Skip to contentSkip to conversation

driver core: Introduce device_find_first_child() helper

PATCH Superseded Andy Shevchenko 3 participants 15 messages Jun 8, 2022

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/2 driver core: Introduce device_find_first_child() helper

There are several places in the kernel where this kind of functionality is being used. Provide a generic helper for such cases.

---
 drivers/base/core.c    | 24 ++++++++++++++++++++++++
 include/linux/device.h |  1 +
 2 files changed, 25 insertions(+)
drivers/base/core.c +24 -0
index 7cd789c4985d..972bfe975cd0 100644
@@ -3832,6 +3832,30 @@ struct device *device_find_child_by_name(struct device *parent,
 }
 EXPORT_SYMBOL_GPL(device_find_child_by_name);
 
+/**
+ * device_find_first_child - device iterator for locating the fist child device.
+ * @parent: parent struct device
+ *
+ * This is similar to the device_find_child() function above, but it
+ * returns a reference to the first child device.
+ *
+ * NOTE: you will need to drop the reference with put_device() after use.
+ */
+struct device *device_find_first_child(struct device *parent)
+{
+	struct klist_iter i;
+	struct device *child;
+
+	if (!parent)
+		return NULL;
+
+	klist_iter_init(&parent->p->klist_children, &i);
+	child = get_device(next_device(&i));
+	klist_iter_exit(&i);
+	return child;
+}
+EXPORT_SYMBOL_GPL(device_find_first_child);
+
 int __init devices_init(void)
 {
 	devices_kset = kset_create_and_add("devices", &device_uevent_ops, NULL);
include/linux/device.h +1 -0
index dc941997795c..20171a4358df 100644
@@ -905,6 +905,7 @@ struct device *device_find_child(struct device *dev, void *data,
 				 int (*match)(struct device *dev, void *data));
 struct device *device_find_child_by_name(struct device *parent,
 					 const char *name);
+struct device *device_find_first_child(struct device *parent);
 int device_rename(struct device *dev, const char *new_name);
 int device_move(struct device *dev, struct device *new_parent,
 		enum dpm_order dpm_order);
Signature · 1 line
2.35.1
#1 lore raw

PATCH 2/2 spi: Use device_find_first_child() instead of custom approach

We have already a helper to get the first child device, use it and drop custom approach.

---
 drivers/spi/spi.c | 9 ++-------
 1 file changed, 2 insertions(+), 7 deletions(-)
drivers/spi/spi.c +2 -7
index ea09d1b42bf6..87dc8773108b 100644
@@ -2613,11 +2613,6 @@ int spi_slave_abort(struct spi_device *spi)
 }
 EXPORT_SYMBOL_GPL(spi_slave_abort);
 
-static int match_true(struct device *dev, void *data)
-{
-	return 1;
-}
-
 static ssize_t slave_show(struct device *dev, struct device_attribute *attr,
 			  char *buf)
 {
@@ -2625,7 +2620,7 @@ static ssize_t slave_show(struct device *dev, struct device_attribute *attr,
 						   dev);
 	struct device *child;
 
-	child = device_find_child(&ctlr->dev, NULL, match_true);
+	child = device_find_first_child(&ctlr->dev);
 	return sprintf(buf, "%s\n",
 		       child ? to_spi_device(child)->modalias : NULL);
 }
@@ -2644,7 +2639,7 @@ static ssize_t slave_store(struct device *dev, struct device_attribute *attr,
 	if (rc != 1 || !name[0])
 		return -EINVAL;
 
-	child = device_find_child(&ctlr->dev, NULL, match_true);
+	child = device_find_first_child(&ctlr->dev);
 	if (child) {
 		/* Remove registered slave */
 		device_unregister(child);
Signature · 1 line
2.35.1
maintainer Reviewed-by on v1 2/2
#2 lore raw

Re: [PATCH v1 2/2] spi: Use device_find_first_child() instead of custom approach

Quoted 5 lines · Andy Shevchenko

We have already a helper to get the first child device, use it and drop custom approach.

Nice cleanup.

