Skip to contentSkip to conversation

thunderbolt: Link USB4 ports to their USB Type-C connectors

PATCH In review Won Chung 4 participants 12 messages Apr 18, 2022 – Apr 29, 2022

Subsystems (3)
lore
Share & tools
lore

Won Chung

author
#0 lore raw

cover thunderbolt: Link USB4 ports to their USB Type-C connectors

Currently, USB port is linked to Type C connector, using the component framework, if they share the same _PLD fields from ACPI table. Type C port-mapper searches for devices with the same _PLD values, and aggregate them as components.

When there is another device that share the same _PLD but does not registers a component, Type C connector (component master) would never be bound due to a component match entry device without a component registered. There exists some cases where USB4 port also shares the same _PLD with USB port and Type C connector, so we need to register a component for USB4 ports too, linking USB4 port with Type C connector. Otherwise, link between USB port and Type C connector would not work either.

Due to the nature of the component framework, all registered components are shared by all component match despite the relevance. MEI subsystems also use the component framework to bind to i915 driver, which try to match components registered by USB ports and USB4 ports. This can be problematic since MEI assumes that there is a driver bound to the component device, while USB4 port does not bind to any drivers. MEI's component match callback functions should handle such case to avoid NULL pointer dereference when USB4 port registers a component.

In summary this patch series
1. Fixes MEI subsystem's component match callbacks to handle a component device without any driver bound
2. Registers a component for USB4 ports to link them to Type C connectors, similar to USB ports.

Heikki Krogerus (1):
thunderbolt: Link USB4 ports to their USB Type-C connectors

Won Chung (1):
misc/mei: Add NULL check to component match callback functions

 .../ABI/testing/sysfs-bus-thunderbolt         | 10 +++++
 drivers/misc/mei/hdcp/mei_hdcp.c              |  2 +-
 drivers/misc/mei/pxp/mei_pxp.c                |  2 +-
 drivers/thunderbolt/usb4_port.c               | 38 +++++++++++++++++++
 4 files changed, 50 insertions(+), 2 deletions(-)
Signature · 1 line
2.36.0.rc0.470.gd361397f0d-goog
author
#1 lore raw

PATCH 1/2 misc/mei: Add NULL check to component match callback functions

Currently, component_match callback functions used in mei refers to the driver name, assuming that the component device being matched has a driver bound. It can cause a NULL pointer dereference when a device without a driver bound registers a component. This is due to the nature of the component framework where all registered components are matched in any component_match callback functions. So even if a component is registered by a totally irrelevant device, that component is also shared to these callbacks for i915 driver.

To prevent totally irrelevant device being matched for i915 and causing a NULL pointer dereference for checking driver name, add a NULL check on dev->driver to check if there is a driver bound before checking the driver name.

In the future, the string compare on the driver name, "i915" may need to be refactored too.

---
 drivers/misc/mei/hdcp/mei_hdcp.c | 2 +-
 drivers/misc/mei/pxp/mei_pxp.c   | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)
drivers/misc/mei/hdcp/mei_hdcp.c +1 -1
index ec2a4fce8581..e889a8bd7ac8 100644
@@ -784,7 +784,7 @@ static int mei_hdcp_component_match(struct device *dev, int subcomponent,
 {
 	struct device *base = data;
 
-	if (strcmp(dev->driver->name, "i915") ||
+	if (!dev->driver || strcmp(dev->driver->name, "i915") ||
 	    subcomponent != I915_COMPONENT_HDCP)
 		return 0;
 
drivers/misc/mei/pxp/mei_pxp.c +1 -1
index f7380d387bab..5c39457e3f53 100644
@@ -131,7 +131,7 @@ static int mei_pxp_component_match(struct device *dev, int subcomponent,
 {
 	struct device *base = data;
 
-	if (strcmp(dev->driver->name, "i915") ||
+	if (!dev->driver || strcmp(dev->driver->name, "i915") ||
 	    subcomponent != I915_COMPONENT_PXP)
 		return 0;
 
Signature · 1 line
2.36.0.rc0.470.gd361397f0d-goog
author
#2 lore raw

Re: [PATCH 1/2] misc/mei: Add NULL check to component match callback functions

Quoted 53 lines · Won Chung

Currently, component_match callback functions used in mei refers to the driver name, assuming that the component device being matched has a driver bound. It can cause a NULL pointer dereference when a device without a driver bound registers a component. This is due to the nature of the component framework where all registered components are matched in any component_match callback functions. So even if a component is registered by a totally irrelevant device, that component is also shared to these callbacks for i915 driver.

