Skip to contentSkip to conversation

Remove superfluous call to call to cdev_del()

RTC Rolf Eike Beer 3 participants 4 messages Sep 21, 2006 Real time clock (RTC)

lore
Share & tools
lore
#0 lore raw

If cdev_add() fails there is no good reason to call cdev_del().

---
commit 7d8c15d5af591c6039a4bba6ab20e4952ed11c1f
tree 8849441bf17c6c8145f09123a320c34572746c7c
parent 10e3d9d489c71aaf7ff86c81178ae5aeefe1ad6f
author Rolf Eike Beer <[email protected]> Thu, 21 Sep 2006 09:42:49 +0200 committer Rolf Eike Beer <[email protected]> Thu, 21 Sep 2006 09:42:49 +0200

 drivers/rtc/rtc-dev.c |    1 -
 1 files changed, 0 insertions(+), 1 deletions(-)
drivers/rtc/rtc-dev.c +0 -1
index 61a5825..062c0ab 100644
@@ -406,7 +406,6 @@ #endif
 	rtc->char_dev.owner = rtc->owner;
 
 	if (cdev_add(&rtc->char_dev, MKDEV(MAJOR(rtc_devt), rtc->id), 1)) {
-		cdev_del(&rtc->char_dev);
 		dev_err(class_dev->dev,
 			"failed to add char device %d:%d\n",
 			MAJOR(rtc_devt), rtc->id);
#1 lore raw

drivers/rtc/rtc-dev.c #endif · line 411

Quoted 11 lines · On Thu, 21 Sep 2006 09:46:06 +0200 Rolf Eike Beer <[email protected]> wrote:

If cdev_add() fails there is no good reason to call cdev_del().

drivers/rtc/rtc-dev.c +0 -1
 	rtc->char_dev.owner = rtc->owner;
 
 	if (cdev_add(&rtc->char_dev, MKDEV(MAJOR(rtc_devt), rtc->id), 1)) {
-		cdev_del(&rtc->char_dev);
 		dev_err(class_dev->dev,
 			"failed to add char device %d:%d\n",
 			MAJOR(rtc_devt), rtc->id);

I'm not sure.. this is drivers/char/raw.c:

 cdev_init(&raw_cdev, &raw_fops);
        if (cdev_add(&raw_cdev, dev, MAX_RAW_MINORS)) {
                kobject_put(&raw_cdev.kobj);
                unregister_chrdev_region(dev, MAX_RAW_MINORS);
                goto error;
        }

in case of failure, it does a kobject_put.
tha same call is done by cdev_del.

other drivers have implemented different error paths. which one is correct?

Signature · 6 lines
 Best regards,

 Alessandro Zummo,
  Tower Technologies - Turin, Italy

  http://www.towertech.it
#2 lore raw
Quoted 25 lines · Alessandro Zummo

On Thu, 21 Sep 2006 09:46:06 +0200

drivers/rtc/rtc-dev.c #endif · line 411

Quoted 11 lines · Rolf Eike Beer

If cdev_add() fails there is no good reason to call cdev_del().

drivers/rtc/rtc-dev.c +0 -1
 	rtc->char_dev.owner = rtc->owner;
 
 	if (cdev_add(&rtc->char_dev, MKDEV(MAJOR(rtc_devt), rtc->id), 1)) {
-		cdev_del(&rtc->char_dev);
 		dev_err(class_dev->dev,
 			"failed to add char device %d:%d\n",
 			MAJOR(rtc_devt), rtc->id);

I'm not sure.. this is drivers/char/raw.c:

 cdev_init(&raw_cdev, &raw_fops);
        if (cdev_add(&raw_cdev, dev, MAX_RAW_MINORS)) {
                kobject_put(&raw_cdev.kobj);
                unregister_chrdev_region(dev, MAX_RAW_MINORS);
                goto error;
        }
Quoted 2 lines

in case of failure, it does a kobject_put.
tha same call is done by cdev_del.

This is unneeded here as it's embedded into struct rtc_device. Jonathan?

Quoted 2 lines

other drivers have implemented different error paths. which one is correct?

Probably half of them are wrong, ugly or both. I think this interface is not very intuitive at all. This two calls needed to set up an embedded cdev are IMHO the best way to keep people confused. At least the (possible) need to call cdev_del() on failed cdev_add() is just weird.

Eike

#3 lore raw
Quoted 26 lines · Rolf Eike Beer
Quoted 25 lines · Alessandro Zummo

On Thu, 21 Sep 2006 09:46:06 +0200

Quoted 11 lines · Rolf Eike Beer

If cdev_add() fails there is no good reason to call cdev_del().

    rtc->char_dev.owner = rtc->owner;

    if (cdev_add(&rtc->char_dev, MKDEV(MAJOR(rtc_devt), rtc->id), 1)) {
-           cdev_del(&rtc->char_dev);
            dev_err(class_dev->dev,
                    "failed to add char device %d:%d\n",
                    MAJOR(rtc_devt), rtc->id);

I'm not sure.. this is drivers/char/raw.c:

 cdev_init(&raw_cdev, &raw_fops);
        if (cdev_add(&raw_cdev, dev, MAX_RAW_MINORS)) {
                kobject_put(&raw_cdev.kobj);
                unregister_chrdev_region(dev, MAX_RAW_MINORS);
                goto error;
        }
Quoted 2 lines
Quoted 2 lines

in case of failure, it does a kobject_put.
tha same call is done by cdev_del.

But cdev_del also tries to do kobj_unmap before doing kobject_put. If cdev_add fails the object is not added to the map so we shoult not try to unmap it (although it does not hurt in the current implementation).

Quoted 3 lines

This is unneeded here as it's embedded into struct rtc_device. Jonathan?

cdev distingueshes between stattically and synamically allocated objects and so it is safe to do kobject_put/cdev_del even on cdevs embedded into other structures.

Quoted 3 lines
Quoted 2 lines

other drivers have implemented different error paths. which one is correct?

raw.c seems to be DTRT.

Quoted 5 lines

Probably half of them are wrong, ugly or both. I think this interface is not very intuitive at all. This two calls needed to set up an embedded cdev are IMHO the best way to keep people confused. At least the (possible) need to call cdev_del() on failed cdev_add() is just weird.

The rule is simple - after cdev_init/cdev_alloc call kobject_put. After successful cdev_add call cdev_del.

Signature · 1 line
Dmitry