Re: Driver reference counting and locking, again

[Patrick Mochel]

On Tue, 16 Jul 2002, Oliver Neukum wrote:

> Am Dienstag, 16. Juli 2002 02:56 schrieb Patrick Mochel:
> > Ok, here is another stab at trying to get reference counting and locking
> > right for drivers.
> >
> > The short of it is that struct device_driver gets an owner field, which
> > should be initialized to THIS_MODULE in the driver.
> 
> In your implementation it would seem that get_driver can suceed while
> remove_driver is already running and the deletion from bus_list has
> already happened.

Yes, you're right. Relative patch appended. You can also pull from 
bk://ldm.bkbits.net/linux-2.5/

	-pat

ChangeSet@1.729, 2002-07-16 10:16:20-07:00, [e-mail]
  driver refcounting:
  Remove driver from global list as early as possible in remove_driver

diffstat results: 
 drivers/base/driver.c |    8 ++++----
 1 files changed, 4 insertions, 4 deletions

diff -Nru a/drivers/base/driver.c b/drivers/base/driver.c
--- a/drivers/base/driver.c	Tue Jul 16 10:16:39 2002
+++ b/drivers/base/driver.c	Tue Jul 16 10:16:39 2002
@@ -86,6 +86,10 @@
 
 void remove_driver(struct device_driver * drv)
 {
+	down(&driver_sem);
+	list_del_init(&drv->g_list);
+	up(&driver_sem);
+
 	write_lock(&drv->bus->lock);
 	list_del_init(&drv->bus_list);
 	write_unlock(&drv->bus->lock);
@@ -93,10 +97,6 @@
 	pr_debug("Unregistering driver '%s' from bus '%s'\n",drv->name,drv->bus->name);
 	driver_detach(drv);
 	driverfs_remove_dir(&drv->dir);
-
-	down(&driver_sem);
-	list_del_init(&drv->g_list);
-	up(&driver_sem);
 
 	if (drv->release)
 		drv->release(drv);