Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20200618222225.102337-1-axelrasmussen@google.com> <20200618222225.102337-2-axelrasmussen@google.com>
In-Reply-To: <20200618222225.102337-2-axelrasmussen@google.com>
From:   Yafang Shao <laoar.shao@gmail.com>
Date:   Tue, 8 Sep 2020 19:41:08 +0800
Message-ID: <CALOAHbA00V8H5jSnaUG_eJ47NptDaz_JLQVOpdGO6yiUf4_fuw@mail.gmail.com>
Subject: Re: [RFC PATCH v3 1/1] mmap_lock: add tracepoints around mmap_lock acquisition
To:     Axel Rasmussen <axelrasmussen@google.com>
Cc:     Michel Lespinasse <walken@google.com>,
        Peter Zijlstra <peterz@infradead.org>,
        David Howells <dhowells@redhat.com>,
        LKML <linux-kernel@vger.kernel.org>,
        Linux MM <linux-mm@kvack.org>,
        Jonathan Adams <jwadams@google.com>,
        David Rientjes <rientjes@google.com>,
        Ying Han <yinghan@google.com>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Fri, Jun 19, 2020 at 6:43 AM Axel Rasmussen <axelrasmussen@google.com> wrote:
>
> The goal is to be able to collect a latency histogram for contended
> mmap_lock acquisitions. This will be used to diagnose slowness observed
> in production workloads, as well as to measure the effect of upcoming
> mmap_lock optimizations like maple trees and range-based locks. The
> "start_locking" and "lock_released" tracepoints can be used to find out
> which process / call stack was waiting on / holding the lock.
>
> The use of tracepoints with histogram triggers for this purpose was
> recommended here: https://marc.info/?l=linux-mm&m=159076475528369&w=2
>
> A new Kconfig is added, because this change requires uninlining some
> functions, adding branches, etc., even if the tracepoints aren't enabled
> at runtime.
>
> The semantics of the "acquire_returned" tracepoint are as follows:
>
> - We leverage a new "contended hook" rwsem API so "_trylock" variants
>   are called first, and only in the contended case is the trace event
>   triggered. This eliminates overhead for the "common" uncontended case.
>
> - "acquire_returned" is triggered even if acquisition fails (e.g. in the
>   case of killable acquires). This is so we can see how long we waited,
>   even if we didn't get the lock in the end. Whether or not the
>   acquisition succeeded is an event parameter.
>
> - The path of the memcg to which the mm_struct belongs is reported in
>   the event, so we can have per-memcg historams. But, note that the
>   stats are *not* hierarchical; if consumers want a histogram for a
>   memcg *and all of its children*, the consumer will need to add up the
>   separate per-memcg stats itself. This is to avoid paying the cost of
>   traversing a series of memcgs *at record time* (more overhead).
>
> - Bucketing based upon the duration (bucket_{lower,upper}) is done in
>   the kernel, as histogram triggers don't have an API to configure such
>   a thing at runtime.
>
> Signed-off-by: Axel Rasmussen <axelrasmussen@google.com>
> ---
>
> Changes since v2:
> - Switched to TRACE_EVENT hist triggers instead of a new hist library.
> - Reduced inlining, as it increased code size with no performance gain.
> - Stopped instrumenting the uncontended case, to reduce overhead. Added
>   additional rwsem.h API surface in order to be able to do this.