To prevent totally irrelevant device being matched for i915 and causing a NULL pointer dereference for checking driver name, add a NULL check on dev->driver to check if there is a driver bound before checking the driver name.

In the future, the string compare on the driver name, "i915" may need to be refactored too.

---
 drivers/misc/mei/hdcp/mei_hdcp.c | 2 +-
 drivers/misc/mei/pxp/mei_pxp.c   | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)
drivers/misc/mei/hdcp/mei_hdcp.c +1 -1
index ec2a4fce8581..e889a8bd7ac8 100644
@@ -784,7 +784,7 @@ static int mei_hdcp_component_match(struct device *dev, int subcomponent,
 {
        struct device *base = data;
 
-       if (strcmp(dev->driver->name, "i915") ||
+       if (!dev->driver || strcmp(dev->driver->name, "i915") ||
            subcomponent != I915_COMPONENT_HDCP)
                return 0;
 
drivers/misc/mei/pxp/mei_pxp.c +1 -2
index f7380d387bab..5c39457e3f53 100644
@@ -131,7 +131,7 @@ static int mei_pxp_component_match(struct device *dev, int subcomponent,
 {
        struct device *base = data;
 
-       if (strcmp(dev->driver->name, "i915") ||
+       if (!dev->driver || strcmp(dev->driver->name, "i915") ||
            subcomponent != I915_COMPONENT_PXP)
                return 0;
 
--
Quoted 2 lines

2.36.0.rc0.470.gd361397f0d-goog

Hi Tomas,

I am adding you to this patch since you are the maintainer of MEI. If this looks okay to you, could you also take a look at the comment thread on the cover letter and give an ack if it is okay to be merged into thunderbolt tree?
https://lore.kernel.org/all/[email protected]/

Thank you,
Won

#3 lore raw

RE: [PATCH 1/2] misc/mei: Add NULL check to component match callback functions

Quoted 19 lines
Quoted 16 lines · Won Chung

Currently, component_match callback functions used in mei refers to the driver name, assuming that the component device being matched has a driver bound. It can cause a NULL pointer dereference when a device without a driver bound registers a component. This is due to the nature of the component framework where all registered components are matched in any component_match callback functions. So even if a component is registered by a totally irrelevant device, that component is also shared to these callbacks for i915 driver.

To prevent totally irrelevant device being matched for i915 and causing a NULL pointer dereference for checking driver name, add a NULL check on
dev->driver to check if there is a driver bound before checking the driver name.

Quoted 6 lines
Quoted 6 lines

In the future, the string compare on the driver name, "i915" may need to be refactored too.

Quoted 49 lines

drivers/misc/mei/hdcp/mei_hdcp.c static int mei_hdcp_component_match(struct

Quoted 11 lines
---
 drivers/misc/mei/hdcp/mei_hdcp.c | 2 +-
 drivers/misc/mei/pxp/mei_pxp.c   | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)
drivers/misc/mei/hdcp/mei_hdcp.c +0 -0
index ec2a4fce8581..e889a8bd7ac8 100644
@@ -784,7 +784,7 @@ static int mei_hdcp_component_match(struct

device

Quoted 23 lines
drivers/misc/mei/hdcp/mei_hdcp.c +1 -1
 *dev, int subcomponent,  {
        struct device *base = data;
 
-       if (strcmp(dev->driver->name, "i915") ||
+       if (!dev->driver || strcmp(dev->driver->name, "i915") ||
            subcomponent != I915_COMPONENT_HDCP)
                return 0;
 
(quoted diff) +0 -0

b/drivers/misc/mei/pxp/mei_pxp.c index f7380d387bab..5c39457e3f53 100644

drivers/misc/mei/pxp/mei_pxp.c +1 -2
@@ -131,7 +131,7 @@ static int mei_pxp_component_match(struct device
 *dev, int subcomponent,  {
        struct device *base = data;
 
