From: Jan Glauber <jglauber@redhat.com>
Subject: [RHEL 5.1 PATCH] OOM killer breaks s390 CMM
Date: Tue, 17 Apr 2007 13:30:57 +0000
Bugzilla: 217968
Message-Id: <1176816657.5737.34.camel@localhost.localdomain>
Changelog: [mm] OOM killer breaks s390 CMM


BZ 217968

The s390 memory balloner (CMM) is broken regarding the OOM killer.
If the system runs out of memory it should first ask CMM to reclaim
memory that is allocated by the balloner before killing processes.

The patch adds a notifier chain to the OOM code and a callback to CMM.
If CMM can free memory the process gets not killed and the allocation
is retried. Although the notifier is a common code change it only 
impacts s390.

Jan

---
 arch/s390/mm/cmm.c   |  155 +++++++++++++++++++++++++++++++--------------------
 include/linux/swap.h |    4 +
 mm/oom_kill.c        |   22 +++++++
 3 files changed, 121 insertions(+), 60 deletions(-)

Index: linux-rhel5/arch/s390/mm/cmm.c
===================================================================
--- linux-rhel5.orig/arch/s390/mm/cmm.c	2006-09-20 05:42:06.000000000 +0200
+++ linux-rhel5/arch/s390/mm/cmm.c	2007-03-06 18:01:03.000000000 +0100
@@ -15,6 +15,7 @@
 #include <linux/sched.h>
 #include <linux/sysctl.h>
 #include <linux/ctype.h>
+#include <linux/swap.h>
 
 #include <asm/pgalloc.h>
 #include <asm/uaccess.h>
@@ -34,17 +35,18 @@ struct cmm_page_array {
 	unsigned long pages[CMM_NR_PAGES];
 };
 
-static long cmm_pages = 0;
-static long cmm_timed_pages = 0;
-static volatile long cmm_pages_target = 0;
-static volatile long cmm_timed_pages_target = 0;
-static long cmm_timeout_pages = 0;
-static long cmm_timeout_seconds = 0;
+static long cmm_pages;
+static long cmm_timed_pages;
+static volatile long cmm_pages_target;
+static volatile long cmm_timed_pages_target;
+static long cmm_timeout_pages;
+static long cmm_timeout_seconds;
+
+static struct cmm_page_array *cmm_page_list;
+static struct cmm_page_array *cmm_timed_page_list;
+static DEFINE_SPINLOCK(cmm_lock);
 
-static struct cmm_page_array *cmm_page_list = NULL;
-static struct cmm_page_array *cmm_timed_page_list = NULL;
-
-static unsigned long cmm_thread_active = 0;
+static unsigned long cmm_thread_active;
 static struct work_struct cmm_thread_starter;
 static wait_queue_head_t cmm_thread_wait;
 static struct timer_list cmm_timer;
@@ -69,58 +71,89 @@ cmm_strtoul(const char *cp, char **endp)
 }
 
 static long
-cmm_alloc_pages(long pages, long *counter, struct cmm_page_array **list)
+cmm_alloc_pages(long nr, long *counter, struct cmm_page_array **list)
 {
-	struct cmm_page_array *pa;
-	unsigned long page;
+	struct cmm_page_array *pa, *npa;
+	unsigned long addr;
 
