| Message ID | 20200901161756.28100-1-uli+renesas@fpond.eu (mailing list archive) |
|---|---|
| State | Changes Requested |
| Headers | show |
| Series | watchdog: da9063: wake up parent ahead of reboot | expand |
On 9/1/20 9:17 AM, Ulrich Hecht wrote: > This patch ensures our parent is awake before a reboot takes place. This > prevents situations in which the I2C host has been suspended and cannot > be safely woken up anymore when it needs to talk to us. > Why not call pm_runtime_get_sync() in da9063_wdt_restart() ? What guarantees that the local notifier is called before the watchdog core's restart notifier ? And what is the point of using the watchdog core's reboot handler if it is bypassed anyway ? Also, why is it not necessary to call pm functions when the watchdog is started, when it is stopped, and during suspend/resume ? Thanks, Guenter > Signed-off-by: Ulrich Hecht <uli+renesas@fpond.eu> > --- > > Hi! > > This is supposed to resolve the issue that came up in the review of > "[PATCH v2] i2c: sh_mobile: implement atomic transfers" that the parent > controller may be suspended when the restart method is called. See > https://www.spinics.net/lists/linux-i2c/msg46367.html for details. > > CU > Uli > > > drivers/watchdog/da9063_wdt.c | 21 +++++++++++++++++++++ > include/linux/mfd/da9063/core.h | 2 ++ > 2 files changed, 23 insertions(+) > > diff --git a/drivers/watchdog/da9063_wdt.c b/drivers/watchdog/da9063_wdt.c > index 423584252606..89718733491e 100644 > --- a/drivers/watchdog/da9063_wdt.c > +++ b/drivers/watchdog/da9063_wdt.c > @@ -18,6 +18,8 @@ > #include <linux/mfd/da9063/registers.h> > #include <linux/mfd/da9063/core.h> > #include <linux/regmap.h> > +#include <linux/pm_runtime.h> > +#include <linux/reboot.h> > > /* > * Watchdog selector to timeout in seconds. > @@ -158,6 +160,21 @@ static int da9063_wdt_set_timeout(struct watchdog_device *wdd, > return ret; > } > > +static int da9063_reboot_notifier(struct notifier_block *nb, > + unsigned long code, void *unused) > +{ > + struct da9063 *da9063 = container_of(nb, struct da9063, reboot_nb); > + > + /* > + * Make sure parent device is running. This cannot be done in the > + * restart handler because it is no longer safe to do runtime PM > + * there. > + */ > + pm_runtime_get_sync(da9063->dev->parent); > + > + return NOTIFY_DONE; > +} > + > static int da9063_wdt_restart(struct watchdog_device *wdd, unsigned long action, > void *data) > { > @@ -233,6 +250,10 @@ static int da9063_wdt_probe(struct platform_device *pdev) > set_bit(WDOG_HW_RUNNING, &wdd->status); > } > > + /* Get early notification of reboot so we can wake up the parent. */ > + da9063->reboot_nb.notifier_call = da9063_reboot_notifier; > + devm_register_reboot_notifier(dev, &da9063->reboot_nb); > + > return devm_watchdog_register_device(dev, wdd); > } > > diff --git a/include/linux/mfd/da9063/core.h b/include/linux/mfd/da9063/core.h > index fa7a43f02f27..9a3283d488b7 100644 > --- a/include/linux/mfd/da9063/core.h > +++ b/include/linux/mfd/da9063/core.h > @@ -85,6 +85,8 @@ struct da9063 { > int chip_irq; > unsigned int irq_base; > struct regmap_irq_chip_data *regmap_irq; > + > + struct notifier_block reboot_nb; > }; > > int da9063_device_init(struct da9063 *da9063, unsigned int irq); >
