Skip to contentSkip to conversation

net: phy: some cleanups following phy_port SFP

PATCH Superseded Maxime Chevallier 3 participants 15 messages Jun 1 – Jun 5

Subsystems (3)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

v2 cover net: phy: some cleanups following phy_port SFP

While posting the v11 of phy_port netlink, sashiko found some pre-existing issues, and following the tentative fix [1], Nicolai found some more :)

Patch 1 and 2 add some cleanup in the phy_probe error paths, to discard all SFP and phy_port that have been set-up when probing fails further down the road.

Patch 3 fixes a sashiko-reported issue that I was able to reproduce, where we hit a deadlock when trying to setup PHY-driven SFP ports while using genphy. This is because the sfp init logic for PHYs is now generic, instead of per-driver, and genphy has the particularity of running its .probe() under RTNL, which clashes with the SFP code.

One more reason to try and cleanup RTNL handling in PHY/SFP :)

Thanks Nicolai for the feedback on V1,

Maxime

[1] : https://lore.kernel.org/r/[email protected]

Maxime Chevallier (3):
net: phy: clean the sfp upstream if phy probing fails
net: phy: remove phy ports upon probe failure
net: phy: don't try to setup PHY-driven SFP cages when using genphy

 drivers/net/phy/phy_device.c | 19 ++++++++++++++++---
 1 file changed, 16 insertions(+), 3 deletions(-)
Signature · 1 line
2.54.0
#1 lore raw

PATCH v2 1/3 net: phy: clean the sfp upstream if phy probing fails

Sashiko reported that we don't call sfp_bus_del_upstream() in the probe failure path, so let's add it, otherwise the sfp-bus is left with a dangling 'upstream' field, that may be used later on during SFP events.

This issue existed before the generic phylib sfp support, back when drivers were calling phy_sfp_probe themselves.

Fixes 298e54fa810e ("net: phy: add core phylib sfp support")

---
V2: Null-ify phydev->sfp_bus upon sfp_bus_add_upstream failure

 drivers/net/phy/phy_device.c | 6 ++++++
 1 file changed, 6 insertions(+)
drivers/net/phy/phy_device.c +6 -0
index 3370eb822017..6ccbfacf7d1d 100644
@@ -1718,6 +1718,9 @@ static int phy_sfp_probe(struct phy_device *phydev)
 
 		ret = sfp_bus_add_upstream(bus, phydev, &sfp_phydev_ops);
 		sfp_bus_put(bus);
+
+		if (ret)
+			phydev->sfp_bus = NULL;
 	}
 
 	if (!ret && phydev->sfp_bus)
@@ -3775,6 +3778,9 @@ static int phy_probe(struct device *dev)
 	return 0;
 
 out:
+	sfp_bus_del_upstream(phydev->sfp_bus);
+	phydev->sfp_bus = NULL;
+
 	if (!phydev->is_on_sfp_module)
 		phy_led_triggers_unregister(phydev);
 
Signature · 1 line
2.54.0
Reviewed-by on v2 1/3
#2 lore raw

Re: [PATCH net v2 1/3] net: phy: clean the sfp upstream if phy probing fails

Quoted 31 lines · Maxime Chevallier

Sashiko reported that we don't call sfp_bus_del_upstream() in the probe failure path, so let's add it, otherwise the sfp-bus is left with a dangling 'upstream' field, that may be used later on during SFP events.

This issue existed before the generic phylib sfp support, back when drivers were calling phy_sfp_probe themselves.

Fixes 298e54fa810e ("net: phy: add core phylib sfp support")

---
V2: Null-ify phydev->sfp_bus upon sfp_bus_add_upstream failure

 drivers/net/phy/phy_device.c | 6 ++++++
 1 file changed, 6 insertions(+)
drivers/net/phy/phy_device.c +3 -0
index 3370eb822017..6ccbfacf7d1d 100644
@@ -1718,6 +1718,9 @@ static int phy_sfp_probe(struct phy_device 
 *phydev)
 
 		ret = sfp_bus_add_upstream(bus, phydev, &sfp_phydev_ops);
 		sfp_bus_put(bus);
+
+		if (ret)
+			phydev->sfp_bus = NULL;
 	}
 
drivers/net/phy/phy_device.c +0 -0
 	if (!ret && phydev->sfp_bus)

drivers/net/phy/phy_device.c static int phy_probe(struct device *dev) · line 3785

Quoted 8 lines
drivers/net/phy/phy_device.c +3 -0
@@ -3775,6 +3778,9 @@ static int phy_probe(struct device *dev)
 	return 0;
 
 out:
+	sfp_bus_del_upstream(phydev->sfp_bus);
+	phydev->sfp_bus = NULL;
+
 	if (!phydev->is_on_sfp_module)
 		phy_led_triggers_unregister(phydev);

