Skip to contentSkip to conversation

drivers: base: revise coredump ABI

PATCH In review Arend van Spriel 4 participants 15 messages Mar 15, 2018 – Apr 7, 2018 Driver core

lore
Share & tools
lore
#0 lore raw

cover drivers: base: revise coredump ABI

Not sure if this can be considered as fix for v4.16. Upon submitting the driver patches I received valuable feedback that I would like to address with this series.

The patch apply to master and driver-core-next of the driver-core repository so you can take it either way.

Arend van Spriel (3):
sysfs: improve devices-coredump description with user-space
perspective
drivers: change struct device_driver::coredump() return type to void
drivers: base: remove check for callback in coredump_store()

 Documentation/ABI/testing/sysfs-devices-coredump | 14 +++++++++-----
 drivers/base/dd.c                                |  3 +--
 include/linux/device.h                           |  5 ++++-
 3 files changed, 14 insertions(+), 8 deletions(-)

--
1.9.1

#1 lore raw

PATCH 1/3 sysfs: improve devices-coredump description with user-space perspective

Instead of referring to kernel internals, describe the ABI from user-space perspective to clarify what can be expected when using it.

---
 Documentation/ABI/testing/sysfs-devices-coredump | 14 +++++++++-----
 1 file changed, 9 insertions(+), 5 deletions(-)
Documentation/ABI/testing/sysfs-devices-coredump +9 -5
index e459368..d5a4c75 100644
@@ -2,9 +2,13 @@ What:		/sys/devices/.../coredump
 Date:		December 2017
 Contact:	Arend van Spriel <[email protected]>
 Description:
-		The /sys/devices/.../coredump attribute is only present when the
-		device is bound to a driver, which provides the .coredump()
-		callback. The attribute is write only. Anything written to this
-		file will trigger the .coredump() callback.
+		When present the /sys/devices/.../coredump attribute can be used
+		to trigger a coredump of the device. The coredump contents are
+		device driver specific and thus vary. The coredump attribute is
+		writeonly. Anything written to this file will trigger creation
+		of the coredump. When the coredump is made available under
+		/sys/class/devcoredump it will generate a uevent. When the
+		coredump can not be successfully generated no ueven will occur.
 
-		Available when CONFIG_DEV_COREDUMP is enabled.
+		Available when CONFIG_DEV_COREDUMP is enabled and the device
+		driver supports coredump generation.

--
1.9.1

#2 lore raw

Re: [PATCH for-4.16 1/3] sysfs: improve devices-coredump description with user-space perspective

Quoted 17 lines · Arend van Spriel

Instead of referring to kernel internals, describe the ABI from user-space perspective to clarify what can be expected when using it.

---
 Documentation/ABI/testing/sysfs-devices-coredump | 14 +++++++++-----
 1 file changed, 9 insertions(+), 5 deletions(-)
Documentation/ABI/testing/sysfs-devices-coredump +0 -1
index e459368..d5a4c75 100644
@@ -2,9 +2,13 @@ What:		/sys/devices/.../coredump
 Date:		December 2017
 Contact:	Arend van Spriel <[email protected]>
 Description:
-		The /sys/devices/.../coredump attribute is only present when the

Documentation/ABI/testing/sysfs-devices-coredump What: /sys/devices/.../coredump · line 11

Quoted 10 lines
Documentation/ABI/testing/sysfs-devices-coredump +7 -3
-		device is bound to a driver, which provides the .coredump()
-		callback. The attribute is write only. Anything written to this
-		file will trigger the .coredump() callback.
+		When present the /sys/devices/.../coredump attribute can be used
+		to trigger a coredump of the device. The coredump contents are
+		device driver specific and thus vary. The coredump attribute is
+		writeonly. Anything written to this file will trigger creation
+		of the coredump. When the coredump is made available under
+		/sys/class/devcoredump it will generate a uevent. When the
+		coredump can not be successfully generated no ueven will occur.

s/ueven/uevent/

Documentation/ABI/testing/sysfs-devices-coredump What: /sys/devices/.../coredump · line 14

Quoted 4 lines
Documentation/ABI/testing/sysfs-devices-coredump +2 -1
-		Available when CONFIG_DEV_COREDUMP is enabled.
+		Available when CONFIG_DEV_COREDUMP is enabled and the device
+		driver supports coredump generation.

