| Message ID | 1445607381-7794-1-git-send-email-soren.brinkmann@xilinx.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Fri, Oct 23, 2015 at 3:36 PM, Soren Brinkmann <soren.brinkmann@xilinx.com> wrote: > The driver uses runtime PM to leverage low power techniques. For > use-cases using GPIO as interrupt the device needs to be in an > appropriate state. > > Reported-by: John Linn <linnj@xilinx.com> > Signed-off-by: Soren Brinkmann <soren.brinkmann@xilinx.com> > Tested-by: John Linn <linnj@xilinx.com> As pointed out by Grygorii in commit aca82d1cbb49af34b69ecd4571a0fe48ad9247c1: The PM runtime API can't be used in atomic contex on -RT even if it's configured as irqsafe. As result, below error report can be seen when PM runtime API called from IRQ chip's callbacks irq_startup/irq_shutdown/irq_set_type, because they are protected by RAW spinlock: (...) The IRQ chip interface defines only two callbacks which are executed in non-atomic contex - irq_bus_lock/irq_bus_sync_unlock, so lets move PM runtime calls there. I.e. these calls are atomic context and it's just luck that it works and this is fragile. Can you please check if you can move it to irq_bus_lock()/irq_sync_unlock() like Grygorii does? Yours, Linus Walleij
On 10/27/2015 05:53 PM, Linus Walleij wrote: > On Fri, Oct 23, 2015 at 3:36 PM, Soren Brinkmann > <soren.brinkmann@xilinx.com> wrote: > >> The driver uses runtime PM to leverage low power techniques. For >> use-cases using GPIO as interrupt the device needs to be in an >> appropriate state. >> >> Reported-by: John Linn <linnj@xilinx.com> >> Signed-off-by: Soren Brinkmann <soren.brinkmann@xilinx.com> >> Tested-by: John Linn <linnj@xilinx.com> > > As pointed out by Grygorii in > commit aca82d1cbb49af34b69ecd4571a0fe48ad9247c1: > > The PM runtime API can't be used in atomic contex on -RT even if > it's configured as irqsafe. As result, below error report can > be seen when PM runtime API called from IRQ chip's callbacks > irq_startup/irq_shutdown/irq_set_type, because they are > protected by RAW spinlock: > (...) > The IRQ chip interface defines only two callbacks which are executed in > non-atomic contex - irq_bus_lock/irq_bus_sync_unlock, so lets move > PM runtime calls there. > > I.e. these calls are atomic context and it's just luck that it works > and this is fragile. > > Can you please check if you can move it to > irq_bus_lock()/irq_sync_unlock() > like Grygorii does? > This patch rises the question not only about PM runtime, but also about gpiochip_irq_reqres()/gpiochip_irq_relres().
On Tue, Oct 27, 2015 at 5:18 PM, Grygorii Strashko <grygorii.strashko@ti.com> wrote: > On 10/27/2015 05:53 PM, Linus Walleij wrote: >> >> On Fri, Oct 23, 2015 at 3:36 PM, Soren Brinkmann >> <soren.brinkmann@xilinx.com> wrote: >> >>> The driver uses runtime PM to leverage low power techniques. For >>> use-cases using GPIO as interrupt the device needs to be in an >>> appropriate state. >>> >>> Reported-by: John Linn <linnj@xilinx.com> >>> Signed-off-by: Soren Brinkmann <soren.brinkmann@xilinx.com> >>> Tested-by: John Linn <linnj@xilinx.com> >> >> >> As pointed out by Grygorii in >> commit aca82d1cbb49af34b69ecd4571a0fe48ad9247c1: >> >> The PM runtime API can't be used in atomic contex on -RT even if >> it's configured as irqsafe. As result, below error report can >> be seen when PM runtime API called from IRQ chip's callbacks >> irq_startup/irq_shutdown/irq_set_type, because they are >> protected by RAW spinlock: >> (...) >> The IRQ chip interface defines only two callbacks which are executed >> in >> non-atomic contex - irq_bus_lock/irq_bus_sync_unlock, so lets move >> PM runtime calls there. >> >> I.e. these calls are atomic context and it's just luck that it works >> and this is fragile. >> >> Can you please check if you can move it to >> irq_bus_lock()/irq_sync_unlock() >> like Grygorii does? >> > > This patch rises the question not only about PM runtime, but also > about gpiochip_irq_reqres()/gpiochip_irq_relres(). Do you mean that these functions contain calls to non-atomic functions? I mainly reacted to this because it was pm_* calls, that you mentioned explicitly in your patch. Yours, Linus Walleij