Thanks
Nicolai

maintainer
#3 lore raw

Re: [PATCH net v2 1/3] net: phy: clean the sfp upstream if phy probing fails

In reply to Maxime Chevallier

This is an AI-generated review of your patch. The human sending this email has considered the AI review valid, or at least plausible.
---
net: phy: clean the sfp upstream if phy probing fails

This commit adds a call to sfp_bus_del_upstream() in the probe failure path to ensure the sfp-bus is not left with a dangling upstream field that could be used during subsequent SFP events.

Quoted 3 lines
drivers/net/phy/phy_device.c +0 -0

[ ... ]

drivers/net/phy/phy_device.c static int phy_probe(struct device *dev) · line 3783

Quoted 7 lines
drivers/net/phy/phy_device.c +3 -0
@@ -3775,6 +3778,9 @@ static int phy_probe(struct device *dev)
 	return 0;
 
 out:
+	sfp_bus_del_upstream(phydev->sfp_bus);
+	phydev->sfp_bus = NULL;
+

[Severity: High]
This is a pre-existing issue, but if a subsequent initialization step fails and we jump to the out label, should we be calling the specific PHY driver's remove callback?

If phydev->drv->probe() succeeds early in phy_probe(), but something like phy_setup_ports() or of_phy_leds() fails later, the execution jumps to this error path. Without calling phydev->drv->remove(phydev), any resources allocated by the driver are leaked.

Also, drivers like dp83640 register the phydev with external subsystems, such as adding it to shared lists like clock->phylist during their probe. Leaving dangling pointers to this unbound phydev in those subsystems could lead to use-after-free crashes later.

Quoted 3 lines
drivers/net/phy/phy_device.c +0 -0
 	if (!phydev->is_on_sfp_module)
 		phy_led_triggers_unregister(phydev);
 
Signature · 1 line
pw-bot: cr
#4 lore raw

Re: [PATCH net v2 1/3] net: phy: clean the sfp upstream if phy probing fails

Quoted 33 lines · Jakub Kicinski

This is an AI-generated review of your patch. The human sending this email has considered the AI review valid, or at least plausible.
---
net: phy: clean the sfp upstream if phy probing fails

This commit adds a call to sfp_bus_del_upstream() in the probe failure path to ensure the sfp-bus is not left with a dangling upstream field that could be used during subsequent SFP events.

Quoted 3 lines
drivers/net/phy/phy_device.c +0 -0

[ ... ]

drivers/net/phy/phy_device.c static int phy_probe(struct device *dev) · line 3783

Quoted 7 lines
drivers/net/phy/phy_device.c +3 -0
@@ -3775,6 +3778,9 @@ static int phy_probe(struct device *dev)
 	return 0;
 
 out:
+	sfp_bus_del_upstream(phydev->sfp_bus);
+	phydev->sfp_bus = NULL;
+

[Severity: High]
This is a pre-existing issue, but if a subsequent initialization step fails and we jump to the out label, should we be calling the specific PHY driver's remove callback?

If phydev->drv->probe() succeeds early in phy_probe(), but something like phy_setup_ports() or of_phy_leds() fails later, the execution jumps to this error path. Without calling phydev->drv->remove(phydev), any resources allocated by the driver are leaked.

Quoted 4 lines

Also, drivers like dp83640 register the phydev with external subsystems, such as adding it to shared lists like clock->phylist during their probe. Leaving dangling pointers to this unbound phydev in those subsystems could lead to use-after-free crashes later.

Good point too. I've sent a V3 already, so if V3 is accepted I'll send a standalone patch for that, or include that in V4 should it be needed :)

Maxime

Quoted 4 lines
Quoted 4 lines
drivers/net/phy/phy_device.c +0 -0
 	if (!phydev->is_on_sfp_module)
 		phy_led_triggers_unregister(phydev);
 
#5 lore raw

PATCH v2 2/3 net: phy: remove phy ports upon probe failure

In reply to Maxime Chevallier

When phy_probe fails, let's clean the phy_ports that were successfully added already.

Fixes 589e934d2735 ("net: phy: Introduce PHY ports representation")
---
 drivers/net/phy/phy_device.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/net/phy/phy_device.c +2 -0
index 6ccbfacf7d1d..a171cbe2a74a 100644
@@ -3778,6 +3778,8 @@ static int phy_probe(struct device *dev)
 	return 0;
 
 out:
+	phy_cleanup_ports(phydev);
+
 	sfp_bus_del_upstream(phydev->sfp_bus);
 	phydev->sfp_bus = NULL;
 
Signature · 1 line
2.54.0
Reviewed-by on v2 2/3
#6 lore raw

Re: [PATCH net v2 2/3] net: phy: remove phy ports upon probe failure

Quoted 15 lines · Maxime Chevallier

When phy_probe fails, let's clean the phy_ports that were successfully added already.

