[V4,01/11] perf/core: Add aux_pause, aux_resume, aux_start_paused

Message ID 20240111081914.3123-2-adrian.hunter@intel.com
State New
Headers
Series perf/core: Add ability for an event to "pause" or "resume" AUX area tracing |

Commit Message

Adrian Hunter Jan. 11, 2024, 8:19 a.m. UTC
  Hardware traces, such as instruction traces, can produce a vast amount of
trace data, so being able to reduce tracing to more specific circumstances
can be useful.

The ability to pause or resume tracing when another event happens, can do
that.

Add ability for an event to "pause" or "resume" AUX area tracing.

Add aux_pause bit to perf_event_attr to indicate that, if the event
happens, the associated AUX area tracing should be paused. Ditto
aux_resume. Do not allow aux_pause and aux_resume to be set together.

Add aux_start_paused bit to perf_event_attr to indicate to an AUX area
event that it should start in a "paused" state.

Add aux_paused to struct perf_event for AUX area events to keep track of
the "paused" state. aux_paused is initialized to aux_start_paused.

Add PERF_EF_PAUSE and PERF_EF_RESUME modes for ->stop() and ->start()
callbacks. Call as needed, during __perf_event_output(). Add
aux_in_pause_resume to struct perf_buffer to prevent races with the NMI
handler. Pause/resume in NMI context will miss out if it coincides with
another pause/resume.

To use aux_pause or aux_resume, an event must be in a group with the AUX
area event as the group leader.

Example (requires Intel PT and tools patches also):

 $ perf record --kcore -e intel_pt/aux-action=start-paused/k,syscalls:sys_enter_newuname/aux-action=resume/,syscalls:sys_exit_newuname/aux-action=pause/ uname
 Linux
 [ perf record: Woken up 1 times to write data ]
 [ perf record: Captured and wrote 0.043 MB perf.data ]
 $ perf script --call-trace
 uname   30805 [000] 24001.058782799: name: 0x7ffc9c1865b0
 uname   30805 [000] 24001.058784424:  psb offs: 0
 uname   30805 [000] 24001.058784424:  cbr: 39 freq: 3904 MHz (139%)
 uname   30805 [000] 24001.058784629: ([kernel.kallsyms])        debug_smp_processor_id
 uname   30805 [000] 24001.058784629: ([kernel.kallsyms])        __x64_sys_newuname
 uname   30805 [000] 24001.058784629: ([kernel.kallsyms])            down_read
 uname   30805 [000] 24001.058784629: ([kernel.kallsyms])                __cond_resched
 uname   30805 [000] 24001.058784629: ([kernel.kallsyms])                preempt_count_add
 uname   30805 [000] 24001.058784629: ([kernel.kallsyms])                    in_lock_functions
 uname   30805 [000] 24001.058784629: ([kernel.kallsyms])                preempt_count_sub
 uname   30805 [000] 24001.058784629: ([kernel.kallsyms])            up_read
 uname   30805 [000] 24001.058784629: ([kernel.kallsyms])                preempt_count_add
 uname   30805 [000] 24001.058784838: ([kernel.kallsyms])                    in_lock_functions
 uname   30805 [000] 24001.058784838: ([kernel.kallsyms])                preempt_count_sub
 uname   30805 [000] 24001.058784838: ([kernel.kallsyms])            _copy_to_user
 uname   30805 [000] 24001.058784838: ([kernel.kallsyms])        syscall_exit_to_user_mode
 uname   30805 [000] 24001.058784838: ([kernel.kallsyms])            syscall_exit_work
 uname   30805 [000] 24001.058784838: ([kernel.kallsyms])                perf_syscall_exit
 uname   30805 [000] 24001.058784838: ([kernel.kallsyms])                    debug_smp_processor_id
 uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                    perf_trace_buf_alloc
 uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                        perf_swevent_get_recursion_context
 uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                            debug_smp_processor_id
 uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                        debug_smp_processor_id
 uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                    perf_tp_event
 uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                        perf_trace_buf_update
 uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                            tracing_gen_ctx_irq_test
 uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                        perf_swevent_event
 uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                            __perf_event_account_interrupt
 uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                                __this_cpu_preempt_check
 uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                            perf_event_output_forward
 uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                                perf_event_aux_pause
 uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                                    ring_buffer_get
 uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                                        __rcu_read_lock
 uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                                        __rcu_read_unlock
 uname   30805 [000] 24001.058785254: ([kernel.kallsyms])                                    pt_event_stop
 uname   30805 [000] 24001.058785254: ([kernel.kallsyms])                                        debug_smp_processor_id
 uname   30805 [000] 24001.058785254: ([kernel.kallsyms])                                        debug_smp_processor_id
 uname   30805 [000] 24001.058785254: ([kernel.kallsyms])                                        native_write_msr
 uname   30805 [000] 24001.058785463: ([kernel.kallsyms])                                        native_write_msr
 uname   30805 [000] 24001.058785639: 0x0