-       if (strcmp(dev->driver->name, "i915") ||
+       if (!dev->driver || strcmp(dev->driver->name, "i915") ||
            subcomponent != I915_COMPONENT_PXP)
                return 0;
 
--
Quoted 2 lines

2.36.0.rc0.470.gd361397f0d-goog

Hi Tomas,

I am adding you to this patch since you are the maintainer of MEI. If this looks okay to you, could you also take a look at the comment thread on the cover letter and give an ack if it is okay to be merged into thunderbolt tree?
https://lore.kernel.org/all/20220418175932.1809770-1-
[email protected]/

Thank you,
Won

author
#4 lore raw

PATCH 2/2 thunderbolt: Link USB4 ports to their USB Type-C connectors

In reply to Won Chung

From: Heikki Krogerus <[email protected]>

Creating a symlink pointing to the correct USB Type-C connector for the on-board USB4 ports when they are created. The link will be created only if the firmware is able to describe the connection between the port and its connector.

---
 .../ABI/testing/sysfs-bus-thunderbolt         | 10 +++++
 drivers/thunderbolt/usb4_port.c               | 38 +++++++++++++++++++
 2 files changed, 48 insertions(+)
Documentation/ABI/testing/sysfs-bus-thunderbolt +10 -0
index b7e87f6c7d47..f7570c240ce8 100644
@@ -293,6 +293,16 @@ Contact:	[email protected]
 Description:	This contains XDomain service specific settings as
 		bitmask. Format: %x
 
+What:		/sys/bus/thunderbolt/devices/usb4_portX/connector
+Date:		April 2022
+Contact:	Heikki Krogerus <[email protected]>
+Description:
+		Symlink to the USB Type-C connector. This link is only
+		created when USB Type-C Connector Class is enabled,
+		and only if the system firmware is capable of
+		describing the connection between a port and its
+		connector.
+
 What:		/sys/bus/thunderbolt/devices/usb4_portX/link
 Date:		Sep 2021
 KernelVersion:	v5.14
drivers/thunderbolt/usb4_port.c +38 -0
index 29e2a4f9c9f5..6b02945624ee 100644
@@ -7,9 +7,37 @@
  */
 
 #include <linux/pm_runtime.h>
+#include <linux/component.h>
+#include <linux/property.h>
 
 #include "tb.h"
 
+static int connector_bind(struct device *dev, struct device *connector, void *data)
+{
+	int ret;
+
+	ret = sysfs_create_link(&dev->kobj, &connector->kobj, "connector");
+	if (ret)
+		return ret;
+
+	ret = sysfs_create_link(&connector->kobj, &dev->kobj, dev_name(dev));
+	if (ret)
+		sysfs_remove_link(&dev->kobj, "connector");
+
+	return ret;
+}
+
+static void connector_unbind(struct device *dev, struct device *connector, void *data)
+{
+	sysfs_remove_link(&connector->kobj, dev_name(dev));
+	sysfs_remove_link(&dev->kobj, "connector");
+}
+
+static const struct component_ops connector_ops = {
+	.bind = connector_bind,
+	.unbind = connector_unbind,
+};
+
 static ssize_t link_show(struct device *dev, struct device_attribute *attr,
 			 char *buf)
 {
@@ -246,6 +274,14 @@ struct usb4_port *usb4_port_device_add(struct tb_port *port)
 		return ERR_PTR(ret);
 	}
 
+	if (dev_fwnode(&usb4->dev)) {
+		ret = component_add(&usb4->dev, &connector_ops);
+		if (ret) {
+			dev_err(&usb4->dev, "failed to add component\n");
+			device_unregister(&usb4->dev);
+		}
+	}
+
 	pm_runtime_no_callbacks(&usb4->dev);
 	pm_runtime_set_active(&usb4->dev);
 	pm_runtime_enable(&usb4->dev);
@@ -265,6 +301,8 @@ struct usb4_port *usb4_port_device_add(struct tb_port *port)
  */
 void usb4_port_device_remove(struct usb4_port *usb4)
 {
+	if (dev_fwnode(&usb4->dev))
+		component_del(&usb4->dev, &connector_ops);
 	device_unregister(&usb4->dev);
 }
 
Signature · 1 line
2.36.0.rc0.470.gd361397f0d-goog
#5 lore raw
In reply to Won Chung

Hi,

Quoted 29 lines · Won Chung

Currently, USB port is linked to Type C connector, using the component framework, if they share the same _PLD fields from ACPI table. Type C port-mapper searches for devices with the same _PLD values, and aggregate them as components.

