Message-ID: <CAJHvVcg6eY0vVtfi8D6D9aus7=5zeP2H7Yc0mY5ofXztSzOFqQ@mail.gmail.com>
Date:   Mon, 21 Sep 2020 09:53:13 -0700
From:   Axel Rasmussen <axelrasmussen@...gle.com>
To:     Yafang Shao <laoar.shao@...il.com>
Cc:     Steven Rostedt <rostedt@...dmis.org>,
        Ingo Molnar <mingo@...hat.com>,
        Andrew Morton <akpm@...ux-foundation.org>,
        Vlastimil Babka <vbabka@...e.cz>,
        Michel Lespinasse <walken@...gle.com>,
        Daniel Jordan <daniel.m.jordan@...cle.com>,
        Davidlohr Bueso <dbueso@...e.de>,
        LKML <linux-kernel@...r.kernel.org>,
        Linux MM <linux-mm@...ck.org>
Subject: Re: [PATCH] mmap_lock: add tracepoints around lock acquisition

On Sun, Sep 20, 2020 at 9:58 PM Yafang Shao <laoar.shao@...il.com> wrote:
>
> On Fri, Sep 18, 2020 at 2:13 AM Axel Rasmussen <axelrasmussen@...gle.com> wrote:
> >
> > The goal of these tracepoints is to be able to debug lock contention
> > issues. This lock is acquired on most (all?) mmap / munmap / page fault
> > operations, so a multi-threaded process which does a lot of these can
> > experience significant contention.
> >
> > We trace just before we start acquisition, when the acquisition returns
> > (whether it succeeded or not), and when the lock is released (or
> > downgraded). The events are broken out by lock type (read / write).
> >
> > The events are also broken out by memcg path. For container-based
> > workloads, users often think of several processes in a memcg as a single
> > logical "task", so collecting statistics at this level is useful.
> >
> > These events *do not* include latency bucket information, which means
> > for a proper latency histogram users will need to use BPF instead of
> > event histograms. The benefit we get from this is simpler code.
> >
> > This patch is a no-op if the Kconfig option is not enabled. If it is,
> > tracepoints are still disabled by default (configurable at runtime);
> > the only fixed cost here is un-inlining a few functions. As best as
> > I've been able to measure, the overhead this introduces is a small
> > fraction of 1%. Actually hooking up the tracepoints to BPF introduces
> > additional overhead, depending on exactly what the BPF program is
> > collecting.
>
> Are there any methods to avoid un-inlining these wrappers ?
>
> For example,
> // include/linux/mmap_lock.h
>
> void mmap_lock_start_trace_wrapper();
> void mmap_lock_acquire_trace_wrapper();
>
> static inline void mmap_write_lock(struct mm_struct *mm)
> {
>     mmap_lock_start_trace_wrapper();
>     down_write(&mm->mmap_lock);
>     mmap_lock_acquire_trace_wrapper();
> }
>
> // mm/mmap_lock.c
> void mmap_lock_start_trace_wrapper()
> {
>     trace_mmap_lock_start();
> }
>
> void mmap_lock_start_trace_wrapper()
> {
>     trace_mmap_lock_acquired();
> }

We can do something like that, but I don't think it would end up being better.

At the end of the day, because the trace stuff cannot be in the
header, we have to add an extra function call one way or the other.
This would just move the call one step further down the call stack.
So, I don't think it would affect performance in the
CONFIG_MMAP_LOCK_STATS + tracepoints not enabled at runtime case.

Also the wrappers aren't quite so simple as this, they need some
parameters to work. (the struct mm_struct, whether it was a read or a
write lock, and whether or not the lock operation succeeded), so it
would mean adding more inlined code, which I think adds up to be a
nontrivial amount since these wrappers are called so often in the
kernel.

If you feel strongly, let me know and I can send a version as you
describe and we can compare the two.

>
>
>
> > ---
> >  include/linux/mmap_lock.h        |  28 +++-
> >  include/trace/events/mmap_lock.h |  73 ++++++++++
> >  mm/Kconfig                       |  17 +++
> >  mm/Makefile                      |   1 +
> >  mm/mmap_lock.c                   | 224 +++++++++++++++++++++++++++++++
> >  5 files changed, 342 insertions(+), 1 deletion(-)
> >  create mode 100644 include/trace/events/mmap_lock.h
> >  create mode 100644 mm/mmap_lock.c
> >
> > diff --git a/include/linux/mmap_lock.h b/include/linux/mmap_lock.h
> > index 0707671851a8..d12aa2ff6c05 100644
> > --- a/include/linux/mmap_lock.h
> > +++ b/include/linux/mmap_lock.h
> > @@ -1,11 +1,35 @@
> >  #ifndef _LINUX_MMAP_LOCK_H
> >  #define _LINUX_MMAP_LOCK_H
> >
> > +#include <linux/lockdep.h>
> > +#include <linux/mm_types.h>
> >  #include <linux/mmdebug.h>
> > +#include <linux/rwsem.h>
> > +#include <linux/types.h>
> >
> >  #define MMAP_LOCK_INITIALIZER(name) \
> >         .mmap_lock = __RWSEM_INITIALIZER((name).mmap_lock),
> >
> > +#ifdef CONFIG_MMAP_LOCK_STATS
> > +
> > +void mmap_init_lock(struct mm_struct *mm);
> > +void mmap_write_lock(struct mm_struct *mm);
> > +void mmap_write_lock_nested(struct mm_struct *mm, int subclass);
> > +int mmap_write_lock_killable(struct mm_struct *mm);
> > +bool mmap_write_trylock(struct mm_struct *mm);
> > +void mmap_write_unlock(struct mm_struct *mm);
> > +void mmap_write_downgrade(struct mm_struct *mm);
> > +void mmap_read_lock(struct mm_struct *mm);
> > +int mmap_read_lock_killable(struct mm_struct *mm);
> > +bool mmap_read_trylock(struct mm_struct *mm);
> > +void mmap_read_unlock(struct mm_struct *mm);
> > +bool mmap_read_trylock_non_owner(struct mm_struct *mm);
> > +void mmap_read_unlock_non_owner(struct mm_struct *mm);
> > +void mmap_assert_locked(struct mm_struct *mm);
> > +void mmap_assert_write_locked(struct mm_struct *mm);
> > +
> > +#else /* !CONFIG_MMAP_LOCK_STATS */
> > +
> >  static inline void mmap_init_lock(struct mm_struct *mm)
> >  {
> >         init_rwsem(&mm->mmap_lock);
> > @@ -63,7 +87,7 @@ static inline void mmap_read_unlock(struct mm_struct *mm)
> >
> >  static inline bool mmap_read_trylock_non_owner(struct mm_struct *mm)
> >  {
> > -       if (down_read_trylock(&mm->mmap_lock)) {
> > +       if (mmap_read_trylock(mm)) {
> >                 rwsem_release(&mm->mmap_lock.dep_map, _RET_IP_);
> >                 return true;
> >         }
> > @@ -87,4 +111,6 @@ static inline void mmap_assert_write_locked(struct mm_struct *mm)
> >         VM_BUG_ON_MM(!rwsem_is_locked(&mm->mmap_lock), mm);
> >  }
> >
> > +#endif /* CONFIG_MMAP_LOCK_STATS */
> > +
> >  #endif /* _LINUX_MMAP_LOCK_H */
> > diff --git a/include/trace/events/mmap_lock.h b/include/trace/events/mmap_lock.h
> > new file mode 100644
> > index 000000000000..549c662e6ed8
> > --- /dev/null
> > +++ b/include/trace/events/mmap_lock.h
> > @@ -0,0 +1,73 @@
> > +/* SPDX-License-Identifier: GPL-2.0 */
> > +#undef TRACE_SYSTEM
> > +#define TRACE_SYSTEM mmap_lock
> > +
> > +#if !defined(_TRACE_MMAP_LOCK_H) || defined(TRACE_HEADER_MULTI_READ)
> > +#define _TRACE_MMAP_LOCK_H
> > +
> > +#include <linux/tracepoint.h>
> > +#include <linux/types.h>
> > +
> > +struct mm_struct;
> > +
> > +DECLARE_EVENT_CLASS(
> > +       mmap_lock_template,
> > +
> > +       TP_PROTO(struct mm_struct *mm, const char *memcg_path, u64 duration,
> > +               bool write, bool success),
> > +
> > +       TP_ARGS(mm, memcg_path, duration, write, success),
> > +
> > +       TP_STRUCT__entry(
> > +               __field(struct mm_struct *, mm)
> > +               __string(memcg_path, memcg_path)
> > +               __field(u64, duration)
> > +               __field(bool, write)
> > +               __field(bool, success)
> > +       ),
> > +
> > +       TP_fast_assign(
