Re: + mm-oom-avoid-printk-iteration-under-rcu.patch added to -mm tree
[ups, I've cut the Cc list somehow, sorry about that]
On Wed 24-07-19 08:27:27, Michal Hocko wrote:
> Andrew,
> I've had some concerns wrt. this patch - especially the additional
> complexity - and I have to say I am not convinced that this is really
> needed. Our past experience in this area suggests that more tricky code
> leads to different corner cases. So I am really reluctant to add more
> complexity without any real world reports.
>
> On Tue 23-07-19 16:14:29, Andrew Morton wrote:
> > From: Tetsuo Handa
> > Subject: mm, oom: avoid printk() iteration under RCU
> >
> > Currently dump_tasks() might call printk() for many thousands times under
> > RCU, which might take many minutes for slow consoles. Therefore, split
> > dump_tasks() into three stages; take a snapshot of possible OOM victim
> > candidates under RCU, dump the snapshot from reschedulable context, and
> > destroy the snapshot.
> >
> > In a future patch, the first stage would be moved to select_bad_process()
> > and the third stage would be moved to after oom_kill_process(), and will
> > simplify refcount handling.
> >
> > Link:
> > http://lkml.kernel.org/r/1563360901-8277-1-git-send-email-penguin-ker...@i-love.sakura.ne.jp
> > Signed-off-by: Tetsuo Handa
> > Cc: Shakeel Butt
> > Cc: Michal Hocko
> > Cc: Roman Gushchin
> > Cc: David Rientjes
> > Signed-off-by: Andrew Morton
> > ---
> >
> > include/linux/sched.h |1
> > mm/oom_kill.c | 67 +++-
> > 2 files changed, 34 insertions(+), 34 deletions(-)
> >
> > --- a/include/linux/sched.h~mm-oom-avoid-printk-iteration-under-rcu
> > +++ a/include/linux/sched.h
> > @@ -1246,6 +1246,7 @@ struct task_struct {
> > #ifdef CONFIG_MMU
> > struct task_struct *oom_reaper_list;
> > #endif
> > + struct list_headoom_victim_list;
> > #ifdef CONFIG_VMAP_STACK
> > struct vm_struct*stack_vm_area;
> > #endif
> > --- a/mm/oom_kill.c~mm-oom-avoid-printk-iteration-under-rcu
> > +++ a/mm/oom_kill.c
> > @@ -377,36 +377,13 @@ static void select_bad_process(struct oo
> > }
> > }
> >
> > -static int dump_task(struct task_struct *p, void *arg)
> > -{
> > - struct oom_control *oc = arg;
> > - struct task_struct *task;
> > -
> > - if (oom_unkillable_task(p))
> > - return 0;
> >
> > - /* p may not have freeable memory in nodemask */
> > - if (!is_memcg_oom(oc) && !oom_cpuset_eligible(p, oc))
> > - return 0;
> > -
> > - task = find_lock_task_mm(p);
> > - if (!task) {
> > - /*
> > -* This is a kthread or all of p's threads have already
> > -* detached their mm's. There's no need to report
> > -* them; they can't be oom killed anyway.
> > -*/
> > - return 0;
> > +static int add_candidate_task(struct task_struct *p, void *arg)
> > +{
> > + if (!oom_unkillable_task(p)) {
> > + get_task_struct(p);
> > + list_add_tail(>oom_victim_list, (struct list_head *) arg);
> > }
> > -
> > - pr_info("[%7d] %5d %5d %8lu %8lu %8ld %8lu %5hd %s\n",
> > - task->pid, from_kuid(_user_ns, task_uid(task)),
> > - task->tgid, task->mm->total_vm, get_mm_rss(task->mm),
> > - mm_pgtables_bytes(task->mm),
> > - get_mm_counter(task->mm, MM_SWAPENTS),
> > - task->signal->oom_score_adj, task->comm);
> > - task_unlock(task);
> > -
> > return 0;
> > }
> >
> > @@ -422,19 +399,41 @@ static int dump_task(struct task_struct
> > */
> > static void dump_tasks(struct oom_control *oc)
> > {
> > - pr_info("Tasks state (memory values in pages):\n");
> > - pr_info("[ pid ] uid tgid total_vm rss pgtables_bytes
> > swapents oom_score_adj name\n");
> > + static LIST_HEAD(list);
> > + struct task_struct *p;
> > + struct task_struct *t;
> >
> > if (is_memcg_oom(oc))
> > - mem_cgroup_scan_tasks(oc->memcg, dump_task, oc);
> > + mem_cgroup_scan_tasks(oc->memcg, add_candidate_task, );
> > else {
> > - struct task_struct *p;
> > -
> > rcu_read_lock();
> > for_each_process(p)
> > - dump_task(p, oc);
> > + add_candidate_task(p, );
> > rcu_read_unlock();
> > }
> > + pr_info("Tasks state (memory values in pages):\n");
> > + pr_info("[ pid ] uid tgid total_vm rss pgtables_bytes
> > swapents oom_score_adj name\n");
> > + list_for_each_entry(p, , oom_victim_list) {
> > + cond_resched();
> > + /* p may not have freeable memory in nodemask */
> > + if (!is_memcg_oom(oc) && !oom_cpuset_eligible(p, oc))
> > + continue;
> > + /* All of p's threads might have already detached their mm's. */
> > + t = find_lock_task_mm(p);
> > + if (!t)
> > + continue;
> > + pr_info("[%7d] %5d %5d %8lu %8lu %8ld %8lu
Re: + mm-oom-avoid-printk-iteration-under-rcu.patch added to -mm tree
Andrew,
I've had some concerns wrt. this patch - especially the additional
complexity - and I have to say I am not convinced that this is really
needed. Our past experience in this area suggests that more tricky code
leads to different corner cases. So I am really reluctant to add more
complexity without any real world reports.