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
---


Changes in V4:
	Rename aux_output_cfg -> aux_action
	Reorder aux_action bits from:
		aux_pause, aux_resume, aux_start_paused
	to:
		aux_start_paused, aux_pause, aux_resume
	Fix aux_action bits __u64 -> __u32


 include/linux/perf_event.h      | 15 +++++++
 include/uapi/linux/perf_event.h | 11 ++++-
 kernel/events/core.c            | 72 +++++++++++++++++++++++++++++++--
 kernel/events/internal.h        |  1 +
 4 files changed, 95 insertions(+), 4 deletions(-)
  

Comments

Namhyung Kim Jan. 19, 2024, 9:40 p.m. UTC | #1
On Thu, Jan 11, 2024 at 12:19 AM Adrian Hunter <adrian.hunter@intelcom> wrote:
>
> Hardware traces, such as instruction traces, can produce a vast amount of
> trace data, so being able to reduce tracing to more specific circumstances
> can be useful.
>
> The ability to pause or resume tracing when another event happens, can do
> that.
>
> Add ability for an event to "pause" or "resume" AUX area tracing.
>
> Add aux_pause bit to perf_event_attr to indicate that, if the event
> happens, the associated AUX area tracing should be paused. Ditto
> aux_resume. Do not allow aux_pause and aux_resume to be set together.
>
> Add aux_start_paused bit to perf_event_attr to indicate to an AUX area
> event that it should start in a "paused" state.
>
> Add aux_paused to struct perf_event for AUX area events to keep track of
> the "paused" state. aux_paused is initialized to aux_start_paused.
>
> Add PERF_EF_PAUSE and PERF_EF_RESUME modes for ->stop() and ->start()
> callbacks. Call as needed, during __perf_event_output(). Add
> aux_in_pause_resume to struct perf_buffer to prevent races with the NMI
> handler. Pause/resume in NMI context will miss out if it coincides with
> another pause/resume.
>
> To use aux_pause or aux_resume, an event must be in a group with the AUX
> area event as the group leader.
>
> Example (requires Intel PT and tools patches also):
>
>  $ perf record --kcore -e intel_pt/aux-action=start-paused/k,syscalls:sys_enter_newuname/aux-action=resume/,syscalls:sys_exit_newuname/aux-action=pause/ uname
>  Linux
>  [ perf record: Woken up 1 times to write data ]
>  [ perf record: Captured and wrote 0.043 MB perf.data ]
>  $ perf script --call-trace
>  uname   30805 [000] 24001.058782799: name: 0x7ffc9c1865b0
>  uname   30805 [000] 24001.058784424:  psb offs: 0
>  uname   30805 [000] 24001.058784424:  cbr: 39 freq: 3904 MHz (139%)
>  uname   30805 [000] 24001.058784629: ([kernel.kallsyms])        debug_smp_processor_id
>  uname   30805 [000] 24001.058784629: ([kernel.kallsyms])        __x64_sys_newuname
>  uname   30805 [000] 24001.058784629: ([kernel.kallsyms])            down_read
>  uname   30805 [000] 24001.058784629: ([kernel.kallsyms])                __cond_resched
>  uname   30805 [000] 24001.058784629: ([kernel.kallsyms])                preempt_count_add
>  uname   30805 [000] 24001.058784629: ([kernel.kallsyms])                    in_lock_functions
>  uname   30805 [000] 24001.058784629: ([kernel.kallsyms])                preempt_count_sub
>  uname   30805 [000] 24001.058784629: ([kernel.kallsyms])            up_read
>  uname   30805 [000] 24001.058784629: ([kernel.kallsyms])                preempt_count_add
>  uname   30805 [000] 24001.058784838: ([kernel.kallsyms])                    in_lock_functions
>  uname   30805 [000] 24001.058784838: ([kernel.kallsyms])                preempt_count_sub
>  uname   30805 [000] 24001.058784838: ([kernel.kallsyms])            _copy_to_user
>  uname   30805 [000] 24001.058784838: ([kernel.kallsyms])        syscall_exit_to_user_mode
>  uname   30805 [000] 24001.058784838: ([kernel.kallsyms])            syscall_exit_work
>  uname   30805 [000] 24001.058784838: ([kernel.kallsyms])                perf_syscall_exit
>  uname   30805 [000] 24001.058784838: ([kernel.kallsyms])                    debug_smp_processor_id
>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                    perf_trace_buf_alloc
>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                        perf_swevent_get_recursion_context
>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                            debug_smp_processor_id
>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                        debug_smp_processor_id
>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                    perf_tp_event
>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                        perf_trace_buf_update
>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                            tracing_gen_ctx_irq_test
>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                        perf_swevent_event
>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                            __perf_event_account_interrupt
>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                                __this_cpu_preempt_check
>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                            perf_event_output_forward
>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                                perf_event_aux_pause
>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                                    ring_buffer_get
>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                                        __rcu_read_lock
>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                                        __rcu_read_unlock
>  uname   30805 [000] 24001.058785254: ([kernel.kallsyms])                                    pt_event_stop
>  uname   30805 [000] 24001.058785254: ([kernel.kallsyms])                                        debug_smp_processor_id
>  uname   30805 [000] 24001.058785254: ([kernel.kallsyms])                                        debug_smp_processor_id
>  uname   30805 [000] 24001.058785254: ([kernel.kallsyms])                                        native_write_msr
>  uname   30805 [000] 24001.058785463: ([kernel.kallsyms])                                        native_write_msr
>  uname   30805 [000] 24001.058785639: 0x0