When there is another device that share the same _PLD but does not registers a component, Type C connector (component master) would never be bound due to a component match entry device without a component registered. There exists some cases where USB4 port also shares the same _PLD with USB port and Type C connector, so we need to register a component for USB4 ports too, linking USB4 port with Type C connector. Otherwise, link between USB port and Type C connector would not work either.

Due to the nature of the component framework, all registered components are shared by all component match despite the relevance. MEI subsystems also use the component framework to bind to i915 driver, which try to match components registered by USB ports and USB4 ports. This can be problematic since MEI assumes that there is a driver bound to the component device, while USB4 port does not bind to any drivers. MEI's component match callback functions should handle such case to avoid NULL pointer dereference when USB4 port registers a component.

In summary this patch series
1. Fixes MEI subsystem's component match callbacks to handle a component device without any driver bound
2. Registers a component for USB4 ports to link them to Type C connectors, similar to USB ports.

Quoted 5 lines

Heikki Krogerus (1):
thunderbolt: Link USB4 ports to their USB Type-C connectors

Won Chung (1):
misc/mei: Add NULL check to component match callback functions

The Thunderbolt patch looks good to me. Do you want me to take the both patches through the Thunderbolt tree or they can go separately? I need an ack from the mei maintainer it goes through my tree.

author
#6 lore raw
Quoted 36 lines · Mika Westerberg

Hi,

Quoted 32 lines · Won Chung

Currently, USB port is linked to Type C connector, using the component framework, if they share the same _PLD fields from ACPI table. Type C port-mapper searches for devices with the same _PLD values, and aggregate them as components.

When there is another device that share the same _PLD but does not registers a component, Type C connector (component master) would never be bound due to a component match entry device without a component registered. There exists some cases where USB4 port also shares the same _PLD with USB port and Type C connector, so we need to register a component for USB4 ports too, linking USB4 port with Type C connector. Otherwise, link between USB port and Type C connector would not work either.

Due to the nature of the component framework, all registered components are shared by all component match despite the relevance. MEI subsystems also use the component framework to bind to i915 driver, which try to match components registered by USB ports and USB4 ports. This can be problematic since MEI assumes that there is a driver bound to the component device, while USB4 port does not bind to any drivers. MEI's component match callback functions should handle such case to avoid NULL pointer dereference when USB4 port registers a component.

In summary this patch series
1. Fixes MEI subsystem's component match callbacks to handle a component device without any driver bound
2. Registers a component for USB4 ports to link them to Type C connectors, similar to USB ports.

Heikki Krogerus (1):
thunderbolt: Link USB4 ports to their USB Type-C connectors

Quoted 6 lines
Quoted 2 lines

Won Chung (1):
misc/mei: Add NULL check to component match callback functions

The Thunderbolt patch looks good to me. Do you want me to take the both patches through the Thunderbolt tree or they can go separately? I need an ack from the mei maintainer it goes through my tree.

Hi Mika,

I would prefer the two patches to go to the same tree since it can cause a crash with only the second patch (tbt). Would that sound okay?

Hey Alexander, would you be the right person to ask for an ack on this?

Thanks,
Won

#7 lore raw

Hi,

Quoted 41 lines · Won Chung
Quoted 39 lines · Mika Westerberg

Hi,

Quoted 34 lines · Won Chung

Currently, USB port is linked to Type C connector, using the component framework, if they share the same _PLD fields from ACPI table. Type C port-mapper searches for devices with the same _PLD values, and aggregate them as components.

When there is another device that share the same _PLD but does not registers a component, Type C connector (component master) would never be bound due to a component match entry device without a component registered. There exists some cases where USB4 port also shares the same _PLD with USB port and Type C connector, so we need to register a component for USB4 ports too, linking USB4 port with Type C connector. Otherwise, link between USB port and Type C connector would not work either.