On 10/27/2015 04:53 PM, Linus Walleij wrote: > On Fri, Oct 23, 2015 at 3:36 PM, Soren Brinkmann > <soren.brinkmann@xilinx.com> wrote: > >> The driver uses runtime PM to leverage low power techniques. For >> use-cases using GPIO as interrupt the device needs to be in an >> appropriate state. >> >> Reported-by: John Linn <linnj@xilinx.com> >> Signed-off-by: Soren Brinkmann <soren.brinkmann@xilinx.com> >> Tested-by: John Linn <linnj@xilinx.com> > > As pointed out by Grygorii in > commit aca82d1cbb49af34b69ecd4571a0fe48ad9247c1: > > The PM runtime API can't be used in atomic contex on -RT even if > it's configured as irqsafe. As result, below error report can > be seen when PM runtime API called from IRQ chip's callbacks > irq_startup/irq_shutdown/irq_set_type, because they are > protected by RAW spinlock: > (...) > The IRQ chip interface defines only two callbacks which are executed in > non-atomic contex - irq_bus_lock/irq_bus_sync_unlock, so lets move > PM runtime calls there. > > I.e. these calls are atomic context and it's just luck that it works > and this is fragile. > > Can you please check if you can move it to > irq_bus_lock()/irq_sync_unlock() > like Grygorii does? That only powers up the chip when the chip is accessed. For proper IRQ operation the chip needs to be powered up though as long as the IRQ is enabled. request_irq() and free_irq() must always be called from sleepable context. The thing is just that request_resource/release_resource are called from within a raw spinlock, which is necessary since otherwise you can't guarantee that they are only called once for shared interrupts. It might make sense to add a separate set of callbacks to the irq_chip struct that are called from the sleepable sections of request_irq()/free_irq() which are meant for power management purposes and which wont have the guarantee that they are only called once for shared IRQs (but are still balanced). Thomas, do you have any thoughts on this? - Lars
On 10/27/2015 06:23 PM, Linus Walleij wrote: > On Tue, Oct 27, 2015 at 5:18 PM, Grygorii Strashko > <grygorii.strashko@ti.com> wrote: >> On 10/27/2015 05:53 PM, Linus Walleij wrote: >>> >>> On Fri, Oct 23, 2015 at 3:36 PM, Soren Brinkmann >>> <soren.brinkmann@xilinx.com> wrote: >>> >>>> The driver uses runtime PM to leverage low power techniques. For >>>> use-cases using GPIO as interrupt the device needs to be in an >>>> appropriate state. >>>> >>>> Reported-by: John Linn <linnj@xilinx.com> >>>> Signed-off-by: Soren Brinkmann <soren.brinkmann@xilinx.com> >>>> Tested-by: John Linn <linnj@xilinx.com> >>> >>> >>> As pointed out by Grygorii in >>> commit aca82d1cbb49af34b69ecd4571a0fe48ad9247c1: >>> >>> The PM runtime API can't be used in atomic contex on -RT even if >>> it's configured as irqsafe. As result, below error report can >>> be seen when PM runtime API called from IRQ chip's callbacks >>> irq_startup/irq_shutdown/irq_set_type, because they are >>> protected by RAW spinlock: >>> (...) >>> The IRQ chip interface defines only two callbacks which are executed >>> in >>> non-atomic contex - irq_bus_lock/irq_bus_sync_unlock, so lets move >>> PM runtime calls there. >>> >>> I.e. these calls are atomic context and it's just luck that it works >>> and this is fragile. >>> >>> Can you please check if you can move it to >>> irq_bus_lock()/irq_sync_unlock() >>> like Grygorii does? >>> >> >> This patch rises the question not only about PM runtime, but also >> about gpiochip_irq_reqres()/gpiochip_irq_relres(). > > Do you mean that these functions contain calls to non-atomic > functions? > Oh. No, I have to be more specific :( if GPIOx driver defines custom .irq_(request|release)_resources() callbacks they will *overwrite* standard GPIOirqchip callbacks. (commit: 8b67a1f "gpio: don't override irq_*_resources() callbacks") As result, such GPIOx driver should *re-implement* the same functionality in its .irq_(request|release)_resources() callbacks as implemented in gpiochip_irq_reqres()/gpiochip_irq_relres(). > I mainly reacted to this because it was pm_* calls, that you > mentioned explicitly in your patch. >