Looks great!  I think this is very similar to what Kees asked in

  https://lore.kernel.org/linux-perf-users/202401091452.B73E21B6C@keescook/

I have a couple of basic questions:
 * Can we do that for regular events too?
 * What's the difference between start/stop and pause/resume?
   (IOW can we do that just using start/stop callbacks?)

Actually I was thinking about dropping samples using a BPF filter
outside the target scope (e.g. a syscall) but it'd be nice if we can
have builtin support for that.

Thanks,
Namhyung

>
> Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
> ---
>
>
> Changes in V4:
>         Rename aux_output_cfg -> aux_action
>         Reorder aux_action bits from:
>                 aux_pause, aux_resume, aux_start_paused
>         to:
>                 aux_start_paused, aux_pause, aux_resume
>         Fix aux_action bits __u64 -> __u32
>
>
>  include/linux/perf_event.h      | 15 +++++++
>  include/uapi/linux/perf_event.h | 11 ++++-
>  kernel/events/core.c            | 72 +++++++++++++++++++++++++++++++--
>  kernel/events/internal.h        |  1 +
>  4 files changed, 95 insertions(+), 4 deletions(-)
>
> diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
> index 5547ba68e6e4..342879168269 100644
> --- a/include/linux/perf_event.h
> +++ b/include/linux/perf_event.h
> @@ -291,6 +291,7 @@ struct perf_event_pmu_context;
>  #define PERF_PMU_CAP_NO_EXCLUDE                        0x0040
>  #define PERF_PMU_CAP_AUX_OUTPUT                        0x0080
>  #define PERF_PMU_CAP_EXTENDED_HW_TYPE          0x0100
> +#define PERF_PMU_CAP_AUX_PAUSE                 0x0200
>
>  struct perf_output_handle;
>
> @@ -363,6 +364,8 @@ struct pmu {
>  #define PERF_EF_START  0x01            /* start the counter when adding    */
>  #define PERF_EF_RELOAD 0x02            /* reload the counter when starting */
>  #define PERF_EF_UPDATE 0x04            /* update the counter when stopping */
> +#define PERF_EF_PAUSE  0x08            /* AUX area event, pause tracing */
> +#define PERF_EF_RESUME 0x10            /* AUX area event, resume tracing */
>
>         /*
>          * Adds/Removes a counter to/from the PMU, can be done inside a
> @@ -402,6 +405,15 @@ struct pmu {
>          *
>          * ->start() with PERF_EF_RELOAD will reprogram the counter
>          *  value, must be preceded by a ->stop() with PERF_EF_UPDATE.
> +        *
> +        * ->stop() with PERF_EF_PAUSE will stop as simply as possible. Will not
> +        * overlap another ->stop() with PERF_EF_PAUSE nor ->start() with
> +        * PERF_EF_RESUME.
> +        *
> +        * ->start() with PERF_EF_RESUME will start as simply as possible but
> +        * only if the counter is not otherwise stopped. Will not overlap
> +        * another ->start() with PERF_EF_RESUME nor ->stop() with
> +        * PERF_EF_PAUSE.
>          */
>         void (*start)                   (struct perf_event *event, int flags);
>         void (*stop)                    (struct perf_event *event, int flags);
> @@ -798,6 +810,9 @@ struct perf_event {
>         /* for aux_output events */
>         struct perf_event               *aux_event;