>
>  include/linux/lockdep.h          |  47 ++++++
>  include/linux/mmap_lock.h        |  27 ++-
>  include/linux/rwsem.h            |  12 ++
>  include/trace/events/mmap_lock.h |  76 +++++++++
>  kernel/locking/rwsem.c           |  64 +++++++
>  mm/Kconfig                       |  19 +++
>  mm/Makefile                      |   1 +
>  mm/mmap_lock.c                   | 281 +++++++++++++++++++++++++++++++
>  8 files changed, 526 insertions(+), 1 deletion(-)
>  create mode 100644 include/trace/events/mmap_lock.h
>  create mode 100644 mm/mmap_lock.c
>
> diff --git a/include/linux/lockdep.h b/include/linux/lockdep.h
> index 8fce5c98a4b0..6de91046437c 100644
> --- a/include/linux/lockdep.h
> +++ b/include/linux/lockdep.h
> @@ -590,6 +590,17 @@ do {                                                               \
>         lock_acquired(&(_lock)->dep_map, _RET_IP_);                     \
>  } while (0)
>
> +#define LOCK_CONTENDED_HOOK(_lock, try, lock, pre, post, arg)  \
> +do {                                                           \
> +       if (!try(_lock)) {                                      \
> +               lock_contended(&(_lock)->dep_map, _RET_IP_);    \
> +               pre(arg);                                       \
> +               lock(_lock);                                    \
> +               post(arg);                                      \
> +       }                                                       \
> +       lock_acquired(&(_lock)->dep_map, _RET_IP_);             \
> +} while (0)
> +
>  #define LOCK_CONTENDED_RETURN(_lock, try, lock)                        \
>  ({                                                             \
>         int ____err = 0;                                        \
> @@ -602,6 +613,21 @@ do {                                                               \
>         ____err;                                                \
>  })
>
> +#define LOCK_CONTENDED_HOOK_RETURN(                            \
> +       _lock, try, lock, pre, post, arg)                       \
> +({                                                             \
> +       int ____err = 0;                                        \
> +       if (!try(_lock)) {                                      \
> +               lock_contended(&(_lock)->dep_map, _RET_IP_);    \
> +               pre(arg);                                       \
> +               ____err = lock(_lock);                          \
> +               post(arg, ____err);                             \
> +       }                                                       \
> +       if (!____err)                                           \
> +               lock_acquired(&(_lock)->dep_map, _RET_IP_);     \
> +       ____err;                                                \
> +})
> +
>  #else /* CONFIG_LOCK_STAT */
>
>  #define lock_contended(lockdep_map, ip) do {} while (0)
> @@ -610,9 +636,30 @@ do {                                                               \
>  #define LOCK_CONTENDED(_lock, try, lock) \
>         lock(_lock)
>
> +#define LOCK_CONTENDED_HOOK(_lock, try, lock, pre, post, arg)  \
> +do {                                                           \
> +       if (!try(_lock)) {                                      \
> +               pre(arg);                                       \
> +               lock(_lock);                                    \
> +               post(arg);                                      \
> +       }                                                       \
> +} while (0)
> +
>  #define LOCK_CONTENDED_RETURN(_lock, try, lock) \
>         lock(_lock)
>
> +#define LOCK_CONTENDED_HOOK_RETURN(                            \
> +       _lock, try, lock, pre, post, arg)                       \
> +({                                                             \
> +       int ____err = 0;                                        \
> +       if (!try(_lock)) {                                      \
> +               pre(arg);                                       \
> +               ____err = lock(_lock);                          \
> +               post(arg, ____err);                             \
> +       }                                                       \
> +       ____err;                                                \
> +})
> +
>  #endif /* CONFIG_LOCK_STAT */
>
>  #ifdef CONFIG_LOCKDEP
> diff --git a/include/linux/mmap_lock.h b/include/linux/mmap_lock.h
> index 0707671851a8..1f61f1eac319 100644