On Tue, Oct 27, 2015 at 6:54 PM, Grygorii Strashko <grygorii.strashko@ti.com> wrote: > [Me] >> Do you mean that these functions contain calls to non-atomic >> functions? >> > > Oh. No, I have to be more specific :( > if GPIOx driver defines custom .irq_(request|release)_resources() callbacks > they will *overwrite* standard GPIOirqchip callbacks. > (commit: 8b67a1f "gpio: don't override irq_*_resources() callbacks") > > As result, such GPIOx driver should *re-implement* the same functionality in > its .irq_(request|release)_resources() callbacks as implemented in > gpiochip_irq_reqres()/gpiochip_irq_relres(). Yes that goes for all drivers not using gpiochip_add_irqchip(). Not everyone uses that ... and for those supplying their own implementations of these functions. Yours, Linus Walleij
On Tue, 2015-10-27 at 05:37PM +0100, Lars-Peter Clausen wrote: > On 10/27/2015 04:53 PM, Linus Walleij wrote: > > On Fri, Oct 23, 2015 at 3:36 PM, Soren Brinkmann > > <soren.brinkmann@xilinx.com> wrote: > > > >> The driver uses runtime PM to leverage low power techniques. For > >> use-cases using GPIO as interrupt the device needs to be in an > >> appropriate state. > >> > >> Reported-by: John Linn <linnj@xilinx.com> > >> Signed-off-by: Soren Brinkmann <soren.brinkmann@xilinx.com> > >> Tested-by: John Linn <linnj@xilinx.com> > > > > As pointed out by Grygorii in > > commit aca82d1cbb49af34b69ecd4571a0fe48ad9247c1: > > > > The PM runtime API can't be used in atomic contex on -RT even if > > it's configured as irqsafe. As result, below error report can > > be seen when PM runtime API called from IRQ chip's callbacks > > irq_startup/irq_shutdown/irq_set_type, because they are > > protected by RAW spinlock: > > (...) > > The IRQ chip interface defines only two callbacks which are executed in > > non-atomic contex - irq_bus_lock/irq_bus_sync_unlock, so lets move > > PM runtime calls there. > > > > I.e. these calls are atomic context and it's just luck that it works > > and this is fragile. > > > > Can you please check if you can move it to > > irq_bus_lock()/irq_sync_unlock() > > like Grygorii does? > > That only powers up the chip when the chip is accessed. For proper IRQ > operation the chip needs to be powered up though as long as the IRQ is > enabled. request_irq() and free_irq() must always be called from sleepable > context. The thing is just that request_resource/release_resource are called > from within a raw spinlock, which is necessary since otherwise you can't > guarantee that they are only called once for shared interrupts. > > It might make sense to add a separate set of callbacks to the irq_chip > struct that are called from the sleepable sections of > request_irq()/free_irq() which are meant for power management purposes and > which wont have the guarantee that they are only called once for shared IRQs > (but are still balanced). Let me try to summarize what I've heard so far: - reqres/relres are called from atomic context, hence must not sleep - pm_runtime API must not be used from atomic context, even when the implementation of the callbacks does not sleep - when overriding regres/relres, a driver must re-implement the default behavior the core provides - bus lock/unlock is not sufficient for this case because it doesn't keep the device on as long as an IRQ is expected - a new pair of gpiochip ops might be helpful Does that summarize the current situation correctly? If so, I'd tend to agree with Lars that we might need another pair of callbacks in the gpiochip struct. Thanks, Sören
