Skip to contentSkip to conversation

net: phy: improve handling of more complex C45 PHY's

PATCH In review Heiner Kallweit 5 participants 8 messages May 30, 2019 – Nov 19, 2019

Subsystems (3)
lore
Share & tools
lore

Heiner Kallweit

author maintainer
#0 lore raw

v2 cover net: phy: improve handling of more complex C45 PHY's

This series tries to address few problematic aspects raised by Russell. Concrete example is the Marvell 88x3310, the changes should be helpful for other complex C45 PHY's too.

v2:
- added patch enabling interrupts also if phylib state machine isn't started
- removed patch dealing with the double link status read This one needs little bit more thinking and will go separately.

Heiner Kallweit (3):
net: phy: enable interrupts when PHY is attached already
net: phy: add callback for custom interrupt handler to struct
phy_driver
net: phy: export phy_queue_state_machine

 drivers/net/phy/phy.c        | 53 +++++++++++++++++++++++-------------
 drivers/net/phy/phy_device.c |  2 +-
 include/linux/phy.h          |  6 +++-
 3 files changed, 40 insertions(+), 21 deletions(-)
Signature · 1 line
2.21.0
author maintainer
#1 lore raw

PATCH v2 1/3 net: phy: enable interrupts when PHY is attached already

This patch is a step towards allowing PHY drivers to handle more interrupt sources than just link change. E.g. several PHY's have built-in temperature monitoring and can raise an interrupt if a temperature threshold is exceeded. We may be interested in such interrupts also if the phylib state machine isn't started. Therefore move enabling interrupts to phy_request_interrupt().

v2:
- patch added to series

---
 drivers/net/phy/phy.c        | 36 ++++++++++++++++++++++--------------
 drivers/net/phy/phy_device.c |  2 +-
 include/linux/phy.h          |  1 +
 3 files changed, 24 insertions(+), 15 deletions(-)
drivers/net/phy/phy.c +22 -14
index e88854292..d90d9863e 100644
@@ -799,10 +799,10 @@ static int phy_enable_interrupts(struct phy_device *phydev)
 }
 
 /**
- * phy_request_interrupt - request interrupt for a PHY device
+ * phy_request_interrupt - request and enable interrupt for a PHY device
  * @phydev: target phy_device struct
  *
- * Description: Request the interrupt for the given PHY.
+ * Description: Request and enable the interrupt for the given PHY.
  *   If this fails, then we set irq to PHY_POLL.
  *   This should only be called with a valid IRQ number.
  */
@@ -817,10 +817,30 @@ void phy_request_interrupt(struct phy_device *phydev)
 		phydev_warn(phydev, "Error %d requesting IRQ %d, falling back to polling\n",
 			    err, phydev->irq);
 		phydev->irq = PHY_POLL;
+	} else {
+		if (phy_enable_interrupts(phydev)) {
+			phydev_warn(phydev, "Can't enable interrupt, falling back to polling\n");
+			phy_free_interrupt(phydev);
+			phydev->irq = PHY_POLL;
+		}
 	}
 }
 EXPORT_SYMBOL(phy_request_interrupt);
 
+/**
+ * phy_free_interrupt - disable and free interrupt for a PHY device
+ * @phydev: target phy_device struct
+ *
+ * Description: Disable and free the interrupt for the given PHY.
+ *   This should only be called with a valid IRQ number.
+ */
+void phy_free_interrupt(struct phy_device *phydev)
+{
+	phy_disable_interrupts(phydev);
+	free_irq(phydev->irq, phydev);
+}
+EXPORT_SYMBOL(phy_free_interrupt);
+
 /**
  * phy_stop - Bring down the PHY link, and stop checking the status
  * @phydev: target phy_device struct
@@ -835,9 +855,6 @@ void phy_stop(struct phy_device *phydev)
 
 	mutex_lock(&phydev->lock);
 
-	if (phy_interrupt_is_valid(phydev))
-		phy_disable_interrupts(phydev);
-
 	phydev->state = PHY_HALTED;
 
 	mutex_unlock(&phydev->lock);
@@ -864,8 +881,6 @@ EXPORT_SYMBOL(phy_stop);
  */
 void phy_start(struct phy_device *phydev)
 {
-	int err;
-
 	mutex_lock(&phydev->lock);
 
 	if (phydev->state != PHY_READY && phydev->state != PHY_HALTED) {
@@ -877,13 +892,6 @@ void phy_start(struct phy_device *phydev)
 	/* if phy was suspended, bring the physical link up again */
 	__phy_resume(phydev);
 