> --- a/include/linux/mmap_lock.h
> +++ b/include/linux/mmap_lock.h
> @@ -1,11 +1,34 @@
>  #ifndef _LINUX_MMAP_LOCK_H
>  #define _LINUX_MMAP_LOCK_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_TRACEPOINTS
> +
> +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_TRACEPOINTS */
> +
>  static inline void mmap_init_lock(struct mm_struct *mm)
>  {
>         init_rwsem(&mm->mmap_lock);
> @@ -63,7 +86,7 @@ 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 +110,6 @@ 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_TRACEPOINTS */
> +
>  #endif /* _LINUX_MMAP_LOCK_H */
> diff --git a/include/linux/rwsem.h b/include/linux/rwsem.h
> index 7e5b2a4eb560..3304a35da4c0 100644
> --- a/include/linux/rwsem.h
> +++ b/include/linux/rwsem.h
> @@ -123,7 +123,13 @@ static inline int rwsem_is_contended(struct rw_semaphore *sem)
>   * lock for reading
>   */
>  extern void down_read(struct rw_semaphore *sem);
> +extern void down_read_contended_hook(struct rw_semaphore *sem,
> +                                    void (*pre)(void *), void (*post)(void *),
> +                                    void *arg);
>  extern int __must_check down_read_killable(struct rw_semaphore *sem);
> +extern int __must_check
> +down_read_killable_contended_hook(struct rw_semaphore *sem, void (*pre)(void *),
> +                                 void (*post)(void *, int), void *arg);
>
>  /*
>   * trylock for reading -- returns 1 if successful, 0 if contention
> @@ -134,7 +140,13 @@ extern int down_read_trylock(struct rw_semaphore *sem);
>   * lock for writing
>   */
>  extern void down_write(struct rw_semaphore *sem);
> +extern void down_write_contended_hook(struct rw_semaphore *sem,
> +                                     void (*pre)(void *), void (*post)(void *),
> +                                     void *arg);
>  extern int __must_check down_write_killable(struct rw_semaphore *sem);
> +extern int __must_check down_write_killable_contended_hook(
> +       struct rw_semaphore *sem, void (*pre)(void *),
> +       void (*post)(void *, int), void *arg);
>
>  /*
>   * trylock for writing -- returns 1 if successful, 0 if contention
> diff --git a/include/trace/events/mmap_lock.h b/include/trace/events/mmap_lock.h
> new file mode 100644
> index 000000000000..d1d4c83b848e
> --- /dev/null
> +++ b/include/trace/events/mmap_lock.h
> @@ -0,0 +1,76 @@
> +/* 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 success, u64 bucket_lower, u64 bucket_upper),
> +
> +       TP_ARGS(mm, memcg_path, duration, success, bucket_lower, bucket_upper),
> +
> +       TP_STRUCT__entry(
> +               __field(struct mm_struct *, mm)
> +               __string(memcg_path, memcg_path)
> +               __field(u64, duration)
> +               __field(bool, success)
> +               __field(u64, bucket_lower)
> +               __field(u64, bucket_upper)
> +       ),
> +
> +       TP_fast_assign(
> +               __entry->mm = mm;
> +               __assign_str(memcg_path, memcg_path);
> +               __entry->duration = duration;
> +               __entry->success = success;
> +               __entry->bucket_lower = bucket_lower;
> +               __entry->bucket_upper = bucket_upper;
> +       ),
> +
> +       TP_printk(
> +               "mm=%p memcg_path=%s duration=%llu success=%s bucket=[%llu, %llu]\n",
> +               __entry->mm,
> +               __get_str(memcg_path),
> +               __entry->duration,
> +               __entry->success ? "true" : "false",
> +               __entry->bucket_lower,
> +               __entry->bucket_upper)
> +       );
> +
> +DEFINE_EVENT(mmap_lock_template, mmap_lock_start_locking,
> +
> +       TP_PROTO(struct mm_struct *mm, const char *memcg_path, u64 duration,
> +                bool success, u64 bucket_lower, u64 bucket_upper),
> +
> +       TP_ARGS(mm, memcg_path, duration, success, bucket_lower, bucket_upper)
> +);
> +
> +DEFINE_EVENT(mmap_lock_template, mmap_lock_acquire_returned,
> +
> +       TP_PROTO(struct mm_struct *mm, const char *memcg_path, u64 duration,
> +                bool success, u64 bucket_lower, u64 bucket_upper),
> +
> +       TP_ARGS(mm, memcg_path, duration, success, bucket_lower, bucket_upper)
> +);
> +
> +DEFINE_EVENT(mmap_lock_template, mmap_lock_released,
> +
> +       TP_PROTO(struct mm_struct *mm, const char *memcg_path, u64 duration,
> +                bool success, u64 bucket_lower, u64 bucket_upper),
> +
> +       TP_ARGS(mm, memcg_path, duration, success, bucket_lower, bucket_upper)