>
> +       /* for AUX area events */
> +       unsigned int                    aux_paused;
> +
>         void (*destroy)(struct perf_event *);
>         struct rcu_head                 rcu_head;
>
> diff --git a/include/uapi/linux/perf_event.h b/include/uapi/linux/perf_event.h
> index 39c6a250dd1b..5f6b3b494184 100644
> --- a/include/uapi/linux/perf_event.h
> +++ b/include/uapi/linux/perf_event.h
> @@ -507,7 +507,16 @@ struct perf_event_attr {
>         __u16   sample_max_stack;
>         __u16   __reserved_2;
>         __u32   aux_sample_size;
> -       __u32   __reserved_3;
> +
> +       union {
> +               __u32   aux_action;
> +               struct {
> +                       __u32   aux_start_paused :  1, /* start AUX area tracing paused */
> +                               aux_pause        :  1, /* on overflow, pause AUX area tracing */
> +                               aux_resume       :  1, /* on overflow, resume AUX area tracing */
> +                               __reserved_3     : 29;
> +               };
> +       };
>
>         /*
>          * User provided data if sigtrap=1, passed back to user via
> diff --git a/kernel/events/core.c b/kernel/events/core.c
> index 9efd0d7775e7..dc9ec2443ac9 100644
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -2097,7 +2097,8 @@ static void perf_put_aux_event(struct perf_event *event)
>
>  static bool perf_need_aux_event(struct perf_event *event)
>  {
> -       return !!event->attr.aux_output || !!event->attr.aux_sample_size;
> +       return event->attr.aux_output || event->attr.aux_sample_size ||
> +              event->attr.aux_pause || event->attr.aux_resume;
>  }
>
>  static int perf_get_aux_event(struct perf_event *event,
> @@ -2122,6 +2123,10 @@ static int perf_get_aux_event(struct perf_event *event,
>             !perf_aux_output_match(event, group_leader))
>                 return 0;
>
> +       if ((event->attr.aux_pause || event->attr.aux_resume) &&
> +           !(group_leader->pmu->capabilities & PERF_PMU_CAP_AUX_PAUSE))
> +               return 0;
> +
>         if (event->attr.aux_sample_size && !group_leader->pmu->snapshot_aux)
>                 return 0;
>
> @@ -7846,6 +7851,47 @@ void perf_prepare_header(struct perf_event_header *header,
>         WARN_ON_ONCE(header->size & 7);
>  }
>
> +static void __perf_event_aux_pause(struct perf_event *event, bool pause)
> +{
> +       if (pause) {
> +               if (!READ_ONCE(event->aux_paused)) {
> +                       WRITE_ONCE(event->aux_paused, 1);
> +                       event->pmu->stop(event, PERF_EF_PAUSE);
> +               }
> +       } else {
> +               if (READ_ONCE(event->aux_paused)) {
> +                       WRITE_ONCE(event->aux_paused, 0);
> +                       event->pmu->start(event, PERF_EF_RESUME);
> +               }
> +       }
> +}
> +
> +static void perf_event_aux_pause(struct perf_event *event, bool pause)
> +{
> +       struct perf_buffer *rb;
> +       unsigned long flags;
> +
> +       if (WARN_ON_ONCE(!event))
> +               return;
> +
> +       rb = ring_buffer_get(event);
> +       if (!rb)
