-
Bug
-
Resolution: Unresolved
-
Undefined
-
None
-
CNV v4.20.1
-
None
-
Quality / Stability / Reliability
-
0.42
-
False
-
-
False
-
None
-
-
Important
-
None
Description of problem:
If qemu-kvm hits a segfault in CNV, we are unable to get the coredump to analyze.
Nov 26 23:40:20 white.shift.home.arpa systemd-coredump[12563]: Resource limits disable core dumping for process 11794 (qemu-kvm).
This is critical for our ability to help customers. Especially for hard to reproduce issues.
If libvirt (PID 24214) segfaults we get a coredump, but not qemu-kvm (24309)
# libvirt $ grep "Max core" /proc/24214/limits Max core file size unlimited unlimited bytes # qemu $ grep "Max core" /proc/24309/limits Max core file size 0 0 bytes
This is because func configureQemuConf(qemuFilename string) does not set max_core by default (so its zero), and newer libvirts do not pass its own limits to qemu-kvm. But CNV does not provide any way to set it. So libvirt launches qemu-kvm with RLIMIT_CORE=0. See qemu: add a max_core setting to qemu.conf for core dump size
Version-Release number of selected component (if applicable):
4.20, 4.19, 4.18, probably some earlier ones too depending on libvirt version?
How reproducible:
Always
Steps to Reproduce:
1. Set https://access.redhat.com/solutions/7081188 2. Start a VM 3. Send SIGSEGV signal to qemu-kvm
Actual results:
Unable to configure the system so we can get a core from qemu-kvm
Expected results:
Configure or allow configuration so we can get a coredump
- relates to
-
RHEL-131553 qemu-kvm segfaults when using block-latency-histogram
-
- New
-