> +);
> +
> +#endif /* _TRACE_MMAP_LOCK_H */
> +
> +/* This part must be outside protection */
> +#include <trace/define_trace.h>
> diff --git a/kernel/locking/rwsem.c b/kernel/locking/rwsem.c
> index f11b9bd3431d..6aabea1cbc5d 100644
> --- a/kernel/locking/rwsem.c
> +++ b/kernel/locking/rwsem.c
> @@ -1495,6 +1495,20 @@ void __sched down_read(struct rw_semaphore *sem)
>  }
>  EXPORT_SYMBOL(down_read);
>
> +/*
> + * lock for reading
> + */
> +void __sched down_read_contended_hook(struct rw_semaphore *sem,
> +                                     void (*pre)(void *),
> +                                     void (*post)(void *), void *arg)
> +{
> +       might_sleep();
> +       rwsem_acquire_read(&sem->dep_map, 0, 0, _RET_IP_);
> +       LOCK_CONTENDED_HOOK(sem, __down_read_trylock, __down_read, pre, post,
> +                           arg);
> +}
> +EXPORT_SYMBOL(down_read_contended_hook);
> +
>  int __sched down_read_killable(struct rw_semaphore *sem)
>  {
>         might_sleep();
> @@ -1509,6 +1523,24 @@ int __sched down_read_killable(struct rw_semaphore *sem)
>  }
>  EXPORT_SYMBOL(down_read_killable);
>
> +int __sched down_read_killable_contended_hook(struct rw_semaphore *sem,
> +                                             void (*pre)(void *),
> +                                             void (*post)(void *, int),
> +                                             void *arg)
> +{
> +       might_sleep();
> +       rwsem_acquire_read(&sem->dep_map, 0, 0, _RET_IP_);
> +
> +       if (LOCK_CONTENDED_HOOK_RETURN(sem, __down_read_trylock,
> +                                      __down_read_killable, pre, post, arg)) {
> +               rwsem_release(&sem->dep_map, _RET_IP_);
> +               return -EINTR;
> +       }
> +
> +       return 0;
> +}
> +EXPORT_SYMBOL(down_read_killable_contended_hook);
> +
>  /*
>   * trylock for reading -- returns 1 if successful, 0 if contention
>   */
> @@ -1533,6 +1565,20 @@ void __sched down_write(struct rw_semaphore *sem)
>  }
>  EXPORT_SYMBOL(down_write);
>
> +/*
> + * lock for writing
> + */
> +void __sched down_write_contended_hook(struct rw_semaphore *sem,
> +                                      void (*pre)(void *),
> +                                      void (*post)(void *), void *arg)
> +{
> +       might_sleep();
> +       rwsem_acquire(&sem->dep_map, 0, 0, _RET_IP_);
> +       LOCK_CONTENDED_HOOK(sem, __down_write_trylock, __down_write, pre, post,
> +                           arg);
> +}
> +EXPORT_SYMBOL(down_write_contended_hook);
> +
>  /*
>   * lock for writing
>   */
> @@ -1551,6 +1597,24 @@ int __sched down_write_killable(struct rw_semaphore *sem)
>  }
>  EXPORT_SYMBOL(down_write_killable);
>
> +int __sched down_write_killable_contended_hook(struct rw_semaphore *sem,
> +                                              void (*pre)(void *),
> +                                              void (*post)(void *, int),
> +                                              void *arg)
> +{
> +       might_sleep();
> +       rwsem_acquire_read(&sem->dep_map, 0, 0, _RET_IP_);
> +
> +       if (LOCK_CONTENDED_HOOK_RETURN(sem, __down_write_trylock,
> +                                      __down_write_killable, pre, post, arg)) {
> +               rwsem_release(&sem->dep_map, _RET_IP_);
> +               return -EINTR;
> +       }
> +
> +       return 0;
> +}
> +EXPORT_SYMBOL(down_write_killable_contended_hook);
> +
>  /*
>   * trylock for writing -- returns 1 if successful, 0 if contention
>   */
> diff --git a/mm/Kconfig b/mm/Kconfig
> index f2104cc0d35c..a88b44ff46e2 100644
> --- a/mm/Kconfig
> +++ b/mm/Kconfig
> @@ -822,6 +822,25 @@ config DEVICE_PRIVATE
>  config FRAME_VECTOR
>         bool
>
> +config MMAP_LOCK_TRACEPOINTS
> +       bool "mmap_lock tracepoints"
> +       select HISTOGRAM
> +       select HIST_TRIGGERS
> +       default n
> +
> +       help
> +         Enable kernel tracepoints around mmap_lock acquisition. These can be
> +         enabled / disabled at runtime, although note there is a small amount
> +         of overhead even before the tracepoints are runtime-enabled. When
> +         histogram triggers are configured at runtime, less than 1% overhead
> +         is added to typical workloads.
> +
> +         This option selects CONFIG_HIST_TRIGGERS, since one of the main
> +         purposes of this feature is to allow lock acquisition latency
> +         histograms to be collected.
> +
> +         If unsure, say "N".
