[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Tue, 12 Dec 2017 09:04:49 +0100
From: Vitaly Kuznetsov <vkuznets@...hat.com>
To: Roman Kagan <rkagan@...tuozzo.com>
Cc: kvm@...r.kernel.org, x86@...nel.org,
Paolo Bonzini <pbonzini@...hat.com>,
Radim Krčmář <rkrcmar@...hat.com>,
Thomas Gleixner <tglx@...utronix.de>,
Ingo Molnar <mingo@...hat.com>,
"H. Peter Anvin" <hpa@...or.com>,
"K. Y. Srinivasan" <kys@...rosoft.com>,
Haiyang Zhang <haiyangz@...rosoft.com>,
Stephen Hemminger <sthemmin@...rosoft.com>,
"Michael Kelley \(EOSG\)" <Michael.H.Kelley@...rosoft.com>,
Andy Lutomirski <luto@...nel.org>,
Mohammed Gamal <mmorsy@...hat.com>,
Cathy Avery <cavery@...hat.com>, linux-kernel@...r.kernel.org,
devel@...uxdriverproject.org
Subject: Re: [PATCH 3/6] x86/hyper-v: reenlightenment notifications support
Roman Kagan <rkagan@...tuozzo.com> writes:
> On Mon, Dec 11, 2017 at 10:56:33AM +0100, Vitaly Kuznetsov wrote:
>> Roman Kagan <rkagan@...tuozzo.com> writes:
>> > On Fri, Dec 08, 2017 at 11:49:57AM +0100, Vitaly Kuznetsov wrote:
>> >> +void register_hv_tsc_update(void (*cb)(void))
>> >> +{
>> >
>> > The function name seems unfortunate. IMHO such a name suggests
>> > registering a callback on a notifier chain (rather than unconditionally
>> > replacing the old callback), and having no other side effects.
>>
>> I see, any suggestion? register_hv_reenlightenment_cb? register_hv_tscchange_cb?
>
> IMHO arm_hv_reenlightenment_cb or arm_hv_tscchange_cb would be better,
> but I'm not very good at giving descriptive names.
>
I would probably try to avoid using 'arm' word in x86 code to assist
poor git-greppers :-) And we actually need a pair of functions
(enable/disable). I will probably go with
set_hv_tscchange_cb()
clear_hv_tscchange_cb()
in v2 unless there's a better suggestion.
>>
>> >
>> >> + struct hv_reenlightenment_control re_ctrl = {
>> >> + .vector = HYPERV_REENLIGHTENMENT_VECTOR,
>> >> + .enabled = 1,
>> >> + .target_vp = hv_vp_index[smp_processor_id()]
>> >> + };
>> >> + struct hv_tsc_emulation_control emu_ctrl = {.enabled = 1};
>> >> +
>> >> + if (!(ms_hyperv.features & HV_X64_ACCESS_REENLIGHTENMENT))
>> >> + return;
>> >
>> > What happens then? L2 guests keep running with their clocks ticking at
>> > a different speed?
>> >
>>
>> In reallity this never happens -- in case nested virtualization is
>> supported reenlightenment is also available. In theory, L0 can emulate
>> TSC acceess for forever after migration.
>
> I would think that Hyper-V only started rdtsc emulation if
> TSC_EMULATION_CONTROL was turned on, which wouldn't happen here.
>
Yes, this is the de-facto behavior I observe with WS2016.
> But indeed, normally this shouldn't be a problem. It may make sense
> just to issue a warning if the feature is unsupported, though.
Will do in v2, thanks.
--
Vitaly
Powered by blists - more mailing lists