What about /sys/class/devcoredump/disabled? Maybe we just need a sysfs-class-devcoredump too, now that there's a formal method for triggering devcoredumps.

Brian

Quoted 3 lines
Documentation/ABI/testing/sysfs-devices-coredump +0 -1
--

1.9.1

#3 lore raw

Re: [PATCH for-4.16 1/3] sysfs: improve devices-coredump description with user-space perspective

Quoted 22 lines · Brian Norris
Quoted 21 lines · Arend van Spriel

Instead of referring to kernel internals, describe the ABI from user-space perspective to clarify what can be expected when using it.

---
  Documentation/ABI/testing/sysfs-devices-coredump | 14 +++++++++-----
  1 file changed, 9 insertions(+), 5 deletions(-)
Documentation/ABI/testing/sysfs-devices-coredump +1 -4
index e459368..d5a4c75 100644
@@ -2,9 +2,13 @@ What:		/sys/devices/.../coredump
  Date:		December 2017
  Contact:	Arend van Spriel <[email protected]>
  Description:
-		The /sys/devices/.../coredump attribute is only present when the
-		device is bound to a driver, which provides the .coredump()
-		callback. The attribute is write only. Anything written to this
-		file will trigger the .coredump() callback.
+		When present the /sys/devices/.../coredump attribute can be used
Quoted 8 lines

Documentation/ABI/testing/sysfs-devices-coredump What: /sys/devices/.../coredump · line 11

Quoted 6 lines
Documentation/ABI/testing/sysfs-devices-coredump +6 -0
+		to trigger a coredump of the device. The coredump contents are
+		device driver specific and thus vary. The coredump attribute is
+		writeonly. Anything written to this file will trigger creation
+		of the coredump. When the coredump is made available under
+		/sys/class/devcoredump it will generate a uevent. When the
+		coredump can not be successfully generated no ueven will occur.

s/ueven/uevent/

crap. I mean: Thanks ;-)

Quoted 8 lines

Documentation/ABI/testing/sysfs-devices-coredump What: /sys/devices/.../coredump · line 14

Quoted 4 lines
Documentation/ABI/testing/sysfs-devices-coredump +2 -1
-		Available when CONFIG_DEV_COREDUMP is enabled.
+		Available when CONFIG_DEV_COREDUMP is enabled and the device
+		driver supports coredump generation.

What about /sys/class/devcoredump/disabled? Maybe we just need a sysfs-class-devcoredump too, now that there's a formal method for triggering devcoredumps.

I am not sure about this. Not my daily routine so I can use some opinions here. Not that yours does not count, but I would like to be sure before respinning.

Regards,
Arend

#4 lore raw

PATCH 2/3 drivers: change struct device_driver::coredump() return type to void

In reply to Arend van Spriel

Upon submitting a patch for mwifiex [1] it was discussed whether this callback function could fail. To keep things simple there is no need for the error code so the driver can do the task synchronous or not without worries. Currently the device driver core already ignores the return value so changing it to void.

[1] https://patchwork.kernel.org/patch/10231933/

---
 include/linux/device.h | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
include/linux/device.h +4 -1
index b093405..f08c25b 100644
@@ -256,6 +256,9 @@ enum probe_type {
  *		automatically.
  * @pm:		Power management operations of the device which matched
  *		this driver.
+ * @coredump:	Called when sysfs entry is written to. The device driver
+ *		is expected to call the dev_coredump API resulting in a
+ *		uevent.
  * @p:		Driver core's private data, no one other than the driver
  *		core can touch this.
  *
@@ -287,7 +290,7 @@ struct device_driver {
 	const struct attribute_group **groups;
 
 	const struct dev_pm_ops *pm;
-	int (*coredump) (struct device *dev);
+	void (*coredump) (struct device *dev);
 
 	struct driver_private *p;
 };

--
1.9.1

#5 lore raw

Re: [PATCH for-4.16 2/3] drivers: change struct device_driver::coredump() return type to void

Quoted 28 lines · Arend van Spriel

Upon submitting a patch for mwifiex [1] it was discussed whether this callback function could fail. To keep things simple there is no need for the error code so the driver can do the task synchronous or not without worries. Currently the device driver core already ignores the return value so changing it to void.

