Hi Sergey,
On 15.08.19 12:17, Sergey Dyasli wrote:
Hi Juergen,
The latest round of testing revealed the following 3 Xen crashes:
1. vcpu_sleep_sync() <-- vlapic_init_sipi_action()
This was seen multiple times. It tends to happen on large Windows Server
VMs (>= 12 vCPUs).
https://paste.debian.net/1095844/
2. vcpu_sleep_sync() <-- vcpu_set_periodic_timer() <-- pv_hypercall()
This was seen on Sles 11.4 PV VM with 32 vCPUs and 6GB RAM
https://paste.debian.net/1095843/
3. ACPI S5: schedule.c#sched_slave()
This one was during shutdown again.
https://paste.debian.net/1095845/
There also were a number of Dom0 soft-lockups (together with RCU stalls)
but unfortunately our crashdump tool was a bit broken so full Dom0 kernel
logs were not captured this time.
And CentOS 5.11 suspend regression (as reported previously) is still
present.
I have found one bug in the v->is_running setting, which might be
related to your observed hangs, but I couldn't verify that.
I have rebased my series with the is_running fix to current staging. You
can find that under
git://github.com/jgross1/xen/ sched-v2b-2019-09-05
This branch includes the needed prerequisite series ("xen/sched: use new
idle scheduler for free cpus"), and some additional debug enhancements
I'd like you to use:
- my series for debugtrace enhancements
- my series for lock-profiling enhancements
- and as the last patch some debugtrace hooks in the scheduler
Please make sure to have CONFIG_DEBUG_TRACE and
CONFIG_DEBUG_LOCK_PROFILE set when building the hypervisor.
When testing please add:
async-show-all=on watchdog=on debugtrace=cpu:16
to the hypervisor boot parameters. This should add a lot of additional
output in case of a crash, which will (hopefully) help to find the
reason(s) for hangs and crashes.
Juergen
___
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel