From Bugzilla Helper:
User-Agent: Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:0.9.4)
Gecko/20011128 Netscape6/6.2.1

Description of problem:
Under ver high IO load (with 40 odd disks connected) on a 4 processor IA-64
machine (Intel's Lion), the machine hangs after about 16-20 hours of run.

Version-Release number of selected component (if applicable):


How reproducible:
Always

Steps to Reproduce:
1.Run IO and memory hog test suite for about 16-20 hours and the kernel hangs.
2.
3.
	

Actual Results:  The debug showed that there is a deadlock condition in kernel
that gets gerneated with 2.4.9-18 (released RH7.2 kernel) 

Processor 1:		

In truncate_inode_pages():

	holds mapping->page_lock
		&
	tries to get page cache lock (pg_lock) (inside truncate_list_pages())

Processor 2:

In reclaim_page():

	gets page cache lock (pg_lock)
		&
	tries to get mapping->page_lock.

The problem is in truncate_inode_pages() where it tries to get the pg_lock and
if it can not get the pg_lock it does not does not do any recovery (i.e. release
the mapping->page_lock and retry).

We have applied the following patch to correct the problem and would like to
know if you see any problem.

--- mm/filemap.c.org    Tue Apr  2 08:25:37 2002
+++ mm/filemap.c        Tue Apr  2 08:38:58 2002
@@ -366,7 +366,10 @@
 
                        pg_lock = PAGECACHE_LOCK(page);
 
-                       if (!spin_trylock(pg_lock)) {
+                       if (!spin_trylock(pg_lock)) {
+                               spin_unlock(&mapping->page_lock);
+                               barrier();
+                               spin_lock(&mapping->page_lock);
                                return 1;
                        }
 



Additional info: