2.5.37: oom stress test crashes immediately

[Manfred Spraul]

This is a multi-part message in MIME format.
--------------070607000906050502080907
Content-Type: text/plain; charset=us-ascii; format=flowed
Content-Transfer-Encoding: 7bit

I've added oom handling into the natsemi network driver, but testing it 
was tricky: I immediately ran into oopses.

The attached patch fails kmalloc and kmem_cache_alloc if

		(jiffies%HZ) < HZ/10

with a 5 minute guaranteed success, for the boot process.

Is that something the kernel should survive? Obviously the computer is 
unusable after the 5 minute grace period, but I didn't expect oopses.

-- 

	Manfred

--------------070607000906050502080907
Content-Type: text/plain;
 name="patch-kmalloc-fail"
Content-Transfer-Encoding: 7bit
Content-Disposition: inline;
 filename="patch-kmalloc-fail"

--- 2.5/mm/slab.c	Sat Sep 21 17:03:15 2002
+++ build-2.5/mm/slab.c	Sun Sep 22 16:59:29 2002
@@ -1568,6 +1568,9 @@
  */
 void * kmem_cache_alloc (kmem_cache_t *cachep, int flags)
 {
+	if (jiffies > HZ*300 && (jiffies % HZ) < HZ/10)
+		return NULL;
+
 	return __kmem_cache_alloc(cachep, flags);
 }
 
@@ -1596,6 +1599,9 @@
 {
 	cache_sizes_t *csizep = cache_sizes;
 
+	if (jiffies > HZ*300 && (jiffies % HZ) < HZ/10)
+		return NULL;
+
 	for (; csizep->cs_size; csizep++) {
 		if (size > csizep->cs_size)
 			continue;

--------------070607000906050502080907--