[1] https://patchwork.kernel.org/patch/10231933/

---
 include/linux/device.h | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
include/linux/device.h +3 -0
index b093405..f08c25b 100644
@@ -256,6 +256,9 @@ enum probe_type {
  *		automatically.
  * @pm:		Power management operations of the device which matched
  *		this driver.
+ * @coredump:	Called when sysfs entry is written to. The device driver
+ *		is expected to call the dev_coredump API resulting in a
+ *		uevent.
  * @p:		Driver core's private data, no one other than the driver
  *		core can touch this.
  *

include/linux/device.h struct device_driver { · line 293

Quoted 5 lines
include/linux/device.h +1 -1
@@ -287,7 +290,7 @@ struct device_driver {
 	const struct attribute_group **groups;
 
 	const struct dev_pm_ops *pm;
-	int (*coredump) (struct device *dev);
+	void (*coredump) (struct device *dev);

Isn't this going to cause build warnings now? Are there no users of this callback function yet?

thanks,

greg k-h

#6 lore raw

Re: [PATCH for-4.16 2/3] drivers: change struct device_driver::coredump() return type to void

Quoted 31 lines · Greg Kroah-Hartman
Quoted 30 lines · Arend van Spriel

Upon submitting a patch for mwifiex [1] it was discussed whether this callback function could fail. To keep things simple there is no need for the error code so the driver can do the task synchronous or not without worries. Currently the device driver core already ignores the return value so changing it to void.

[1] https://patchwork.kernel.org/patch/10231933/

---
 include/linux/device.h | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
include/linux/device.h +3 -0
index b093405..f08c25b 100644
@@ -256,6 +256,9 @@ enum probe_type {
  *           automatically.
  * @pm:              Power management operations of the device which matched
  *           this driver.
+ * @coredump:        Called when sysfs entry is written to. The device driver
+ *           is expected to call the dev_coredump API resulting in a
+ *           uevent.
  * @p:               Driver core's private data, no one other than the driver
  *           core can touch this.
  *
@@ -287,7 +290,7 @@ struct device_driver {
      const struct attribute_group **groups;
 
Quoted 6 lines

include/linux/device.h struct device_driver { · line 293

Quoted 3 lines
include/linux/device.h +1 -1
      const struct dev_pm_ops *pm;
-     int (*coredump) (struct device *dev);
+     void (*coredump) (struct device *dev);

Isn't this going to cause build warnings now? Are there no users of this callback function yet?

Hi Greg,

I submitted driver patches for the 4.17 kernel and from that discussion we concluded it would be good to change to void return type. So those driver patches were dropped. The caller of the callback in drivers/base/dd.c does not use the return value so from that side there is no issue. So my motivation for asking to consider this for 4.16 is so I can resubmit the driver patches for 4.17 if there is still time before the merge window.

Regards,
Arend

#7 lore raw

Re: [PATCH for-4.16 2/3] drivers: change struct device_driver::coredump() return type to void

Quoted 42 lines · Arend van Spriel
Quoted 31 lines · Greg Kroah-Hartman
Quoted 30 lines · Arend van Spriel

Upon submitting a patch for mwifiex [1] it was discussed whether this callback function could fail. To keep things simple there is no need for the error code so the driver can do the task synchronous or not without worries. Currently the device driver core already ignores the return value so changing it to void.

[1] https://patchwork.kernel.org/patch/10231933/

---
 include/linux/device.h | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
include/linux/device.h +3 -0
index b093405..f08c25b 100644
@@ -256,6 +256,9 @@ enum probe_type {
  *           automatically.
  * @pm:              Power management operations of the device which matched
  *           this driver.
+ * @coredump:        Called when sysfs entry is written to. The device driver
+ *           is expected to call the dev_coredump API resulting in a
+ *           uevent.
  * @p:               Driver core's private data, no one other than the driver
  *           core can touch this.
  *
@@ -287,7 +290,7 @@ struct device_driver {
      const struct attribute_group **groups;
 
Quoted 6 lines

include/linux/device.h struct device_driver { · line 293

Quoted 3 lines
include/linux/device.h +1 -1
      const struct dev_pm_ops *pm;
-     int (*coredump) (struct device *dev);
+     void (*coredump) (struct device *dev);

Isn't this going to cause build warnings now? Are there no users of this callback function yet?

