Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20200416085627.1882-1-clay@daemons.net> <6fef3a00-6c18-b775-d1b4-dfd692261bd3@ti.com>
 <20200420093610.GA28162@arctic-shiba-lx>
In-Reply-To: <20200420093610.GA28162@arctic-shiba-lx>
From:   Arnd Bergmann <arnd@arndb.de>
Date:   Mon, 20 Apr 2020 16:38:32 +0200
Message-ID: <CAK8P3a36ZxNJxUS4UzrwJiMx8UrgYPkcv4X6yYw7EC4jRBbbGQ@mail.gmail.com>
Subject: Re: [PATCH] net: cpts: Condition WARN_ON on PTP_1588_CLOCK
To:     Clay McClure <clay@daemons.net>
Cc:     Grygorii Strashko <grygorii.strashko@ti.com>,
        "David S. Miller" <davem@davemloft.net>,
        Sekhar Nori <nsekhar@ti.com>,
        Richard Cochran <richardcochran@gmail.com>,
        Networking <netdev@vger.kernel.org>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Mon, Apr 20, 2020 at 11:38 AM Clay McClure <clay@daemons.net> wrote:
> On Thu, Apr 16, 2020 at 02:11:45PM +0300, Grygorii Strashko wrote:
>
> > > CPTS_MOD merely implies PTP_1588_CLOCK; it is possible to build cpts
> > > without PTP clock support. In that case, ptp_clock_register() returns
> > > NULL and we should not WARN_ON(cpts->clock) when downing the interface.
> > > The ptp_*() functions are stubbed without PTP_1588_CLOCK, so it's safe
> > > to pass them a null pointer.
> >
> > Could you explain the purpose of the exercise (Enabling CPTS with
> > PTP_1588_CLOCK disabled), pls?
>
> Hardware timestamping with a free-running PHC _almost_ works without
> PTP_1588_CLOCK, but since PHC rollover is handled by the PTP kworker
> in this driver the timestamps end up not being monotonic.
>
> And of course the moment you want to syntonize/synchronize the PHC with
> another clock (say, CLOCK_REALTIME), you'll need a PTP clock device. So
> you're right, there's not much point in building CPTS_MOD without
> PTP_1588_CLOCK.
>
> Given that, I wonder why all the Ethernet drivers seem to just `imply`
> PTP_1588_CLOCK, rather than `depends on` it?

I suspect we should move all of them back. This was an early user
of 'imply', but the meaning of that keyword has now changed
in the latest Kconfig.

> diff --git a/drivers/net/ethernet/ti/cpts.c b/drivers/net/ethernet/ti/cpts.c
> index 10ad706dda53..70b15039cd37 100644
> --- a/drivers/net/ethernet/ti/cpts.c
> +++ b/drivers/net/ethernet/ti/cpts.c
> @@ -462,8 +462,8 @@ int cpts_register(struct cpts *cpts)
>         timecounter_init(&cpts->tc, &cpts->cc, ktime_get_real_ns());
>
>         cpts->clock = ptp_clock_register(&cpts->info, cpts->dev);
> -       if (IS_ERR(cpts->clock)) {
> -               err = PTR_ERR(cpts->clock);
> +       if (IS_ERR_OR_NULL(cpts->clock)) {
> +               err = cpts->clock ? PTR_ERR(cpts->clock) : -EOPNOTSUPP;
>                 cpts->clock = NULL;
>                 goto err_ptp;

Something else is wrong if you need IS_ERR_OR_NULL(). Any
kernel interface should either return an negative error code when
something goes wrong, or should return NULL for all errors, but
not mix the two.

        Arnd