> +
>  config ARCH_USES_HIGH_VMA_FLAGS
>         bool
>  config ARCH_HAS_PKEYS
> diff --git a/mm/Makefile b/mm/Makefile
> index 6e9d46b2efc9..14860f7b9984 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_TRACEPOINTS) += mmap_lock.o
> diff --git a/mm/mmap_lock.c b/mm/mmap_lock.c
> new file mode 100644
> index 000000000000..d0734d6325ea
> --- /dev/null
> +++ b/mm/mmap_lock.c
> @@ -0,0 +1,281 @@
> +// SPDX-License-Identifier: GPL-2.0
> +#define CREATE_TRACE_POINTS
> +#include <trace/events/mmap_lock.h>
> +
> +#include <linux/cgroup.h>
> +#include <linux/kernel.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>
> +
> +static const u64 mmap_lock_buckets[] = {
> +       250,            /* 250 ns */
> +       375,            /* 375 ns */
> +       500,            /* 500 ns */
> +       1000,           /* 1 us */
> +       10000,          /* 10 us */
> +       100000,         /* 100 us */
> +       500000,         /* 500 us */
> +       1000000,        /* 1 ms */
> +       5000000,        /* 5 ms */
> +       10000000,       /* 10 ms */
> +       50000000,       /* 50 ms */
> +       100000000,      /* 100 ms */
> +       500000000,      /* 500 ms */
> +       1000000000,     /* 1 s */
> +       5000000000UL,   /* 5 s */
> +       10000000000UL,  /* 10 s */
> +       ~0              /* > 10s */
> +};
> +
> +static int find_bucket(u64 duration)
> +{
> +       int i, lower, upper;
> +
> +       lower = 0;
> +       upper = ARRAY_SIZE(mmap_lock_buckets) - 1;
> +       while (lower < upper) {
> +               /* Can't realistically overflow, number of buckets < 2**63. */
> +               i = (lower + upper) / 2;
> +               if (duration <= mmap_lock_buckets[i])
> +                       upper = i;
> +               else
> +                       lower = i + 1;
> +       }
> +       return upper;
> +}
> +
> +#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 cirtuclar
> + * dependency between linux/mmap_lock.h and trace/events/mmap_lock.h.
> + */
> +
> +static void trace_start_locking(struct mm_struct *mm)
> +{
> +       TRACE_MMAP_LOCK_EVENT(start_locking, mm, 0, true, 0,
> +                             mmap_lock_buckets[0]);
> +}
> +
> +static void trace_acquire_returned(struct mm_struct *mm, u64 start_time_ns,
> +                                  bool success)
> +{
> +       u64 duration = sched_clock() - start_time_ns;
> +       int i = find_bucket(duration);
> +
> +       TRACE_MMAP_LOCK_EVENT(acquire_returned, mm, duration, success,
> +                             i > 0 ? mmap_lock_buckets[i - 1] + 1 : 0,
> +                             mmap_lock_buckets[i]);
> +}
> +
> +static void trace_released(struct mm_struct *mm)
> +{
> +       TRACE_MMAP_LOCK_EVENT(released, mm, 0, true, 0, mmap_lock_buckets[0]);
> +}
> +
> +struct trace_context {
> +       struct mm_struct *mm;
> +       u64 start_time_ns;
> +};
> +
> +static inline void trace_pre(void *c)
> +{
> +       ((struct trace_context *) c)->start_time_ns = sched_clock();
> +}
> +
> +static void trace_post_return(void *c, int ret)
> +{
> +       struct trace_context *context = (struct trace_context *)c;
> +
> +       trace_acquire_returned(context->mm, context->start_time_ns, ret == 0);
> +}
> +
> +static inline void trace_post(void *c)
> +{
> +       trace_post_return(c, 0);
> +}
> +
> +static inline bool trylock_impl(struct mm_struct *mm,
> +                               int (*trylock)(struct rw_semaphore *))
> +{
> +       trace_start_locking(mm);
> +       return trylock(&mm->mmap_lock) != 0;
> +}
> +
> +static inline void lock_impl(struct mm_struct *mm,
> +                            void (*lock)(struct rw_semaphore *,
> +                                         void (*)(void *), void (*)(void *),
> +                                         void *))
> +{
> +       struct trace_context context = { .mm = mm, .start_time_ns = 0 };
> +
> +       trace_start_locking(mm);
> +       lock(&mm->mmap_lock, trace_pre, trace_post, &context);
> +}
> +
> +static inline int lock_return_impl(struct mm_struct *mm,
> +                                  int (*lock)(struct rw_semaphore *,
> +                                              void (*)(void *),
> +                                              void (*)(void *, int), void *))
> +{
> +       struct trace_context context = { .mm = mm, .start_time_ns = 0 };
> +
> +       trace_start_locking(mm);
> +       return lock(&mm->mmap_lock, trace_pre, trace_post_return, &context);
> +}
> +
> +void mmap_init_lock(struct mm_struct *mm)
> +{
> +       init_rwsem(&mm->mmap_lock);
> +}
> +EXPORT_SYMBOL_GPL(mmap_init_lock);
> +
> +void mmap_write_lock(struct mm_struct *mm)
> +{
> +       lock_impl(mm, down_write_contended_hook);
> +}
> +EXPORT_SYMBOL_GPL(mmap_write_lock);
> +
> +void mmap_write_lock_nested(struct mm_struct *mm, int subclass)
> +{
> +       /*
> +        * Unlike other functions, we don't need the contended hook API here.
