Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <1659939208-22454-1-git-send-email-zhaoyang.huang@unisoc.com> <cb20ece8-3053-1b6c-5820-08663e5c1acc@bytedance.com>
In-Reply-To: <cb20ece8-3053-1b6c-5820-08663e5c1acc@bytedance.com>
From:   Zhaoyang Huang <huangzhaoyang@gmail.com>
Date:   Mon, 8 Aug 2022 17:39:39 +0800
Message-ID: <CAGWkznEFEiV4xchHCKTh39zKbpgGR_vSZw_R1KgTiEDaQ6E-Rg@mail.gmail.com>
Subject: Re: [RFC PATCH] psi: introduce PSI UNINTERRUPTIBLE
To:     Chengming Zhou <zhouchengming@bytedance.com>
Cc:     "zhaoyang.huang" <zhaoyang.huang@unisoc.com>,
        Johannes Weiner <hannes@cmpxchg.org>,
        Suren Baghdasaryan <surenb@google.com>,
        LKML <linux-kernel@vger.kernel.org>,
        Ke Wang <ke.wang@unisoc.com>, xuewen.yan@unisoc.com
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Mon, Aug 8, 2022 at 5:12 PM Chengming Zhou
<zhouchengming@bytedance.com> wrote:
>
> On 2022/8/8 14:13, zhaoyang.huang wrote:
> > From: Zhaoyang Huang <zhaoyang.huang@unisoc.com>
> >
> > Uninterruptible sleep has not been monitored as an important system status yet.
> > Imagin that a set of psi triggers are created for monitoring a special group, while
> > get nothing high for none of the pressures, which could be the processes within
> > are stock in some given resources and turn to be UN status. Introduce PSI_UN as
> > a sub-type among PSI system here.
>
> Hello,
>
> The problem is that not all TASK_UNINTERRUPTIBLE task means stalled on some
> shared resource, like many schedule_timeout() paths.
Thanks for heads up. The aim is to distinguish where the processes go
(on or off cpu? waiting for anything?). It could be deemed as PSI_MEM
like property which is not a precise running time for memory things.
Furthermore, we do have method to make it be precise.
>
> Thanks.
>
> >
> > Signed-off-by: Zhaoyang Huang <zhaoyang.huang@unisoc.com>
> > ---
> >  include/linux/psi_types.h | 11 ++++++++---
> >  kernel/sched/psi.c        | 10 ++++++++++
> >  kernel/sched/stats.h      |  6 +++++-
> >  3 files changed, 23 insertions(+), 4 deletions(-)
> >
> > diff --git a/include/linux/psi_types.h b/include/linux/psi_types.h
> > index c7fe7c0..8cc1979 100644
> > --- a/include/linux/psi_types.h
> > +++ b/include/linux/psi_types.h
> > @@ -15,6 +15,7 @@ enum psi_task_count {
> >       NR_IOWAIT,
> >       NR_MEMSTALL,
> >       NR_RUNNING,
> > +     NR_UNINTERRUPTIBLE,
> >       /*
> >        * This can't have values other than 0 or 1 and could be
> >        * implemented as a bit flag. But for now we still have room
> > @@ -32,7 +33,7 @@ enum psi_task_count {
> >        * threads and memstall ones.