-	/* make sure interrupts are enabled for the PHY */
-	if (phy_interrupt_is_valid(phydev)) {
-		err = phy_enable_interrupts(phydev);
-		if (err < 0)
-			goto out;
-	}
-
 	phydev->state = PHY_UP;
 
 	phy_start_machine(phydev);
drivers/net/phy/phy_device.c +1 -1
index 5d288da9a..d71b3ed52 100644
@@ -1013,7 +1013,7 @@ void phy_disconnect(struct phy_device *phydev)
 		phy_stop(phydev);
 
 	if (phy_interrupt_is_valid(phydev))
-		free_irq(phydev->irq, phydev);
+		phy_free_interrupt(phydev);
 
 	phydev->adjust_link = NULL;
 
include/linux/phy.h +1 -0
index 7180b1d1e..72e1196f9 100644
@@ -1147,6 +1147,7 @@ int phy_ethtool_ksettings_set(struct phy_device *phydev,
 			      const struct ethtool_link_ksettings *cmd);
 int phy_mii_ioctl(struct phy_device *phydev, struct ifreq *ifr, int cmd);
 void phy_request_interrupt(struct phy_device *phydev);
+void phy_free_interrupt(struct phy_device *phydev);
 void phy_print_status(struct phy_device *phydev);
 int phy_set_max_speed(struct phy_device *phydev, u32 max_speed);
 void phy_remove_link_mode(struct phy_device *phydev, u32 link_mode);
Signature · 1 line
2.21.0
maintainer Reviewed-by on v2 1/3
#2 lore raw

Re: [PATCH net-next v2 1/3] net: phy: enable interrupts when PHY is attached already

Quoted 11 lines · Heiner Kallweit

This patch is a step towards allowing PHY drivers to handle more interrupt sources than just link change. E.g. several PHY's have built-in temperature monitoring and can raise an interrupt if a temperature threshold is exceeded. We may be interested in such interrupts also if the phylib state machine isn't started. Therefore move enabling interrupts to phy_request_interrupt().

v2:
- patch added to series

Andrew

author maintainer
#3 lore raw

PATCH v2 2/3 net: phy: add callback for custom interrupt handler to struct phy_driver

In reply to Heiner Kallweit

The phylib interrupt handler handles link change events only currently. However PHY drivers may want to use other interrupt sources too, e.g. to report temperature monitoring events. Therefore add a callback to struct phy_driver allowing PHY drivers to implement a custom interrupt handler.

---
 drivers/net/phy/phy.c | 9 +++++++--
 include/linux/phy.h   | 3 +++
 2 files changed, 10 insertions(+), 2 deletions(-)
drivers/net/phy/phy.c +7 -2
index d90d9863e..068f0a126 100644
@@ -772,8 +772,13 @@ static irqreturn_t phy_interrupt(int irq, void *phy_dat)
 	if (phydev->drv->did_interrupt && !phydev->drv->did_interrupt(phydev))
 		return IRQ_NONE;
 
-	/* reschedule state queue work to run as soon as possible */
-	phy_trigger_machine(phydev);
+	if (phydev->drv->handle_interrupt) {
+		if (phydev->drv->handle_interrupt(phydev))
+			goto phy_err;
+	} else {
+		/* reschedule state queue work to run as soon as possible */
+		phy_trigger_machine(phydev);
+	}
 
 	if (phy_clear_interrupt(phydev))
 		goto phy_err;
include/linux/phy.h +3 -0
index 72e1196f9..16cd33915 100644
@@ -537,6 +537,9 @@ struct phy_driver {
 	 */
 	int (*did_interrupt)(struct phy_device *phydev);
 
+	/* Override default interrupt handling */
+	int (*handle_interrupt)(struct phy_device *phydev);
+
 	/* Clears up any memory if needed */
 	void (*remove)(struct phy_device *phydev);
 
Signature · 1 line
2.21.0
#4 lore raw

Re: [PATCH net-next v2 2/3] net: phy: add callback for custom interrupt handler to struct phy_driver

Hi,

this is an old thread and I know its already applied. But I'd like to hear your opinion on the following problem below.