Quoted 41 lines
---
 drivers/spi/spi.c | 9 ++-------
 1 file changed, 2 insertions(+), 7 deletions(-)
drivers/spi/spi.c +2 -8
index ea09d1b42bf6..87dc8773108b 100644
@@ -2613,11 +2613,6 @@ int spi_slave_abort(struct spi_device *spi)
 }
 EXPORT_SYMBOL_GPL(spi_slave_abort);
 
-static int match_true(struct device *dev, void *data)
-{
-       return 1;
-}
-
 static ssize_t slave_show(struct device *dev, struct device_attribute *attr,
                          char *buf)
 {
@@ -2625,7 +2620,7 @@ static ssize_t slave_show(struct device *dev, struct device_attribute *attr,
                                                   dev);
        struct device *child;
 
-       child = device_find_child(&ctlr->dev, NULL, match_true);
+       child = device_find_first_child(&ctlr->dev);
        return sprintf(buf, "%s\n",
                       child ? to_spi_device(child)->modalias : NULL);
 }
@@ -2644,7 +2639,7 @@ static ssize_t slave_store(struct device *dev, struct device_attribute *attr,
        if (rc != 1 || !name[0])
                return -EINVAL;
 
-       child = device_find_child(&ctlr->dev, NULL, match_true);
+       child = device_find_first_child(&ctlr->dev);
        if (child) {
                /* Remove registered slave */
                device_unregister(child);
--

2.35.1

#3 lore raw

Re: [PATCH v1 2/2] spi: Use device_find_first_child() instead of custom approach

In reply to Andy Shevchenko
Quoted 25 lines · Andy Shevchenko

We have already a helper to get the first child device, use it and drop custom approach.

---
 drivers/spi/spi.c | 9 ++-------
 1 file changed, 2 insertions(+), 7 deletions(-)
drivers/spi/spi.c +0 -5
index ea09d1b42bf6..87dc8773108b 100644
@@ -2613,11 +2613,6 @@ int spi_slave_abort(struct spi_device *spi)
 }
 EXPORT_SYMBOL_GPL(spi_slave_abort);
 
-static int match_true(struct device *dev, void *data)
-{
-	return 1;
-}
-
 static ssize_t slave_show(struct device *dev, struct device_attribute *attr,
 			  char *buf)
 {

drivers/spi/spi.c static ssize_t slave_show(struct device *dev, struct device_attribute *attr, · line 2626

Quoted 8 lines
drivers/spi/spi.c +1 -1
@@ -2625,7 +2620,7 @@ static ssize_t slave_show(struct device *dev, struct device_attribute *attr,
 						   dev);
 	struct device *child;
 
-	child = device_find_child(&ctlr->dev, NULL, match_true);
+	child = device_find_first_child(&ctlr->dev);
 	return sprintf(buf, "%s\n",
 		       child ? to_spi_device(child)->modalias : NULL);
 }

Horrible naming convention asside, what is this really showing? I do not see this documented in Documentation/ABI/ anywhere, so can it just be dropped entirely?

Ah, it's in Documentation/spi/spi-summary.rst not where it belongs...

Looks like "any" of the child devices could match here, so it's just finding the first one by default. So you aren't explicitly asking for the real first device, you could return the last one as well, and it would still work as there is just "one" device in this list from what I can tell.

So is does this really deserve a new driver core api call?

thanks,

greg k-h

#4 lore raw

Re: [PATCH v1 2/2] spi: Use device_find_first_child() instead of custom approach

Quoted 41 lines · Greg Kroah-Hartman
Quoted 25 lines · Andy Shevchenko

We have already a helper to get the first child device, use it and drop custom approach.

---
 drivers/spi/spi.c | 9 ++-------
 1 file changed, 2 insertions(+), 7 deletions(-)
drivers/spi/spi.c +0 -5
index ea09d1b42bf6..87dc8773108b 100644
@@ -2613,11 +2613,6 @@ int spi_slave_abort(struct spi_device *spi)
 }
 EXPORT_SYMBOL_GPL(spi_slave_abort);
 
-static int match_true(struct device *dev, void *data)
-{
-	return 1;
-}
-
 static ssize_t slave_show(struct device *dev, struct device_attribute *attr,
 			  char *buf)
 {

drivers/spi/spi.c static ssize_t slave_show(struct device *dev, struct device_attribute *attr, · line 2626

Quoted 8 lines
drivers/spi/spi.c +1 -1
@@ -2625,7 +2620,7 @@ static ssize_t slave_show(struct device *dev, struct device_attribute *attr,
 						   dev);
 	struct device *child;
 
-	child = device_find_child(&ctlr->dev, NULL, match_true);
+	child = device_find_first_child(&ctlr->dev);
 	return sprintf(buf, "%s\n",
 		       child ? to_spi_device(child)->modalias : NULL);
 }

Horrible naming convention asside, what is this really showing? I do not see this documented in Documentation/ABI/ anywhere, so can it just be dropped entirely?