> >        */
> >       NR_MEMSTALL_RUNNING,
> > -     NR_PSI_TASK_COUNTS = 5,
> > +     NR_PSI_TASK_COUNTS = 6,
> >  };
> >
> >  /* Task state bitmasks */
> > @@ -41,13 +42,15 @@ enum psi_task_count {
> >  #define TSK_RUNNING  (1 << NR_RUNNING)
> >  #define TSK_ONCPU    (1 << NR_ONCPU)
> >  #define TSK_MEMSTALL_RUNNING (1 << NR_MEMSTALL_RUNNING)
> > +#define TSK_UNINTERRUPTIBLE  (1 << NR_UNINTERRUPTIBLE)
> >
> >  /* Resources that workloads could be stalled on */
> >  enum psi_res {
> >       PSI_IO,
> >       PSI_MEM,
> >       PSI_CPU,
> > -     NR_PSI_RESOURCES = 3,
> > +     PSI_UN,
> > +     NR_PSI_RESOURCES = 4,
> >  };
> >
> >  /*
> > @@ -63,9 +66,11 @@ enum psi_states {
> >       PSI_MEM_FULL,
> >       PSI_CPU_SOME,
> >       PSI_CPU_FULL,
> > +     PSI_UN_SOME,
> > +     PSI_UN_FULL,
> >       /* Only per-CPU, to weigh the CPU in the global average: */
> >       PSI_NONIDLE,
> > -     NR_PSI_STATES = 7,
> > +     NR_PSI_STATES = 9,
> >  };
> >
> >  enum psi_aggregators {
> > diff --git a/kernel/sched/psi.c b/kernel/sched/psi.c
> > index a337f3e..a37b4a4 100644
> > --- a/kernel/sched/psi.c
> > +++ b/kernel/sched/psi.c
> > @@ -231,6 +231,10 @@ static bool test_state(unsigned int *tasks, enum psi_states state)
> >               return unlikely(tasks[NR_RUNNING] > tasks[NR_ONCPU]);
> >       case PSI_CPU_FULL:
> >               return unlikely(tasks[NR_RUNNING] && !tasks[NR_ONCPU]);
> > +     case PSI_UN_SOME:
> > +             return unlikely(tasks[NR_UNINTERRUPTIBLE]);
> > +     case PSI_UN_FULL:
> > +             return unlikely(tasks[NR_UNINTERRUPTIBLE] && !tasks[NR_RUNNING]);
> >       case PSI_NONIDLE:
> >               return tasks[NR_IOWAIT] || tasks[NR_MEMSTALL] ||
> >                       tasks[NR_RUNNING];
> > @@ -683,6 +687,12 @@ static void record_times(struct psi_group_cpu *groupc, u64 now)
> >                       groupc->times[PSI_CPU_FULL] += delta;
> >       }
> >
> > +     if (groupc->state_mask & (1 << PSI_UN_SOME)) {
> > +             groupc->times[PSI_UN_SOME] += delta;
> > +             if (groupc->state_mask & (1 << PSI_UN_FULL))
> > +                     groupc->times[PSI_UN_FULL] += delta;
> > +     }
> > +
> >       if (groupc->state_mask & (1 << PSI_NONIDLE))
> >               groupc->times[PSI_NONIDLE] += delta;
> >  }
> > diff --git a/kernel/sched/stats.h b/kernel/sched/stats.h
> > index baa839c..bf98829 100644
> > --- a/kernel/sched/stats.h
> > +++ b/kernel/sched/stats.h
> > @@ -132,6 +132,7 @@ static inline void psi_enqueue(struct task_struct *p, bool wakeup)
> >               if (p->in_iowait)
> >                       clear |= TSK_IOWAIT;
> >       }
> > +     clear |= TSK_UNINTERRUPTIBLE;
> >
> >       psi_task_change(p, clear, set);
> >  }
> > @@ -139,6 +140,7 @@ static inline void psi_enqueue(struct task_struct *p, bool wakeup)
> >  static inline void psi_dequeue(struct task_struct *p, bool sleep)
> >  {
> >       int clear = TSK_RUNNING;
> > +     int set = 0;
> >
> >       if (static_branch_likely(&psi_disabled))
> >               return;
> > @@ -154,8 +156,10 @@ static inline void psi_dequeue(struct task_struct *p, bool sleep)
> >
> >       if (p->in_memstall)
> >               clear |= (TSK_MEMSTALL | TSK_MEMSTALL_RUNNING);
> > +     if (READ_ONCE(p->__state) & TASK_UNINTERRUPTIBLE)
> > +             set = TSK_UNINTERRUPTIBLE;
> >
> > -     psi_task_change(p, clear, 0);
> > +     psi_task_change(p, clear, set);
> >  }
> >
> >  static inline void psi_ttwu_dequeue(struct task_struct *p)