| Message ID | 1351238130-7158-1-git-send-email-tim.niemeyer@corscience.de (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Hi, On Fri, Oct 26, 2012 at 09:55:30AM +0200, Tim Niemeyer wrote: > Adds support for configuring the omap-gpio driver use autosuspend for > runtime power management. This can reduce the latency in using it by > not suspending the device immediately on idle. If another access takes > place before the autosuspend timeout (2 secs), the call to resume the > device can return immediately saving some save/ restore cycles. > > I use a gpio to monitor a spi transfer which occurs every 250µs. The > suspend overhead is to high, so almost every second transfer is lost. > This patch fixes that. > > Signed-off-by: Tim Niemeyer <tim.niemeyer@corscience.de> > --- > drivers/gpio/gpio-omap.c | 23 ++++++++++++++++++----- > 1 files changed, 18 insertions(+), 5 deletions(-) > > diff --git a/drivers/gpio/gpio-omap.c b/drivers/gpio/gpio-omap.c > index 94cbc84..92f48cb 100644 > --- a/drivers/gpio/gpio-omap.c > +++ b/drivers/gpio/gpio-omap.c > @@ -31,6 +31,7 @@ > #include <asm/mach/irq.h> > > #define OFF_MODE 1 > +#define GPIO_AUTOSUSPEND_TIMEOUT 2000 > > static LIST_HEAD(omap_gpio_list); > > @@ -628,8 +629,10 @@ static void omap_gpio_free(struct gpio_chip *chip, unsigned offset) > * If this is the last gpio to be freed in the bank, > * disable the bank module. > */ > - if (!bank->mod_usage) > - pm_runtime_put(bank->dev); > + if (!bank->mod_usage) { while at that I would drop this bank->mod_usage nonsense and let power core handle reference counting. > + pm_runtime_mark_last_busy(bank->dev); > + pm_runtime_put_autosuspend(bank->dev); > + } > } > > /* > @@ -715,7 +718,8 @@ static void gpio_irq_handler(unsigned int irq, struct irq_desc *desc) > exit: > if (!unmasked) > chained_irq_exit(chip, desc); > - pm_runtime_put(bank->dev); > + pm_runtime_mark_last_busy(bank->dev); > + pm_runtime_put_autosuspend(bank->dev); > } > > static void gpio_irq_shutdown(struct irq_data *d) > @@ -1132,6 +1136,8 @@ static int __devinit omap_gpio_probe(struct platform_device *pdev) > > platform_set_drvdata(pdev, bank); > > + pm_runtime_use_autosuspend(bank->dev); > + pm_runtime_set_autosuspend_delay(bank->dev, GPIO_AUTOSUSPEND_TIMEOUT); > pm_runtime_enable(bank->dev); > pm_runtime_irq_safe(bank->dev); > pm_runtime_get_sync(bank->dev); > @@ -1146,7 +1152,8 @@ static int __devinit omap_gpio_probe(struct platform_device *pdev) > if (bank->loses_context) > bank->get_context_loss_count = pdata->get_context_loss_count; > > - pm_runtime_put(bank->dev); > + pm_runtime_mark_last_busy(bank->dev); > + pm_runtime_put_autosuspend(bank->dev); > > list_add_tail(&bank->node, &omap_gpio_list); > > @@ -1333,7 +1340,13 @@ void omap2_gpio_prepare_for_idle(int pwr_mode) > > bank->power_mode = pwr_mode; > > - pm_runtime_put_sync_suspend(bank->dev); > + /* direct pm_runtime on pwroff */ > + if (pwr_mode) you also need braces here. > + pm_runtime_put_sync_suspend(bank->dev); > + else { > + pm_runtime_mark_last_busy(bank->dev); > + pm_runtime_put_sync_autosuspend(bank->dev); > + } > } > } > > -- > 1.7.2.5 > > -- > To unsubscribe from this list: send the line "unsubscribe linux-omap" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html
Am Freitag, den 26.10.2012, 11:03 +0300 schrieb Felipe Balbi: > Hi, > > On Fri, Oct 26, 2012 at 09:55:30AM +0200, Tim Niemeyer wrote: > > Adds support for configuring the omap-gpio driver use autosuspend for > > runtime power management. This can reduce the latency in using it by > > not suspending the device immediately on idle. If another access takes > > place before the autosuspend timeout (2 secs), the call to resume the > > device can return immediately saving some save/ restore cycles. > > > > I use a gpio to monitor a spi transfer which occurs every 250µs. The > > suspend overhead is to high, so almost every second