[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <352a7f910269daf1a7ff57ea4a41a306d6981b21.camel@infradead.org>
Date: Wed, 03 Jul 2024 11:40:16 +0100
From: David Woodhouse <dwmw2@...radead.org>
To: Peter Hilber <peter.hilber@...nsynergy.com>,
linux-kernel@...r.kernel.org, virtualization@...ts.linux.dev,
linux-arm-kernel@...ts.infradead.org, linux-rtc@...r.kernel.org, "Ridoux,
Julien" <ridouxj@...zon.com>, virtio-dev@...ts.linux.dev, "Luu, Ryan"
<rluu@...zon.com>, "Chashper, David" <chashper@...zon.com>
Cc: "Christopher S. Hall" <christopher.s.hall@...el.com>, Jason Wang
<jasowang@...hat.com>, John Stultz <jstultz@...gle.com>, "Michael S.
Tsirkin" <mst@...hat.com>, netdev@...r.kernel.org, Richard Cochran
<richardcochran@...il.com>, Stephen Boyd <sboyd@...nel.org>, Thomas
Gleixner <tglx@...utronix.de>, Xuan Zhuo <xuanzhuo@...ux.alibaba.com>, Marc
Zyngier <maz@...nel.org>, Mark Rutland <mark.rutland@....com>, Daniel
Lezcano <daniel.lezcano@...aro.org>, Alessandro Zummo
<a.zummo@...ertech.it>, Alexandre Belloni <alexandre.belloni@...tlin.com>
Subject: Re: [RFC PATCH v2] ptp: Add vDSO-style vmclock support
On Wed, 2024-07-03 at 11:56 +0200, Peter Hilber wrote:
> On 02.07.24 20:40, David Woodhouse wrote:
> > On 2 July 2024 19:12:00 BST, Peter Hilber <peter.hilber@...nsynergy.com> wrote:
> > > On 02.07.24 18:39, David Woodhouse wrote:
> > > > To clarify then, the main types are
> > > >
> > > > VIRTIO_RTC_CLOCK_UTC == 0
> > > > VIRTIO_RTC_CLOCK_TAI == 1
> > > > VIRTIO_RTC_CLOCK_MONOTONIC == 2
> > > > VIRTIO_RTC_CLOCK_SMEARED_UTC == 3
> > > >
> > > > And the subtypes are *only* for the case of
> > > > VIRTIO_RTC_CLOCK_SMEARED_UTC. They include
> > > >
> > > > VIRTIO_RTC_SUBTYPE_STRICT
> > > > VIRTIO_RTC_SUBTYPE_UNDEFINED /* or whatever you want to call it */
> > > > VIRTIO_RTC_SUBTYPE_SMEAR_NOON_LINEAR
> > > > VIRTIO_RTC_SUBTYPE_UTC_SLS /* if it's worth doing this one */
> > > >
> > > > Is that what we just agreed on?
> > > >
> > > >
> > >
> > > This is a misunderstanding. My idea was that the main types are
> > >
> > > > VIRTIO_RTC_CLOCK_UTC == 0
> > > > VIRTIO_RTC_CLOCK_TAI == 1
> > > > VIRTIO_RTC_CLOCK_MONOTONIC == 2
> > > > VIRTIO_RTC_CLOCK_SMEARED_UTC == 3
> > >
> > > VIRTIO_RTC_CLOCK_MAYBE_SMEARED_UTC == 4
> > >
> > > The subtypes would be (1st for clocks other than
> > > VIRTIO_RTC_CLOCK_SMEARED_UTC, 2nd to last for
> > > VIRTIO_RTC_CLOCK_SMEARED_UTC):
> > >
> > > #define VIRTIO_RTC_SUBTYPE_STRICT 0
> > > #define VIRTIO_RTC_SUBTYPE_SMEAR_NOON_LINEAR 1
> > > #define VIRTIO_RTC_SUBTYPE_SMEAR_UTC_SLS 2
> > >
> >
> > Thanks. I really do think that from the guest point of view there's
> > really no distinction between "maybe smeared" and "undefined
> > smearing", and have a preference for using the latter form, which
> > is the key difference there?
> >
> > Again though, not a hill for me to die on.
>
> I have no issue with staying with "undefined smearing", so would you agree
> to something like
>
> VIRTIO_RTC_CLOCK_SMEAR_UNDEFINED_UTC == 4
>
> (or another name if you prefer)?
Well, the point of contention was really whether that was a *type* or a
*subtype*.
Either way, it's a "precision clock" telling its consumer that the
device *itself* doesn't really know what time is being exposed. Which
seems like a bizarre thing to support.