> > +               __entry->mm = mm;
> > +               __assign_str(memcg_path, memcg_path);
> > +               __entry->duration = duration;
> > +               __entry->write = write;
> > +               __entry->success = success;
> > +       ),
> > +
> > +       TP_printk(
> > +               "mm=%p memcg_path=%s duration=%llu write=%s success=%s\n",
> > +               __entry->mm,
> > +               __get_str(memcg_path),
> > +               __entry->duration,
> > +               __entry->write ? "true" : "false",
> > +               __entry->success ? "true" : "false")
> > +       );
> > +
> > +DEFINE_EVENT(mmap_lock_template, mmap_lock_start_locking,
> > +
> > +       TP_PROTO(struct mm_struct *mm, const char *memcg_path, u64 duration,
> > +               bool write, bool success),
> > +
> > +       TP_ARGS(mm, memcg_path, duration, write, success)
> > +);
> > +
> > +DEFINE_EVENT(mmap_lock_template, mmap_lock_acquire_returned,
> > +
> > +       TP_PROTO(struct mm_struct *mm, const char *memcg_path, u64 duration,
> > +               bool write, bool success),
> > +
> > +       TP_ARGS(mm, memcg_path, duration, write, success)
> > +);
> > +
> > +DEFINE_EVENT(mmap_lock_template, mmap_lock_released,
> > +
> > +       TP_PROTO(struct mm_struct *mm, const char *memcg_path, u64 duration,
> > +               bool write, bool success),
> > +
> > +       TP_ARGS(mm, memcg_path, duration, write, success)
> > +);
> > +
> > +#endif /* _TRACE_MMAP_LOCK_H */
> > +
> > +/* This part must be outside protection */
> > +#include <trace/define_trace.h>
> > diff --git a/mm/Kconfig b/mm/Kconfig
> > index 6c974888f86f..b602df8bcee0 100644
> > --- a/mm/Kconfig
> > +++ b/mm/Kconfig
> > @@ -818,6 +818,23 @@ config DEVICE_PRIVATE
> >  config FRAME_VECTOR
> >         bool
> >
> > +config MMAP_LOCK_STATS
> > +       bool "mmap_lock stats / instrumentation"
> > +       select HISTOGRAM
> > +       default n
> > +
> > +       help
> > +         Enables tracepoints around mmap_lock (start aquiring, acquire
> > +         returned, and released), which are off by default + controlled at
> > +         runtime. These can be used for deeper debugging of contention
> > +         issues, via e.g. BPF.
> > +
> > +         This option has a small (small fraction of 1%) fixed overhead
> > +         even if tracepoints aren't actually in use at runtime, since it
> > +         requires un-inlining some functions.
> > +
> > +         If unsure, say "n".
> > +
> >  config ARCH_USES_HIGH_VMA_FLAGS
> >         bool
> >  config ARCH_HAS_PKEYS
> > diff --git a/mm/Makefile b/mm/Makefile
> > index d5649f1c12c0..eb6ed855a002 100644
> > --- a/mm/Makefile
> > +++ b/mm/Makefile
> > @@ -121,3 +121,4 @@ obj-$(CONFIG_MEMFD_CREATE) += memfd.o
> >  obj-$(CONFIG_MAPPING_DIRTY_HELPERS) += mapping_dirty_helpers.o
> >  obj-$(CONFIG_PTDUMP_CORE) += ptdump.o
> >  obj-$(CONFIG_PAGE_REPORTING) += page_reporting.o
> > +obj-$(CONFIG_MMAP_LOCK_STATS) += mmap_lock.o
> > diff --git a/mm/mmap_lock.c b/mm/mmap_lock.c
> > new file mode 100644
> > index 000000000000..1624f90164c0
> > --- /dev/null
> > +++ b/mm/mmap_lock.c
> > @@ -0,0 +1,224 @@
> > +// SPDX-License-Identifier: GPL-2.0
> > +#define CREATE_TRACE_POINTS
> > +#include <trace/events/mmap_lock.h>
> > +
> > +#include <linux/cgroup.h>
> > +#include <linux/memcontrol.h>
> > +#include <linux/mmap_lock.h>
> > +#include <linux/percpu.h>
> > +#include <linux/smp.h>
> > +#include <linux/trace_events.h>
> > +#include <linux/sched/clock.h>
> > +
> > +#ifdef CONFIG_MEMCG
> > +
> > +DEFINE_PER_CPU(char[MAX_FILTER_STR_VAL], trace_memcg_path);
> > +
> > +/*
> > + * Write the given mm_struct's memcg path to a percpu buffer, and return a
> > + * pointer to it. If the path cannot be determined, the buffer will contain the
> > + * empty string.
