Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:8 as permitted sender) client-ip=2620:137:e000::3:8;
MIME-Version: 1.0
References: <20231127153901.6399-1-maimon.sagi@gmail.com> <ZWUwTnWEHipJqHnk@hoboy.vegasvil.org>
 <CANDhNCq=iV2_1bzaP=BYuUwJtNsQBdjuYqUUtsiLc-MCNURJ6w@mail.gmail.com>
In-Reply-To: <CANDhNCq=iV2_1bzaP=BYuUwJtNsQBdjuYqUUtsiLc-MCNURJ6w@mail.gmail.com>
From:   Sagi Maimon <maimon.sagi@gmail.com>
Date:   Thu, 30 Nov 2023 17:45:08 +0200
Message-ID: <CAMuE1bG1ShKVakUH5f41_NNixBZRTUgXW8-=BWZxfNde6ohbVw@mail.gmail.com>
Subject: Re: [PATCH v2] posix-timers: add multi_clock_gettime system call
To:     John Stultz <jstultz@google.com>
Cc:     Richard Cochran <richardcochran@gmail.com>, reibax@gmail.com,
        davem@davemloft.net, rrameshbabu@nvidia.com,
        netdev@vger.kernel.org, linux-kernel@vger.kernel.org,
        maheshb@google.com, Thomas Gleixner <tglx@linutronix.de>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

Hi John,
Thanks for your notes.

On Tue, Nov 28, 2023 at 2:46=E2=80=AFAM John Stultz <jstultz@google.com> wr=
ote:
>
> On Mon, Nov 27, 2023 at 4:12=E2=80=AFPM Richard Cochran
> <richardcochran@gmail.com> wrote:
> >
> > On Mon, Nov 27, 2023 at 05:39:01PM +0200, Sagi Maimon wrote:
> > >  Some user space applications need to read some clocks.
> > >  Each read requires moving from user space to kernel space.
> > >  This asymmetry causes the measured offset to have a significant erro=
r.
> >
> > Adding time/clock gurus (jstultz, tglx) on CC for visibility...
> >
>
> Thanks for the heads up! (though, "guru" is just the noise I make
> standing up these days)
>
> > >  Introduce a new system call multi_clock_gettime, which can be used t=
o measure
> > >  the offset between multiple clocks, from variety of types: PHC, virt=
ual PHC
> > >  and various system clocks (CLOCK_REALTIME, CLOCK_MONOTONIC, etc).
> > >  The offset includes the total time that the driver needs to read the=
 clock
> > >  timestamp.
>
> This last bit about "offset includes the total time that the driver
> needs to read the clock" is a bit confusing. It seems to suggest there
> would be start/stop bookend timestamps so you could bound how long it
> took to read all the clocks, but I don't see that in the patch.
>
You are right it is a little confusing, I will remove it from the commit .

> > >  New system call allows the reading of a list of clocks - up to PTP_M=
AX_CLOCKS.
> > >  Supported clocks IDs: PHC, virtual PHC and various system clocks.
> > >  Up to PTP_MAX_SAMPLES times (per clock) in a single system call read=
.
> > >  The system call returns n_clocks timestamps for each measurement:
> > >  - clock 0 timestamp
> > >  - ...
> > >  - clock n timestamp
> > >
> > > Signed-off-by: Sagi Maimon <maimon.sagi@gmail.com>
>
> Overally, while I understand the intent, I'm pretty hesitant on it
> (and "__ptp_multi_clock_get multi_clk_get" has me squinting to find
> the actual space amongst all the underscores :).
>
struct __ptp_multi_clock_get __user * ptp_multi_clk_get has many
underscores indeed :-)
I will rename it, if you have any naming suggestions, I will be glad to hea=
r it.

> If the overhead of reading clockids individually is too much, it seems
> like the next thing will be folks wanting to export multiple raw
> hardware counter values so the counter->ns transformation doesn't get
> inbetween each hw clock read, which this interface wouldn't solve, so
> we'd have to add yet another interface.
>
future raw HW counter read interface:
=E2=80=A2 For PHC - reading raw HW counter is driver dependent, and will
require changes in many existing PTP drivers.
=E2=80=A2 For System clock it is possible but with much more code changes a=
nd
the improvements seems to be small.
=E2=80=A2 The issue you introduced can be reduced (but not completely
overcome) by user space usage of the interface.
For example, to minimize the difference between clock X and clock Y,
users can call  multi_clock_gettime  with 3 clock reads : X, Y, and X
again.
Considering gain (thin extra accuracy) vs pain (a lot of code changes,
also for system clocks) and needs =E2=80=93 we chose to settle with the
current suggested interface.

> Also, I wonder if trying to get multiple clocks in one read seems
> similar to something uio_ring might help with? Though I can't say I'm
> very savvy with uio_ring. Have folks looked into that?
>
I am also not an expert with the uio_ring/liburing, but I researched a
little and it seems it is mainly used for IO operations and support
only few of them (IORING_OP_READV, IORING_OP_SENDMSG, etc.)
I am not sure how to implement it for consecutive clock_gettime system
calls and if it can be done at all.
Even if it can be done, it adds a lot of complexity to the user space
code and the use in one generic system call is more elegant in my
opinion.
For example: Looking at the existing ioctl PTP_SYS_OFFSET,
theoretically it can also be implemented by uio_ring, but it is still
a more elegant solution.
> thanks
> -john