Due to the nature of the component framework, all registered components are shared by all component match despite the relevance. MEI subsystems also use the component framework to bind to i915 driver, which try to match components registered by USB ports and USB4 ports. This can be problematic since MEI assumes that there is a driver bound to the component device, while USB4 port does not bind to any drivers. MEI's component match callback functions should handle such case to avoid NULL pointer dereference when USB4 port registers a component.

In summary this patch series
1. Fixes MEI subsystem's component match callbacks to handle a component device without any driver bound
2. Registers a component for USB4 ports to link them to Type C connectors, similar to USB ports.

Heikki Krogerus (1):
thunderbolt: Link USB4 ports to their USB Type-C connectors

Won Chung (1):
misc/mei: Add NULL check to component match callback functions

Quoted 8 lines
Quoted 3 lines

The Thunderbolt patch looks good to me. Do you want me to take the both patches through the Thunderbolt tree or they can go separately? I need an ack from the mei maintainer it goes through my tree.

Hi Mika,

I would prefer the two patches to go to the same tree since it can cause a crash with only the second patch (tbt). Would that sound okay?

Sounds good to me.

#8 lore raw
Quoted 41 lines
Quoted 14 lines
Quoted 14 lines
Quoted 14 lines

Currently, USB port is linked to Type C connector, using the component framework, if they share the same _PLD fields from ACPI table. Type C port-mapper searches for devices with the same _PLD values, and aggregate them as components.

When there is another device that share the same _PLD but does not registers a component, Type C connector (component master) would never be bound due to a component match entry device without a component registered. There exists some cases where USB4 port also shares the same _PLD with USB port and Type C connector, so we need to register a component for USB4 ports too, linking USB4 port with Type C connector. Otherwise, link between USB port and Type C connector would not work either.

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines

Due to the nature of the component framework, all registered components are shared by all component match despite the relevance. MEI subsystems also use the component framework to bind to i915 driver, which try to match components registered by USB ports and USB4 ports. This can be problematic since MEI assumes that there is a driver bound to the component device, while USB4 port does not bind to any drivers. MEI's component match callback functions should handle such case to avoid

NULL

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

pointer dereference when USB4 port registers a component.

In summary this patch series
1. Fixes MEI subsystem's component match callbacks to handle a

component

Quoted 14 lines
Quoted 13 lines
Quoted 9 lines

device without any driver bound
2. Registers a component for USB4 ports to link them to Type C connectors, similar to USB ports.

Heikki Krogerus (1):
thunderbolt: Link USB4 ports to their USB Type-C connectors

Won Chung (1):
misc/mei: Add NULL check to component match callback functions

The Thunderbolt patch looks good to me. Do you want me to take the both patches through the Thunderbolt tree or they can go separately? I need an ack from the mei maintainer it goes through my tree.

Quoted 6 lines
Quoted 4 lines

Hi Mika,

I would prefer the two patches to go to the same tree since it can cause a crash with only the second patch (tbt). Would that sound okay?

Sounds good to me.

Tomas is MEI maintainer and should ack MEI patch

Signature · 2 lines
Thanks,
Sasha
author
#9 lore raw
Quoted 47 lines · Alexander
Quoted 47 lines
Quoted 15 lines
Quoted 15 lines
Quoted 15 lines

Currently, USB port is linked to Type C connector, using the component framework, if they share the same _PLD fields from ACPI table. Type C port-mapper searches for devices with the same _PLD values, and aggregate them as components.

When there is another device that share the same _PLD but does not registers a component, Type C connector (component master) would never be bound due to a component match entry device without a component registered. There exists some cases where USB4 port also shares the same _PLD with USB port and Type C connector, so we need to register a component for USB4 ports too, linking USB4 port with Type C connector. Otherwise, link between USB port and Type C connector would not work either.

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines

Due to the nature of the component framework, all registered components are shared by all component match despite the relevance. MEI subsystems also use the component framework to bind to i915 driver, which try to match components registered by USB ports and USB4 ports. This can be problematic since MEI assumes that there is a driver bound to the component device, while USB4 port does not bind to any drivers. MEI's component match callback functions should handle such case to avoid

NULL

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

pointer dereference when USB4 port registers a component.