Fixes 589e934d2735 ("net: phy: Introduce PHY ports representation")
---
 drivers/net/phy/phy_device.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/net/phy/phy_device.c +0 -0
index 6ccbfacf7d1d..a171cbe2a74a 100644

drivers/net/phy/phy_device.c static int phy_probe(struct device *dev) · line 3784

Quoted 7 lines
drivers/net/phy/phy_device.c +2 -0
@@ -3778,6 +3778,8 @@ static int phy_probe(struct device *dev)
 	return 0;
 
 out:
+	phy_cleanup_ports(phydev);
+
 	sfp_bus_del_upstream(phydev->sfp_bus);
 	phydev->sfp_bus = NULL;

Thanks
Nicolai

#7 lore raw

Re: [PATCH net v2 2/3] net: phy: remove phy ports upon probe failure

Hi Nicolai,

Quoted 18 lines · Nicolai Buchwitz
Quoted 17 lines · Maxime Chevallier

When phy_probe fails, let's clean the phy_ports that were successfully added already.

Fixes 589e934d2735 ("net: phy: Introduce PHY ports representation")

---
 drivers/net/phy/phy_device.c | 2 ++

 1 file changed, 2 insertions(+)
drivers/net/phy/phy_device.c +0 -0
index 6ccbfacf7d1d..a171cbe2a74a 100644
@@ -3778,6 +3778,8 @@ static int phy_probe(struct device *dev)
      return 0;
 
Quoted 7 lines
Quoted 5 lines

 out:
+    phy_cleanup_ports(phydev);
+
     sfp_bus_del_upstream(phydev->sfp_bus);
     phydev->sfp_bus = NULL;

I'll be moving phy_cleanup_ports(phydev); after the sfp cleanup, can I keep your review tag ?

Maxime

Quoted 3 lines

Thanks
Nicolai

#8 lore raw

Re: [PATCH net v2 2/3] net: phy: remove phy ports upon probe failure

Quoted 29 lines · Maxime Chevallier

Hi Nicolai,

Quoted 18 lines · Nicolai Buchwitz
Quoted 17 lines · Maxime Chevallier

When phy_probe fails, let's clean the phy_ports that were successfully added already.

Fixes 589e934d2735 ("net: phy: Introduce PHY ports representation")

---
 drivers/net/phy/phy_device.c | 2 ++

 1 file changed, 2 insertions(+)
drivers/net/phy/phy_device.c +0 -0
index 6ccbfacf7d1d..a171cbe2a74a 100644
@@ -3778,6 +3778,8 @@ static int phy_probe(struct device *dev)
      return 0;
 
Quoted 7 lines
Quoted 5 lines

 out:
+    phy_cleanup_ports(phydev);
+
     sfp_bus_del_upstream(phydev->sfp_bus);
     phydev->sfp_bus = NULL;

Quoted 2 lines

I'll be moving phy_cleanup_ports(phydev); after the sfp cleanup, can I keep your review tag ?

Sure.

Thanks,
Nicolai

Quoted 1 line

[...]

maintainer
#9 lore raw

Re: [PATCH net v2 2/3] net: phy: remove phy ports upon probe failure

In reply to Maxime Chevallier

This is an AI-generated review of your patch. The human sending this email has considered the AI review valid, or at least plausible.
---
net: phy: remove phy ports upon probe failure

When phy_probe fails, let's clean the phy_ports that were successfully added already.

drivers/net/phy/phy_device.c static int phy_probe(struct device *dev) · line 3785

Quoted 13 lines
drivers/net/phy/phy_device.c +2 -0
index 6ccbfacf7d1d2..a171cbe2a74a1 100644
@@ -3778,6 +3778,8 @@ static int phy_probe(struct device *dev)
 	return 0;
 
 out:
+	phy_cleanup_ports(phydev);
+
 	sfp_bus_del_upstream(phydev->sfp_bus);
 	phydev->sfp_bus = NULL;
 

[Severity: Critical]
Is it safe to call phy_cleanup_ports() before sfp_bus_del_upstream() here?

sfp_bus_del_upstream() unregisters the upstream SFP callbacks and acts as a synchronization barrier by acquiring rtnl_lock(). Before it is called, the SFP bus is still active and can asynchronously fire events like phy_sfp_module_insert().

