[PATCH, RFC] uio BKL removal

Previous thread: [PATCH] exit signals: use of uninitialized field notify_count by Steve VanDeBogart on Tuesday, August 26, 2008 - 3:14 pm. (6 messages)

Next thread: drivers/char/random.c line 728 BUG by Aaron Straus on Tuesday, August 26, 2008 - 3:59 pm. (12 messages)
From: Jonathan Corbet
Date: Tuesday, August 26, 2008 - 4:15 pm

I had a moment to dedicate to the BKL removal cause, so I went for the
UIO driver, which seemed simple.  The main thing I found was that there
was locking around some idr accesses, but not all, so I filled that
in.  With that in place, removing the BKL from uio_open() seems safe,
especially since none of the in-tree UIO drivers have open() or
release() methods.  

(Incidentally, I don't see how uio_pdrv.c could ever work - who initializes
uioinfo?)

If there's no complaints, I'll feed this into linux-next via the
bkl-removal tree.

Thanks,

jon

---
UIO: BKL removal

Remove the BKL from the UIO driver, and add complete locking where needed
to serialize idr accesses.

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

diff --git a/drivers/uio/uio.c b/drivers/uio/uio.c
index 3a6934b..4f28f4b 100644
--- a/drivers/uio/uio.c
+++ b/drivers/uio/uio.c
@@ -47,6 +47,9 @@ static struct uio_class {
 	struct class *class;
 } *uio_class;
 
+/* Protect idr accesses */
+static DEFINE_MUTEX(minor_lock);
+
 /*
  * attributes
  */
@@ -231,7 +234,6 @@ static void uio_dev_del_attributes(struct uio_device *idev)
 
 static int uio_get_minor(struct uio_device *idev)
 {
-	static DEFINE_MUTEX(minor_lock);
 	int retval = -ENOMEM;
 	int id;
 
@@ -253,7 +255,9 @@ exit:
 
 static void uio_free_minor(struct uio_device *idev)
 {
+	mutex_lock(&minor_lock);
 	idr_remove(&uio_idr, idev->minor);
+	mutex_unlock(&minor_lock);
 }
 
 /**
@@ -297,8 +301,9 @@ static int uio_open(struct inode *inode, struct file *filep)
 	struct uio_listener *listener;
 	int ret = 0;
 
-	lock_kernel();
+	mutex_lock(&minor_lock);
 	idev = idr_find(&uio_idr, iminor(inode));
+	mutex_unlock(&minor_lock);
 	if (!idev) {
 		ret = -ENODEV;
 		goto out;
@@ -324,18 +329,15 @@ static int uio_open(struct inode *inode, struct file *filep)
 		if (ret)
 			goto err_infoopen;
 	}
-	unlock_kernel();
 	return 0;
 
 err_infoopen:
-
 	kfree(listener);
-err_alloc_listener:
 
+err_alloc_listener:
 ...
From: Greg KH
Date: Tuesday, August 26, 2008 - 5:06 pm

Looks good to me, feel free to add an:
	Acked-by: Greg Kroah-Hartman <gregkh@suse.de>
to it.

greg k-h
--

From: Hans J. Koch
Date: Wednesday, August 27, 2008 - 1:08 am

That has to be done somewhere in your board specific file, e.g. for ARM
boards in arch/arm/mach-xxx/board-something.c - the same place where you

Seems right.

--

From: Alexey Dobriyan
Date: Wednesday, August 27, 2008 - 1:45 pm

You can rename it to uio_idr_mutex (it's mutex, dammit!) and put next to
uio_idr and avoid comment ;-) Or if locking around _pre_get() isn't

Indeed, BKL protects nothing because of GFP_KERNEL allocation.

--

Previous thread: [PATCH] exit signals: use of uninitialized field notify_count by Steve VanDeBogart on Tuesday, August 26, 2008 - 3:14 pm. (6 messages)

Next thread: drivers/char/random.c line 728 BUG by Aaron Straus on Tuesday, August 26, 2008 - 3:59 pm. (12 messages)