> +               return;
> +
> +       local_irq_save(flags);
> +       /* Guard against NMI, NMI loses here */
> +       if (READ_ONCE(rb->aux_in_pause_resume))
> +               goto out_restore;
> +       WRITE_ONCE(rb->aux_in_pause_resume, 1);
> +       barrier();
> +       __perf_event_aux_pause(event, pause);
> +       barrier();
> +       WRITE_ONCE(rb->aux_in_pause_resume, 0);
> +out_restore:
> +       local_irq_restore(flags);
> +       ring_buffer_put(rb);
> +}
> +
>  static __always_inline int
>  __perf_event_output(struct perf_event *event,
>                     struct perf_sample_data *data,
> @@ -7859,6 +7905,9 @@ __perf_event_output(struct perf_event *event,
>         struct perf_event_header header;
>         int err;
>
> +       if (event->attr.aux_pause)
> +               perf_event_aux_pause(event->aux_event, true);
> +
>         /* protect the callchain buffers */
>         rcu_read_lock();
>
> @@ -7875,6 +7924,10 @@ __perf_event_output(struct perf_event *event,
>
>  exit:
>         rcu_read_unlock();
> +
> +       if (event->attr.aux_resume)
> +               perf_event_aux_pause(event->aux_event, false);
> +
>         return err;
>  }
>
> @@ -12014,10 +12067,23 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu,
>         }
>
>         if (event->attr.aux_output &&
> -           !(pmu->capabilities & PERF_PMU_CAP_AUX_OUTPUT)) {
> +           (!(pmu->capabilities & PERF_PMU_CAP_AUX_OUTPUT) ||
> +            event->attr.aux_pause || event->attr.aux_resume)) {
> +               err = -EOPNOTSUPP;
> +               goto err_pmu;
> +       }
> +
> +       if (event->attr.aux_pause && event->attr.aux_resume) {
> +               err = -EINVAL;
> +               goto err_pmu;
> +       }
> +
> +       if (event->attr.aux_start_paused &&
> +           !(pmu->capabilities & PERF_PMU_CAP_AUX_PAUSE)) {
>                 err = -EOPNOTSUPP;
>                 goto err_pmu;
>         }
> +       event->aux_paused = event->attr.aux_start_paused;
>
>         if (cgroup_fd != -1) {
>                 err = perf_cgroup_connect(cgroup_fd, event, attr, group_leader);
> @@ -12814,7 +12880,7 @@ perf_event_create_kernel_counter(struct perf_event_attr *attr, int cpu,
>          * Grouping is not supported for kernel events, neither is 'AUX',
>          * make sure the caller's intentions are adjusted.
>          */
> -       if (attr->aux_output)
> +       if (attr->aux_output || attr->aux_action)
>                 return ERR_PTR(-EINVAL);
>
>         event = perf_event_alloc(attr, cpu, task, NULL, NULL,
> diff --git a/kernel/events/internal.h b/kernel/events/internal.h
> index 5150d5f84c03..3320f78117dc 100644
> --- a/kernel/events/internal.h
> +++ b/kernel/events/internal.h
> @@ -51,6 +51,7 @@ struct perf_buffer {
>         void                            (*free_aux)(void *);
>         refcount_t                      aux_refcount;
>         int                             aux_in_sampling;
> +       int                             aux_in_pause_resume;
>         void                            **aux_pages;
>         void                            *aux_priv;
>
> --
> 2.34.1
>
  
Adrian Hunter Jan. 23, 2024, 10:09 a.m. UTC | #2
On 19/01/24 23:40, Namhyung Kim wrote:
> On Thu, Jan 11, 2024 at 12:19 AM Adrian Hunter <adrian.hunter@intel.com> wrote:
>>
>> Hardware traces, such as instruction traces, can produce a vast amount of
>> trace data, so being able to reduce tracing to more specific circumstances
>> can be useful.
>>
>> The ability to pause or resume tracing when another event happens, can do
>> that.
>>
>> Add ability for an event to "pause" or "resume" AUX area tracing.
>>
>> Add aux_pause bit to perf_event_attr to indicate that, if the event
>> happens, the associated AUX area tracing should be paused. Ditto
>> aux_resume. Do not allow aux_pause and aux_resume to be set together.
>>
>> Add aux_start_paused bit to perf_event_attr to indicate to an AUX area
>> event that it should start in a "paused" state.
>>
>> Add aux_paused to struct perf_event for AUX area events to keep track of
>> the "paused" state. aux_paused is initialized to aux_start_paused.
>>
>> Add PERF_EF_PAUSE and PERF_EF_RESUME modes for ->stop() and ->start()
>> callbacks. Call as needed, during __perf_event_output(). Add
>> aux_in_pause_resume to struct perf_buffer to prevent races with the NMI
>> handler. Pause/resume in NMI context will miss out if it coincides with
>> another pause/resume.
>>
>> To use aux_pause or aux_resume, an event must be in a group with the AUX
>> area event as the group leader.
>>
>> Example (requires Intel PT and tools patches also):
>>
>>  $ perf record --kcore -e intel_pt/aux-action=start-paused/k,syscalls:sys_enter_newuname/aux-action=resume/,syscalls:sys_exit_newuname/aux-action=pause/ uname
>>  Linux
>>  [ perf record: Woken up 1 times to write data ]
>>  [ perf record: Captured and wrote 0.043 MB perf.data ]
>>  $ perf script --call-trace
>>  uname   30805 [000] 24001.058782799: name: 0x7ffc9c1865b0
>>  uname   30805 [000] 24001.058784424:  psb offs: 0
>>  uname   30805 [000] 24001.058784424:  cbr: 39 freq: 3904 MHz (139%)
>>  uname   30805 [000] 24001.058784629: ([kernel.kallsyms])        debug_smp_processor_id
>>  uname   30805 [000] 24001.058784629: ([kernel.kallsyms])        __x64_sys_newuname
>>  uname   30805 [000] 24001.058784629: ([kernel.kallsyms])            down_read
>>  uname   30805 [000] 24001.058784629: ([kernel.kallsyms])                __cond_resched
>>  uname   30805 [000] 24001.058784629: ([kernel.kallsyms])                preempt_count_add
>>  uname   30805 [000] 24001.058784629: ([kernel.kallsyms])                    in_lock_functions
>>  uname   30805 [000] 24001.058784629: ([kernel.kallsyms])                preempt_count_sub
>>  uname   30805 [000] 24001.058784629: ([kernel.kallsyms])            up_read
>>  uname   30805 [000] 24001.058784629: ([kernel.kallsyms])                preempt_count_add
>>  uname   30805 [000] 24001.058784838: ([kernel.kallsyms])                    in_lock_functions
>>  uname   30805 [000] 24001.058784838: ([kernel.kallsyms])                preempt_count_sub
>>  uname   30805 [000] 24001.058784838: ([kernel.kallsyms])            _copy_to_user
>>  uname   30805 [000] 24001.058784838: ([kernel.kallsyms])        syscall_exit_to_user_mode
>>  uname   30805 [000] 24001.058784838: ([kernel.kallsyms])            syscall_exit_work
>>  uname   30805 [000] 24001.058784838: ([kernel.kallsyms])                perf_syscall_exit
>>  uname   30805 [000] 24001.058784838: ([kernel.kallsyms])                    debug_smp_processor_id
>>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                    perf_trace_buf_alloc
>>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                        perf_swevent_get_recursion_context
>>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                            debug_smp_processor_id
>>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                        debug_smp_processor_id
>>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                    perf_tp_event
>>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                        perf_trace_buf_update
>>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                            tracing_gen_ctx_irq_test
>>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                        perf_swevent_event
>>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                            __perf_event_account_interrupt
>>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                                __this_cpu_preempt_check
>>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                            perf_event_output_forward
>>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                                perf_event_aux_pause
>>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                                    ring_buffer_get
>>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                                        __rcu_read_lock
>>  uname   30805 [000] 24001.058785046: ([kernel.kallsyms])                                        __rcu_read_unlock
>>  uname   30805 [000] 24001.058785254: ([kernel.kallsyms])                                    pt_event_stop
>>  uname   30805 [000] 24001.058785254: ([kernel.kallsyms])                                        debug_smp_processor_id
>>  uname   30805 [000] 24001.058785254: ([kernel.kallsyms])                                        debug_smp_processor_id
>>  uname   30805 [000] 24001.058785254: ([kernel.kallsyms])                                        native_write_msr
>>  uname   30805 [000] 24001.058785463: ([kernel.kallsyms])                                        native_write_msr
>>  uname   30805 [000] 24001.058785639: 0x0
> 
> Looks great!  I think this is very similar to what Kees asked in
> 
>   https://lore.kernel.org/linux-perf-users/202401091452.B73E21B6C@keescook/

Sometimes a precisely-defined workload is needed, just so that
running it repeatedly does not produce results that vary too much
to tell whether one software version is better than another.

> 
> I have a couple of basic questions:
>  * Can we do that for regular events too?

That would be much more complicated.  The current implementation
can only pause / resume 1 event, the group leader, and it has to
be supported by the PMU callbacks.

>  * What's the difference between start/stop and pause/resume?
>    (IOW can we do that just using start/stop callbacks?)

It is using start / stop callbacks, albeit with a different mode
parameter.  However pause / resume is not allowed unless the event
has been started and not stopped, so it is a different state.

> 
> Actually I was thinking about dropping samples using a BPF filter
> outside the target scope (e.g. a syscall) but it'd be nice if we can
> have builtin support for that.

In general, I would have thought that capturing samples does not
produce so much data that it cannot be filtered in post-processing.
Looking at the email thread from above, that seems to be what
Arnaldo has proposed.

AUX area tracing is different in this regard.  Intel PT can produce
more trace data than can be written out in time, so data will be
lost for large traces.  Also post-processing takes a long time, so
less data captured helps a lot there also.
  

Patch

diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index 5547ba68e6e4..342879168269 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -291,6 +291,7 @@  struct perf_event_pmu_context;
 #define PERF_PMU_CAP_NO_EXCLUDE			0x0040
 #define PERF_PMU_CAP_AUX_OUTPUT			0x0080
 #define PERF_PMU_CAP_EXTENDED_HW_TYPE		0x0100
+#define PERF_PMU_CAP_AUX_PAUSE			0x0200
 
 struct perf_output_handle;
 
@@ -363,6 +364,8 @@  struct pmu {
 #define PERF_EF_START	0x01		/* start the counter when adding    */
 #define PERF_EF_RELOAD	0x02		/* reload the counter when starting */
 #define PERF_EF_UPDATE	0x04		/* update the counter when stopping */
+#define PERF_EF_PAUSE	0x08		/* AUX area event, pause tracing */
+#define PERF_EF_RESUME	0x10		/* AUX area event, resume tracing */
 
 	/*
 	 * Adds/Removes a counter to/from the PMU, can be done inside a
@@ -402,6 +405,15 @@  struct pmu {
 	 *
 	 * ->start() with PERF_EF_RELOAD will reprogram the counter
 	 *  value, must be preceded by a ->stop() with PERF_EF_UPDATE.
+	 *
+	 * ->stop() with PERF_EF_PAUSE will stop as simply as possible. Will not
+	 * overlap another ->stop() with PERF_EF_PAUSE nor ->start() with
+	 * PERF_EF_RESUME.
+	 *
+	 * ->start() with PERF_EF_RESUME will start as simply as possible but
+	 * only if the counter is not otherwise stopped. Will not overlap
+	 * another ->start() with PERF_EF_RESUME nor ->stop() with
+	 * PERF_EF_PAUSE.
 	 */
 	void (*start)			(struct perf_event *event, int flags);
 	void (*stop)			(struct perf_event *event, int flags);
@@ -798,6 +810,9 @@  struct perf_event {
 	/* for aux_output events */
 	struct perf_event		*aux_event;
 
+	/* for AUX area events */
+	unsigned int			aux_paused;
+
 	void (*destroy)(struct perf_event *);
 	struct rcu_head			rcu_head;
 
diff --git a/include/uapi/linux/perf_event.h b/include/uapi/linux/perf_event.h
index 39c6a250dd1b..5f6b3b494184 100644
--- a/include/uapi/linux/perf_event.h
+++ b/include/uapi/linux/perf_event.h
@@ -507,7 +507,16 @@  struct perf_event_attr {
 	__u16	sample_max_stack;
 	__u16	__reserved_2;
 	__u32	aux_sample_size;
-	__u32	__reserved_3;
+
+	union {
+		__u32	aux_action;
+		struct {
+			__u32	aux_start_paused :  1, /* start AUX area tracing paused */
+				aux_pause        :  1, /* on overflow, pause AUX area tracing */
+				aux_resume       :  1, /* on overflow, resume AUX area tracing */
+				__reserved_3     : 29;
+		};
+	};
 
 	/*
 	 * User provided data if sigtrap=1, passed back to user via
diff --git a/kernel/events/core.c b/kernel/events/core.c
index 9efd0d7775e7..dc9ec2443ac9 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -2097,7 +2097,8 @@  static void perf_put_aux_event(struct perf_event *event)
 
 static bool perf_need_aux_event(struct perf_event *event)
 {
-	return !!event->attr.aux_output || !!event->attr.aux_sample_size;
+	return event->attr.aux_output || event->attr.aux_sample_size ||
+	       event->attr.aux_pause || event->attr.aux_resume;
 }
 
 static int perf_get_aux_event(struct perf_event *event,
@@ -2122,6 +2123,10 @@  static int perf_get_aux_event(struct perf_event *event,
 	    !perf_aux_output_match(event, group_leader))
 		return 0;
 
+	if ((event->attr.aux_pause || event->attr.aux_resume) &&
+	    !(group_leader->pmu->capabilities & PERF_PMU_CAP_AUX_PAUSE))
+		return 0;
+
 	if (event->attr.aux_sample_size && !group_leader->pmu->snapshot_aux)
 		return 0;
 
@@ -7846,6 +7851,47 @@  void perf_prepare_header(struct perf_event_header *header,
 	WARN_ON_ONCE(header->size & 7);
 }
 
+static void __perf_event_aux_pause(struct perf_event *event, bool pause)
+{
+	if (pause) {
+		if (!READ_ONCE(event->aux_paused)) {
+			WRITE_ONCE(event->aux_paused, 1);
+			event->pmu->stop(event, PERF_EF_PAUSE);
+		}
+	} else {
+		if (READ_ONCE(event->aux_paused)) {
+			WRITE_ONCE(event->aux_paused, 0);
+			event->pmu->start(event, PERF_EF_RESUME);
+		}
+	}
+}
+
+static void perf_event_aux_pause(struct perf_event *event, bool pause)
+{
+	struct perf_buffer *rb;
+	unsigned long flags;
+
+	if (WARN_ON_ONCE(!event))
+		return;
+
+	rb = ring_buffer_get(event);
+	if (!rb)
+		return;
+
+	local_irq_save(flags);
+	/* Guard against NMI, NMI loses here */
+	if (READ_ONCE(rb->aux_in_pause_resume))
+		goto out_restore;
+	WRITE_ONCE(rb->aux_in_pause_resume, 1);
+	barrier();
+	__perf_event_aux_pause(event, pause);
+	barrier();
+	WRITE_ONCE(rb->aux_in_pause_resume, 0);
+out_restore:
+	local_irq_restore(flags);
+	ring_buffer_put(rb);
+}
+
 static __always_inline int
 __perf_event_output(struct perf_event *event,
 		    struct perf_sample_data *data,
@@ -7859,6 +7905,9 @@  __perf_event_output(struct perf_event *event,
 	struct perf_event_header header;
 	int err;
 
+	if (event->attr.aux_pause)
+		perf_event_aux_pause(event->aux_event, true);
+
 	/* protect the callchain buffers */
 	rcu_read_lock();
 
@@ -7875,6 +7924,10 @@  __perf_event_output(struct perf_event *event,
 
 exit:
 	rcu_read_unlock();
+
+	if (event->attr.aux_resume)
+		perf_event_aux_pause(event->aux_event, false);
+
 	return err;
 }
 
@@ -12014,10 +12067,23 @@  perf_event_alloc(struct perf_event_attr *attr, int cpu,
 	}
 
 	if (event->attr.aux_output &&
-	    !(pmu->capabilities & PERF_PMU_CAP_AUX_OUTPUT)) {
+	    (!(pmu->capabilities & PERF_PMU_CAP_AUX_OUTPUT) ||
+	     event->attr.aux_pause || event->attr.aux_resume)) {
+		err = -EOPNOTSUPP;
+		goto err_pmu;
+	}
+
+	if (event->attr.aux_pause && event->attr.aux_resume) {
+		err = -EINVAL;
+		goto err_pmu;
+	}
+
+	if (event->attr.aux_start_paused &&
+	    !(pmu->capabilities & PERF_PMU_CAP_AUX_PAUSE)) {
 		err = -EOPNOTSUPP;
 		goto err_pmu;
 	}
+	event->aux_paused = event->attr.aux_start_paused;
 
 	if (cgroup_fd != -1) {
 		err = perf_cgroup_connect(cgroup_fd, event, attr, group_leader);
@@ -12814,7 +12880,7 @@  perf_event_create_kernel_counter(struct perf_event_attr *attr, int cpu,
 	 * Grouping is not supported for kernel events, neither is 'AUX',
 	 * make sure the caller's intentions are adjusted.
 	 */
-	if (attr->aux_output)
+	if (attr->aux_output || attr->aux_action)
 		return ERR_PTR(-EINVAL);
 
 	event = perf_event_alloc(attr, cpu, task, NULL, NULL,
diff --git a/kernel/events/internal.h b/kernel/events/internal.h
index 5150d5f84c03..3320f78117dc 100644
--- a/kernel/events/internal.h
+++ b/kernel/events/internal.h
@@ -51,6 +51,7 @@  struct perf_buffer {
 	void				(*free_aux)(void *);
 	refcount_t			aux_refcount;
 	int				aux_in_sampling;
+	int				aux_in_pause_resume;
 	void				**aux_pages;
 	void				*aux_priv;