-	pa = *list;
-	while (pages) {
-		page = __get_free_page(GFP_NOIO);
-		if (!page)
+	while (nr) {
+		addr = __get_free_page(GFP_NOIO);
+		if (!addr)
 			break;
+		spin_lock(&cmm_lock);
+		pa = *list;
 		if (!pa || pa->index >= CMM_NR_PAGES) {
 			/* Need a new page for the page list. */
-			pa = (struct cmm_page_array *)
+			spin_unlock(&cmm_lock);
+			npa = (struct cmm_page_array *)
 				__get_free_page(GFP_NOIO);
-			if (!pa) {
-				free_page(page);
+			if (!npa) {
+				free_page(addr);
 				break;
 			}
-			pa->next = *list;
-			pa->index = 0;
-			*list = pa;
+			spin_lock(&cmm_lock);
+			pa = *list;
+			if (!pa || pa->index >= CMM_NR_PAGES) {
+				npa->next = pa;
+				npa->index = 0;
+				pa = npa;
+				*list = pa;
+			} else
+				free_page((unsigned long) npa);
 		}
-		diag10(page);
-		pa->pages[pa->index++] = page;
+		diag10(addr);
+		pa->pages[pa->index++] = addr;
 		(*counter)++;
-		pages--;
+		spin_unlock(&cmm_lock);
+		nr--;
 	}
-	return pages;
+	return nr;
 }
 
-static void
-cmm_free_pages(long pages, long *counter, struct cmm_page_array **list)
+static long
+cmm_free_pages(long nr, long *counter, struct cmm_page_array **list)
 {
 	struct cmm_page_array *pa;
-	unsigned long page;
+	unsigned long addr;
 
+	spin_lock(&cmm_lock);
 	pa = *list;
-	while (pages) {
+	while (nr) {
 		if (!pa || pa->index <= 0)
 			break;
-		page = pa->pages[--pa->index];
+		addr = pa->pages[--pa->index];
 		if (pa->index == 0) {
 			pa = pa->next;
 			free_page((unsigned long) *list);
 			*list = pa;
 		}
-		free_page(page);
+		free_page(addr);
 		(*counter)--;
-		pages--;
+		nr--;
 	}
+	spin_unlock(&cmm_lock);
+	return nr;
 }
 
+static int cmm_oom_notify(struct notifier_block *self,
+			  unsigned long dummy, void *parm)
+{
+	unsigned long *freed = parm;
+	long nr = 256;
+
+	nr = cmm_free_pages(nr, &cmm_timed_pages, &cmm_timed_page_list);
+	if (nr > 0)
+		nr = cmm_free_pages(nr, &cmm_pages, &cmm_page_list);
+	cmm_pages_target = cmm_pages;
+	cmm_timed_pages_target = cmm_timed_pages;
+	*freed += 256 - nr;
+	return NOTIFY_OK;
+}
+
+static struct notifier_block cmm_oom_nb = {
+	.notifier_call = cmm_oom_notify
+};
+
 static int
 cmm_thread(void *dummy)
 {
@@ -193,21 +226,21 @@ cmm_set_timer(void)
 static void
 cmm_timer_fn(unsigned long ignored)
 {
-	long pages;
+	long nr;
 
-	pages = cmm_timed_pages_target - cmm_timeout_pages;
-	if (pages < 0)
+	nr = cmm_timed_pages_target - cmm_timeout_pages;
+	if (nr < 0)
 		cmm_timed_pages_target = 0;
 	else
-		cmm_timed_pages_target = pages;
+		cmm_timed_pages_target = nr;
 	cmm_kick_thread();
 	cmm_set_timer();
 }
 
 void
-cmm_set_pages(long pages)
+cmm_set_pages(long nr)
 {
-	cmm_pages_target = pages;
+	cmm_pages_target = nr;
 	cmm_kick_thread();
 }
 
@@ -218,9 +251,9 @@ cmm_get_pages(void)
 }
 
 void
-cmm_add_timed_pages(long pages)
+cmm_add_timed_pages(long nr)
 {
-	cmm_timed_pages_target += pages;
+	cmm_timed_pages_target += nr;
 	cmm_kick_thread();
 }
 
