[PATCH] mm, oom: OOM sysrq should always kill a process

[PATCH] mm, oom: OOM sysrq should always kill a process

[Jann Horn]

The OOM kill sysrq (alt+sysrq+F) should allow the user to kill the
process with the highest OOM badness with a single execution.

However, at the moment, the OOM kill can bail out if an OOM notifier
(e.g. the i915 one) says that it reclaimed a tiny amount of memory
from somewhere. That's probably not what the user wants.

As documented in struct oom_control, order == -1 means the oom kill is
required by sysrq. So check for that, and if it's true, don't bail out
no matter what the OOM notifiers say.

Signed-off-by: Jann Horn <jannh@google.com>
---
 mm/oom_kill.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/mm/oom_kill.c b/mm/oom_kill.c
index 1ddabefcfb5a..dc645cbc6e0d 100644
--- a/mm/oom_kill.c
+++ b/mm/oom_kill.c
@@ -1051,13 +1051,14 @@ EXPORT_SYMBOL_GPL(unregister_oom_notifier);
 bool out_of_memory(struct oom_control *oc)
 {
 	unsigned long freed = 0;
+	bool sysrq_forced = oc->order == -1;
 
 	if (oom_killer_disabled)
 		return false;
 
 	if (!is_memcg_oom(oc)) {
 		blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
-		if (freed > 0)
+		if (freed > 0 && !sysrq_forced)
 			/* Got some memory back in the last second. */
 			return true;
 	}

base-commit: c9e6606c7fe92b50a02ce51dda82586ebdf99b48
-- 
2.34.1.448.ga2b2bfdf31-goog

Re: [PATCH] mm, oom: OOM sysrq should always kill a process

[Michal Hocko]

On Wed 05-01-22 18:51:15, Jann Horn wrote:
> The OOM kill sysrq (alt+sysrq+F) should allow the user to kill the
> process with the highest OOM badness with a single execution.
> 
> However, at the moment, the OOM kill can bail out if an OOM notifier
> (e.g. the i915 one) says that it reclaimed a tiny amount of memory
> from somewhere. That's probably not what the user wants.
> 
> As documented in struct oom_control, order == -1 means the oom kill is
> required by sysrq. So check for that, and if it's true, don't bail out
> no matter what the OOM notifiers say.

I agree that it is suboptimal to disable sysrq+f because of notfiers
because the OOM invocation is not a direct result of the OOM situation
but rather an admin will. We already kill a new task even if an oom
victim is still pending.

> Signed-off-by: Jann Horn <jannh@google.com>

with a minor update as below
Acked-by: Michal Hocko <mhocko@suse.com>

> ---
>  mm/oom_kill.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/mm/oom_kill.c b/mm/oom_kill.c
> index 1ddabefcfb5a..dc645cbc6e0d 100644
> --- a/mm/oom_kill.c
> +++ b/mm/oom_kill.c
> @@ -1051,13 +1051,14 @@ EXPORT_SYMBOL_GPL(unregister_oom_notifier);
>  bool out_of_memory(struct oom_control *oc)
>  {
>  	unsigned long freed = 0;
> +	bool sysrq_forced = oc->order == -1;
>  
>  	if (oom_killer_disabled)
>  		return false;
>  
>  	if (!is_memcg_oom(oc)) {
>  		blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
> -		if (freed > 0)
> +		if (freed > 0 && !sysrq_forced)
>  			/* Got some memory back in the last second. */
>  			return true;
>  	}

is_sysrq_oom(oc) is a more appropriate way to check this.

> 
> base-commit: c9e6606c7fe92b50a02ce51dda82586ebdf99b48
> -- 
> 2.34.1.448.ga2b2bfdf31-goog

-- 
Michal Hocko
SUSE Labs