On Tue, 27 Oct 2015, Lars-Peter Clausen wrote: > On 10/27/2015 04:53 PM, Linus Walleij wrote: > > On Fri, Oct 23, 2015 at 3:36 PM, Soren Brinkmann > > <soren.brinkmann@xilinx.com> wrote: > > > >> The driver uses runtime PM to leverage low power techniques. For > >> use-cases using GPIO as interrupt the device needs to be in an > >> appropriate state. > >> > >> Reported-by: John Linn <linnj@xilinx.com> > >> Signed-off-by: Soren Brinkmann <soren.brinkmann@xilinx.com> > >> Tested-by: John Linn <linnj@xilinx.com> > > > > As pointed out by Grygorii in > > commit aca82d1cbb49af34b69ecd4571a0fe48ad9247c1: > > > > The PM runtime API can't be used in atomic contex on -RT even if > > it's configured as irqsafe. As result, below error report can > > be seen when PM runtime API called from IRQ chip's callbacks > > irq_startup/irq_shutdown/irq_set_type, because they are > > protected by RAW spinlock: > > (...) > > The IRQ chip interface defines only two callbacks which are executed in > > non-atomic contex - irq_bus_lock/irq_bus_sync_unlock, so lets move > > PM runtime calls there. > > > > I.e. these calls are atomic context and it's just luck that it works > > and this is fragile. > > > > Can you please check if you can move it to > > irq_bus_lock()/irq_sync_unlock() > > like Grygorii does? > > That only powers up the chip when the chip is accessed. For proper IRQ > operation the chip needs to be powered up though as long as the IRQ is > enabled. request_irq() and free_irq() must always be called from sleepable > context. The thing is just that request_resource/release_resource are called > from within a raw spinlock, which is necessary since otherwise you can't > guarantee that they are only called once for shared interrupts. > > It might make sense to add a separate set of callbacks to the irq_chip > struct that are called from the sleepable sections of > request_irq()/free_irq() which are meant for power management purposes and > which wont have the guarantee that they are only called once for shared IRQs > (but are still balanced). > > Thomas, do you have any thoughts on this? If you want to keep the chip powered as long as an interrupt is enabled, then having a irq chip callback might be the proper solution. Thanks, tglx
diff --git a/drivers/gpio/gpio-zynq.c b/drivers/gpio/gpio-zynq.c index 8abeacac5885..0d1669416112 100644 --- a/drivers/gpio/gpio-zynq.c +++ b/drivers/gpio/gpio-zynq.c @@ -471,6 +471,22 @@ static int zynq_gpio_set_wake(struct irq_data *data, unsigned int on) return 0; } +int zynq_gpio_irq_request_resources(struct irq_data *data) +{ + struct zynq_gpio *gpio = irq_data_get_irq_chip_data(data); + struct device *dev = gpio->chip.dev; + + return pm_runtime_get_sync(dev); +} + +void zynq_gpio_irq_release_resources(struct irq_data *data) +{ + struct zynq_gpio *gpio = irq_data_get_irq_chip_data(data); + struct device *dev = gpio->chip.dev; + + pm_runtime_put(dev); +} + /* irq chip descriptor */ static struct irq_chip zynq_gpio_level_irqchip = { .name = DRIVER_NAME, @@ -480,6 +496,8 @@ static struct irq_chip zynq_gpio_level_irqchip = { .irq_unmask = zynq_gpio_irq_unmask, .irq_set_type = zynq_gpio_set_irq_type, .irq_set_wake = zynq_gpio_set_wake, + .irq_request_resources = zynq_gpio_irq_request_resources, + .irq_release_resources = zynq_gpio_irq_release_resources, .flags = IRQCHIP_EOI_THREADED | IRQCHIP_EOI_IF_HANDLED | IRQCHIP_MASK_ON_SUSPEND, }; @@ -492,6 +510,8 @@ static struct irq_chip zynq_gpio_edge_irqchip = { .irq_unmask = zynq_gpio_irq_unmask, .irq_set_type = zynq_gpio_set_irq_type, .irq_set_wake = zynq_gpio_set_wake, + .irq_request_resources = zynq_gpio_irq_request_resources, + .irq_release_resources = zynq_gpio_irq_release_resources, .flags = IRQCHIP_MASK_ON_SUSPEND, };