From: Avi Kivity <avi@redhat.com>
To: Jeremy Fitzhardinge <jeremy@goop.org>
Cc: Dan Magenheimer <dan.magenheimer@oracle.com>,
Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>,
kurt.hackel@oracle.com, the arch/x86 maintainers <x86@kernel.org>,
Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
Glauber de Oliveira Costa <gcosta@redhat.com>,
Xen-devel <xen-devel@lists.xensource.com>,
Keir Fraser <keir.fraser@eu.citrix.com>,
Zach Brown <zach.brown@oracle.com>,
Chris Mason <chris.mason@oracle.com>
Subject: Re: [Xen-devel] Re: [PATCH 3/5] x86/pvclock: add vsyscall implementation
Date: Mon, 12 Oct 2009 20:29:57 +0200 [thread overview]
Message-ID: <4AD375A5.8050205@redhat.com> (raw)
In-Reply-To: <4AD3738B.6050200@goop.org>
On 10/12/2009 08:20 PM, Jeremy Fitzhardinge wrote:
> On 10/10/09 11:10, Avi Kivity wrote:
>
>> On 10/10/2009 02:24 AM, Jeremy Fitzhardinge wrote:
>>
>>> On 10/07/09 03:25, Avi Kivity wrote:
>>>
>>>
>>>> def try_pvclock_vtime():
>>>> tsc, p0 = rdtscp()
>>>> v0 = pvclock[p0].version
>>>> tsc, p = rdtscp()
>>>> t = pvclock_time(pvclock[p], tsc)
>>>> if p != p0 or pvclock[p].version != v0:
>>>> raise Exception("Processor or timebased change under our feet")
>>>> return t
>>>>
> There's a second problem: If the time_info gets updated between the
> first rdtscp and the first version fetch, then we won't have a
> consistent tsc,time_info pair. You could check if tsc_timestamp is>
> tsc, but that won't necessarily work on save/restore/migrate.
>
Good catch. Doesn't that invalidate rdtscp based vgettimeofday on
non-virt as well (assuming p == cpu)?
> I suppose that works if you assume that:
>
> 1. every task->vcpu migration is associated with a hv/guest context
> switch, and
> 2. every hv/guest context switch is a write barrier
>
> I guess 2 is a given, but I can at least imagine cases where 1 might not
> be true. Maybe. It all seems very subtle.
>
What is 1 exactly? task switching to another vcpu? that doesn't incur
hypervisor involvement. vcpu moving to another cpu? That does.
> And I don't really see a gain. You avoid maintaining a second version
> number, but at the cost of two rdtscps. In my measurements, the whole
> vsyscall takes around 100ns to run, and a single rdtsc takes about 30,
> so 30% of total. Unlike rdtsc, rdtscp is documented as being ordered in
> the instruction stream, and so will take at least as long; two of them
> will completely blow the vsyscall execution time.
>
I agree, let's stick with the rdtscpless implementation.
--
I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.
WARNING: multiple messages have this Message-ID (diff)
From: Avi Kivity <avi@redhat.com>
To: Jeremy Fitzhardinge <jeremy@goop.org>
Cc: Dan Magenheimer <dan.magenheimer@oracle.com>,
Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>,
kurt.hackel@oracle.com, the arch/x86 maintainers <x86@kernel.org>,
Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
Glauber de Oliveira Costa <gcosta@redhat.com>,
Xen-devel <xen-devel@lists.xensource.com>,
Keir Fraser <keir.fraser@eu.citrix.com>,
Zach Brown <zach.brown@oracle.com>,
Chris Mason <chris.mason@oracle.com>
Subject: Re: Re: [PATCH 3/5] x86/pvclock: add vsyscall implementation
Date: Mon, 12 Oct 2009 20:29:57 +0200 [thread overview]
Message-ID: <4AD375A5.8050205@redhat.com> (raw)
In-Reply-To: <4AD3738B.6050200@goop.org>
On 10/12/2009 08:20 PM, Jeremy Fitzhardinge wrote:
> On 10/10/09 11:10, Avi Kivity wrote:
>
>> On 10/10/2009 02:24 AM, Jeremy Fitzhardinge wrote:
>>
>>> On 10/07/09 03:25, Avi Kivity wrote:
>>>
>>>
>>>> def try_pvclock_vtime():
>>>> tsc, p0 = rdtscp()
>>>> v0 = pvclock[p0].version
>>>> tsc, p = rdtscp()
>>>> t = pvclock_time(pvclock[p], tsc)
>>>> if p != p0 or pvclock[p].version != v0:
>>>> raise Exception("Processor or timebased change under our feet")
>>>> return t
>>>>
> There's a second problem: If the time_info gets updated between the
> first rdtscp and the first version fetch, then we won't have a
> consistent tsc,time_info pair. You could check if tsc_timestamp is>
> tsc, but that won't necessarily work on save/restore/migrate.
>
Good catch. Doesn't that invalidate rdtscp based vgettimeofday on
non-virt as well (assuming p == cpu)?