Hi Greg,

Quoted 7 lines

I submitted driver patches for the 4.17 kernel and from that discussion we concluded it would be good to change to void return type. So those driver patches were dropped. The caller of the callback in drivers/base/dd.c does not use the return value so from that side there is no issue. So my motivation for asking to consider this for 4.16 is so I can resubmit the driver patches for 4.17 if there is still time before the merge window.

It's too late for 4.16 for this, and I would queue it up in my tree now but I don't want to cause any build warnings in linux-next from it. So how about I submit something like this right after 4.17-rc1 is out, where the function signature is changed _and_ all definitions of that function are changed at the same time to keep everything sane at once?

Can you send me such a patch right before -rc1 is out base on Linus's tree? That should give everyone enough time to get the things merged, right?

Or is there no in-flight patches to use this yet, and I can queue it up now for -rc1 as no build warnings will happen?

thanks,

greg k-h

#8 lore raw

Re: [PATCH for-4.16 2/3] drivers: change struct device_driver::coredump() return type to void

+ Marcel

Quoted 57 lines · Greg Kroah-Hartman
Quoted 42 lines · Arend van Spriel
Quoted 31 lines · Greg Kroah-Hartman
Quoted 30 lines · Arend van Spriel

Upon submitting a patch for mwifiex [1] it was discussed whether this callback function could fail. To keep things simple there is no need for the error code so the driver can do the task synchronous or not without worries. Currently the device driver core already ignores the return value so changing it to void.

[1] https://patchwork.kernel.org/patch/10231933/

---
 include/linux/device.h | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
include/linux/device.h +3 -0
index b093405..f08c25b 100644
@@ -256,6 +256,9 @@ enum probe_type {
  *           automatically.
  * @pm:              Power management operations of the device which matched
  *           this driver.
+ * @coredump:        Called when sysfs entry is written to. The device driver
+ *           is expected to call the dev_coredump API resulting in a
+ *           uevent.
  * @p:               Driver core's private data, no one other than the driver
  *           core can touch this.
  *
@@ -287,7 +290,7 @@ struct device_driver {
      const struct attribute_group **groups;
 
Quoted 6 lines

include/linux/device.h struct device_driver { · line 293

Quoted 3 lines
include/linux/device.h +1 -1
      const struct dev_pm_ops *pm;
-     int (*coredump) (struct device *dev);
+     void (*coredump) (struct device *dev);

Isn't this going to cause build warnings now? Are there no users of this callback function yet?

Hi Greg,

Quoted 7 lines

I submitted driver patches for the 4.17 kernel and from that discussion we concluded it would be good to change to void return type. So those driver patches were dropped. The caller of the callback in drivers/base/dd.c does not use the return value so from that side there is no issue. So my motivation for asking to consider this for 4.16 is so I can resubmit the driver patches for 4.17 if there is still time before the merge window.

It's too late for 4.16 for this, and I would queue it up in my tree now but I don't want to cause any build warnings in linux-next from it. So how about I submit something like this right after 4.17-rc1 is out, where the function signature is changed _and_ all definitions of that function are changed at the same time to keep everything sane at once?

Quoted 6 lines

Can you send me such a patch right before -rc1 is out base on Linus's tree? That should give everyone enough time to get the things merged, right?

Or is there no in-flight patches to use this yet, and I can queue it up now for -rc1 as no build warnings will happen?

I had patches for wireless and bluetooth. The latter was already applied by Marcel, but I asked to revert that. However, I just checked bt-next to be sure and it is still there.

@Marcel: Can you revert commit 118c193d9699 ("btmrvl: support sysfs initiated firmware coredump") [1]

With that out of the way there are no in-flight patches yet.