Quoted 19 lines

The phylib interrupt handler handles link change events only currently. However PHY drivers may want to use other interrupt sources too, e.g. to report temperature monitoring events. Therefore add a callback to struct phy_driver allowing PHY drivers to implement a custom interrupt handler.

---
 drivers/net/phy/phy.c | 9 +++++++--
 include/linux/phy.h   | 3 +++
 2 files changed, 10 insertions(+), 2 deletions(-)
drivers/net/phy/phy.c +0 -0
index d90d9863e..068f0a126 100644

drivers/net/phy/phy.c static irqreturn_t phy_interrupt(int irq, void · line 779

Quoted 10 lines
drivers/net/phy/phy.c +3 -2
@@ -772,8 +772,13 @@ static irqreturn_t phy_interrupt(int irq, void 
 *phy_dat)
 	if (phydev->drv->did_interrupt && 
 !phydev->drv->did_interrupt(phydev))
 		return IRQ_NONE;
 
-	/* reschedule state queue work to run as soon as possible */
-	phy_trigger_machine(phydev);
+	if (phydev->drv->handle_interrupt) {
+		if (phydev->drv->handle_interrupt(phydev))
+			goto phy_err;

There are PHYs which clears the interrupt already by reading the interrupt status register. To do something useful in handle_interrupt() I have to read the interrupt status register, thus clearing the pending interrupts.

drivers/net/phy/phy.c static irqreturn_t phy_interrupt(int irq, void *phy_dat) · line 784

Quoted 7 lines
drivers/net/phy/phy.c +4 -0
+	} else {
+		/* reschedule state queue work to run as soon as possible */
+		phy_trigger_machine(phydev);
+	}
 
drivers/net/phy/phy.c +0 -0
 	if (phy_clear_interrupt(phydev))
 		goto phy_err;

But here the interrupts are cleared again, which means we might loose interrupt causes in between.

I could think of two different fixes:
(1) handle_interrupt() has to take care to clear the interrupts and
skip the phy_clear_interrupt() above.
(2) handle_interrupt() might return a special return code which skips
the phy_clear_interrupt

TBH, I'd prefer (1) but I don't know if it is allowed to change semantics afterwards. (Also, I've found no driver where handle_interrupt() is actually used for now?)

-michael

#5 lore raw

Re: [PATCH net-next v2 2/3] net: phy: add callback for custom interrupt handler to struct phy_driver

Quoted 57 lines · Michael Walle

Hi,

this is an old thread and I know its already applied. But I'd like to hear your opinion on the following problem below.

Quoted 19 lines

The phylib interrupt handler handles link change events only currently. However PHY drivers may want to use other interrupt sources too, e.g. to report temperature monitoring events. Therefore add a callback to struct phy_driver allowing PHY drivers to implement a custom interrupt handler.

---
 drivers/net/phy/phy.c | 9 +++++++--
 include/linux/phy.h   | 3 +++
 2 files changed, 10 insertions(+), 2 deletions(-)
drivers/net/phy/phy.c +0 -0
index d90d9863e..068f0a126 100644

drivers/net/phy/phy.c static irqreturn_t phy_interrupt(int irq, void · line 778

Quoted 9 lines
drivers/net/phy/phy.c +3 -2
@@ -772,8 +772,13 @@ static irqreturn_t phy_interrupt(int irq, void
 *phy_dat)
 	if (phydev->drv->did_interrupt && !phydev->drv->did_interrupt(phydev))
 		return IRQ_NONE;
 
-	/* reschedule state queue work to run as soon as possible */
-	phy_trigger_machine(phydev);
+	if (phydev->drv->handle_interrupt) {
+		if (phydev->drv->handle_interrupt(phydev))
+			goto phy_err;

There are PHYs which clears the interrupt already by reading the interrupt status register. To do something useful in handle_interrupt() I have to read the interrupt status register, thus clearing the pending interrupts.