> I suppose that works if you assume that:
>
> 1. every task->vcpu migration is associated with a hv/guest context
> switch, and
> 2. every hv/guest context switch is a write barrier
>
> I guess 2 is a given, but I can at least imagine cases where 1 might not
> be true. Maybe. It all seems very subtle.
>
What is 1 exactly? task switching to another vcpu? that doesn't incur
hypervisor involvement. vcpu moving to another cpu? That does.
> And I don't really see a gain. You avoid maintaining a second version
> number, but at the cost of two rdtscps. In my measurements, the whole
> vsyscall takes around 100ns to run, and a single rdtsc takes about 30,
> so 30% of total. Unlike rdtsc, rdtscp is documented as being ordered in
> the instruction stream, and so will take at least as long; two of them
> will completely blow the vsyscall execution time.
>
I agree, let's stick with the rdtscpless implementation.
--
I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.
next prev parent reply other threads:[~2009-10-12 18:31 UTC|newest]
Thread overview: 119+ messages / expand[flat|nested] mbox.gz Atom feed top
2009-10-06 0:50 [PATCH RFC] Extending pvclock down to usermode for vsyscall Jeremy Fitzhardinge
2009-10-06 0:50 ` Jeremy Fitzhardinge
2009-10-06 0:50 ` [PATCH 1/5] x86/pvclock: make sure rdtsc doesn't speculate out of region Jeremy Fitzhardinge
2009-10-06 0:50 ` Jeremy Fitzhardinge
2009-10-06 0:50 ` [PATCH 2/5] x86/pvclock: no need to use strong read barriers in pvclock_get_time_values Jeremy Fitzhardinge
2009-10-06 0:50 ` Jeremy Fitzhardinge
2009-10-06 0:50 ` [PATCH 3/5] x86/pvclock: add vsyscall implementation Jeremy Fitzhardinge
2009-10-06 0:50 ` Jeremy Fitzhardinge
2009-10-06 9:04 ` Avi Kivity
2009-10-06 9:04 ` Avi Kivity
2009-10-06 14:19 ` Dan Magenheimer
2009-10-06 14:19 ` Dan Magenheimer
2009-10-06 15:11 ` Avi Kivity
2009-10-06 15:11 ` Avi Kivity
2009-10-06 18:46 ` [Xen-devel] " Jeremy Fitzhardinge
2009-10-06 18:46 ` Jeremy Fitzhardinge
2009-10-07 10:25 ` [Xen-devel] " Avi Kivity
2009-10-07 10:25 ` Avi Kivity
2009-10-07 19:29 ` [Xen-devel] " Jeremy Fitzhardinge
2009-10-07 19:29 ` Jeremy Fitzhardinge
2009-10-07 20:09 ` [Xen-devel] " Avi Kivity
2009-10-07 20:09 ` Avi Kivity
2009-10-07 21:19 ` [Xen-devel] " Jeremy Fitzhardinge
2009-10-07 21:19 ` Jeremy Fitzhardinge
2009-10-07 21:37 ` [Xen-devel] " Avi Kivity
2009-10-07 21:37 ` Avi Kivity
2009-10-07 21:51 ` [Xen-devel] " Jeremy Fitzhardinge
2009-10-07 21:51 ` Jeremy Fitzhardinge
2009-10-07 21:53 ` [Xen-devel] " Avi Kivity
2009-10-07 21:53 ` Avi Kivity
2009-10-07 20:48 ` [Xen-devel] " Dan Magenheimer
2009-10-07 20:48 ` Dan Magenheimer
2009-10-07 21:08 ` [Xen-devel] " Avi Kivity
2009-10-07 21:08 ` Avi Kivity
2009-10-07 22:36 ` [Xen-devel] " Dan Magenheimer
2009-10-07 22:36 ` Dan Magenheimer
2009-10-10 0:24 ` [Xen-devel] " Jeremy Fitzhardinge
2009-10-10 0:24 ` Jeremy Fitzhardinge
2009-10-10 18:10 ` [Xen-devel] " Avi Kivity
2009-10-10 18:10 ` Avi Kivity
2009-10-12 18:20 ` [Xen-devel] " Jeremy Fitzhardinge
2009-10-12 18:20 ` Jeremy Fitzhardinge
2009-10-12 18:29 ` Avi Kivity [this message]
2009-10-12 18:29 ` Avi Kivity
2009-10-12 19:13 ` [Xen-devel] " Jeremy Fitzhardinge
2009-10-12 19:13 ` Jeremy Fitzhardinge
2009-10-13 6:39 ` [Xen-devel] " Avi Kivity
2009-10-13 6:39 ` Avi Kivity
2009-10-13 20:00 ` [Xen-devel] " Jeremy Fitzhardinge
2009-10-13 20:00 ` Jeremy Fitzhardinge
