Attachment 299452 Details for Bug 429103 – linux-2.6.18-xen.hg 498:0637d22ed554 backported to 2.6.9-67.0.7.EL (simple s/mutex_unlock/up/ required)

[patch] linux-2.6.18-xen.hg 498:0637d22ed554 backported to 2.6.9-67.0.7.EL (simple s/mutex_unlock/up/ required)

linux-2.6.18-xen-498-0637d22ed554 (text/plain), 2.30 KB, created by Ian Campbell on 2008-03-28 11:23:09 UTC

(hide)

Description:

Filename:

MIME Type:

Creator: Ian Campbell

Created: 2008-03-28 11:23:09 UTC

Size: 2.30 KB

patch

obsolete

># HG changeset patch
># User Ian Campbell <ian.campbell@citrix.com>
># Date 1206697762 0
># Node ID 0637d22ed554eb0b9cecc7a74fa0a77a7c456497
># Parent  fdb998e79aba45e27948047f680bf70ca5dddbd9
>Avoid deadlock when unregistering a xenbus watch.
>
>Watch handlers which run in a separate thread (XBWF_new_thread) should
>run without the xenbus_mutex held since kthread_run can block waiting
>for memory which causes a deadlock if further watches need to be
>unregistered in order to activate the swap device on resume.
>
>XBWF_new_thread cannot be safely unregistered anyway since the mutex
>only protects thread startup.
>
>Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
>linux-2.6.18-xen changeset: 498:0637d22ed554eb0b9cecc7a74fa0a77a7c456497
>linux-2.6.18-xen date: Fri Mar 28 09:49:22 2008 +0000
>
>Index: linux-2.6.9-67.0.7.EL/drivers/xen/xenbus/xenbus_xs.c
>===================================================================
>--- linux-2.6.9-67.0.7.EL.orig/drivers/xen/xenbus/xenbus_xs.c	2008-03-28 11:07:53.000000000 +0000
>+++ linux-2.6.9-67.0.7.EL/drivers/xen/xenbus/xenbus_xs.c	2008-03-28 11:10:52.000000000 +0000
>@@ -624,6 +624,8 @@
> 	char token[sizeof(watch) * 2 + 1];
> 	int err;
> 
>+	BUG_ON(watch->flags & XBWF_new_thread);
>+
> 	sprintf(token, "%lX", (long)watch);
> 
> 	down_read(&xs_state.suspend_mutex);
>@@ -730,16 +732,29 @@
> 			list_del(ent);
> 		spin_unlock(&watch_events_lock);
> 
>-		if (ent != &watch_events) {
>-			msg = list_entry(ent, struct xs_stored_msg, list);
>-			if (msg->u.watch.handle->flags & XBWF_new_thread)
>-				kthread_run(xenwatch_handle_callback,
>-					    msg, "xenwatch_cb");
>-			else
>-				xenwatch_handle_callback(msg);
>+		if (ent == &watch_events) {
>+			up(&xenwatch_mutex);
>+			continue;
> 		}
> 
>-		up(&xenwatch_mutex);
>+		msg = list_entry(ent, struct xs_stored_msg, list);
>+
>+		/*
>+		 * Unlock the mutex before running an XBWF_new_thread
>+		 * handler. kthread_run can block which can deadlock
>+		 * against unregister_xenbus_watch() if we need to
>+		 * unregister other watches in order to make
>+		 * progress. This can occur on resume before the swap
>+		 * device is attached.
>+		 */
>+		if (msg->u.watch.handle->flags & XBWF_new_thread) {
>+			up(&xenwatch_mutex);
>+			kthread_run(xenwatch_handle_callback,
>+				    msg, "xenwatch_cb");
>+		} else {
>+			xenwatch_handle_callback(msg);
>+			up(&xenwatch_mutex);
>+		}
> 	}
> 
> 	return 0;

Actions: View | Diff

Attachments on bug 429103: 291978 | 291979 | 292483 | 298729 | 298730 | 299451 | 299452 | 306545