Quoted 7 lines
drivers/net/phy/phy.c +4 -0
+	} else {
+		/* reschedule state queue work to run as soon as possible */
+		phy_trigger_machine(phydev);
+	}
 
 	if (phy_clear_interrupt(phydev))

goto phy_err;

But here the interrupts are cleared again, which means we might loose interrupt causes in between.

I could think of two different fixes:
(1) handle_interrupt() has to take care to clear the interrupts and skip
the phy_clear_interrupt() above.
(2) handle_interrupt() might return a special return code which skips the
phy_clear_interrupt

Quoted 3 lines

TBH, I'd prefer (1) but I don't know if it is allowed to change semantics afterwards. (Also, I've found no driver where handle_interrupt() is actually used for now?)

I made the argument at the time that phylib should stop being a middle- layer, but instead let PHY drivers take care of interrupt handling themselves, just like we do elsewhere in the kernel. I think your case just shows that trying to keep the interrupt handling structured inside phylib and trying to make all PHYs fit is just going to be painful.

Signature · 3 lines
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 12.1Mbps down 622kbps up
According to speedtest.net: 11.9Mbps down 500kbps up
author maintainer
#6 lore raw

PATCH v2 3/3 net: phy: export phy_queue_state_machine

In reply to Heiner Kallweit

We face the issue that link change interrupt and link status may be reported by different PHY layers. As a result the link change interrupt may occur before the link status changes. Export phy_queue_state_machine to allow PHY drivers to specify a delay between link status change interrupt and link status check.

v2:
- change jiffies parameter type to unsigned long

---
 drivers/net/phy/phy.c | 8 +++++---
 include/linux/phy.h   | 2 +-
 2 files changed, 6 insertions(+), 4 deletions(-)
drivers/net/phy/phy.c +5 -3
index 068f0a126..74dfd424c 100644
@@ -29,6 +29,8 @@
 #include <linux/uaccess.h>
 #include <linux/atomic.h>
 
+#define PHY_STATE_TIME	HZ
+
 #define PHY_STATE_STR(_state)			\
 	case PHY_##_state:			\
 		return __stringify(_state);	\
@@ -478,12 +480,12 @@ int phy_mii_ioctl(struct phy_device *phydev, struct ifreq *ifr, int cmd)
 }
 EXPORT_SYMBOL(phy_mii_ioctl);
 
-static void phy_queue_state_machine(struct phy_device *phydev,
-				    unsigned int secs)
+void phy_queue_state_machine(struct phy_device *phydev, unsigned long jiffies)
 {
 	mod_delayed_work(system_power_efficient_wq, &phydev->state_queue,
-			 secs * HZ);
+			 jiffies);
 }
+EXPORT_SYMBOL(phy_queue_state_machine);
 
 static void phy_trigger_machine(struct phy_device *phydev)
 {
include/linux/phy.h +1 -1
index 16cd33915..dc4b51060 100644
@@ -188,7 +188,6 @@ static inline const char *phy_modes(phy_interface_t interface)
 
 
 #define PHY_INIT_TIMEOUT	100000
-#define PHY_STATE_TIME		1
 #define PHY_FORCE_TIMEOUT	10
 
 #define PHY_MAX_ADDR	32
@@ -1140,6 +1139,7 @@ int phy_driver_register(struct phy_driver *new_driver, struct module *owner);
 int phy_drivers_register(struct phy_driver *new_driver, int n,
 			 struct module *owner);
 void phy_state_machine(struct work_struct *work);
+void phy_queue_state_machine(struct phy_device *phydev, unsigned long jiffies);
 void phy_mac_interrupt(struct phy_device *phydev);
 void phy_start_machine(struct phy_device *phydev);
 void phy_stop_machine(struct phy_device *phydev);
Signature · 1 line
2.21.0
maintainer
#7 lore raw
In reply to Heiner Kallweit

From: Heiner Kallweit <[email protected]>
Date: Thu, 30 May 2019 15:08:09 +0200

Quoted 9 lines

This series tries to address few problematic aspects raised by Russell. Concrete example is the Marvell 88x3310, the changes should be helpful for other complex C45 PHY's too.

v2:
- added patch enabling interrupts also if phylib state machine isn't started
- removed patch dealing with the double link status read This one needs little bit more thinking and will go separately.

Series applied.