arm64/ascend: Add new enable_oom_killer interface for oom contrl

ascend inclusion
category: feature
bugzilla: NA
CVE: NA

-------------------------------------------------

Support disable oom-killer, and report oom events to bbox
vm.enable_oom_killer:
	0: disable oom killer
	1: enable oom killer (default,compatible with mainline)
	2: disable oom killer and panic on oom
Signed-off-by: NWeilong Chen <chenweilong@huawei.com>
Reviewed-by: NKefeng Wang <wangkefeng.wang@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

arch/arm64/Kconfig

@@ -1335,6 +1335,17 @@ config ASCEND_DVPP_MMAP
 	  special memory for DvPP processor, the new flag is only valid for Ascend
 	  platform.
 
+config ASCEND_OOM
+	bool "Enable support for disable oom killer"
+	default y
+	help
+	  In some cases we hopes that the oom will not kill the process when it occurs,
+	  be able to notify the black box to report the event, and be able to trigger
+	  the panic to locate the problem.
+	  vm.enable_oom_killer:
+		0: disable oom killer
+		1: enable oom killer (default,compatible with mainline)
+		2: disable oom killer and panic on oom
 endif
 
 endmenu

include/linux/oom.h

@@ -117,4 +117,15 @@ extern struct task_struct *find_lock_task_mm(struct task_struct *p);
 extern int sysctl_oom_dump_tasks;
 extern int sysctl_oom_kill_allocating_task;
 extern int sysctl_panic_on_oom;
+
+#ifdef CONFIG_ASCEND_OOM
+#define HISI_OOM_TYPE_NOMEM		0
+#define HISI_OOM_TYPE_OVERCOMMIT	1
+#define HISI_OOM_TYPE_CGROUP		2
+
+extern int sysctl_enable_oom_killer;
+extern int register_hisi_oom_notifier(struct notifier_block *nb);
+extern int hisi_oom_notifier_call(unsigned long val, void *v);
+extern int unregister_hisi_oom_notifier(struct notifier_block *nb);
+#endif
 #endif /* _INCLUDE_LINUX_OOM_H */

kernel/sysctl.c

@@ -1264,6 +1264,18 @@ static struct ctl_table vm_table[] = {
 		.extra1		= &zero,
 		.extra2		= &two,
 	},
+#ifdef CONFIG_ASCEND_OOM
+	{
+		/* 0: diasable, 1: enable, 2: disable and panic on oom */
+		.procname	= "enable_oom_killer",
+		.data		= &sysctl_enable_oom_killer,
+		.maxlen		= sizeof(sysctl_enable_oom_killer),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= &zero,
+		.extra2		= &two,
+	},
+#endif
 	{
 		.procname	= "oom_kill_allocating_task",
 		.data		= &sysctl_oom_kill_allocating_task,

mm/memcontrol.c

@@ -1729,6 +1729,9 @@ static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int
 		current->memcg_in_oom = memcg;
 		current->memcg_oom_gfp_mask = mask;
 		current->memcg_oom_order = order;
+#ifdef CONFIG_ASCEND_OOM
+		hisi_oom_notifier_call(HISI_OOM_TYPE_CGROUP, NULL);
+#endif
 
 		return OOM_ASYNC;
 	}
@@ -1802,6 +1805,9 @@ bool mem_cgroup_oom_synchronize(bool handle)
 		mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask,
 					 current->memcg_oom_order);
 	} else {
+#ifdef CONFIG_ASCEND_OOM
+		hisi_oom_notifier_call(HISI_OOM_TYPE_CGROUP, NULL);
+#endif
 		schedule();
 		mem_cgroup_unmark_under_oom(memcg);
 		finish_wait(&memcg_oom_waitq, &owait.wait);

mm/oom_kill.c

@@ -52,6 +52,9 @@
 int sysctl_panic_on_oom;
 int sysctl_oom_kill_allocating_task;
 int sysctl_oom_dump_tasks = 1;
+#ifdef CONFIG_ASCEND_OOM
+int sysctl_enable_oom_killer = 1;
+#endif
 
 /*
  * Serializes oom killer invocations (out_of_memory()) from all contexts to
@@ -1047,6 +1050,42 @@ int unregister_oom_notifier(struct notifier_block *nb)
 }
 EXPORT_SYMBOL_GPL(unregister_oom_notifier);
 
+#ifdef CONFIG_ASCEND_OOM
+static BLOCKING_NOTIFIER_HEAD(hisi_oom_notify_list);
+
+int register_hisi_oom_notifier(struct notifier_block *nb)
+{
+	return blocking_notifier_chain_register(&hisi_oom_notify_list, nb);
+}
+EXPORT_SYMBOL_GPL(register_hisi_oom_notifier);
+
+static unsigned long last_jiffies;
+int hisi_oom_notifier_call(unsigned long val, void *v)
+{
+	/* when enable oom killer, just return */
+	if (sysctl_enable_oom_killer == 1)
+		return 0;
+
+	/* Print time interval to 10 seconds */
+	if (time_after(jiffies, last_jiffies + 10 * HZ)) {
+		pr_err("OOM_NOTIFIER: oom type %lu\n", val);
+		dump_stack();
+		show_mem(SHOW_MEM_FILTER_NODES, NULL);
+		dump_tasks(NULL, 0);
+		last_jiffies = jiffies;
+	}
+
+	return blocking_notifier_call_chain(&hisi_oom_notify_list, val, v);
+}
+EXPORT_SYMBOL_GPL(hisi_oom_notifier_call);
+
+int unregister_hisi_oom_notifier(struct notifier_block *nb)
+{
+	return blocking_notifier_chain_unregister(&hisi_oom_notify_list, nb);
+}
+EXPORT_SYMBOL_GPL(unregister_hisi_oom_notifier);
+#endif
+
 /**
  * out_of_memory - kill the "best" process when we run out of memory
  * @oc: pointer to struct oom_control
@@ -1060,10 +1099,27 @@ bool out_of_memory(struct oom_control *oc)
 {
 	unsigned long freed = 0;
 	enum oom_constraint constraint = CONSTRAINT_NONE;
+#ifdef CONFIG_ASCEND_OOM
+	unsigned long oom_type;
+#endif
 
 	if (oom_killer_disabled)
 		return false;
 
+#ifdef CONFIG_ASCEND_OOM
+	if (sysctl_enable_oom_killer == 0 || sysctl_enable_oom_killer == 2) {
+		if (is_memcg_oom(oc))
+			oom_type = HISI_OOM_TYPE_CGROUP;
+		else
+			oom_type = HISI_OOM_TYPE_NOMEM;
+
+		hisi_oom_notifier_call(oom_type, NULL);
+		if (unlikely(sysctl_enable_oom_killer == 2))
+			panic("Out of memory, panic by sysctl_enable_oom_killer");
+		return false;
+	}
+#endif
+
 	if (!is_memcg_oom(oc)) {
 		blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
 		if (freed > 0)

mm/util.c

@@ -17,6 +17,9 @@
 
 #include <asm/sections.h>
 #include <linux/uaccess.h>
+#ifdef CONFIG_ASCEND_OOM
+#include <linux/oom.h>
+#endif
 
 #include "internal.h"
 
@@ -726,6 +729,9 @@ int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
 	if (percpu_counter_read_positive(&vm_committed_as) < allowed)
 		return 0;
 error:
+#ifdef CONFIG_ASCEND_OOM
+	hisi_oom_notifier_call(HISI_OOM_TYPE_OVERCOMMIT, NULL);
+#endif
 	vm_unacct_memory(pages);
 
 	return -ENOMEM;