Skip to contentSkip to conversation

driver core: fw_devlink: Don't create device links for devices not on a bus

PATCH In review Saravana Kannan 3 participants 11 messages Aug 31, 2021 – Sep 2, 2021 Driver core

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 driver core: fw_devlink: Don't create device links for devices not on a bus

Devices that are not on a bus will never have a driver bound to it. So, fw_devlink should not create device links for it as it can cause probe issues[1] or sync_state() call back issues[2].

[1] - https://lore.kernel.org/lkml/CAGETcx_xJCqOWtwZ9Ee2+0sPGNLM5=F=djtbdYENkAYZa0ynqQ@mail.gmail.com/
[2] - https://lore.kernel.org/lkml/CAPDyKFo9Bxremkb1dDrr4OcXSpE0keVze94Cm=zrkOVxHHxBmQ@mail.gmail.com/

Fixes f9aa460672c9 ("driver core: Refactor fw_devlink feature")
---
 drivers/base/core.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
drivers/base/core.c +16 -0
index f6360490a4a3..304a06314656 100644
@@ -1719,8 +1719,24 @@ static int fw_devlink_create_devlink(struct device *con,
 	struct device *sup_dev;
 	int ret = 0;
 
+	/*
+	 * If a consumer device is not on a bus (i.e. a driver will never bind
+	 * to it), it doesn't make sense for fw_devlink to create device links
+	 * for it.
+	 */
+	if (con->bus == NULL)
+		return -EINVAL;
+
 	sup_dev = get_dev_from_fwnode(sup_handle);
 	if (sup_dev) {
+		/*
+		 * If a supplier device is not on a bus (i.e. a driver will
+		 * never bind to it), it doesn't make sense for fw_devlink to
+		 * create device links for it.
+		 */
+		if (sup_dev->bus == NULL)
+			return -EINVAL;
+
 		/*
 		 * If it's one of those drivers that don't actually bind to
 		 * their device using driver core, then don't wait on this
Signature · 1 line
2.33.0.259.gc128427fd7-goog
#1 lore raw
Quoted 20 lines · Saravana Kannan

Devices that are not on a bus will never have a driver bound to it. So, fw_devlink should not create device links for it as it can cause probe issues[1] or sync_state() call back issues[2].

[1] - https://lore.kernel.org/lkml/CAGETcx_xJCqOWtwZ9Ee2+0sPGNLM5=F=djtbdYENkAYZa0ynqQ@mail.gmail.com/
[2] - https://lore.kernel.org/lkml/CAPDyKFo9Bxremkb1dDrr4OcXSpE0keVze94Cm=zrkOVxHHxBmQ@mail.gmail.com/

Fixes f9aa460672c9 ("driver core: Refactor fw_devlink feature")
---
 drivers/base/core.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
drivers/base/core.c +0 -0
index f6360490a4a3..304a06314656 100644
@@ -1719,8 +1719,24 @@ static int fw_devlink_create_devlink(struct device *con,
 	struct device *sup_dev;

drivers/base/core.c static int fw_devlink_create_devlink(struct device *con, · line 1729

Quoted 10 lines
drivers/base/core.c +8 -0
 	int ret = 0;
 
+	/*
+	 * If a consumer device is not on a bus (i.e. a driver will never bind
+	 * to it), it doesn't make sense for fw_devlink to create device links
+	 * for it.
+	 */
+	if (con->bus == NULL)
+		return -EINVAL;
+

Why would a device not be on a bus that has to do with a driver needing it? What types of devices are these?

thanks,

greg k-h

#2 lore raw
Quoted 24 lines · Greg Kroah-Hartman
Quoted 22 lines · Saravana Kannan

Devices that are not on a bus will never have a driver bound to it. So, fw_devlink should not create device links for it as it can cause probe issues[1] or sync_state() call back issues[2].

[1] - https://lore.kernel.org/lkml/CAGETcx_xJCqOWtwZ9Ee2+0sPGNLM5=F=djtbdYENkAYZa0ynqQ@mail.gmail.com/
[2] - https://lore.kernel.org/lkml/CAPDyKFo9Bxremkb1dDrr4OcXSpE0keVze94Cm=zrkOVxHHxBmQ@mail.gmail.com/

Fixes f9aa460672c9 ("driver core: Refactor fw_devlink feature")
---
 drivers/base/core.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
drivers/base/core.c +0 -0
index f6360490a4a3..304a06314656 100644
@@ -1719,8 +1719,24 @@ static int fw_devlink_create_devlink(struct device *con,
      struct device *sup_dev;
      int ret = 0;
 
Quoted 11 lines

drivers/base/core.c static int fw_devlink_create_devlink(struct device *con, · line 1729

Quoted 8 lines
drivers/base/core.c +8 -0
+     /*
+      * If a consumer device is not on a bus (i.e. a driver will never bind
+      * to it), it doesn't make sense for fw_devlink to create device links
+      * for it.
+      */
+     if (con->bus == NULL)
+             return -EINVAL;
+

Why would a device not be on a bus that has to do with a driver needing it? What types of devices are these?

