scsi-hd-dr0,format=qcow2,cache=none \ \ -device virtio-net-pci,netdev=net0,id=nic0,mac=52:54:00:c4:e7:84 \ -netdev tap,id=net0,script=/etc/qemu-ifup,downscript=/etc/qemu-ifdown,vhost=on \ 2.check vcpu in the guest with command lscpu: Architecture: ppc64le Byte Order: Little Endian CPU(s): 4 On-line CPU(s) list: 0-3 Thread(s) per core: 2 Core(s) per socket: 2 Socket(s): 1 NUMA node(s): 1 Model: IBM pSeries (emulated by qemu) L1d cache: 64K L1i cache: 32K NUMA node0 CPU(s): 0-3 3.change vpcu command to "-smp 6,sockets=1,cores=3,threads=4" then check in guest: Architecture: ppc64le Byte Order: Little Endian CPU(s): 8 On-line CPU(s) list: 0-5 Off-line CPU(s) list: 6,7 Thread(s) per core: 3 Core(s) per socket: 2 Socket(s): 1 NUMA node(s): 1 Model: IBM pSeries (emulated by qemu) L1d cache: 64K L1i cache: 32K NUMA node0 CPU(s): 0-5 Actual result: socket/cores/threads in guest is 1/2/2 socket/cores/threads in guest is 1/2/3 Expected result: socket/cores/threads in guest is 1/4/1 socket/cores/threads in guest is 1/3/2 Verified the issue on the latest build: Version-Release number of selected component (if applicable): Guest kernel: 3.10.0-327.el7.ppc64le qemu-kvm-rhev:qemu-kvm-rhev-2.6.0-4.el7.ppc64le Host kernel:3.10.0-418.el7.ppc64le Steps to Verify: 1. start a guest with command: /usr/libexec/qemu-kvm \ -name xuma-vm \ -smp 4,sockets=1,cores=4,threads=2 \ -m 8192 \ -realtime mlock=off \ -monitor stdio \ -rtc base=localtime,clock=host \ -boot strict=on \ -vnc 0:59 \ -qmp tcp:0:9999,server,nowait \ \ -device virtio-scsi-pci,bus=pci.0,addr=0x5 \ \ -device scsi-hd,id=scsi-hd0,drive=scsi-hd-dr0,bootindex=0 \ -drive file=RHEL-7.2.qcow2,if=none,id=scsi-hd-dr0,format=qcow2,cache=none \ \ -device virtio-net-pci,netdev=net0,id=nic0,mac=52:54:00:c4:e7:84 \ -netdev tap,id=net0,script=/etc/qemu-ifup,downscript=/etc/qemu-ifdown,vhost=on \ 2.change vpcu command to "-smp 6,sockets=1,cores=3,threads=4" then run the commands: Results: can't run the commands and display message as following: qemu-kvm: cpu topology: sockets (1) * cores (4) * threads (2) > maxcpus (4) qemu-kvm: cpu topology: sockets (1) * cores (3) * threads (4) > maxcpus (6) so the bus was fixed. Since the problem described in this bug report should be resolved in a recent advisory, it has been closed with a resolution of ERRATA. For information on the advisory, and where to find the updated files, follow the link below. If the solution does not work for you, open a new bug report. https://rhn.redhat.com/errata/RHBA-2016-2673.html |
Description of problem: Define wrong cpu topology and qemu-kvm does not expose correct cpu topology to guest as expected. Test two scenarios: A: -smp 4,sockets=1,cores=4,threads=2 B: -smp 6,sockets=1,cores=3,threads=4 Version-Release number of selected component (if applicable): libvirt-1.2.16-1.el7.ppc64le qemu-kvm-rhev-2.3.0-5.el7.ppc64le kernel-3.10.0-282.el7.ppc64le How reproducible: 100% Steps to Reproduce: Scenario A: 1.Edit XML with below setting: ....... <vcpu placement='static'>4</vcpu> ...... <cpu> <topology sockets='1' cores='4' threads='2'/> </cpu> 2. check qemu process, and parameters are set correctly by libvirt. # ps -ef|grep qemu-kvm qemu 21296 ... -smp 4,sockets=1,cores=4,threads=2 ...... 3. Use following formula to calculate supposed socket/cores/threads value: <== Is this current qemu realization? smp = n (argument to -smp) nr_threads = smp / (cores * sockets) nr_cores = cores smp = n which is 4 nr_threads = 4 /(4 * 1) which is 1 nr_cores = cores which is 4 So the expected socket/cores/threads in guest is 1/4/1 4.login guest: [root@localhost ~]# lscpu Architecture: ppc64le Byte Order: Little Endian CPU(s): 4 On-line CPU(s) list: 0-3 Thread(s) per core: 2 Core(s) per socket: 2 Socket(s): 1 NUMA node(s): 1 Model: IBM pSeries (emulated by qemu) L1d cache: 64K L1i cache: 32K NUMA node0 CPU(s): 0-3 Actual result: socket/cores/threads in guest is 1/2/2 The #cores is changed which is unexpected. Expected result: socket/cores/threads in guest is 1/4/1 Scenario B： 1. Use below setting: ....... <vcpu placement='static'>6</vcpu> ...... <cpu> <topology sockets='1' cores='3' threads='4'/> </cpu> 2. Check qemu command line and is correct -smp 6,sockets=1,cores=3,threads=4 3. Check in guest [root@localhost ~]# lscpu Architecture: ppc64le Byte Order: Little Endian CPU(s): 8 On-line CPU(s) list: 0-5 Off-line CPU(s) list: 6,7 Thread(s) per core: 3 Core(s) per socket: 2 Socket(s): 1 NUMA node(s): 1 Model: IBM pSeries (emulated by qemu) L1d cache: 64K L1i cache: 32K NUMA node0 CPU(s): 0-5 Actual result: socket/cores/threads in guest is 1/2/3 CPU(s) :8 Off-line CPU(s) list: 6,7 Expected result: CPU(s): 6 No offline CPU socket/cores/threads in guest is 1/3/2 Additional information: -- see attachment for libvirtd.log