Ah, it's in Documentation/spi/spi-summary.rst not where it belongs...

Quoted 7 lines

Looks like "any" of the child devices could match here, so it's just finding the first one by default. So you aren't explicitly asking for the real first device, you could return the last one as well, and it would still work as there is just "one" device in this list from what I can tell.

So is does this really deserve a new driver core api call?

As I said I noticed more places like this (*) and the problem is that I can't simply use device_match_any() because of the different prototype.

I agree that all thing should be using _any instead of _first.

*) e.g. ptp_ocp.

Signature · 2 lines
With Best Regards,
Andy Shevchenko
#5 lore raw

Re: [PATCH v1 2/2] spi: Use device_find_first_child() instead of custom approach

Quoted 48 lines · Andy Shevchenko
Quoted 47 lines · Greg Kroah-Hartman
Quoted 25 lines · Andy Shevchenko

We have already a helper to get the first child device, use it and drop custom approach.

---
 drivers/spi/spi.c | 9 ++-------
 1 file changed, 2 insertions(+), 7 deletions(-)
drivers/spi/spi.c +0 -5
index ea09d1b42bf6..87dc8773108b 100644
@@ -2613,11 +2613,6 @@ int spi_slave_abort(struct spi_device *spi)
 }
 EXPORT_SYMBOL_GPL(spi_slave_abort);
 
-static int match_true(struct device *dev, void *data)
-{
-	return 1;
-}
-
 static ssize_t slave_show(struct device *dev, struct device_attribute *attr,
 			  char *buf)
 {

drivers/spi/spi.c static ssize_t slave_show(struct device *dev, struct device_attribute *attr, · line 2626

Quoted 8 lines
drivers/spi/spi.c +1 -1
@@ -2625,7 +2620,7 @@ static ssize_t slave_show(struct device *dev, struct device_attribute *attr,
 						   dev);
 	struct device *child;
 
-	child = device_find_child(&ctlr->dev, NULL, match_true);
+	child = device_find_first_child(&ctlr->dev);
 	return sprintf(buf, "%s\n",
 		       child ? to_spi_device(child)->modalias : NULL);
 }

Horrible naming convention asside, what is this really showing? I do not see this documented in Documentation/ABI/ anywhere, so can it just be dropped entirely?

Ah, it's in Documentation/spi/spi-summary.rst not where it belongs...

Looks like "any" of the child devices could match here, so it's just finding the first one by default. So you aren't explicitly asking for the real first device, you could return the last one as well, and it would still work as there is just "one" device in this list from what I can tell.

Quoted 4 lines
Quoted 1 line

So is does this really deserve a new driver core api call?

As I said I noticed more places like this (*) and the problem is that I can't simply use device_match_any() because of the different prototype.

Why not exactly? match_true() above and device_match_any() have the same signature from what I can tell:
static int match_true(struct device *dev, void *data) int device_match_any(struct device *dev, const void *unused)

What am I missing, the const?

Quoted 1 line

I agree that all thing should be using _any instead of _first.

Yes, so let's fix it please, don't propagate bad patterns.

thanks,

greg k-h

#6 lore raw

Re: [PATCH v1 2/2] spi: Use device_find_first_child() instead of custom approach

Quoted 1 line · Greg Kroah-Hartman

On Wed, Jun 08, 2022 at 02:49:14PM +0300, Andy Shevchenko wrote:

...

Quoted 6 lines

Why not exactly? match_true() above and device_match_any() have the same signature from what I can tell:
static int match_true(struct device *dev, void *data) int device_match_any(struct device *dev, const void *unused)

What am I missing, the const?

Yep! Compiler is very unhappy about it.

Quoted 3 lines
Quoted 1 line

I agree that all thing should be using _any instead of _first.

Yes, so let's fix it please, don't propagate bad patterns.

Will do, thanks!

Signature · 2 lines
With Best Regards,
Andy Shevchenko
#7 lore raw
In reply to Andy Shevchenko
Quoted 31 lines · Andy Shevchenko

There are several places in the kernel where this kind of functionality is being used. Provide a generic helper for such cases.

---
 drivers/base/core.c    | 24 ++++++++++++++++++++++++
 include/linux/device.h |  1 +
 2 files changed, 25 insertions(+)
drivers/base/core.c +13 -0
index 7cd789c4985d..972bfe975cd0 100644
@@ -3832,6 +3832,30 @@ struct device *device_find_child_by_name(struct device *parent,
 }
 EXPORT_SYMBOL_GPL(device_find_child_by_name);
 
+/**
+ * device_find_first_child - device iterator for locating the fist child device.
+ * @parent: parent struct device
+ *
+ * This is similar to the device_find_child() function above, but it
+ * returns a reference to the first child device.
+ *
+ * NOTE: you will need to drop the reference with put_device() after use.
+ */
+struct device *device_find_first_child(struct device *parent)
+{
+       struct klist_iter i;
+       struct device *child;

drivers/base/core.c struct device *device_find_child_by_name(struct device *parent, · line 3857

Quoted 10 lines
drivers/base/core.c +10 -0
+
+       if (!parent)
+               return NULL;
+
+       klist_iter_init(&parent->p->klist_children, &i);
+       child = get_device(next_device(&i));
+       klist_iter_exit(&i);
+       return child;
+}
+EXPORT_SYMBOL_GPL(device_find_first_child);

I would define it as

static int match_first(struct device *dev, void *)

{
       return 1;
}

struct device *device_find_first_child(struct device *parent)

{
        return device_find_first_child(parent, NULL, match_first);
}

EXPORT_SYMBOL_GPL(device_find_first_child);

which is not that much more overhead.

Quoted 19 lines
drivers/base/core.c +1 -0
+
 int __init devices_init(void)
 {
        devices_kset = kset_create_and_add("devices", &device_uevent_ops, NULL);
include/linux/device.h +1 -1
index dc941997795c..20171a4358df 100644
@@ -905,6 +905,7 @@ struct device *device_find_child(struct device *dev, void *data,
                                 int (*match)(struct device *dev, void *data));
 struct device *device_find_child_by_name(struct device *parent,
                                         const char *name);
+struct device *device_find_first_child(struct device *parent);
 int device_rename(struct device *dev, const char *new_name);
 int device_move(struct device *dev, struct device *new_parent,
                enum dpm_order dpm_order);
--

2.35.1

#8 lore raw
Quoted 2 lines · Rafael J. Wysocki

On Tue, Jun 7, 2022 at 10:22 PM Andy Shevchenko
<[email protected]> wrote:

...

Quoted 7 lines

I would define it as

static int match_first(struct device *dev, void *)

{
       return 1;
}
Quoted 7 lines

struct device *device_find_first_child(struct device *parent)

{
        return device_find_first_child(parent, NULL, match_first);
}

EXPORT_SYMBOL_GPL(device_find_first_child);

which is not that much more overhead.

With this we actually may simply provide a match function and it will make the clean ups (like patch 2 in the series) almost the same without introducing a device core call.

Something like

int device_match_any_for_find(struct device *dev, void *unused)

{
	return 1;
}

As I replied to Greg it's pity we can't use device_match_any()...

Signature · 2 lines
With Best Regards,
Andy Shevchenko
#9 lore raw
Quoted 25 lines · Andy Shevchenko
Quoted 2 lines · Rafael J. Wysocki

On Tue, Jun 7, 2022 at 10:22 PM Andy Shevchenko
<[email protected]> wrote:

...

Quoted 7 lines

I would define it as

static int match_first(struct device *dev, void *)

{
       return 1;
}
Quoted 7 lines

struct device *device_find_first_child(struct device *parent)

{
        return device_find_first_child(parent, NULL, match_first);
}

EXPORT_SYMBOL_GPL(device_find_first_child);

which is not that much more overhead.

With this we actually may simply provide a match function and it will make the clean ups (like patch 2 in the series) almost the same without introducing a device core call.

Quoted 8 lines

Something like

int device_match_any_for_find(struct device *dev, void *unused)

{
	return 1;
}

As I replied to Greg it's pity we can't use device_match_any()...

int device_match_any(struct device *dev, const void *unused)

How is that not ok to use here?

thanks,

greg k-h

#10 lore raw
Quoted 34 lines · Greg Kroah-Hartman
Quoted 32 lines · Andy Shevchenko
Quoted 2 lines · Rafael J. Wysocki

On Tue, Jun 7, 2022 at 10:22 PM Andy Shevchenko
<[email protected]> wrote:

...

Quoted 7 lines

I would define it as

static int match_first(struct device *dev, void *)

{
       return 1;
}
Quoted 7 lines

struct device *device_find_first_child(struct device *parent)

{
        return device_find_first_child(parent, NULL, match_first);
}

EXPORT_SYMBOL_GPL(device_find_first_child);

which is not that much more overhead.

With this we actually may simply provide a match function and it will make the clean ups (like patch 2 in the series) almost the same without introducing a device core call.