Regards,
Arend

[1] https://git.kernel.org/pub/scm/linux/kernel/git/bluetooth/bluetooth-next.git/commit/?id=118c193d

#9 lore raw

Re: [PATCH for-4.16 2/3] drivers: change struct device_driver::coredump() return type to void

In reply to Greg Kroah-Hartman
Quoted 57 lines · Greg Kroah-Hartman
Quoted 42 lines · Arend van Spriel
Quoted 31 lines · Greg Kroah-Hartman
Quoted 30 lines · Arend van Spriel

Upon submitting a patch for mwifiex [1] it was discussed whether this callback function could fail. To keep things simple there is no need for the error code so the driver can do the task synchronous or not without worries. Currently the device driver core already ignores the return value so changing it to void.

[1] https://patchwork.kernel.org/patch/10231933/

---
 include/linux/device.h | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
include/linux/device.h +3 -0
index b093405..f08c25b 100644
@@ -256,6 +256,9 @@ enum probe_type {
  *           automatically.
  * @pm:              Power management operations of the device which matched
  *           this driver.
+ * @coredump:        Called when sysfs entry is written to. The device driver
+ *           is expected to call the dev_coredump API resulting in a
+ *           uevent.
  * @p:               Driver core's private data, no one other than the driver
  *           core can touch this.
  *
@@ -287,7 +290,7 @@ struct device_driver {
      const struct attribute_group **groups;
 
Quoted 6 lines

include/linux/device.h struct device_driver { · line 293

Quoted 3 lines
include/linux/device.h +1 -1
      const struct dev_pm_ops *pm;
-     int (*coredump) (struct device *dev);
+     void (*coredump) (struct device *dev);

Isn't this going to cause build warnings now? Are there no users of this callback function yet?

Hi Greg,

Quoted 7 lines

I submitted driver patches for the 4.17 kernel and from that discussion we concluded it would be good to change to void return type. So those driver patches were dropped. The caller of the callback in drivers/base/dd.c does not use the return value so from that side there is no issue. So my motivation for asking to consider this for 4.16 is so I can resubmit the driver patches for 4.17 if there is still time before the merge window.

It's too late for 4.16 for this, and I would queue it up in my tree now but I don't want to cause any build warnings in linux-next from it. So how about I submit something like this right after 4.17-rc1 is out, where the function signature is changed _and_ all definitions of that function are changed at the same time to keep everything sane at once?

Quoted 6 lines

Can you send me such a patch right before -rc1 is out base on Linus's tree? That should give everyone enough time to get the things merged, right?

Or is there no in-flight patches to use this yet, and I can queue it up now for -rc1 as no build warnings will happen?

Hi Greg,

Are we good regarding this patch. I have assured there are not in-flight patches.

Regards,
Arend

#10 lore raw

Re: [PATCH for-4.16 2/3] drivers: change struct device_driver::coredump() return type to void

Quoted 63 lines · Arend van Spriel
Quoted 61 lines · Greg Kroah-Hartman
Quoted 42 lines · Arend van Spriel
Quoted 31 lines · Greg Kroah-Hartman
Quoted 30 lines · Arend van Spriel

Upon submitting a patch for mwifiex [1] it was discussed whether this callback function could fail. To keep things simple there is no need for the error code so the driver can do the task synchronous or not without worries. Currently the device driver core already ignores the return value so changing it to void.

[1] https://patchwork.kernel.org/patch/10231933/

---
 include/linux/device.h | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
include/linux/device.h +3 -0
index b093405..f08c25b 100644
@@ -256,6 +256,9 @@ enum probe_type {
  *           automatically.
  * @pm:              Power management operations of the device which matched
  *           this driver.
+ * @coredump:        Called when sysfs entry is written to. The device driver
+ *           is expected to call the dev_coredump API resulting in a
+ *           uevent.
  * @p:               Driver core's private data, no one other than the driver
  *           core can touch this.
  *
@@ -287,7 +290,7 @@ struct device_driver {
      const struct attribute_group **groups;
 
Quoted 6 lines

include/linux/device.h struct device_driver { · line 293

Quoted 3 lines
include/linux/device.h +1 -1
      const struct dev_pm_ops *pm;
-     int (*coredump) (struct device *dev);
+     void (*coredump) (struct device *dev);

Isn't this going to cause build warnings now? Are there no users of this callback function yet?

Hi Greg,

Quoted 7 lines

I submitted driver patches for the 4.17 kernel and from that discussion we concluded it would be good to change to void return type. So those driver patches were dropped. The caller of the callback in drivers/base/dd.c does not use the return value so from that side there is no issue. So my motivation for asking to consider this for 4.16 is so I can resubmit the driver patches for 4.17 if there is still time before the merge window.