In summary this patch series
1. Fixes MEI subsystem's component match callbacks to handle a

component

Quoted 18 lines
Quoted 7 lines
Quoted 7 lines

device without any driver bound
2. Registers a component for USB4 ports to link them to Type C connectors, similar to USB ports.

Heikki Krogerus (1):
thunderbolt: Link USB4 ports to their USB Type-C connectors

Quoted 6 lines
Quoted 2 lines

Won Chung (1):
misc/mei: Add NULL check to component match callback functions

The Thunderbolt patch looks good to me. Do you want me to take the both patches through the Thunderbolt tree or they can go separately? I need an ack from the mei maintainer it goes through my tree.

Hi Mika,

I would prefer the two patches to go to the same tree since it can cause a crash with only the second patch (tbt). Would that sound okay?

Quoted 9 lines
Quoted 1 line

Sounds good to me.

Tomas is MEI maintainer and should ack MEI patch

--
Thanks,
Sasha

Hi Tomas,

Can you take a look at this patch series and give an ack if it looks okay to be merged into Mika's tree?
One of the patches adds NULL checks in MEI:
https://lore.kernel.org/all/[email protected]/

Thank you,
Won

author
#10 lore raw
Quoted 62 lines · Won Chung
Quoted 47 lines · Alexander
Quoted 47 lines
Quoted 15 lines
Quoted 15 lines
Quoted 15 lines

Currently, USB port is linked to Type C connector, using the component framework, if they share the same _PLD fields from ACPI table. Type C port-mapper searches for devices with the same _PLD values, and aggregate them as components.

When there is another device that share the same _PLD but does not registers a component, Type C connector (component master) would never be bound due to a component match entry device without a component registered. There exists some cases where USB4 port also shares the same _PLD with USB port and Type C connector, so we need to register a component for USB4 ports too, linking USB4 port with Type C connector. Otherwise, link between USB port and Type C connector would not work either.

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines

Due to the nature of the component framework, all registered components are shared by all component match despite the relevance. MEI subsystems also use the component framework to bind to i915 driver, which try to match components registered by USB ports and USB4 ports. This can be problematic since MEI assumes that there is a driver bound to the component device, while USB4 port does not bind to any drivers. MEI's component match callback functions should handle such case to avoid

NULL

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

pointer dereference when USB4 port registers a component.

In summary this patch series
1. Fixes MEI subsystem's component match callbacks to handle a

component

Quoted 18 lines
Quoted 7 lines
Quoted 7 lines

device without any driver bound
2. Registers a component for USB4 ports to link them to Type C connectors, similar to USB ports.

Heikki Krogerus (1):
thunderbolt: Link USB4 ports to their USB Type-C connectors

Quoted 6 lines
Quoted 2 lines

Won Chung (1):
misc/mei: Add NULL check to component match callback functions

The Thunderbolt patch looks good to me. Do you want me to take the both patches through the Thunderbolt tree or they can go separately? I need an ack from the mei maintainer it goes through my tree.

Hi Mika,

I would prefer the two patches to go to the same tree since it can cause a crash with only the second patch (tbt). Would that sound okay?

Quoted 9 lines
Quoted 1 line

Sounds good to me.

Tomas is MEI maintainer and should ack MEI patch

--
Thanks,
Sasha

Hi Tomas,

Quoted 7 lines

Can you take a look at this patch series and give an ack if it looks okay to be merged into Mika's tree?
One of the patches adds NULL checks in MEI:
https://lore.kernel.org/all/[email protected]/

Thank you,
Won

Hi Mika,

Tomas has given an ack on the first patch on MEI:
https://lore.kernel.org/all/[email protected]/

Should I resend the patch with an additional "Acked-by" tag? Or could that be added as the patch is merged into the thunderbolt tree?

Thank you,
Won

#11 lore raw

Hi,

Quoted 7 lines · Won Chung

Hi Mika,

Tomas has given an ack on the first patch on MEI:
https://lore.kernel.org/all/[email protected]/

Should I resend the patch with an additional "Acked-by" tag? Or could that be added as the patch is merged into the thunderbolt tree?

Both applied to thunderbolt.git/next, thanks!