transfer is lost. > > This patch fixes that. > > > > Signed-off-by: Tim Niemeyer <tim.niemeyer@corscience.de> > > --- > > drivers/gpio/gpio-omap.c | 23 ++++++++++++++++++----- > > 1 files changed, 18 insertions(+), 5 deletions(-) > > > > diff --git a/drivers/gpio/gpio-omap.c b/drivers/gpio/gpio-omap.c > > index 94cbc84..92f48cb 100644 > > --- a/drivers/gpio/gpio-omap.c > > +++ b/drivers/gpio/gpio-omap.c > > @@ -31,6 +31,7 @@ > > #include <asm/mach/irq.h> > > > > #define OFF_MODE 1 > > +#define GPIO_AUTOSUSPEND_TIMEOUT 2000 > > > > static LIST_HEAD(omap_gpio_list); > > > > @@ -628,8 +629,10 @@ static void omap_gpio_free(struct gpio_chip *chip, unsigned offset) > > * If this is the last gpio to be freed in the bank, > > * disable the bank module. > > */ > > - if (!bank->mod_usage) > > - pm_runtime_put(bank->dev); > > + if (!bank->mod_usage) { > > while at that I would drop this bank->mod_usage nonsense and let > power core handle reference counting. I looked at it, but i'm unsure about the GPIO_MOD_CTRL_BIT. The bank->mod_usage counter prevents omap_gpio_free() to disable the hole bank while another gpio is still in use. > > + pm_runtime_mark_last_busy(bank->dev); > > + pm_runtime_put_autosuspend(bank->dev); > > + } > > } > > > > /* > > @@ -715,7 +718,8 @@ static void gpio_irq_handler(unsigned int irq, struct irq_desc *desc) > > exit: > > if (!unmasked) > > chained_irq_exit(chip, desc); > > - pm_runtime_put(bank->dev); > > + pm_runtime_mark_last_busy(bank->dev); > > + pm_runtime_put_autosuspend(bank->dev); > > } > > > > static void gpio_irq_shutdown(struct irq_data *d) > > @@ -1132,6 +1136,8 @@ static int __devinit omap_gpio_probe(struct platform_device *pdev) > > > > platform_set_drvdata(pdev, bank); > > > > + pm_runtime_use_autosuspend(bank->dev); > > + pm_runtime_set_autosuspend_delay(bank->dev, GPIO_AUTOSUSPEND_TIMEOUT); > > pm_runtime_enable(bank->dev); > > pm_runtime_irq_safe(bank->dev); > > pm_runtime_get_sync(bank->dev); > > @@ -1146,7 +1152,8 @@ static int __devinit omap_gpio_probe(struct platform_device *pdev) > > if (bank->loses_context) > > bank->get_context_loss_count = pdata->get_context_loss_count; > > > > - pm_runtime_put(bank->dev); > > + pm_runtime_mark_last_busy(bank->dev); > > + pm_runtime_put_autosuspend(bank->dev); > > > > list_add_tail(&bank->node, &omap_gpio_list); > > > > @@ -1333,7 +1340,13 @@ void omap2_gpio_prepare_for_idle(int pwr_mode) > > > > bank->power_mode = pwr_mode; > > > > - pm_runtime_put_sync_suspend(bank->dev); > > + /* direct pm_runtime on pwroff */ > > + if (pwr_mode) > > you also need braces here. Yes. I resend this later. > > + pm_runtime_put_sync_suspend(bank->dev); > > + else { > > + pm_runtime_mark_last_busy(bank->dev); > > + pm_runtime_put_sync_autosuspend(bank->dev); > > + } > > } > > } > > > > -- > > 1.7.2.5 > > > > -- > > To unsubscribe from this list: send the line "unsubscribe linux-omap" in > > the body of a message to majordomo@vger.kernel.org > > More majordomo info at http://vger.kernel.org/majordomo-info.html >
On Fri, Oct 26, 2012 at 12:42:35PM +0200, Tim Niemeyer wrote: > Am Freitag, den 26.10.2012, 11:03 +0300 schrieb Felipe Balbi: > > Hi, > > > > On Fri, Oct 26, 2012 at 09:55:30AM +0200, Tim Niemeyer wrote: > > > Adds support for configuring the omap-gpio driver use autosuspend for > > > runtime power management. This can reduce the latency in using it by > > > not suspending the device immediately on idle. If another access takes > > > place before the autosuspend timeout (2 secs), the call to resume the > > > device can return immediately saving some save/ restore cycles. > > > > > > I use a gpio to monitor a spi transfer which occurs every 250µs. The > > > suspend overhead is to high, so almost every second transfer is lost. > > > This patch fixes that. > > > > > > Signed-off-by: Tim Niemeyer <tim.niemeyer@corscience.de> > > > --- > > > drivers/gpio/gpio-omap.c | 23 ++++++++++++++++++----- > > > 1 files changed, 18 insertions(+), 5 deletions(-) > > > > > > diff --git a/drivers/gpio/gpio-omap.c b/drivers/gpio/gpio-omap.c > > > index 94cbc84..92f48cb 100644 > > > --- a/drivers/gpio/gpio-omap.c > > > +++ b/drivers/gpio/gpio-omap.c > > > @@ -31,6 +31,7 @@ > > > #include <asm/mach/irq.h> > > > > > > #define OFF_MODE 1 > > > +#define GPIO_AUTOSUSPEND_TIMEOUT 2000 > > > > > > static LIST_HEAD(omap_gpio_list); > > > > > > @@ -628,8 +629,10 @@ static void omap_gpio_free(struct gpio_chip *chip, unsigned offset) > > > * If this is the last gpio to be freed in the bank, > > > * disable the bank module. > > > */ > > > - if (!bank->mod_usage) > > > - pm_runtime_put(bank->dev); > > > + if (!bank->mod_usage) { > > > > while at that I would drop this bank->mod_usage nonsense and let > > power core handle reference counting. > I looked at it, but i'm unsure about the GPIO_MOD_CTRL_BIT. The > bank->mod_usage counter prevents omap_gpio_free() to disable the hole > bank while another gpio is still in use. pm core's usage count will do the same thing. If you request 6 gpios, you will have 6 pm_runtime_get(). If you free 5 of those, you will have 5 pm_runtime_put_autosuspend() which will only decrement the counter and do nothing else.
diff --git a/drivers/gpio/gpio-omap.c b/drivers/gpio/gpio-omap.c index 94cbc84..92f48cb 100644 --- a/drivers/gpio/gpio-omap.c +++ b/drivers/gpio/gpio-omap.c @@ -31,6 +31,7 @@ #include <asm/mach/irq.h> #define OFF_MODE 1 +#define GPIO_AUTOSUSPEND_TIMEOUT 2000 static LIST_HEAD(omap_gpio_list); @@ -628,8 +629,10 @@ static void omap_gpio_free(struct gpio_chip *chip, unsigned offset) * If this is the last gpio to be freed in the bank, * disable the bank module. */ - if (!bank->mod_usage) - pm_runtime_put(bank->dev); + if (!bank->mod_usage) { + pm_runtime_mark_last_busy(bank->dev); + pm_runtime_put_autosuspend(bank->dev); + } } /* @@ -715,7 +718,8 @@ static void gpio_irq_handler(unsigned int irq, struct irq_desc *desc) exit: if (!unmasked) chained_irq_exit(chip, desc); - pm_runtime_put(bank->dev); + pm_runtime_mark_last_busy(bank->dev); + pm_runtime_put_autosuspend(bank->dev); } static void gpio_irq_shutdown(struct irq_data *d) @@ -1132,6 +1136,8 @@ static int __devinit omap_gpio_probe(struct platform_device *pdev) platform_set_drvdata(pdev, bank); + pm_runtime_use_autosuspend(bank->dev); + pm_runtime_set_autosuspend_delay(bank->dev, GPIO_AUTOSUSPEND_TIMEOUT); pm_runtime_enable(bank->dev); pm_runtime_irq_safe(bank->dev); pm_runtime_get_sync(bank->dev); @@ -1146,7 +1152,8 @@ static int __devinit omap_gpio_probe(struct platform_device *pdev) if (bank->loses_context) bank->get_context_loss_count = pdata->get_context_loss_count; - pm_runtime_put(bank->dev); + pm_runtime_mark_last_busy(bank->dev); + pm_runtime_put_autosuspend(bank->dev); list_add_tail(&bank->node, &omap_gpio_list); @@ -1333,7 +1340,13 @@ void omap2_gpio_prepare_for_idle(int pwr_mode) bank->power_mode = pwr_mode; - pm_runtime_put_sync_suspend(bank->dev); + /* direct pm_runtime on pwroff */ + if (pwr_mode) + pm_runtime_put_sync_suspend(bank->dev); + else { + pm_runtime_mark_last_busy(bank->dev); + pm_runtime_put_sync_autosuspend(bank->dev); + } } }
Adds support for configuring the omap-gpio driver use autosuspend for runtime power management. This can reduce the latency in using it by not suspending the device immediately on idle. If another access takes place before the autosuspend timeout (2 secs), the call to resume the device can return immediately saving some save/ restore cycles. I use a gpio to monitor a spi transfer which occurs every 250µs. The suspend overhead is to high, so almost every second transfer is lost. This patch fixes that. Signed-off-by: Tim Niemeyer <tim.niemeyer@corscience.de> --- drivers/gpio/gpio-omap.c | 23 ++++++++++++++++++----- 1 files changed, 18 insertions(+), 5 deletions(-)