2009-10-14 12:32 ` [Xen-devel] " Avi Kivity
2009-10-14 12:32 ` Avi Kivity
2009-10-15 19:17 ` [Xen-devel] " Jeremy Fitzhardinge
2009-10-15 19:17 ` Jeremy Fitzhardinge
2009-10-27 17:29 ` [Xen-devel] " Dan Magenheimer
2009-10-27 17:29 ` Dan Magenheimer
2009-10-27 18:20 ` [Xen-devel] " Jeremy Fitzhardinge
2009-10-27 18:20 ` Jeremy Fitzhardinge
2009-10-28 5:52 ` [Xen-devel] " Avi Kivity
2009-10-28 5:52 ` Avi Kivity
2009-10-28 9:29 ` [Xen-devel] " Glauber Costa
2009-10-28 9:34 ` Avi Kivity
2009-10-28 9:34 ` Avi Kivity
2009-10-28 17:47 ` [Xen-devel] " Jeremy Fitzhardinge
2009-10-28 17:47 ` Jeremy Fitzhardinge
2009-10-29 12:13 ` [Xen-devel] " Avi Kivity
2009-10-29 12:13 ` Avi Kivity
2009-10-29 13:03 ` [Xen-devel] " Chris Mason
2009-10-29 13:03 ` Chris Mason
2009-10-29 14:46 ` [Xen-devel] " Dan Magenheimer
2009-10-29 14:46 ` Dan Magenheimer
2009-10-29 15:07 ` [Xen-devel] " Avi Kivity
2009-10-29 15:07 ` Avi Kivity
2009-10-29 15:55 ` [Xen-devel] " Dan Magenheimer
2009-10-29 15:55 ` Dan Magenheimer
2009-10-29 16:15 ` [Xen-devel] " Dan Magenheimer
2009-10-29 16:15 ` Dan Magenheimer
2009-11-01 9:28 ` [Xen-devel] " Avi Kivity
2009-11-01 9:28 ` Avi Kivity
2009-11-02 15:28 ` [Xen-devel] " Dan Magenheimer
2009-11-02 15:28 ` Dan Magenheimer
2009-11-02 15:41 ` [Xen-devel] " Avi Kivity
2009-11-02 15:41 ` Avi Kivity
2009-11-01 9:32 ` [Xen-devel] " Avi Kivity
2009-11-01 9:32 ` Avi Kivity
2009-11-02 15:46 ` [Xen-devel] " Dan Magenheimer
2009-11-02 15:46 ` Dan Magenheimer
2009-11-03 5:12 ` [Xen-devel] " Avi Kivity
2009-11-03 5:12 ` Avi Kivity
2009-11-04 20:30 ` [Xen-devel] " Dan Magenheimer
2009-11-04 20:30 ` Dan Magenheimer
2009-11-05 6:47 ` [Xen-devel] " Avi Kivity
2009-11-05 6:47 ` Avi Kivity
2009-11-05 14:52 ` [Xen-devel] " Dan Magenheimer
2009-11-05 14:52 ` Dan Magenheimer
2009-11-05 15:07 ` [Xen-devel] " Keir Fraser
2009-11-05 15:07 ` Keir Fraser
2009-11-04 21:19 ` [Xen-devel] " john stultz
2009-11-04 21:19 ` john stultz
2009-11-04 21:28 ` Dan Magenheimer
2009-11-04 21:28 ` Dan Magenheimer
2009-11-05 0:02 ` [Xen-devel] " john stultz
2009-11-05 0:02 ` john stultz
2009-11-05 0:45 ` [Xen-devel] " Dan Magenheimer
2009-11-05 0:45 ` Dan Magenheimer
2009-10-30 23:30 ` pvclock implementation in pv_ops kernel: why not __native_read_tsc()? Dan Magenheimer
2009-10-31 1:17 ` Jeremy Fitzhardinge
2009-10-06 0:50 ` [PATCH 4/5] x86/fixmap: add a predicate for usermode fixmaps Jeremy Fitzhardinge
2009-10-06 0:50 ` Jeremy Fitzhardinge
2009-10-06 10:23 ` [Xen-devel] " Jan Beulich
2009-10-06 10:23 ` Jan Beulich
2009-10-06 18:47 ` [Xen-devel] " Jeremy Fitzhardinge
2009-10-06 18:47 ` Jeremy Fitzhardinge
2009-10-06 0:50 ` [PATCH 5/5] xen/time: add pvclock_clocksource_vread support Jeremy Fitzhardinge
2009-10-06 0:50 ` Jeremy Fitzhardinge
2009-10-06 10:28 ` [Xen-devel] " Jan Beulich
2009-10-06 10:28 ` Jan Beulich
2009-10-06 18:48 ` [Xen-devel] " Jeremy Fitzhardinge
2009-10-06 18:48 ` Jeremy Fitzhardinge
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=4AD375A5.8050205@redhat.com \
--to=avi@redhat.com \
--cc=chris.mason@oracle.com \
--cc=dan.magenheimer@oracle.com \
--cc=gcosta@redhat.com \
--cc=jeremy.fitzhardinge@citrix.com \
--cc=jeremy@goop.org \
--cc=keir.fraser@eu.citrix.com \
--cc=kurt.hackel@oracle.com \
--cc=linux-kernel@vger.kernel.org \
--cc=x86@kernel.org \
--cc=xen-devel@lists.xensource.com \
--cc=zach.brown@oracle.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.