I'm not sure what you are asking, because it seems like a question you'd know the answer to. What I'm trying to say is that bus_probe_device() is obviously only possible if the device is on a bus. And fw_devlink creates managed device links that track if a device is bound to a driver. So it doesn't make sense to create the device link if the device is not on a bus.

There are plenty of examples for this -- like all the devices that get added to a class? For example the platform bus or the mdio bus itself are devices that are added to a class instead of a bus.

Not sure if I answered your question. Let me know.

-Saravana

#3 lore raw
Quoted 45 lines · Saravana Kannan
Quoted 24 lines · Greg Kroah-Hartman
Quoted 22 lines · Saravana Kannan

Devices that are not on a bus will never have a driver bound to it. So, fw_devlink should not create device links for it as it can cause probe issues[1] or sync_state() call back issues[2].

[1] - https://lore.kernel.org/lkml/CAGETcx_xJCqOWtwZ9Ee2+0sPGNLM5=F=djtbdYENkAYZa0ynqQ@mail.gmail.com/
[2] - https://lore.kernel.org/lkml/CAPDyKFo9Bxremkb1dDrr4OcXSpE0keVze94Cm=zrkOVxHHxBmQ@mail.gmail.com/

Fixes f9aa460672c9 ("driver core: Refactor fw_devlink feature")
---
 drivers/base/core.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
drivers/base/core.c +0 -0
index f6360490a4a3..304a06314656 100644
@@ -1719,8 +1719,24 @@ static int fw_devlink_create_devlink(struct device *con,
      struct device *sup_dev;
      int ret = 0;
 
Quoted 11 lines

drivers/base/core.c static int fw_devlink_create_devlink(struct device *con, · line 1729

Quoted 8 lines
drivers/base/core.c +8 -0
+     /*
+      * If a consumer device is not on a bus (i.e. a driver will never bind
+      * to it), it doesn't make sense for fw_devlink to create device links
+      * for it.
+      */
+     if (con->bus == NULL)
+             return -EINVAL;
+

Why would a device not be on a bus that has to do with a driver needing it? What types of devices are these?

I'm not sure what you are asking, because it seems like a question you'd know the answer to. What I'm trying to say is that bus_probe_device() is obviously only possible if the device is on a bus. And fw_devlink creates managed device links that track if a device is bound to a driver. So it doesn't make sense to create the device link if the device is not on a bus.

Quoted 5 lines

There are plenty of examples for this -- like all the devices that get added to a class? For example the platform bus or the mdio bus itself are devices that are added to a class instead of a bus.

Not sure if I answered your question. Let me know.

Sorry, that was a vague question.

I was trying to say, "how can a device that is not attached to a bus at all even be called for this?" Who would be making that type of connection, what subsystem is causing this to happen?

thanks,

greg k-h

#4 lore raw
Quoted 51 lines · Greg Kroah-Hartman
Quoted 49 lines · Saravana Kannan
Quoted 24 lines · Greg Kroah-Hartman
Quoted 22 lines · Saravana Kannan

Devices that are not on a bus will never have a driver bound to it. So, fw_devlink should not create device links for it as it can cause probe issues[1] or sync_state() call back issues[2].

[1] - https://lore.kernel.org/lkml/CAGETcx_xJCqOWtwZ9Ee2+0sPGNLM5=F=djtbdYENkAYZa0ynqQ@mail.gmail.com/
[2] - https://lore.kernel.org/lkml/CAPDyKFo9Bxremkb1dDrr4OcXSpE0keVze94Cm=zrkOVxHHxBmQ@mail.gmail.com/

Fixes f9aa460672c9 ("driver core: Refactor fw_devlink feature")
---
 drivers/base/core.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
drivers/base/core.c +0 -0
index f6360490a4a3..304a06314656 100644
@@ -1719,8 +1719,24 @@ static int fw_devlink_create_devlink(struct device *con,
      struct device *sup_dev;
      int ret = 0;
 
Quoted 11 lines

drivers/base/core.c static int fw_devlink_create_devlink(struct device *con, · line 1729

Quoted 8 lines
drivers/base/core.c +8 -0
+     /*
+      * If a consumer device is not on a bus (i.e. a driver will never bind
+      * to it), it doesn't make sense for fw_devlink to create device links
+      * for it.
+      */
+     if (con->bus == NULL)
+             return -EINVAL;
+

Why would a device not be on a bus that has to do with a driver needing it? What types of devices are these?

