Let's drop the use of pm_runtime_irq_safe() for 8250_omap. The use of
pm_runtime_irq_safe() is not nice as it takes a permanent usage count on
the parent device.
We can finally drop pm_runtime_irq_safe() safely as the kernel now knows
when the uart port tx is active. This changed with commit 84a9582fd203
("serial: core: Start managing serial controllers to enable runtime PM").
For serial port rx, we already use Linux generic wakeirqs for 8250_omap.
To drop pm_runtime_irq_safe(), we need to add handling for shallow idle
state where the port hardware may already be awake and an IO interrupt
happens. We also need to replace the serial8250_rpm sync calls in the
interrupt handlers with async runtime PM calls.
Note that omap8250_irq() calls omap_8250_dma_handle_irq(), so we don't
need separate runtime PM calls in omap_8250_dma_handle_irq().
While at it, let's also add the missing line break to the end of
omap8250_runtime_resume() to group the calls.
Signed-off-by: Tony Lindgren <[email protected]>
---
Changes since v1:
- Fix build warning if CONFIG_SERIAL_8250_DMA is not set as reported by
kernel test robot
---
drivers/tty/serial/8250/8250_omap.c | 29 +++++++++++++++++++++--------
1 file changed, 21 insertions(+), 8 deletions(-)
diff --git a/drivers/tty/serial/8250/8250_omap.c b/drivers/tty/serial/8250/8250_omap.c
--- a/drivers/tty/serial/8250/8250_omap.c
+++ b/drivers/tty/serial/8250/8250_omap.c
@@ -8,6 +8,7 @@
*
*/
+#include <linux/atomic.h>
#include <linux/clk.h>
#include <linux/device.h>
#include <linux/io.h>
@@ -130,6 +131,7 @@ struct omap8250_priv {
u8 tx_trigger;
u8 rx_trigger;
+ atomic_t active;
bool is_suspending;
int wakeirq;
int wakeups_enabled;
@@ -632,14 +634,23 @@ static irqreturn_t omap8250_irq(int irq, void *dev_id)
unsigned int iir, lsr;
int ret;
+ pm_runtime_get_noresume(port->dev);
+
+ /* Shallow idle state wake-up to an IO interrupt? */
+ if (atomic_add_unless(&priv->active, 1, 1)) {
+ priv->latency = priv->calc_latency;
+ schedule_work(&priv->qos_work);
+ }
+
#ifdef CONFIG_SERIAL_8250_DMA
if (up->dma) {
ret = omap_8250_dma_handle_irq(port);
+ pm_runtime_mark_last_busy(port->dev);
+ pm_runtime_put(port->dev);
return IRQ_RETVAL(ret);
}
#endif
- serial8250_rpm_get(up);
lsr = serial_port_in(port, UART_LSR);
iir = serial_port_in(port, UART_IIR);
ret = serial8250_handle_irq(port, iir);
@@ -676,7 +687,8 @@ static irqreturn_t omap8250_irq(int irq, void *dev_id)
schedule_delayed_work(&up->overrun_backoff, delay);
}
- serial8250_rpm_put(up);
+ pm_runtime_mark_last_busy(port->dev);
+ pm_runtime_put(port->dev);
return IRQ_RETVAL(ret);
}
@@ -1270,11 +1282,8 @@ static int omap_8250_dma_handle_irq(struct uart_port *port)
u16 status;
u8 iir;
- serial8250_rpm_get(up);
-
iir = serial_port_in(port, UART_IIR);
if (iir & UART_IIR_NO_INT) {
- serial8250_rpm_put(up);
return IRQ_HANDLED;
}
@@ -1305,7 +1314,6 @@ static int omap_8250_dma_handle_irq(struct uart_port *port)
uart_unlock_and_check_sysrq(port);
- serial8250_rpm_put(up);
return 1;
}
@@ -1503,8 +1511,6 @@ static int omap8250_probe(struct platform_device *pdev)
if (!of_get_available_child_count(pdev->dev.of_node))
pm_runtime_set_autosuspend_delay(&pdev->dev, -1);
- pm_runtime_irq_safe(&pdev->dev);
-
pm_runtime_get_sync(&pdev->dev);
omap_serial_fill_features_erratas(&up, priv);
@@ -1743,6 +1749,7 @@ static int omap8250_runtime_suspend(struct device *dev)
priv->latency = PM_QOS_CPU_LATENCY_DEFAULT_VALUE;
schedule_work(&priv->qos_work);
+ atomic_set(&priv->active, 0);
return 0;
}
@@ -1752,6 +1759,10 @@ static int omap8250_runtime_resume(struct device *dev)
struct omap8250_priv *priv = dev_get_drvdata(dev);
struct uart_8250_port *up = NULL;
+ /* Did the hardware wake to a device IO interrupt before a wakeirq? */
+ if (atomic_read(&priv->active))
+ return 0;
+
if (priv->line >= 0)
up = serial8250_get_port(priv->line);
@@ -1767,8 +1778,10 @@ static int omap8250_runtime_resume(struct device *dev)
uart_port_unlock_irq(&up->port);
}
+ atomic_set(&priv->active, 1);
priv->latency = priv->calc_latency;
schedule_work(&priv->qos_work);
+
return 0;
}
--
2.42.0
On Wed, Oct 04, 2023 at 09:26:48AM +0300, Tony Lindgren wrote:
> Let's drop the use of pm_runtime_irq_safe() for 8250_omap. The use of
> pm_runtime_irq_safe() is not nice as it takes a permanent usage count on
> the parent device.
>
> We can finally drop pm_runtime_irq_safe() safely as the kernel now knows
> when the uart port tx is active. This changed with commit 84a9582fd203
> ("serial: core: Start managing serial controllers to enable runtime PM").
>
> For serial port rx, we already use Linux generic wakeirqs for 8250_omap.
>
> To drop pm_runtime_irq_safe(), we need to add handling for shallow idle
> state where the port hardware may already be awake and an IO interrupt
> happens. We also need to replace the serial8250_rpm sync calls in the
> interrupt handlers with async runtime PM calls.
>
> Note that omap8250_irq() calls omap_8250_dma_handle_irq(), so we don't
> need separate runtime PM calls in omap_8250_dma_handle_irq().
>
> While at it, let's also add the missing line break to the end of
> omap8250_runtime_resume() to group the calls.
Yoo-hoo! A few years only to get to the point :-)
Reviewed-by: Andy Shevchenko <[email protected]>
--
With Best Regards,
Andy Shevchenko