@@ -231,9 +264,9 @@ cmm_get_timed_pages(void)
 }
 
 void
-cmm_set_timeout(long pages, long seconds)
+cmm_set_timeout(long nr, long seconds)
 {
-	cmm_timeout_pages = pages;
+	cmm_timeout_pages = nr;
 	cmm_timeout_seconds = seconds;
 	cmm_set_timer();
 }
@@ -261,7 +294,7 @@ cmm_pages_handler(ctl_table *ctl, int wr
 		  void __user *buffer, size_t *lenp, loff_t *ppos)
 {
 	char buf[16], *p;
-	long pages;
+	long nr;
 	int len;
 
 	if (!*lenp || (*ppos && !write)) {
@@ -276,17 +309,17 @@ cmm_pages_handler(ctl_table *ctl, int wr
 			return -EFAULT;
 		buf[sizeof(buf) - 1] = '\0';
 		cmm_skip_blanks(buf, &p);
-		pages = cmm_strtoul(p, &p);
+		nr = cmm_strtoul(p, &p);
 		if (ctl == &cmm_table[0])
-			cmm_set_pages(pages);
+			cmm_set_pages(nr);
 		else
-			cmm_add_timed_pages(pages);
+			cmm_add_timed_pages(nr);
 	} else {
 		if (ctl == &cmm_table[0])
-			pages = cmm_get_pages();
+			nr = cmm_get_pages();
 		else
-			pages = cmm_get_timed_pages();
-		len = sprintf(buf, "%ld\n", pages);
+			nr = cmm_get_timed_pages();
+		len = sprintf(buf, "%ld\n", nr);
 		if (len > *lenp)
 			len = *lenp;
 		if (copy_to_user(buffer, buf, len))
@@ -302,7 +335,7 @@ cmm_timeout_handler(ctl_table *ctl, int 
 		    void __user *buffer, size_t *lenp, loff_t *ppos)
 {
 	char buf[64], *p;
-	long pages, seconds;
+	long nr, seconds;
 	int len;
 
 	if (!*lenp || (*ppos && !write)) {
@@ -317,10 +350,10 @@ cmm_timeout_handler(ctl_table *ctl, int 
 			return -EFAULT;
 		buf[sizeof(buf) - 1] = '\0';
 		cmm_skip_blanks(buf, &p);
-		pages = cmm_strtoul(p, &p);
+		nr = cmm_strtoul(p, &p);
 		cmm_skip_blanks(p, &p);
 		seconds = cmm_strtoul(p, &p);
-		cmm_set_timeout(pages, seconds);
+		cmm_set_timeout(nr, seconds);
 	} else {
 		len = sprintf(buf, "%ld %ld\n",
 			      cmm_timeout_pages, cmm_timeout_seconds);
@@ -373,7 +406,7 @@ static struct ctl_table cmm_dir_table[] 
 static void
 cmm_smsg_target(char *from, char *msg)
 {
-	long pages, seconds;
+	long nr, seconds;
 