It's too late for 4.16 for this, and I would queue it up in my tree now but I don't want to cause any build warnings in linux-next from it. So how about I submit something like this right after 4.17-rc1 is out, where the function signature is changed _and_ all definitions of that function are changed at the same time to keep everything sane at once?

Can you send me such a patch right before -rc1 is out base on Linus's tree? That should give everyone enough time to get the things merged, right?

Quoted 7 lines
Quoted 2 lines

Or is there no in-flight patches to use this yet, and I can queue it up now for -rc1 as no build warnings will happen?

Hi Greg,

Are we good regarding this patch. I have assured there are not in-flight patches.

All of my stuff is now in Linus's tree, so check there :)

thanks,

greg k-h

#11 lore raw

Re: [PATCH for-4.16 2/3] drivers: change struct device_driver::coredump() return type to void

Op vr 6 apr. 2018 4:46 PM schreef Greg Kroah-Hartman <[email protected]>:

Quoted 68 lines · Arend van Spriel
Quoted 66 lines
Quoted 63 lines · Greg Kroah-Hartman
Quoted 42 lines · Arend van Spriel
Quoted 31 lines · Greg Kroah-Hartman
Quoted 30 lines · Arend van Spriel

Upon submitting a patch for mwifiex [1] it was discussed whether this callback function could fail. To keep things simple there is no need for the error code so the driver can do the task synchronous or not without worries. Currently the device driver core already ignores the return value so changing it to void.

[1] https://patchwork.kernel.org/patch/10231933/

---
 include/linux/device.h | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
include/linux/device.h +3 -0
index b093405..f08c25b 100644
@@ -256,6 +256,9 @@ enum probe_type {
  *           automatically.
  * @pm:              Power management operations of the device which matched
  *           this driver.
+ * @coredump:        Called when sysfs entry is written to. The device driver
+ *           is expected to call the dev_coredump API resulting in a
+ *           uevent.
  * @p:               Driver core's private data, no one other than the driver
  *           core can touch this.
  *
@@ -287,7 +290,7 @@ struct device_driver {
      const struct attribute_group **groups;
 
Quoted 6 lines

include/linux/device.h struct device_driver { · line 293

Quoted 3 lines
include/linux/device.h +1 -1
      const struct dev_pm_ops *pm;
-     int (*coredump) (struct device *dev);
+     void (*coredump) (struct device *dev);

Isn't this going to cause build warnings now? Are there no users of this callback function yet?

Hi Greg,

Quoted 7 lines

I submitted driver patches for the 4.17 kernel and from that discussion we concluded it would be good to change to void return type. So those driver patches were dropped. The caller of the callback in drivers/base/dd.c does not use the return value so from that side there is no issue. So my motivation for asking to consider this for 4.16 is so I can resubmit the driver patches for 4.17 if there is still time before the merge window.

It's too late for 4.16 for this, and I would queue it up in my tree now but I don't want to cause any build warnings in linux-next from it. So how about I submit something like this right after 4.17-rc1 is out, where the function signature is changed _and_ all definitions of that function are changed at the same time to keep everything sane at once?

Can you send me such a patch right before -rc1 is out base on Linus's tree? That should give everyone enough time to get the things merged, right?

Or is there no in-flight patches to use this yet, and I can queue it up now for -rc1 as no build warnings will happen?

Quoted 6 lines
Quoted 4 lines

Hi Greg,

Are we good regarding this patch. I have assured there are not in-flight patches.

All of my stuff is now in Linus's tree, so check there :)

Hi Greg,

That should have been obvious to me with merge window. Anyway, I checked and it seems this particular patch got lost somehow. The other two patches in the series are in Linus's tree. There was only one in-flight patch in bt-next and had it removed with you on Cc:. Probably got lost in your daily email storm :-p

Regards,
Arend

#12 lore raw

Re: [PATCH for-4.16 2/3] drivers: change struct device_driver::coredump() return type to void

Quoted 77 lines · Arend van Spriel

Op vr 6 apr. 2018 4:46 PM schreef Greg Kroah-Hartman <[email protected]>:

Quoted 68 lines · Arend van Spriel
Quoted 66 lines
Quoted 63 lines · Greg Kroah-Hartman
Quoted 42 lines · Arend van Spriel
Quoted 31 lines · Greg Kroah-Hartman
Quoted 30 lines · Arend van Spriel

Upon submitting a patch for mwifiex [1] it was discussed whether this callback function could fail. To keep things simple there is no need for the error code so the driver can do the task synchronous or not without worries. Currently the device driver core already ignores the return value so changing it to void.