But I think I've constructed an argument which persuades me to your
point of view that *if* we permit it, it should be a primary type...
A clock can *either* be UTC, *or* it can be monotonic. The whole point
of smearing is to produce a monotonic clock, of course.
VIRTIO_RTC_CLOCK_UTC is UTC. It is not monotonic.
VIRTIO_RTC_CLOCK_SMEARED is, presumably, monotonic (and I think we
should explicitly require that to be true in virtio-rtc).
But VIRTIO_RTC_CLOCK_MAYBE_SMEARED is the worst of both worlds. It is
neither known to be correct UTC, *nor* is it known to be monotonic. So
(again, if we permit it at all) I think it probably does make sense for
that to be a primary type.
This is what I currently have for 'struct vmclock_abi' that I'd like to
persuade you to adopt. I need to tweak it some more, for at least the
following reasons, as well as any more you can see:
• size isn't big enough for 64KiB pages
• Should be explicitly little-endian
• Does it need esterror as well as maxerror?
• Why is maxerror in picoseconds? It's the only use of that unit
• Where do the clock_status values come from? Do they make sense?
• Are signed integers OK? (I think so!).
/*
* This structure provides a vDSO-style clock to VM guests, exposing the
* relationship (or lack thereof) between the CPU clock (TSC, timebase, arch
* counter, etc.) and real time. It is designed to address the problem of
* live migration, which other clock enlightenments do not.
*
* When a guest is live migrated, this affects the clock in two ways.
*
* First, even between identical hosts the actual frequency of the underlying
* counter will change within the tolerances of its specification (typically
* ±50PPM, or 4 seconds a day). This frequency also varies over time on the
* same host, but can be tracked by NTP as it generally varies slowly. With
* live migration there is a step change in the frequency, with no warning.
*
* Second, there may be a step change in the value of the counter itself, as
* its accuracy is limited by the precision of the NTP synchronization on the
* source and destination hosts.
*
* So any calibration (NTP, PTP, etc.) which the guest has done on the source
* host before migration is invalid, and needs to be redone on the new host.
*
* In its most basic mode, this structure provides only an indication to the
* guest that live migration has occurred. This allows the guest to know that
* its clock is invalid and take remedial action. For applications that need
* reliable accurate timestamps (e.g. distributed databases), the structure
* can be mapped all the way to userspace. This allows the application to see
* directly for itself that the clock is disrupted and take appropriate
* action, even when using a vDSO-style method to get the time instead of a
* system call.
*
* In its more advanced mode. this structure can also be used to expose the
* precise relationship of the CPU counter to real time, as calibrated by the
* host. This means that userspace applications can have accurate time
* immediately after live migration, rather than having to pause operations
* and wait for NTP to recover. This mode does, of course, rely on the
* counter being reliable and consistent across CPUs.
*
* Note that this must be true UTC, never with smeared leap seconds. If a
* guest wishes to construct a smeared clock, it can do so. Presenting a
* smeared clock through this interface would be problematic because it
* actually messes with the apparent counter *period*. A linear smearing
* of 1 ms per second would effectively tweak the counter period by 1000PPM
* at the start/end of the smearing period, while a sinusoidal smear would
* basically be impossible to represent.
*
* This structure is offered with the intent that it be adopted into the
* nascent virtio-rtc standard, as a virtio-rtc that does not address the live
* migration problem seems a little less than fit for purpose. For that
* reason, certain fields use precisely the same numeric definitions as in
* the virtio-rtc proposal. The structure can also be exposed through an ACPI
* device with the CID "VMCLOCK", modelled on the "VMGENID" device except for
* the fact that it uses a real _CRS to convey the address of the structure
* (which should be a full page, to allow for mapping directly to userspace).
*/
#ifndef __VMCLOCK_ABI_H__
#define __VMCLOCK_ABI_H__
#ifdef __KERNEL__
#include <linux/types.h>
#else
#include <stdint.h>
#endif
struct vmclock_abi {
uint64_t magic;
#define VMCLOCK_MAGIC 0x4b4c4356 /* "VCLK" */
uint16_t size; /* Size of page containing this structure */
uint16_t version; /* 1 */
/* Sequence lock. Low bit means an update is in progress. */
uint32_t seq_count;
uint32_t flags;
/* Indicates that the tai_offset_sec field is valid */
#define VMCLOCK_FLAG_TAI_OFFSET_VALID (1 << 0)
/*
* Optionally used to notify guests of pending maintenance events.
* A guest may wish to remove itself from service if an event is
* coming up. Two flags indicate the rough imminence of the event.