I'm not sure what you are asking, because it seems like a question you'd know the answer to. What I'm trying to say is that bus_probe_device() is obviously only possible if the device is on a bus. And fw_devlink creates managed device links that track if a device is bound to a driver. So it doesn't make sense to create the device link if the device is not on a bus.

There are plenty of examples for this -- like all the devices that get added to a class? For example the platform bus or the mdio bus itself are devices that are added to a class instead of a bus.

Quoted 6 lines
Quoted 1 line

Not sure if I answered your question. Let me know.

Sorry, that was a vague question.

I was trying to say, "how can a device that is not attached to a bus at all even be called for this?"

This code path is called for all device_add()s.

Quoted 2 lines

Who would be making that type of
connection, what subsystem is causing this to happen?

fw_devlink trying to interpret DT dependencies can sometimes end up in this situation for some uncommon cases. But it's not helping [1] or [2], so for now let's drop this patch. I'll come back to this if I need this to fix any real issue.

-Saravana

#5 lore raw

[...]

Quoted 4 lines

fw_devlink trying to interpret DT dependencies can sometimes end up in this situation for some uncommon cases. But it's not helping [1] or [2], so for now let's drop this patch. I'll come back to this if I need this to fix any real issue.

Just to be clear, the issue is really there, with the psci power domain, for example.

I am working with an ARM vendor on a platform that has a child node below the common soc node and which child node is a consumer of the psci power domain provider node. We fail to get the ->sync_state() to be called, for reasons that have been explained by now, I think.

The reason for explaining the problem with a superficial example was to try to easier point out the problem.

That said, I would certainly appreciate your help to get this fixed, in one way or the other.

Quoted 2 lines

-Saravana

Kind regards
Uffe

#6 lore raw
In reply to Saravana Kannan
Quoted 7 lines · Saravana Kannan

Devices that are not on a bus will never have a driver bound to it. So, fw_devlink should not create device links for it as it can cause probe issues[1] or sync_state() call back issues[2].

[1] - https://lore.kernel.org/lkml/CAGETcx_xJCqOWtwZ9Ee2+0sPGNLM5=F=djtbdYENkAYZa0ynqQ@mail.gmail.com/
[2] - https://lore.kernel.org/lkml/CAPDyKFo9Bxremkb1dDrr4OcXSpE0keVze94Cm=zrkOVxHHxBmQ@mail.gmail.com/

Unfortunately, this doesn't fix my problem in [2].

When the "soctest" device is initialized, via of_platform_populate(), it will be attached to the platform bus, hence the check for the bus pointer that you suggest to add below, doesn't have an impact on my use case. I still get the below in the log:

"platform soctest: Linked as a sync state only consumer to pm_domain_test"

Kind regards
Uffe

Quoted 40 lines
Fixes f9aa460672c9 ("driver core: Refactor fw_devlink feature")
---
 drivers/base/core.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
drivers/base/core.c +16 -1
index f6360490a4a3..304a06314656 100644
@@ -1719,8 +1719,24 @@ static int fw_devlink_create_devlink(struct device *con,
        struct device *sup_dev;
        int ret = 0;
 
+       /*
+        * If a consumer device is not on a bus (i.e. a driver will never bind
+        * to it), it doesn't make sense for fw_devlink to create device links
+        * for it.
+        */
+       if (con->bus == NULL)
+               return -EINVAL;
+
        sup_dev = get_dev_from_fwnode(sup_handle);
        if (sup_dev) {
+               /*
+                * If a supplier device is not on a bus (i.e. a driver will
+                * never bind to it), it doesn't make sense for fw_devlink to
+                * create device links for it.
+                */
+               if (sup_dev->bus == NULL)
+                       return -EINVAL;
+
                /*
                 * If it's one of those drivers that don't actually bind to
                 * their device using driver core, then don't wait on this
--

2.33.0.259.gc128427fd7-goog

#7 lore raw
Quoted 10 lines · Ulf Hansson
Quoted 7 lines · Saravana Kannan

Devices that are not on a bus will never have a driver bound to it. So, fw_devlink should not create device links for it as it can cause probe issues[1] or sync_state() call back issues[2].

