Re: [External] Re: [RESEND RFC: timer passthrough 0/9] Support timer passthrough for VM

[Zhimin Feng <fengzhimin@bytedance.com>]

Hi

The host timer would be saved when cpu entry the non-root mode, and it 
would be restored when cpu entry the root mode. So the guest doesn't the 
host timer.

The host timer would be written to the preemption timer in non-root 
mode. When the host timer is expired(preemption timer value is '0'), the 
preemption timer would trigger immediate VMExit, so the host timer would 
be handled in the preemption timer handler.

Thanks!

Zhimin

在 2021/2/9 上午2:13, Konrad Rzeszutek Wilk 写道:
> On Fri, Feb 05, 2021 at 06:03:08PM +0800, Zhimin Feng wrote:
>> The main motivation for this patch is to improve the performance of VM.
>> This patch series introduces how to enable the timer passthrough in
>> non-root mode.
> Nice! Those are impressive numbers!
>
>> The main idea is to offload the host timer to the preemtion timer in
>> non-root mode. Through doing this, guest can write tscdeadline msr directly
>> in non-root mode and host timer isn't lost. If CPU is in root mode,
>> guest timer is switched to software timer.
> I am sorry - but I am having a hard time understanding the sentence
> above so let me ask some specific questions.
>
> - How do you protect against the guest DoS-ing the host and mucking with
>    the host timer?
>
> - As in can you explain how the host can still continue scheduling it's
>    own quanta?
>
> And one more - what happens with Live Migration? I would assume that
> becomes a no-go anymore unless you swap in the guest timer back in? So
> we end up emulating the MSR again?
>
> Thanks!
>
>> Testing on Intel(R) Xeon(R) Platinum 8260 server.
>>
>> The guest OS is Debian(kernel: 4.19.28). The specific configuration is
>>   is as follows: 8 cpu, 16GB memory, guest idle=poll
>> memcached in guest(memcached -d -t 8 -u root)
>>
>> I use the memtier_benchmark tool to test performance
>> (memtier_benchmark -P memcache_text -s guest_ip -c 16 -t 32
>>   --key-maximum=10000000000 --random-data --data-size-range=64-128 -p 11211
>>   --generate-keys --ratio 5:1 --test-time=500)
>>
>> Total Ops can be improved 25% and Avg.Latency can be improved 20% when
>> the timer-passthrough is enabled.
>>
>> =============================================================
>>                 | Enable timer-passth | Disable timer-passth |
>> =============================================================
>> Totals Ops/sec |    514869.67        |     411766.67        |
>> -------------------------------------------------------------
>> Avg.Latency    |    0.99483          |     1.24294          |
>> =============================================================
>>
>>
>> Zhimin Feng (9):
>>    KVM: vmx: hook set_next_event for getting the host tscd
>>    KVM: vmx: enable host lapic timer offload preemtion timer
>>    KVM: vmx: enable passthrough timer to guest
>>    KVM: vmx: enable passth timer switch to sw timer
>>    KVM: vmx: use tsc_adjust to enable tsc_offset timer passthrough
>>    KVM: vmx: check enable_timer_passth strictly
>>    KVM: vmx: save the initial value of host tscd
>>    KVM: vmx: Dynamically open or close the timer-passthrough for pre-vm
>>    KVM: vmx: query the state of timer-passth for vm
>>
>>   arch/x86/include/asm/kvm_host.h |  27 ++++
>>   arch/x86/kvm/lapic.c            |   1 +
>>   arch/x86/kvm/vmx/vmx.c          | 331 +++++++++++++++++++++++++++++++++++++++-
>>   arch/x86/kvm/x86.c              |  26 +++-
>>   include/linux/kvm_host.h        |   1 +
>>   include/uapi/linux/kvm.h        |   3 +
>>   kernel/time/tick-common.c       |   1 +
>>   tools/include/uapi/linux/kvm.h  |   3 +
>>   virt/kvm/kvm_main.c             |   1 +
>>   9 files changed, 389 insertions(+), 5 deletions(-)
>>
>> -- 
>> 2.11.0
>>