> > + *
> > + * Note: buffers are allocated per-cpu to avoid locking, so preemption must be
> > + * disabled by the caller before calling us, and re-enabled only after the
> > + * caller is done with the pointer.
> > + */
> > +static const char *get_mm_memcg_path(struct mm_struct *mm)
> > +{
> > +       struct mem_cgroup *memcg = get_mem_cgroup_from_mm(mm);
> > +
> > +       if (memcg != NULL && likely(memcg->css.cgroup != NULL)) {
> > +               char *buf = this_cpu_ptr(trace_memcg_path);
> > +
> > +               cgroup_path(memcg->css.cgroup, buf, MAX_FILTER_STR_VAL);
> > +               return buf;
> > +       }
> > +       return "";
> > +}
> > +
> > +#define TRACE_MMAP_LOCK_EVENT(type, mm, ...)                                   \
> > +       do {                                                                   \
> > +               if (trace_mmap_lock_##type##_enabled()) {                      \
> > +                       get_cpu();                                             \
> > +                       trace_mmap_lock_##type(mm, get_mm_memcg_path(mm),      \
> > +                                              ##__VA_ARGS__);                 \
> > +                       put_cpu();                                             \
> > +               }                                                              \
> > +       } while (0)
> > +
> > +#else /* !CONFIG_MEMCG */
> > +
> > +#define TRACE_MMAP_LOCK_EVENT(type, mm, ...)                                   \
> > +       trace_mmap_lock_##type(mm, "", ##__VA_ARGS__)
> > +
> > +#endif /* CONFIG_MEMCG */
> > +
> > +/*
> > + * Trace calls must be in a separate file, as otherwise there's a circuclar
> > + * dependency between linux/mmap_lock.h and trace/events/mmap_lock.h.
> > + */
> > +
> > +static void trace_start_locking(struct mm_struct *mm, bool write)
> > +{
> > +       TRACE_MMAP_LOCK_EVENT(start_locking, mm, 0, write, true);
> > +}
> > +
> > +static void trace_acquire_returned(struct mm_struct *mm, u64 start_time_ns,
> > +                                  bool write, bool success)
> > +{
> > +       TRACE_MMAP_LOCK_EVENT(acquire_returned, mm,
> > +                             sched_clock() - start_time_ns, write, success);
> > +}
> > +
> > +static void trace_released(struct mm_struct *mm, bool write)
> > +{
> > +       TRACE_MMAP_LOCK_EVENT(released, mm, 0, write, true);
> > +}
> > +
> > +static bool trylock_impl(struct mm_struct *mm,
> > +                        int (*trylock)(struct rw_semaphore *), bool write)
> > +{
> > +       bool ret;
> > +
> > +       trace_start_locking(mm, write);
> > +       ret = trylock(&mm->mmap_lock) != 0;
> > +       /* Avoid calling sched_clock() for trylocks; assume duration = 0. */
> > +       TRACE_MMAP_LOCK_EVENT(acquire_returned, mm, 0, write, ret);
> > +       return ret;
> > +}
> > +
> > +static inline void lock_impl(struct mm_struct *mm,
> > +                            void (*lock)(struct rw_semaphore *), bool write)
> > +{
> > +       u64 start_time_ns;
> > +
> > +       trace_start_locking(mm, write);
> > +       start_time_ns = sched_clock();
> > +       lock(&mm->mmap_lock);
> > +       trace_acquire_returned(mm, start_time_ns, write, true);
> > +}
> > +
> > +static inline int lock_return_impl(struct mm_struct *mm,
> > +                                  int (*lock)(struct rw_semaphore *),
> > +                                  bool write)
> > +{
> > +       u64 start_time_ns;
> > +       int ret;
> > +
> > +       trace_start_locking(mm, write);
> > +       start_time_ns = sched_clock();
> > +       ret = lock(&mm->mmap_lock);