> On 09/01/2020 6:48 PM Guenter Roeck <linux@roeck-us.net> wrote: > > > On 9/1/20 9:17 AM, Ulrich Hecht wrote: > > This patch ensures our parent is awake before a reboot takes place. This > > prevents situations in which the I2C host has been suspended and cannot > > be safely woken up anymore when it needs to talk to us. > > > > Why not call pm_runtime_get_sync() in da9063_wdt_restart() ? It's no longer safe to do so at that point because the restart handler runs in atomic context. The unpleasant details can be found at https://www.spinics.net/lists/linux-i2c/msg46367.html > What > guarantees that the local notifier is called before the watchdog core's > restart notifier ? It would seem to me that a reboot notifier that triggers after the restart notifier (which, if successfully handled, does not return) would be rather pointless. > And what is the point of using the watchdog core's > reboot handler if it is bypassed anyway ? I don't think it's bypassed, it should be possible to register several reboot notifiers per device. It would also be possible to do this stuff in the core's reboot handler, but I have chosen not to do so because the core seems to avoid dealing with PM, which I assume is a design decision. Also, it would affect all other drivers with a restart handler, with (to me) unforeseeable effects. What do you think? > Also, why is it not necessary to call pm functions when the watchdog is > started, when it is stopped, and during suspend/resume ? These cases are handled automatically, AFAIK. CU Uli
diff --git a/drivers/watchdog/da9063_wdt.c b/drivers/watchdog/da9063_wdt.c index 423584252606..89718733491e 100644 --- a/drivers/watchdog/da9063_wdt.c +++ b/drivers/watchdog/da9063_wdt.c @@ -18,6 +18,8 @@ #include <linux/mfd/da9063/registers.h> #include <linux/mfd/da9063/core.h> #include <linux/regmap.h> +#include <linux/pm_runtime.h> +#include <linux/reboot.h> /* * Watchdog selector to timeout in seconds. @@ -158,6 +160,21 @@ static int da9063_wdt_set_timeout(struct watchdog_device *wdd, return ret; } +static int da9063_reboot_notifier(struct notifier_block *nb, + unsigned long code, void *unused) +{ + struct da9063 *da9063 = container_of(nb, struct da9063, reboot_nb); + + /* + * Make sure parent device is running. This cannot be done in the + * restart handler because it is no longer safe to do runtime PM + * there. + */ + pm_runtime_get_sync(da9063->dev->parent); + + return NOTIFY_DONE; +} + static int da9063_wdt_restart(struct watchdog_device *wdd, unsigned long action, void *data) { @@ -233,6 +250,10 @@ static int da9063_wdt_probe(struct platform_device *pdev) set_bit(WDOG_HW_RUNNING, &wdd->status); } + /* Get early notification of reboot so we can wake up the parent. */ + da9063->reboot_nb.notifier_call = da9063_reboot_notifier; + devm_register_reboot_notifier(dev, &da9063->reboot_nb); + return devm_watchdog_register_device(dev, wdd); } diff --git a/include/linux/mfd/da9063/core.h b/include/linux/mfd/da9063/core.h index fa7a43f02f27..9a3283d488b7 100644 --- a/include/linux/mfd/da9063/core.h +++ b/include/linux/mfd/da9063/core.h @@ -85,6 +85,8 @@ struct da9063 { int chip_irq; unsigned int irq_base; struct regmap_irq_chip_data *regmap_irq; + + struct notifier_block reboot_nb; }; int da9063_device_init(struct da9063 *da9063, unsigned int irq);
This patch ensures our parent is awake before a reboot takes place. This prevents situations in which the I2C host has been suspended and cannot be safely woken up anymore when it needs to talk to us. Signed-off-by: Ulrich Hecht <uli+renesas@fpond.eu> --- Hi! This is supposed to resolve the issue that came up in the review of "[PATCH v2] i2c: sh_mobile: implement atomic transfers" that the parent controller may be suspended when the restart method is called. See https://www.spinics.net/lists/linux-i2c/msg46367.html for details. CU Uli drivers/watchdog/da9063_wdt.c | 21 +++++++++++++++++++++ include/linux/mfd/da9063/core.h | 2 ++ 2 files changed, 23 insertions(+)