*/
#define VMCLOCK_FLAG_DISRUPTION_SOON (1 << 1) /* About a day */
#define VMCLOCK_FLAG_DISRUPTION_IMMINENT (1 << 2) /* About an hour */
/* Indicates that the utc_time_maxerror_picosec field is valid */
#define VMCLOCK_FLAG_UTC_MAXERROR_VALID (1 << 3)
/* Indicates counter_period_error_rate_frac_sec is valid */
#define VMCLOCK_FLAG_PERIOD_ERROR_VALID (1 << 4)
/*
* This field changes to another non-repeating value when the CPU
* counter is disrupted, for example on live migration. This lets
* the guest know that it should discard any calibration it has
* performed of the counter against external sources (NTP/PTP/etc.).
*/
uint64_t disruption_marker;
uint8_t clock_status;
#define VMCLOCK_STATUS_UNKNOWN 0
#define VMCLOCK_STATUS_INITIALIZING 1
#define VMCLOCK_STATUS_SYNCHRONIZED 2
#define VMCLOCK_STATUS_FREERUNNING 3
#define VMCLOCK_STATUS_UNRELIABLE 4
uint8_t counter_id; /* Matches VIRTIO_RTC_COUNTER_xxx */
#define VMCLOCK_COUNTER_ARM_VCNT 0
#define VMCLOCK_COUNTER_X86_TSC 1
#define VMCLOCK_COUNTER_INVALID 0xff
/*
* By providing the offset from UTC to TAI, the guest can know both
* UTC and TAI reliably, whichever is indicated in the time_type
* field. Valid if VMCLOCK_FLAG_TAI_OFFSET_VALID is set in flags.
*/
int16_t tai_offset_sec;
/*
* What time is exposed in the time_sec/time_frac_sec fields?
*/
uint8_t time_type; /* Matches VIRTIO_RTC_TYPE_xxx */
#define VMCLOCK_TIME_UTC 0 /* Since 1970-01-01 00:00:00z */
#define VMCLOCK_TIME_TAI 1 /* Since 1970-01-01 00:00:00z */
#define VMCLOCK_TIME_MONOTONIC 2 /* Since undefined epoch */
#define VMCLOCK_TIME_INVALID_SMEARED 3 /* Not supported */
#define VMCLOCK_TIME_INVALID_MAYBE_SMEARED 4 /* Not supported */
/*
* The time exposed through this device is never smeared. This field
* corresponds to the 'subtype' field in virtio-rtc, which indicates
* the smearing method. However in this case it provides a *hint* to
* the guest operating system, such that *if* the guest OS wants to
* provide its users with an alternative clock which does not follow
* the POSIX CLOCK_REALTIME standard, it may do so in a fashion
* consistent with the other systems in the nearby environment.
*/
uint8_t leap_second_smearing_hint; /* Matches VIRTIO_RTC_SUBTYPE_xxx */
#define VMCLOCK_SMEARING_STRICT 0
#define VMCLOCK_SMEARING_NOON_LINEAR 1
#define VMCLOCK_SMEARING_UTC_SLS 2
/* Bit shift for counter_period_frac_sec and its error rate */
uint8_t counter_period_shift;
/*
* Unlike in NTP, this can indicate a leap second in the past. This
* is needed to allow guests to derive an imprecise clock with
* smeared leap seconds for themselves, as some modes of smearing
* need the adjustments to continue even after the moment at which
* the leap second should have occurred.
*/
uint8_t leap_indicator; /* Matches VIRTIO_RTC_LEAP_xxx */
#define VMCLOCK_LEAP_NONE 0
#define VMCLOCK_LEAP_PRE_POS 1
#define VMCLOCK_LEAP_PRE_NEG 2
#define VMCLOCK_LEAP_POS 3
#define VMCLOCK_LEAP_NEG 4
uint64_t leapsecond_tai_sec; /* Since 1970-01-01 00:00:00z */
/*
* Paired values of counter and UTC at a given point in time.
*/
uint64_t counter_value;
uint64_t time_sec;
uint64_t time_frac_sec;
/*
* Counter frequency, and error margin. The unit of these fields is
* seconds >> (64 + counter_period_shift)
*/
uint64_t counter_period_frac_sec;
uint64_t counter_period_error_rate_frac_sec;
/* Error margin of UTC reading above (± picoseconds) */
uint64_t utc_time_maxerror_picosec;
};
#endif /* __VMCLOCK_ABI_H__ */
Download attachment "smime.p7s" of type "application/pkcs7-signature" (5965 bytes)
Powered by blists - more mailing lists