Something like

int device_match_any_for_find(struct device *dev, void *unused)

{
      return 1;
}
Quoted 5 lines
Quoted 1 line

As I replied to Greg it's pity we can't use device_match_any()...

int device_match_any(struct device *dev, const void *unused)

How is that not ok to use here?

Because of the const that will be frowned upon by the compiler.

We need to define another device_match_any_relaxed() taking (void *) as the second argument for this.

#11 lore raw
Quoted 38 lines · Rafael J. Wysocki
Quoted 36 lines · Greg Kroah-Hartman
Quoted 33 lines · Andy Shevchenko
Quoted 2 lines · Rafael J. Wysocki

On Tue, Jun 7, 2022 at 10:22 PM Andy Shevchenko
<[email protected]> wrote:

...

Quoted 7 lines

I would define it as

static int match_first(struct device *dev, void *)

{
       return 1;
}
Quoted 7 lines

struct device *device_find_first_child(struct device *parent)

{
        return device_find_first_child(parent, NULL, match_first);
}

EXPORT_SYMBOL_GPL(device_find_first_child);

which is not that much more overhead.

With this we actually may simply provide a match function and it will make the clean ups (like patch 2 in the series) almost the same without introducing a device core call.

Something like

int device_match_any_for_find(struct device *dev, void *unused)

{
      return 1;
}

As I replied to Greg it's pity we can't use device_match_any()...

Quoted 8 lines
Quoted 3 lines

int device_match_any(struct device *dev, const void *unused)

How is that not ok to use here?

Because of the const that will be frowned upon by the compiler.

We need to define another device_match_any_relaxed() taking (void *) as the second argument for this.

Or we could cast it away :)

#12 lore raw
Quoted 42 lines · Greg Kroah-Hartman
Quoted 40 lines · Rafael J. Wysocki
Quoted 38 lines · Greg Kroah-Hartman
Quoted 25 lines · Andy Shevchenko
Quoted 2 lines · Rafael J. Wysocki

On Tue, Jun 7, 2022 at 10:22 PM Andy Shevchenko
<[email protected]> wrote:

...

Quoted 7 lines

I would define it as

static int match_first(struct device *dev, void *)

{
       return 1;
}
Quoted 7 lines

struct device *device_find_first_child(struct device *parent)

{
        return device_find_first_child(parent, NULL, match_first);
}

EXPORT_SYMBOL_GPL(device_find_first_child);

which is not that much more overhead.

With this we actually may simply provide a match function and it will make the clean ups (like patch 2 in the series) almost the same without introducing a device core call.

Quoted 8 lines

Something like

int device_match_any_for_find(struct device *dev, void *unused)

{
      return 1;
}

As I replied to Greg it's pity we can't use device_match_any()...

int device_match_any(struct device *dev, const void *unused)

Quoted 8 lines
Quoted 6 lines
Quoted 1 line

How is that not ok to use here?

Because of the const that will be frowned upon by the compiler.

We need to define another device_match_any_relaxed() taking (void *) as the second argument for this.

Or we could cast it away :)

I'm not sure what exactly you mean. The function pointer signature doesn't match here.

#13 lore raw
In reply to Andy Shevchenko
Quoted 20 lines · Andy Shevchenko
Quoted 2 lines · Rafael J. Wysocki

On Tue, Jun 7, 2022 at 10:22 PM Andy Shevchenko
<[email protected]> wrote:

...

Quoted 13 lines

I would define it as

static int match_first(struct device *dev, void *)

{
       return 1;
}

struct device *device_find_first_child(struct device *parent)

{
        return device_find_first_child(parent, NULL, match_first);
}

EXPORT_SYMBOL_GPL(device_find_first_child);

Quoted 5 lines
Quoted 1 line

which is not that much more overhead.

With this we actually may simply provide a match function and it will make the clean ups (like patch 2 in the series) almost the same without introducing a device core call.

That works too, but IMO it would be a bit cleaner to have the wrapper defined as a proper function.

Quoted 9 lines

Something like

int device_match_any_for_find(struct device *dev, void *unused)

{
        return 1;
}

As I replied to Greg it's pity we can't use device_match_any()...

Well, that only is a matter of adding one more variant of _match_any_ ...

#14 lore raw
In reply to Andy Shevchenko
Quoted 2 lines · Andy Shevchenko

There are several places in the kernel where this kind of functionality is being used. Provide a generic helper for such cases.

This feels really wrong/broken.

There should not be any specific ordering of children in the tree. What subsystem relies on this such that they require this?

thanks,

greg k-h