On Tue 23-07-19 16:14:29, Andrew Morton wrote:
> From: Tetsuo Handa
> Subject: mm, oom: avoid printk() iteration under RCU
>
> Currently dump_tasks() might call printk() for many thousands times under
> RCU, which might take many minutes for slow consoles. Therefore, split
> dump_tasks() into three stages; take a snapshot of possible OOM victim
> candidates under RCU, dump the snapshot from reschedulable context, and
> destroy the snapshot.
>
> In a future patch, the first stage would be moved to select_bad_process()
> and the third stage would be moved to after oom_kill_process(), and will
> simplify refcount handling.
>
> Link:
> http://lkml.kernel.org/r/1563360901-8277-1-git-send-email-penguin-ker...@i-love.sakura.ne.jp
> Signed-off-by: Tetsuo Handa
> Cc: Shakeel Butt
> Cc: Michal Hocko
> Cc: Roman Gushchin
> Cc: David Rientjes
> Signed-off-by: Andrew Morton
> ---
>
> include/linux/sched.h |1
> mm/oom_kill.c | 67 +++-
> 2 files changed, 34 insertions(+), 34 deletions(-)
>
> --- a/include/linux/sched.h~mm-oom-avoid-printk-iteration-under-rcu
> +++ a/include/linux/sched.h
> @@ -1246,6 +1246,7 @@ struct task_struct {
> #ifdef CONFIG_MMU
> struct task_struct *oom_reaper_list;
> #endif
> + struct list_headoom_victim_list;
> #ifdef CONFIG_VMAP_STACK
> struct vm_struct*stack_vm_area;
> #endif
> --- a/mm/oom_kill.c~mm-oom-avoid-printk-iteration-under-rcu
> +++ a/mm/oom_kill.c
> @@ -377,36 +377,13 @@ static void select_bad_process(struct oo
> }
> }
>
> -static int dump_task(struct task_struct *p, void *arg)
> -{
> - struct oom_control *oc = arg;
> - struct task_struct *task;
> -
> - if (oom_unkillable_task(p))
> - return 0;
>
> - /* p may not have freeable memory in nodemask */
> - if (!is_memcg_oom(oc) && !oom_cpuset_eligible(p, oc))
> - return 0;
> -
> - task = find_lock_task_mm(p);
> - if (!task) {
> - /*
> - * This is a kthread or all of p's threads have already
> - * detached their mm's. There's no need to report
> - * them; they can't be oom killed anyway.
> - */
> - return 0;
> +static int add_candidate_task(struct task_struct *p, void *arg)
> +{
> + if (!oom_unkillable_task(p)) {
> + get_task_struct(p);
> + list_add_tail(>oom_victim_list, (struct list_head *) arg);
> }
> -
> - pr_info("[%7d] %5d %5d %8lu %8lu %8ld %8lu %5hd %s\n",
> - task->pid, from_kuid(_user_ns, task_uid(task)),
> - task->tgid, task->mm->total_vm, get_mm_rss(task->mm),
> - mm_pgtables_bytes(task->mm),
> - get_mm_counter(task->mm, MM_SWAPENTS),
> - task->signal->oom_score_adj, task->comm);
> - task_unlock(task);
> -
> return 0;
> }
>
> @@ -422,19 +399,41 @@ static int dump_task(struct task_struct
> */
> static void dump_tasks(struct oom_control *oc)
> {
> - pr_info("Tasks state (memory values in pages):\n");
> - pr_info("[ pid ] uid tgid total_vm rss pgtables_bytes
> swapents oom_score_adj name\n");
> + static LIST_HEAD(list);
> + struct task_struct *p;
> + struct task_struct *t;
>
> if (is_memcg_oom(oc))
> - mem_cgroup_scan_tasks(oc->memcg, dump_task, oc);
> + mem_cgroup_scan_tasks(oc->memcg, add_candidate_task, );
> else {
> - struct task_struct *p;
> -
> rcu_read_lock();
> for_each_process(p)
> - dump_task(p, oc);
> + add_candidate_task(p, );
> rcu_read_unlock();
> }
> + pr_info("Tasks state (memory values in pages):\n");
> + pr_info("[ pid ] uid tgid total_vm rss pgtables_bytes
> swapents oom_score_adj name\n");
> + list_for_each_entry(p, , oom_victim_list) {
> + cond_resched();
> + /* p may not have freeable memory in nodemask */
> + if (!is_memcg_oom(oc) && !oom_cpuset_eligible(p, oc))
> + continue;
> + /* All of p's threads might have already detached their mm's. */
> + t = find_lock_task_mm(p);
> + if (!t)
> + continue;
> + pr_info("[%7d] %5d %5d %8lu %8lu %8ld %8lu %5hd %s\n",
> + t->pid, from_kuid(_user_ns, task_uid(t)),
> + t->tgid, t->mm->total_vm, get_mm_rss(t->mm),
> + mm_pgtables_bytes(t->mm),
> +