> > +       trace_acquire_returned(mm, start_time_ns, write, ret == 0);
> > +       return ret;
> > +}
> > +
> > +static inline void unlock_impl(struct mm_struct *mm,
> > +                              void (*unlock)(struct rw_semaphore *),
> > +                              bool write)
> > +{
> > +       unlock(&mm->mmap_lock);
> > +       trace_released(mm, write);
> > +}
> > +
> > +void mmap_init_lock(struct mm_struct *mm)
> > +{
> > +       init_rwsem(&mm->mmap_lock);
> > +}
> > +
> > +void mmap_write_lock(struct mm_struct *mm)
> > +{
> > +       lock_impl(mm, down_write, true);
> > +}
> > +EXPORT_SYMBOL(mmap_write_lock);
> > +
> > +void mmap_write_lock_nested(struct mm_struct *mm, int subclass)
> > +{
> > +       u64 start_time_ns;
> > +
> > +       trace_start_locking(mm, true);
> > +       start_time_ns = sched_clock();
> > +       down_write_nested(&mm->mmap_lock, subclass);
> > +       trace_acquire_returned(mm, start_time_ns, true, true);
> > +}
> > +EXPORT_SYMBOL(mmap_write_lock_nested);
> > +
> > +int mmap_write_lock_killable(struct mm_struct *mm)
> > +{
> > +       return lock_return_impl(mm, down_write_killable, true);
> > +}
> > +EXPORT_SYMBOL(mmap_write_lock_killable);
> > +
> > +bool mmap_write_trylock(struct mm_struct *mm)
> > +{
> > +       return trylock_impl(mm, down_write_trylock, true);
> > +}
> > +EXPORT_SYMBOL(mmap_write_trylock);
> > +
> > +void mmap_write_unlock(struct mm_struct *mm)
> > +{
> > +       unlock_impl(mm, up_write, true);
> > +}
> > +EXPORT_SYMBOL(mmap_write_unlock);
> > +
> > +void mmap_write_downgrade(struct mm_struct *mm)
> > +{
> > +       downgrade_write(&mm->mmap_lock);
> > +       TRACE_MMAP_LOCK_EVENT(acquire_returned, mm, 0, false, true);
> > +}
> > +EXPORT_SYMBOL(mmap_write_downgrade);
> > +
> > +void mmap_read_lock(struct mm_struct *mm)
> > +{
> > +       lock_impl(mm, down_read, false);
> > +}
> > +EXPORT_SYMBOL(mmap_read_lock);
> > +
> > +int mmap_read_lock_killable(struct mm_struct *mm)
> > +{
> > +       return lock_return_impl(mm, down_read_killable, false);
> > +}
> > +EXPORT_SYMBOL(mmap_read_lock_killable);
> > +
> > +bool mmap_read_trylock(struct mm_struct *mm)
> > +{
> > +       return trylock_impl(mm, down_read_trylock, false);
> > +}
> > +EXPORT_SYMBOL(mmap_read_trylock);
> > +
> > +void mmap_read_unlock(struct mm_struct *mm)
> > +{
> > +       unlock_impl(mm, up_read, false);
> > +}
> > +EXPORT_SYMBOL(mmap_read_unlock);
> > +
> > +bool mmap_read_trylock_non_owner(struct mm_struct *mm)
> > +{
> > +       if (mmap_read_trylock(mm)) {
> > +               rwsem_release(&mm->mmap_lock.dep_map, _RET_IP_);
> > +               trace_released(mm, false);
> > +               return true;
> > +       }
> > +       return false;
> > +}
> > +EXPORT_SYMBOL(mmap_read_trylock_non_owner);
> > +
> > +void mmap_read_unlock_non_owner(struct mm_struct *mm)
> > +{
> > +       up_read_non_owner(&mm->mmap_lock);
> > +       trace_released(mm, false);
> > +}
> > +EXPORT_SYMBOL(mmap_read_unlock_non_owner);
> > +
> > +void mmap_assert_locked(struct mm_struct *mm)
> > +{
> > +       lockdep_assert_held(&mm->mmap_lock);
> > +       VM_BUG_ON_MM(!rwsem_is_locked(&mm->mmap_lock), mm);
> > +}
> > +EXPORT_SYMBOL(mmap_assert_locked);
> > +
> > +void mmap_assert_write_locked(struct mm_struct *mm)
> > +{
> > +       lockdep_assert_held_write(&mm->mmap_lock);
> > +       VM_BUG_ON_MM(!rwsem_is_locked(&mm->mmap_lock), mm);
> > +}
> > +EXPORT_SYMBOL(mmap_assert_write_locked);
> > --
> > 2.28.0.618.gf4bc123cb7-goog
> >
>
>
> --
> Thanks
> Yafang

Powered by blists - more mailing lists