Date:   Thu, 28 Feb 2019 01:26:16 +0100
From:   Niklas Cassel <niklas.cassel@...aro.org>
To:     Rob Herring <robh@...nel.org>
Cc:     Alexander Graf <agraf@...e.de>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        Ulf Hansson <ulf.hansson@...aro.org>,
        "Rafael J. Wysocki" <rjw@...ysocki.net>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        "open list:THERMAL" <linux-pm@...r.kernel.org>
Subject: Re: broken probe deferred for power-domains

On Wed, Feb 27, 2019 at 05:32:14PM -0600, Rob Herring wrote:
> On Tue, Feb 26, 2019 at 8:18 PM Niklas Cassel <niklas.cassel@...aro.org> wrote:
> >
> > Hello Rob,
> >
> > Your patch e01afc325025 ("PM / Domains: Stop deferring probe
> > at the end of initcall") breaks deferred probe for power domains.
> 
> With all the dependencies built-in, right?

Yep

> 
> What board in case I have one?

You probably do :) Dragonboard 410c.

It is reproducible very easily with branch: initcalls-for-robh
from this git:
https://git.linaro.org/people/niklas.cassel/kernel.git

ARCH=arm64 make defconfig

I recommend you to use trace_printk() when debugging drivers/base/dd.c,
since I've seen that printk can affect the timing so much that the issue
goes away :) (So don't forget to enable CONFIG_FTRACE).

> 
> > The patch looks like this:
> >
> > +++ b/drivers/base/power/domain.c
> > @@ -2253,7 +2253,7 @@ static int __genpd_dev_pm_attach(struct device *dev, struct device_node *np,
> >                 mutex_unlock(&gpd_list_lock);
> >                 dev_dbg(dev, "%s() failed to find PM domain: %ld\n",
> >                         __func__, PTR_ERR(pd));
> > -               return -EPROBE_DEFER;
> > +               return driver_deferred_probe_check_state(dev);
> >         }
> >
> >
> > Having two drivers (both using module_platform_driver),
> > one being a PD provider and one being a PD consumer.
> >
> > Before your patch:
> > The PD consumer driver calls dev_pm_domain_attach(),
> > and gets EPROBE_DEFER until the PD provider driver
> > has been probed successfully.
> >
> > (The PD provider driver needs some regulators, so it
> > is only successfully probed after the regulator driver
> > has been probed successfully.)
> >
> > Anyway, dev_pm_domain_attach() returned success after
> > the some deferred probes.
> >
> >
> > After your patch:
> > dev_pm_domain_attach() return ENODEV,
> > which comes from driver_deferred_probe_check_state().
> > Since it returns ENODEV rather than EPROBE_DEFER,
> > the PD consumer driver fails to probe.
> >
> >
> > The problem is related to your other patch 25b4e70dcce9
> > ("driver core: allow stopping deferred probe after init").
> >
> > driver_deferred_probe_check_state() returns ENODEV if
> > initcalls_done == true.
> >
> > initcalls_done is set from late_initcall(deferred_probe_initcall),
> > in drivers/base/dd.c:
> >        driver_deferred_probe_trigger();
> >        flush_work(&deferred_probe_work);
> >        initcalls_done = true;
> >
> > This does not seem very robust, since
> >
> > #1 It does not handle the case where two drivers have been
> > deferred (put in the deferred probe pending list),
> > where additionally one of the drivers has to be probed
> > before the other.
> >
> > (We would need to call driver_deferred_probe_trigger() + flush_work()
> > at least twice to handle this.)
> >
> > #2 Since this code is run from late_initcall(),
> > initcalls_done might get set before other drivers using late_initcall()
> > have even had a chance to run.
> 
> IMO, we should not have drivers using late_initcall. We need some
> level just to do things at the end of boot. The same fragility exists
> with the clock and regulator disabling.
> 
> > I can imagine that a driver using late_initcall() + EPROBE_DEFER
> > will absolutely not work with this code.
> >
> >
> > This patch fixes #1, but not #2.
> > However, I assume that even this change would not work if we have 3
> > drivers, where each driver a > b > c has to be probed, in that order.
> > (and all of them were placed in the deferred probe pending list).
> 
> I thought a successful probe would trigger a retry too. I need to look
> at it again.

That is true, a successful probe does trigger a retry,
driver_bound() has a call to driver_deferred_probe_trigger(),
which calls schedule_work(&deferred_probe_work).

I guess what is missing is a call to flush_work(&deferred_probe_work)
after the work has been scheduled.

But since the flush_work() waits, I guess the additional flush_work()
has to be in deferred_probe_initcall(), and not in driver_bound().


Kind regards,
Niklas

> 
> Maybe it would be more robust to re-trigger probe until the pending
> list doesn't change. Then we could handle any length of dependencies.
> 
> >
> > Suggestions and patches are welcome.
> >
> >
> > diff --git a/drivers/base/dd.c b/drivers/base/dd.c
> > index a823f469e53f..3443cb78be9b 100644
> > --- a/drivers/base/dd.c
> > +++ b/drivers/base/dd.c
> > @@ -288,7 +288,6 @@ static int deferred_probe_initcall(void)
> >         driver_deferred_probe_trigger();
> >         /* Sort as many dependencies as possible before exiting initcalls */
> >         flush_work(&deferred_probe_work);
> > -       initcalls_done = true;
> >
> >         /*
> >          * Trigger deferred probe again, this time we won't defer anything
> > @@ -297,6 +296,8 @@ static int deferred_probe_initcall(void)
> >         driver_deferred_probe_trigger();
> >         flush_work(&deferred_probe_work);
> >
> > +       initcalls_done = true;
> > +
> >         if (deferred_probe_timeout > 0) {
> >                 schedule_delayed_work(&deferred_probe_timeout_work,
> >                         deferred_probe_timeout * HZ);
> >
> >
> >
> > Kind regards,
> > Niklas

Powered by blists - more mailing lists