[1] - https://lore.kernel.org/lkml/CAGETcx_xJCqOWtwZ9Ee2+0sPGNLM5=F=djtbdYENkAYZa0ynqQ@mail.gmail.com/
[2] - https://lore.kernel.org/lkml/CAPDyKFo9Bxremkb1dDrr4OcXSpE0keVze94Cm=zrkOVxHHxBmQ@mail.gmail.com/

Quoted 6 lines

Unfortunately, this doesn't fix my problem in [2].

When the "soctest" device is initialized, via of_platform_populate(), it will be attached to the platform bus, hence the check for the bus pointer that you suggest to add below, doesn't have an impact on my use case. I still get the below in the log:

*face palm* Right. I forgot that. I just read "bus" and my mind went to busses added as devices. It apparently also didn't help [1] which is surprising to me. I'll dig into that separately. I'll look into fixing this. The annoying part is that some devices have compatible property that's both "simple-bus" and some other string that a driver actually matches with.

-Saravana

Quoted 46 lines

"platform soctest: Linked as a sync state only consumer to pm_domain_test"

Kind regards
Uffe

Quoted 40 lines
Fixes f9aa460672c9 ("driver core: Refactor fw_devlink feature")
---
 drivers/base/core.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
drivers/base/core.c +16 -1
index f6360490a4a3..304a06314656 100644
@@ -1719,8 +1719,24 @@ static int fw_devlink_create_devlink(struct device *con,
        struct device *sup_dev;
        int ret = 0;
 
+       /*
+        * If a consumer device is not on a bus (i.e. a driver will never bind
+        * to it), it doesn't make sense for fw_devlink to create device links
+        * for it.
+        */
+       if (con->bus == NULL)
+               return -EINVAL;
+
        sup_dev = get_dev_from_fwnode(sup_handle);
        if (sup_dev) {
+               /*
+                * If a supplier device is not on a bus (i.e. a driver will
+                * never bind to it), it doesn't make sense for fw_devlink to
+                * create device links for it.
+                */
+               if (sup_dev->bus == NULL)
+                       return -EINVAL;
+
                /*
                 * If it's one of those drivers that don't actually bind to
                 * their device using driver core, then don't wait on this
--

2.33.0.259.gc128427fd7-goog

#8 lore raw
Quoted 19 lines · Saravana Kannan
Quoted 10 lines · Ulf Hansson
Quoted 7 lines · Saravana Kannan

Devices that are not on a bus will never have a driver bound to it. So, fw_devlink should not create device links for it as it can cause probe issues[1] or sync_state() call back issues[2].

[1] - https://lore.kernel.org/lkml/CAGETcx_xJCqOWtwZ9Ee2+0sPGNLM5=F=djtbdYENkAYZa0ynqQ@mail.gmail.com/
[2] - https://lore.kernel.org/lkml/CAPDyKFo9Bxremkb1dDrr4OcXSpE0keVze94Cm=zrkOVxHHxBmQ@mail.gmail.com/

Quoted 6 lines

Unfortunately, this doesn't fix my problem in [2].

When the "soctest" device is initialized, via of_platform_populate(), it will be attached to the platform bus, hence the check for the bus pointer that you suggest to add below, doesn't have an impact on my use case. I still get the below in the log:

Quoted 6 lines

*face palm* Right. I forgot that. I just read "bus" and my mind went to busses added as devices. It apparently also didn't help [1] which is surprising to me. I'll dig into that separately. I'll look into fixing this. The annoying part is that some devices have compatible property that's both "simple-bus" and some other string that a driver actually matches with.

Yes, that is my view of the problem as well.

So perhaps we should do a more fine grained check for when the "simple-bus" compatible is present in the node, and then don't create a fw_devlink if we reach an ascendant with only this compatible?

Would that work, do you think?

Kind regards
Uffe

Quoted 49 lines

-Saravana

Quoted 46 lines

"platform soctest: Linked as a sync state only consumer to pm_domain_test"

Kind regards
Uffe

Quoted 40 lines
Fixes f9aa460672c9 ("driver core: Refactor fw_devlink feature")
---
 drivers/base/core.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
drivers/base/core.c +16 -1
index f6360490a4a3..304a06314656 100644
@@ -1719,8 +1719,24 @@ static int fw_devlink_create_devlink(struct device *con,
        struct device *sup_dev;
        int ret = 0;
 
