On 05/07/2018 02:01 PM, Johannes Weiner wrote:
>
> Signed-off-by: Johannes Weiner <hannes@xxxxxxxxxxx>
> ---
> Documentation/accounting/psi.txt | 73 ++++++
> include/linux/psi.h | 27 ++
> include/linux/psi_types.h | 84 ++++++
> include/linux/sched.h | 10 +
> include/linux/sched/stat.h | 10 +-
> init/Kconfig | 16 ++
> kernel/fork.c | 4 +
> kernel/sched/Makefile | 1 +
> kernel/sched/core.c | 3 +
> kernel/sched/psi.c | 424 +++++++++++++++++++++++++++++++
> kernel/sched/sched.h | 166 ++++++------
> kernel/sched/stats.h | 91 ++++++-
> mm/compaction.c | 5 +
> mm/filemap.c | 15 +-
> mm/page_alloc.c | 10 +
> mm/vmscan.c | 13 +
> 16 files changed, 859 insertions(+), 93 deletions(-)
> create mode 100644 Documentation/accounting/psi.txt
> create mode 100644 include/linux/psi.h
> create mode 100644 include/linux/psi_types.h
> create mode 100644 kernel/sched/psi.c
>
> diff --git a/Documentation/accounting/psi.txt b/Documentation/accounting/psi.txt
> new file mode 100644
> index 000000000000..e051810d5127
> --- /dev/null
> +++ b/Documentation/accounting/psi.txt
> @@ -0,0 +1,73 @@
Looks good to me.
> diff --git a/kernel/sched/psi.c b/kernel/sched/psi.c
> new file mode 100644
> index 000000000000..052c529a053b
> --- /dev/null
> +++ b/kernel/sched/psi.c
> @@ -0,0 +1,424 @@
> +/*
> + * Measure workload productivity impact from overcommitting CPU, memory, IO
> + *
> + * Copyright (c) 2017 Facebook, Inc.
> + * Author: Johannes Weiner <hannes@xxxxxxxxxxx>
> + *
> + * Implementation
> + *
> + * Task states -- running, iowait, memstall -- are tracked through the
> + * scheduler and aggregated into a system-wide productivity state. The
> + * ratio between the times spent in productive states and delays tells
> + * us the overall productivity of the workload.
> + *
> + * The ratio is tracked in decaying time averages over 10s, 1m, 5m
> + * windows. Cumluative stall times are tracked and exported as well to
Cumulative
> + * allow detection of latency spikes and custom time averaging.
> + *
> + * Multiple CPUs
> + *
> + * To avoid cache contention, times are tracked local to the CPUs. To
> + * get a comprehensive view of a system or cgroup, we have to consider
> + * the fact that CPUs could be unevenly loaded or even entirely idle
> + * if the workload doesn't have enough threads. To avoid artifacts
> + * caused by that, when adding up the global pressure ratio, the
> + * CPU-local ratios are weighed according to their non-idle time:
> + *
> + * Time the CPU had stalled tasks Time the CPU was non-idle
> + * ------------------------------ * ---------------------------
> + * Walltime Time all CPUs were non-idle
> + */
> +
> +/**
> + * psi_memstall_leave - mark the end of an memory stall section
end of a memory
> + * @flags: flags to handle nested memdelay sections
> + *
> + * Marks the calling task as no longer stalled due to lack of memory.
> + */
> +void psi_memstall_leave(unsigned long *flags)
> +{
--
~Randy