Because these SFP events locklessly iterate over the phydev->ports list via phy_get_sfp_port(), and phy_cleanup_ports() mutates the list and frees port structures without lock protection:

Could this concurrent access lead to list corruption and a use-after-free?

This same incorrect ordering is a pre-existing issue in the phy_remove() teardown path, but this patch introduces it into the phy_probe() error path.

Signature · 1 line
pw-bot: cr
#10 lore raw

Re: [PATCH net v2 2/3] net: phy: remove phy ports upon probe failure

Quoted 30 lines · Jakub Kicinski

This is an AI-generated review of your patch. The human sending this email has considered the AI review valid, or at least plausible.
---
net: phy: remove phy ports upon probe failure

When phy_probe fails, let's clean the phy_ports that were successfully added already.

drivers/net/phy/phy_device.c static int phy_probe(struct device *dev) · line 3785

Quoted 13 lines
drivers/net/phy/phy_device.c +2 -0
index 6ccbfacf7d1d2..a171cbe2a74a1 100644
@@ -3778,6 +3778,8 @@ static int phy_probe(struct device *dev)
 	return 0;
 
 out:
+	phy_cleanup_ports(phydev);
+
 	sfp_bus_del_upstream(phydev->sfp_bus);
 	phydev->sfp_bus = NULL;
 

[Severity: Critical]
Is it safe to call phy_cleanup_ports() before sfp_bus_del_upstream() here?

sfp_bus_del_upstream() unregisters the upstream SFP callbacks and acts as a synchronization barrier by acquiring rtnl_lock(). Before it is called, the SFP bus is still active and can asynchronously fire events like phy_sfp_module_insert().

Quoted 8 lines

Because these SFP events locklessly iterate over the phydev->ports list via phy_get_sfp_port(), and phy_cleanup_ports() mutates the list and frees port structures without lock protection:

Could this concurrent access lead to list corruption and a use-after-free?

This same incorrect ordering is a pre-existing issue in the phy_remove() teardown path, but this patch introduces it into the phy_probe() error path.

Heh true indeed... Let me respin then, I'll fix the other callsite that has the same problem

Thanks,

Maxime

#11 lore raw

PATCH v2 3/3 net: phy: don't try to setup PHY-driven SFP cages when using genphy

In reply to Maxime Chevallier

We don't have support for PHY-driver SFP cages with the genphy code.

On top of that, it was found by sashiko that running sfp_bus_add_upstream() for genphy deadlocks, as for genphy the PHY probing runs under RTNL, which isn't the case for non-genphy drivers.

This problem was reproduced, and does lead to a deadlock on RTNL.

Before the blamed commit, the phy_sfp_probe() call was made by individual PHY drivers, so there was no way to get to the SFP probing path when using genphy.

Let's therefore only run phy_sfp_probe when not using genphy.

Fixes bad869b5e41a ("net: phy: Only rely on phy_port for PHY-driven SFP")
---
 drivers/net/phy/phy_device.c | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)
drivers/net/phy/phy_device.c +8 -3
index a171cbe2a74a..8eace58e9f12 100644
@@ -3512,9 +3512,14 @@ static int phy_setup_ports(struct phy_device *phydev)
 	if (ret)
 		return ret;
 
-	ret = phy_sfp_probe(phydev);
-	if (ret)
-		goto out;
+	/* We don't support SFP with genphy drivers. Also, genphy driver binding
+	 * occurs with RTNL help, wich will deadlock the sfp_bus_add_upstream().
+	 */
+	if (!phydev->is_genphy_driven) {
+		ret = phy_sfp_probe(phydev);
+		if (ret)
+			goto out;
+	}
 
 	if (phydev->n_ports < phydev->max_n_ports) {
 		ret = phy_default_setup_single_port(phydev);
Signature · 1 line
2.54.0
Reviewed-by on v2 3/3
#12 lore raw

Re: [PATCH net v2 3/3] net: phy: don't try to setup PHY-driven SFP cages when using genphy

Quoted 36 lines · Maxime Chevallier

We don't have support for PHY-driver SFP cages with the genphy code.

On top of that, it was found by sashiko that running sfp_bus_add_upstream() for genphy deadlocks, as for genphy the PHY probing runs under RTNL, which isn't the case for non-genphy drivers.

This problem was reproduced, and does lead to a deadlock on RTNL.

Before the blamed commit, the phy_sfp_probe() call was made by individual PHY drivers, so there was no way to get to the SFP probing path when using genphy.

Let's therefore only run phy_sfp_probe when not using genphy.