 	if (strlen(sender) > 0 && strcmp(from, sender) != 0)
 		return;
@@ -382,27 +415,27 @@ cmm_smsg_target(char *from, char *msg)
 	if (strncmp(msg, "SHRINK", 6) == 0) {
 		if (!cmm_skip_blanks(msg + 6, &msg))
 			return;
-		pages = cmm_strtoul(msg, &msg);
+		nr = cmm_strtoul(msg, &msg);
 		cmm_skip_blanks(msg, &msg);
 		if (*msg == '\0')
-			cmm_set_pages(pages);
+			cmm_set_pages(nr);
 	} else if (strncmp(msg, "RELEASE", 7) == 0) {
 		if (!cmm_skip_blanks(msg + 7, &msg))
 			return;
-		pages = cmm_strtoul(msg, &msg);
+		nr = cmm_strtoul(msg, &msg);
 		cmm_skip_blanks(msg, &msg);
 		if (*msg == '\0')
-			cmm_add_timed_pages(pages);
+			cmm_add_timed_pages(nr);
 	} else if (strncmp(msg, "REUSE", 5) == 0) {
 		if (!cmm_skip_blanks(msg + 5, &msg))
 			return;
-		pages = cmm_strtoul(msg, &msg);
+		nr = cmm_strtoul(msg, &msg);
 		if (!cmm_skip_blanks(msg, &msg))
 			return;
 		seconds = cmm_strtoul(msg, &msg);
 		cmm_skip_blanks(msg, &msg);
 		if (*msg == '\0')
-			cmm_set_timeout(pages, seconds);
+			cmm_set_timeout(nr, seconds);
 	}
 }
 #endif
@@ -418,6 +451,7 @@ cmm_init (void)
 #ifdef CONFIG_CMM_IUCV
 	smsg_register_callback(SMSG_PREFIX, cmm_smsg_target);
 #endif
+	register_oom_notifier(&cmm_oom_nb);
 	INIT_WORK(&cmm_thread_starter, (void *) cmm_start_thread, NULL);
 	init_waitqueue_head(&cmm_thread_wait);
 	init_timer(&cmm_timer);
@@ -427,6 +461,7 @@ cmm_init (void)
 static void
 cmm_exit(void)
 {
+	unregister_oom_notifier(&cmm_oom_nb);
 	cmm_free_pages(cmm_pages, &cmm_pages, &cmm_page_list);
 	cmm_free_pages(cmm_timed_pages, &cmm_timed_pages, &cmm_timed_page_list);
 #ifdef CONFIG_CMM_PROC
Index: linux-rhel5/include/linux/swap.h
===================================================================
--- linux-rhel5.orig/include/linux/swap.h	2007-03-05 18:23:35.000000000 +0100
+++ linux-rhel5/include/linux/swap.h	2007-03-06 18:01:03.000000000 +0100
@@ -12,6 +12,8 @@
 
 struct bio;
 
+struct notifier_block;
+
 #define SWAP_FLAG_PREFER	0x8000	/* set if swap priority specified */
 #define SWAP_FLAG_PRIO_MASK	0x7fff
 #define SWAP_FLAG_PRIO_SHIFT	0
@@ -158,6 +160,8 @@ struct swap_list_t {
 
 /* linux/mm/oom_kill.c */
 extern void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, int order);
+extern int register_oom_notifier(struct notifier_block *nb);
+extern int unregister_oom_notifier(struct notifier_block *nb);
 
 /* linux/mm/memory.c */
 extern void swapin_readahead(swp_entry_t, unsigned long, struct vm_area_struct *);
Index: linux-rhel5/mm/oom_kill.c
===================================================================
--- linux-rhel5.orig/mm/oom_kill.c	2007-03-05 18:23:36.000000000 +0100
+++ linux-rhel5/mm/oom_kill.c	2007-03-06 18:01:03.000000000 +0100
@@ -21,6 +21,8 @@
 #include <linux/timex.h>
 #include <linux/jiffies.h>
 #include <linux/cpuset.h>
+#include <linux/module.h>
+#include <linux/notifier.h>
 
 int sysctl_panic_on_oom;
 /* #define DEBUG */
@@ -404,6 +406,20 @@ out_unlock:
 	return ret;
 }
 
+static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
+
+int register_oom_notifier(struct notifier_block *nb)
+{
+	return blocking_notifier_chain_register(&oom_notify_list, nb);
+}
+EXPORT_SYMBOL_GPL(register_oom_notifier);
+
+int unregister_oom_notifier(struct notifier_block *nb)
+{
+	return blocking_notifier_chain_unregister(&oom_notify_list, nb);
+}
+EXPORT_SYMBOL_GPL(unregister_oom_notifier);
+
 /**
  * out_of_memory - kill the "best" process when we run out of memory
  *
@@ -416,6 +432,12 @@ void out_of_memory(struct zonelist *zone
 {
 	struct task_struct *p;
 	unsigned long points = 0;
+	unsigned long freed = 0;
+
+	blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
+	if (freed > 0)
+		/* Got some memory back in the last second. */
+		return;
 
 	if (printk_ratelimit()) {
 		printk(KERN_WARNING "%s invoked oom-killer: "