Re: [PATCH v2 0/3] KVM: VMX: Fix for kexec VMCLEAR and VMXON cleanup

From: Baoquan He
Date: Wed Apr 08 2020 - 21:20:21 EST


On 04/08/20 at 09:44pm, Vitaly Kuznetsov wrote:
> Baoquan He <bhe@xxxxxxxxxx> writes:
>
> > On 04/07/20 at 02:04pm, Vitaly Kuznetsov wrote:
> >> Baoquan He <bhe@xxxxxxxxxx> writes:
> >>
> >> >
> >> > The trace is here.
> >> >
> >> > [ 132.480817] RIP: 0010:crash_vmclear_local_loaded_vmcss+0x57/0xd0 [kvm_intel]
> >>
> >> This is a known bug,
> >>
> >> https://lore.kernel.org/kvm/20200401081348.1345307-1-vkuznets@xxxxxxxxxx/
> >
> > Thanks for telling, Vitaly.
> >
> > I tested your patch, it works.
> >
> > One thing is I noticed a warning message when your patch is applied. When
> > I changed back to revert this patchset, didn't found this message. I didn't
> > look into the detail of network core code and the kvm vmx code, maybe it's
> > not relevant.
> >
> >
> > [ 3708.629234] Type was not set for devlink port.
> > [ 3708.629258] WARNING: CPU: 3 PID: 60 at net/core/devlink.c:7164 devlink_port_type_warn+0x11/0x20
> > [ 3708.632328] Modules linked in: rfkill sunrpc intel_powerclamp coretemp kvm_intel kvm irqbypass intel_cstate iTCO_wdt hpwdt intel_uncore gpio_ich iTCO_vendor_support pcspkr ipmi_ssif hpilo lpc_ich ipmi_si ipmi_devintf ipmi_msghandler acpi_power_meter pcc_cpufreq i7core_edac ip_tables xfs libcrc32c radeon i2c_algo_bit drm_kms_helper cec ttm crc32c_intel serio_raw drm ata_generic pata_acpi mlx4_core bnx2 hpsa scsi_transport_sas
> > [ 3708.640782] CPU: 3 PID: 60 Comm: kworker/3:1 Kdump: loaded Tainted: G I 5.6.0+ #1
> > [ 3708.642715] Hardware name: HP ProLiant DL380 G6, BIOS P62 08/16/2015
> > [ 3708.644222] Workqueue: events devlink_port_type_warn
> > [ 3708.645349] RIP: 0010:devlink_port_type_warn+0x11/0x20
>
> What's in the patchset you're testing? Is it Sean's series + my patch,
> or just my patch? In case it's the later I'm having hard times trying to
> see how this can be related, but in case it's the former the fact that
> we do stuff a little bit differently on kexec may actually be triggering
> the issue above. I still think that it's not causing it, just
> triggering.

I am testing on Linus's tree, this patchset is already there. I just
reverted these patchset, or apply your patch on top of it. Both of them
works. The devlink warning message is not related to this issue because
I found it too when this patchset are reverted.

While I would suggest adding kexec@xxxxxxxxxxxxxxxxxxx when code changes
are related to kexec/kdump since we usually watch this mailing list.
LKML contains too many mails, we may miss this kind of change, have to
debug and test again. Thanks.

Baoquan