> +        * This is because we don't care about breaking out the {un,}contended
> +        * cases.
> +        *
> +        * This function is only called once on fork (nowhere else), so 1)
> +        * performance isn't as important, and 2) we expect to never experience
> +        * contention.
> +        */
> +       u64 start_time_ns = sched_clock();
> +
> +       down_write_nested(&mm->mmap_lock, subclass);
> +       trace_acquire_returned(mm, start_time_ns, true);
> +}
> +EXPORT_SYMBOL_GPL(mmap_write_lock_nested);
> +
> +int mmap_write_lock_killable(struct mm_struct *mm)
> +{
> +       return lock_return_impl(mm, down_write_killable_contended_hook);
> +}
> +EXPORT_SYMBOL_GPL(mmap_write_lock_killable);
> +
> +bool mmap_write_trylock(struct mm_struct *mm)
> +{
> +       return trylock_impl(mm, down_write_trylock);
> +}
> +EXPORT_SYMBOL_GPL(mmap_write_trylock);
> +
> +void mmap_write_unlock(struct mm_struct *mm)
> +{
> +       up_write(&mm->mmap_lock);
> +       trace_released(mm);
> +}
> +EXPORT_SYMBOL_GPL(mmap_write_unlock);
> +
> +void mmap_write_downgrade(struct mm_struct *mm)
> +{
> +       downgrade_write(&mm->mmap_lock);
> +}
> +EXPORT_SYMBOL_GPL(mmap_write_downgrade);
> +
> +void mmap_read_lock(struct mm_struct *mm)
> +{
> +       lock_impl(mm, down_read_contended_hook);
> +}
> +EXPORT_SYMBOL_GPL(mmap_read_lock);
> +
> +int mmap_read_lock_killable(struct mm_struct *mm)
> +{
> +       return lock_return_impl(mm, down_read_killable_contended_hook);
> +}
> +EXPORT_SYMBOL_GPL(mmap_read_lock_killable);
> +
> +bool mmap_read_trylock(struct mm_struct *mm)
> +{
> +       return trylock_impl(mm, down_read_trylock);
> +}
> +EXPORT_SYMBOL_GPL(mmap_read_trylock);
> +
> +void mmap_read_unlock(struct mm_struct *mm)
> +{
> +       up_read(&mm->mmap_lock);
> +       trace_released(mm);
> +}
> +EXPORT_SYMBOL_GPL(mmap_read_unlock);
> +
> +bool mmap_read_trylock_non_owner(struct mm_struct *mm)
> +{
> +       if (trylock_impl(mm, down_read_trylock)) {
> +               rwsem_release(&mm->mmap_lock.dep_map, _RET_IP_);
> +               return true;
> +       }
> +       return false;
> +}
> +EXPORT_SYMBOL_GPL(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);
> +}
> +EXPORT_SYMBOL_GPL(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_GPL(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_GPL(mmap_assert_write_locked);
> --
> 2.27.0.111.gc72c7da667-goog
>
>

Hi Axel,

Any updates on this patchset ?

I think adding tracepoints to trace the mmap_lock would be helpful to
us, which can help us identify which one holds the mmap_lock too long
and why it takes too long.


-- 
Thanks
Yafang