Some PHY drivers can reliably report link-down events via IRQs,
but may fail to generate reliable link-up IRQs. To support such
cases, polling is often needed - but only selectively.

Extend get_next_update_time() so drivers can return PHY_STATE_IRQ
to indicate that polling is not needed and IRQs are sufficient.
This allows finer control over PHY state machine behavior.

Introduce PHY_STATE_IRQ (UINT_MAX) as a sentinel value, and move
PHY_STATE_TIME to phy.h to allow consistent use across the codebase.

This change complements the previous patch enabling polling when
get_next_update_time() is present.

Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
---
changes v2:
- this patch is added
---
 drivers/net/phy/phy.c | 18 +++++++++++-------
 include/linux/phy.h   | 10 +++++++++-
 2 files changed, 20 insertions(+), 8 deletions(-)

diff --git a/drivers/net/phy/phy.c b/drivers/net/phy/phy.c
index 13df28445f02..faf9a48d3b6f 100644
--- a/drivers/net/phy/phy.c
+++ b/drivers/net/phy/phy.c
@@ -39,8 +39,6 @@
 #include "phylib-internal.h"
 #include "phy-caps.h"
 
-#define PHY_STATE_TIME	HZ
-
 #define PHY_STATE_STR(_state)			\
 	case PHY_##_state:			\
 		return __stringify(_state);	\
@@ -1575,16 +1573,22 @@ static enum phy_state_work _phy_state_machine(struct phy_device *phydev)
 	phy_process_state_change(phydev, old_state);
 
 	/* Only re-schedule a PHY state machine change if we are polling the
-	 * PHY, if PHY_MAC_INTERRUPT is set, then we will be moving
-	 * between states from phy_mac_interrupt().
+	 * PHY. If PHY_MAC_INTERRUPT is set or get_next_update_time() returns
+	 * PHY_STATE_IRQ, then we rely on interrupts for state changes.
 	 *
 	 * In state PHY_HALTED the PHY gets suspended, so rescheduling the
 	 * state machine would be pointless and possibly error prone when
 	 * called from phy_disconnect() synchronously.
 	 */
-	if (phy_polling_mode(phydev) && phy_is_started(phydev))
-		phy_queue_state_machine(phydev,
-					phy_get_next_update_time(phydev));
+	if (phy_polling_mode(phydev) && phy_is_started(phydev)) {
+		unsigned int next_time = phy_get_next_update_time(phydev);
+
+		/* Drivers returning PHY_STATE_IRQ opt out of polling.
+		 * Use IRQ-only mode by not re-queuing the state machine.
+		 */
+		if (next_time != PHY_STATE_IRQ)
+			phy_queue_state_machine(phydev, next_time);
+	}
 
 	return state_work;
 }
diff --git a/include/linux/phy.h b/include/linux/phy.h
index 3d4e5c41235e..d92258e3ac1a 100644
--- a/include/linux/phy.h
+++ b/include/linux/phy.h
@@ -66,6 +66,10 @@ extern const int phy_basic_ports_array[3];
 #define PHY_ALWAYS_CALL_SUSPEND	0x00000008
 #define MDIO_DEVICE_IS_PHY	0x80000000
 
+#define PHY_STATE_TIME		HZ
+/* disable polling, rely on IRQs */
+#define PHY_STATE_IRQ		UINT_MAX
+
 /**
  * enum phy_interface_t - Interface Mode definitions
  *
@@ -1257,7 +1261,11 @@ struct phy_driver {
 	 * dynamically adjust polling intervals based on link state or other
 	 * conditions.
 	 *
-	 * Returns the time in jiffies until the next update event.
+	 * Returning PHY_STATE_IRQ disables polling and indicates that the
+	 * driver relies solely on IRQs for link state changes.
+	 *
+	 * Returns the time in jiffies until the next update event, or
+	 * PHY_STATE_IRQ to disable polling.
 	 */
 	unsigned int (*get_next_update_time)(struct phy_device *dev);
 };
-- 
2.39.5

>  	/* Only re-schedule a PHY state machine change if we are polling the
> -	 * PHY, if PHY_MAC_INTERRUPT is set, then we will be moving
> -	 * between states from phy_mac_interrupt().
> +	 * PHY. If PHY_MAC_INTERRUPT is set or get_next_update_time() returns
> +	 * PHY_STATE_IRQ, then we rely on interrupts for state changes.
>  	 *
>  	 * In state PHY_HALTED the PHY gets suspended, so rescheduling the
>  	 * state machine would be pointless and possibly error prone when
>  	 * called from phy_disconnect() synchronously.
>  	 */
> -	if (phy_polling_mode(phydev) && phy_is_started(phydev))
> -		phy_queue_state_machine(phydev,
> -					phy_get_next_update_time(phydev));
> +	if (phy_polling_mode(phydev) && phy_is_started(phydev)) {
> +		unsigned int next_time = phy_get_next_update_time(phydev);
> +
> +		/* Drivers returning PHY_STATE_IRQ opt out of polling.
> +		 * Use IRQ-only mode by not re-queuing the state machine.
> +		 */
> +		if (next_time != PHY_STATE_IRQ)
> +			phy_queue_state_machine(phydev, next_time);
> +	}

How does this interact with update_stats()?

phy_polling_mode() returns true because the update_stats() op is
implemented. phy_get_next_update_time() returns PHY_STATE_IRQ, because
the PHY is in a state where interrupts works, and then the statistics
overflow.

It seems like this code needs to be somehow made part of
phy_polling_mode(), so that it has the full picture of why polling is
being used.

    Andrew

---
pw-bot: cr

On Wed, Jul 09, 2025 at 04:10:48PM +0200, Andrew Lunn wrote:
> >  	/* Only re-schedule a PHY state machine change if we are polling the
> > -	 * PHY, if PHY_MAC_INTERRUPT is set, then we will be moving
> > -	 * between states from phy_mac_interrupt().
> > +	 * PHY. If PHY_MAC_INTERRUPT is set or get_next_update_time() returns
> > +	 * PHY_STATE_IRQ, then we rely on interrupts for state changes.
> >  	 *
> >  	 * In state PHY_HALTED the PHY gets suspended, so rescheduling the
> >  	 * state machine would be pointless and possibly error prone when
> >  	 * called from phy_disconnect() synchronously.
> >  	 */
> > -	if (phy_polling_mode(phydev) && phy_is_started(phydev))
> > -		phy_queue_state_machine(phydev,
> > -					phy_get_next_update_time(phydev));
> > +	if (phy_polling_mode(phydev) && phy_is_started(phydev)) {
> > +		unsigned int next_time = phy_get_next_update_time(phydev);
> > +
> > +		/* Drivers returning PHY_STATE_IRQ opt out of polling.
> > +		 * Use IRQ-only mode by not re-queuing the state machine.
> > +		 */
> > +		if (next_time != PHY_STATE_IRQ)
> > +			phy_queue_state_machine(phydev, next_time);
> > +	}
> 
> How does this interact with update_stats()?
> 
> phy_polling_mode() returns true because the update_stats() op is
> implemented. phy_get_next_update_time() returns PHY_STATE_IRQ, because
> the PHY is in a state where interrupts works, and then the statistics
> overflow.
> 
> It seems like this code needs to be somehow made part of
> phy_polling_mode(), so that it has the full picture of why polling is
> being used.

Ah, good point! I forgot about it.

-- 
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |