Re: [PATCH] LinuxPPS - PPS support for Linux

[Roland Dreier <rdreier@cisco.com>]

A few comments:

 > +		dev_err(port->dev, "PPS support disabled due port \"%s\" is "
 > +					"in polling mode\n",

I think "because" instead of "due" is closer to standard English.

 > +		printk(KERN_ERR "pps: %s: too much PPS sources in the system\n",
 > +					info->name);

Similarly should be "many" instead of "much".

 > +	/* Get new ID for the new PPS source */
 > +	if (idr_pre_get(&pps_idr, GFP_KERNEL) == 0) {
 > +		err = -ENOMEM;
 > +		goto kfree_pps;
 > +	}
 > +
 > +	spin_lock_irq(&idr_lock);
 > +	err = idr_get_new(&pps_idr, pps, &id);
 > +	spin_unlock_irq(&idr_lock);
 > +
 > +	if (err < 0)
 > +		goto kfree_pps;

You usually can handle idr_get_new() returning -EAGAIN by jumping back
to the idr_pre_get(), to handle someone else coming in and stealing
the memory you just preallocated.  In this case it may not matter
since it's pretty unlikely that a lot of contexts are using the idr at
the same time.  But anyway...

 > +void pps_unregister_source(int source)
 > ...
 > +	wait_event(pps->usage_queue, atomic_read(&pps->usage) == 0);
 > +
 > +	pps_sysfs_remove_source_entry(pps);
 > +	pps_unregister_cdev(pps);
 > +	kfree(pps);

This reference counting looks dubious to me... later on in the code
you have:

 > +static int pps_cdev_open(struct inode *inode, struct file *file)
 > +{
 > +	struct pps_device *pps = container_of(inode->i_cdev,
 > +						struct pps_device, cdev);
 > +
 > +	/* Lock the PPS source against (possible) deregistration */
 > +	atomic_inc(&pps->usage);

with no locking, so I see no reason why the atomic_inc() couldn't
happen right after the wait_event() sees a count of 0 and lets the
deregistration continue.  Which would lead to use-after-free.

 - R.