Re: [patch 1/1] uml:fix ubd deadlock on SMP

[Chris Wright <chrisw@osdl.org>]

* blaisorblade_spam@yahoo.it (blaisorblade_spam@yahoo.it) wrote:
> 
> Trivial: don't lock the queue spinlock when called from the request function.
> Since the faulty function must use spinlock in another case, double-case it.
> And since we will never use both functions together, let no object code be
> shared between them.

Why not add a helper which locks around the core function.  Then either
call helper or core function directly depending on locking needs?

Smth. along the lines of below.

===== arch/um/drivers/ubd_kern.c 1.36 vs edited =====
--- 1.36/arch/um/drivers/ubd_kern.c	2004-08-24 02:08:18 -07:00
+++ edited/arch/um/drivers/ubd_kern.c	2004-09-08 11:06:54 -07:00
@@ -396,14 +396,20 @@
  */
 int intr_count = 0;
 
-static void ubd_finish(struct request *req, int error)
+static inline void ubd_finish(struct request *req, int error)
+{
+ 	spin_lock(&ubd_io_lock);
+	__ubd_finish(req, error);
+	spin_unlock(&ubd_io_lock);
+}
+
+/* call ubd_finish if you need to serialize */
+static void __ubd_finish(struct request *req, int error)
 {
 	int nsect;
 
 	if(error){
- 		spin_lock(&ubd_io_lock);
 		end_request(req, 0);
- 		spin_unlock(&ubd_io_lock);
 		return;
 	}
 	nsect = req->current_nr_sectors;
@@ -412,9 +418,7 @@
 	req->errors = 0;
 	req->nr_sectors -= nsect;
 	req->current_nr_sectors = 0;
-	spin_lock(&ubd_io_lock);
 	end_request(req, 1);
-	spin_unlock(&ubd_io_lock);
 }
 
 static void ubd_handler(void)
-- 
Linux Security Modules     http://lsm.immunix.org     http://lsm.bkbits.net