+       /*
+        * If a consumer device is not on a bus (i.e. a driver will never bind
+        * to it), it doesn't make sense for fw_devlink to create device links
+        * for it.
+        */
+       if (con->bus == NULL)
+               return -EINVAL;
+
        sup_dev = get_dev_from_fwnode(sup_handle);
        if (sup_dev) {
+               /*
+                * If a supplier device is not on a bus (i.e. a driver will
+                * never bind to it), it doesn't make sense for fw_devlink to
+                * create device links for it.
+                */
+               if (sup_dev->bus == NULL)
+                       return -EINVAL;
+
                /*
                 * If it's one of those drivers that don't actually bind to
                 * their device using driver core, then don't wait on this
--

2.33.0.259.gc128427fd7-goog

#9 lore raw
Quoted 28 lines · Ulf Hansson
Quoted 19 lines · Saravana Kannan
Quoted 10 lines · Ulf Hansson
Quoted 7 lines · Saravana Kannan

Devices that are not on a bus will never have a driver bound to it. So, fw_devlink should not create device links for it as it can cause probe issues[1] or sync_state() call back issues[2].

[1] - https://lore.kernel.org/lkml/CAGETcx_xJCqOWtwZ9Ee2+0sPGNLM5=F=djtbdYENkAYZa0ynqQ@mail.gmail.com/
[2] - https://lore.kernel.org/lkml/CAPDyKFo9Bxremkb1dDrr4OcXSpE0keVze94Cm=zrkOVxHHxBmQ@mail.gmail.com/

Quoted 6 lines

Unfortunately, this doesn't fix my problem in [2].

When the "soctest" device is initialized, via of_platform_populate(), it will be attached to the platform bus, hence the check for the bus pointer that you suggest to add below, doesn't have an impact on my use case. I still get the below in the log:

Quoted 6 lines

*face palm* Right. I forgot that. I just read "bus" and my mind went to busses added as devices. It apparently also didn't help [1] which is surprising to me. I'll dig into that separately. I'll look into fixing this. The annoying part is that some devices have compatible property that's both "simple-bus" and some other string that a driver actually matches with.

Quoted 4 lines

Yes, that is my view of the problem as well.

So perhaps we should do a more fine grained check for when the "simple-bus" compatible is present in the node,

Exactly. Do you want to take a stab at this? There are too many things I want to work on, so if you can do this one, that'd be nice.

Quoted 2 lines

and then don't create
a fw_devlink if we reach an ascendant with only this compatible?

And you can achieve that by setting this flag for that DT node:
fwnode->flags |= FWNODE_FLAG_NOT_DEVICE;

-Saravana

#10 lore raw
Quoted 30 lines · Saravana Kannan
Quoted 28 lines · Ulf Hansson
Quoted 25 lines · Saravana Kannan
Quoted 10 lines · Ulf Hansson
Quoted 7 lines · Saravana Kannan

Devices that are not on a bus will never have a driver bound to it. So, fw_devlink should not create device links for it as it can cause probe issues[1] or sync_state() call back issues[2].

[1] - https://lore.kernel.org/lkml/CAGETcx_xJCqOWtwZ9Ee2+0sPGNLM5=F=djtbdYENkAYZa0ynqQ@mail.gmail.com/
[2] - https://lore.kernel.org/lkml/CAPDyKFo9Bxremkb1dDrr4OcXSpE0keVze94Cm=zrkOVxHHxBmQ@mail.gmail.com/

Quoted 6 lines

Unfortunately, this doesn't fix my problem in [2].

When the "soctest" device is initialized, via of_platform_populate(), it will be attached to the platform bus, hence the check for the bus pointer that you suggest to add below, doesn't have an impact on my use case. I still get the below in the log:

*face palm* Right. I forgot that. I just read "bus" and my mind went to busses added as devices. It apparently also didn't help [1] which is surprising to me. I'll dig into that separately. I'll look into fixing this. The annoying part is that some devices have compatible property that's both "simple-bus" and some other string that a driver actually matches with.

Quoted 7 lines
Quoted 4 lines

Yes, that is my view of the problem as well.

So perhaps we should do a more fine grained check for when the "simple-bus" compatible is present in the node,

Exactly. Do you want to take a stab at this? There are too many things I want to work on, so if you can do this one, that'd be nice.

Nevermind. I already did most of it and it needs a little bit more massaging. I'll send out a patch later.

-Saravana

Quoted 8 lines
Quoted 3 lines

and then don't create
a fw_devlink if we reach an ascendant with only this compatible?

And you can achieve that by setting this flag for that DT node:
fwnode->flags |= FWNODE_FLAG_NOT_DEVICE;

-Saravana