[1] https://patchwork.kernel.org/patch/10231933/

---
 include/linux/device.h | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
include/linux/device.h +3 -0
index b093405..f08c25b 100644
@@ -256,6 +256,9 @@ enum probe_type {
  *           automatically.
  * @pm:              Power management operations of the device which matched
  *           this driver.
+ * @coredump:        Called when sysfs entry is written to. The device driver
+ *           is expected to call the dev_coredump API resulting in a
+ *           uevent.
  * @p:               Driver core's private data, no one other than the driver
  *           core can touch this.
  *
@@ -287,7 +290,7 @@ struct device_driver {
      const struct attribute_group **groups;
 
Quoted 6 lines

include/linux/device.h struct device_driver { · line 293

Quoted 3 lines
include/linux/device.h +1 -1
      const struct dev_pm_ops *pm;
-     int (*coredump) (struct device *dev);
+     void (*coredump) (struct device *dev);

Isn't this going to cause build warnings now? Are there no users of this callback function yet?

Hi Greg,

Quoted 7 lines

I submitted driver patches for the 4.17 kernel and from that discussion we concluded it would be good to change to void return type. So those driver patches were dropped. The caller of the callback in drivers/base/dd.c does not use the return value so from that side there is no issue. So my motivation for asking to consider this for 4.16 is so I can resubmit the driver patches for 4.17 if there is still time before the merge window.

It's too late for 4.16 for this, and I would queue it up in my tree now but I don't want to cause any build warnings in linux-next from it. So how about I submit something like this right after 4.17-rc1 is out, where the function signature is changed _and_ all definitions of that function are changed at the same time to keep everything sane at once?

Can you send me such a patch right before -rc1 is out base on Linus's tree? That should give everyone enough time to get the things merged, right?

Or is there no in-flight patches to use this yet, and I can queue it up now for -rc1 as no build warnings will happen?

Quoted 6 lines
Quoted 4 lines

Hi Greg,

Are we good regarding this patch. I have assured there are not in-flight patches.

All of my stuff is now in Linus's tree, so check there :)

Quoted 7 lines

Hi Greg,

That should have been obvious to me with merge window. Anyway, I checked and it seems this particular patch got lost somehow. The other two patches in the series are in Linus's tree. There was only one in-flight patch in bt-next and had it removed with you on Cc:. Probably got lost in your daily email storm :-p

Ah, sorry, now I remember. Yeah, this is long-gone from my tree, care to resend this and I'll get it merged after 4.17-rc1 is out?

thanks,

greg k-h

#13 lore raw

PATCH 3/3 drivers: base: remove check for callback in coredump_store()

In reply to Arend van Spriel

The check for the .coredump() callback in coredump_store() is redundant. It is already assured the device driver implements the callback upon creating the coredump sysfs entry.

---
 drivers/base/dd.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
drivers/base/dd.c +1 -2
index de6fd09..c9f5408 100644
@@ -292,8 +292,7 @@ static ssize_t coredump_store(struct device *dev, struct device_attribute *attr,
 			    const char *buf, size_t count)
 {
 	device_lock(dev);
-	if (dev->driver->coredump)
-		dev->driver->coredump(dev);
+	dev->driver->coredump(dev);
 	device_unlock(dev);
 
 	return count;

--
1.9.1

#14 lore raw
In reply to Arend van Spriel
Quoted 6 lines · Arend van Spriel

Not sure if this can be considered as fix for v4.16. Upon submitting the driver patches I received valuable feedback that I would like to address with this series.

The patch apply to master and driver-core-next of the driver-core repository so you can take it either way.

We can wait for 4.17-rc1 for these, they are not fixing regressions.

thanks,

greg k-h