Fixes bad869b5e41a ("net: phy: Only rely on phy_port for PHY-driven SFP")
---
 drivers/net/phy/phy_device.c | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)
drivers/net/phy/phy_device.c +1 -3
index a171cbe2a74a..8eace58e9f12 100644
@@ -3512,9 +3512,14 @@ static int phy_setup_ports(struct phy_device 
 *phydev)
 	if (ret)
 		return ret;
 
-	ret = phy_sfp_probe(phydev);
-	if (ret)
-		goto out;
+	/* We don't support SFP with genphy drivers. Also, genphy driver 

binding

drivers/net/phy/phy_device.c +1 -0
+	 * occurs with RTNL help, wich will deadlock the 

sfp_bus_add_upstream().

drivers/net/phy/phy_device.c static int phy_setup_ports(struct phy_device *phydev) · line 3525

Quoted 9 lines
drivers/net/phy/phy_device.c +6 -0
+	 */
+	if (!phydev->is_genphy_driven) {
+		ret = phy_sfp_probe(phydev);
+		if (ret)
+			goto out;
+	}
 
 	if (phydev->n_ports < phydev->max_n_ports) {
 		ret = phy_default_setup_single_port(phydev);

Thanks
Nicolai

maintainer
#13 lore raw

Re: [PATCH net v2 3/3] net: phy: don't try to setup PHY-driven SFP cages when using genphy

In reply to Maxime Chevallier

This is an AI-generated review of your patch. The human sending this email has considered the AI review valid, or at least plausible.
---
net: phy: don't try to setup PHY-driven SFP cages when using genphy

This change skips phy_sfp_probe() when the genphy driver is bound, avoiding an RTNL deadlock in sfp_bus_add_upstream() since genphy binding runs under RTNL while non-genphy driver binding does not.

drivers/net/phy/phy_device.c static int phy_setup_ports(struct phy_device *phydev) · line 3516

Quoted 13 lines
drivers/net/phy/phy_device.c +2 -3
index a171cbe2a74a..8eace58e9f12 100644
@@ -3512,9 +3512,14 @@ static int phy_setup_ports(struct phy_device *phydev)
 	if (ret)
 		return ret;
 
-	ret = phy_sfp_probe(phydev);
-	if (ret)
-		goto out;
+	/* We don't support SFP with genphy drivers. Also, genphy driver binding
+	 * occurs with RTNL help, wich will deadlock the sfp_bus_add_upstream().
                          ^^^^

[Low]
There is a typo in the new comment, "wich" should be "which".

Quoted 9 lines
drivers/net/phy/phy_device.c +6 -0
+	 */
+	if (!phydev->is_genphy_driven) {
+		ret = phy_sfp_probe(phydev);
+		if (ret)
+			goto out;
+	}
 
 	if (phydev->n_ports < phydev->max_n_ports) {
 		ret = phy_default_setup_single_port(phydev);
#14 lore raw

Re: [PATCH net v2 3/3] net: phy: don't try to setup PHY-driven SFP cages when using genphy

Quoted 24 lines · Jakub Kicinski

This is an AI-generated review of your patch. The human sending this email has considered the AI review valid, or at least plausible.
---
net: phy: don't try to setup PHY-driven SFP cages when using genphy

This change skips phy_sfp_probe() when the genphy driver is bound, avoiding an RTNL deadlock in sfp_bus_add_upstream() since genphy binding runs under RTNL while non-genphy driver binding does not.

drivers/net/phy/phy_device.c static int phy_setup_ports(struct phy_device *phydev) · line 3516

Quoted 13 lines
drivers/net/phy/phy_device.c +2 -3
index a171cbe2a74a..8eace58e9f12 100644
@@ -3512,9 +3512,14 @@ static int phy_setup_ports(struct phy_device *phydev)
 	if (ret)
 		return ret;
 
-	ret = phy_sfp_probe(phydev);
-	if (ret)
-		goto out;
+	/* We don't support SFP with genphy drivers. Also, genphy driver binding
+	 * occurs with RTNL help, wich will deadlock the sfp_bus_add_upstream().
                          ^^^^
Quoted 2 lines

[Low]
There is a typo in the new comment, "wich" should be "which".

Sure thing, I need to respin anyway, let me fix this :)

Maxime

Quoted 10 lines
Quoted 10 lines
drivers/net/phy/phy_device.c +6 -0
+	 */
+	if (!phydev->is_genphy_driven) {
+		ret = phy_sfp_probe(phydev);
+		if (ret)
+			goto out;
+	}
 
 	if (phydev->n_ports < phydev->max_n_ports) {
 		ret = phy_default_setup_single_port(phydev);