Re: [RFC PATCH 1/7] perf/trace: Add notification for perf trace events

[Punit Agrawal]

Punit Agrawal <punit.agrawal@xxxxxxx> writes:

> Add a mechanism to notify listeners about perf trace event state
> changes. This enables listeners to take actions requiring the event
> context (e.g., attached process).
>
> The notification mechanism can be used to reduce trace point based
> profiling overhead by enabling/disabling hardware traps for specific
> contexts (e.g., virtual machines).
>
> Signed-off-by: Punit Agrawal <punit.agrawal@xxxxxxx>
> Cc: Steven Rostedt <rostedt@xxxxxxxxxxx>
> Cc: Ingo Molnar <mingo@xxxxxxxxxx>

While I respin the series addressing comments on the arm64 architectural
bits, I'd appreciate any feedback on this patch as it forms the basis of
the rest of the series.

Thanks,
Punit

> ---
>  include/linux/trace_events.h    |  3 +++
>  kernel/trace/trace_event_perf.c | 24 ++++++++++++++++++++++++
>  2 files changed, 27 insertions(+)
>
> diff --git a/include/linux/trace_events.h b/include/linux/trace_events.h
> index be00761..5924032 100644
> --- a/include/linux/trace_events.h
> +++ b/include/linux/trace_events.h
> @@ -505,6 +505,9 @@ perf_trace_buf_submit(void *raw_data, int size, int rctx, u16 type,
>  {
>  	perf_tp_event(type, count, raw_data, size, regs, head, rctx, task);
>  }
> +
> +extern int perf_trace_notifier_register(struct notifier_block *nb);
> +extern int perf_trace_notifier_unregister(struct notifier_block *nb);
>  #endif
>  
>  #endif /* _LINUX_TRACE_EVENT_H */
> diff --git a/kernel/trace/trace_event_perf.c b/kernel/trace/trace_event_perf.c
> index 562fa69..9aaaacf 100644
> --- a/kernel/trace/trace_event_perf.c
> +++ b/kernel/trace/trace_event_perf.c
> @@ -6,10 +6,12 @@
>   */
>  
>  #include <linux/module.h>
> +#include <linux/notifier.h>
>  #include <linux/kprobes.h>
>  #include "trace.h"
>  
>  static char __percpu *perf_trace_buf[PERF_NR_CONTEXTS];
> +static RAW_NOTIFIER_HEAD(perf_trace_notifier_list);
>  
>  /*
>   * Force it to be aligned to unsigned long to avoid misaligned accesses
> @@ -86,6 +88,26 @@ static int perf_trace_event_perm(struct trace_event_call *tp_event,
>  	return 0;
>  }
>  
> +int perf_trace_notifier_register(struct notifier_block *nb)
> +{
> +	return raw_notifier_chain_register(&perf_trace_notifier_list, nb);
> +}
> +
> +int perf_trace_notifier_unregister(struct notifier_block *nb)
> +{
> +	return raw_notifier_chain_unregister(&perf_trace_notifier_list, nb);
> +}
> +
> +static void perf_trace_notify(enum trace_reg event, struct perf_event *p_event)
> +{
> +	/*
> +	 * We use raw notifiers here as we are called with the
> +	 * event_mutex held.
> +	 */
> +	raw_notifier_call_chain(&perf_trace_notifier_list,
> +				     event, p_event);
> +}
> +
>  static int perf_trace_event_reg(struct trace_event_call *tp_event,
>  				struct perf_event *p_event)
>  {
> @@ -176,6 +198,7 @@ out:
>  static int perf_trace_event_open(struct perf_event *p_event)
>  {
>  	struct trace_event_call *tp_event = p_event->tp_event;
> +	perf_trace_notify(TRACE_REG_PERF_OPEN, p_event);
>  	return tp_event->class->reg(tp_event, TRACE_REG_PERF_OPEN, p_event);
>  }
>  
> @@ -183,6 +206,7 @@ static void perf_trace_event_close(struct perf_event *p_event)
>  {
>  	struct trace_event_call *tp_event = p_event->tp_event;
>  	tp_event->class->reg(tp_event, TRACE_REG_PERF_CLOSE, p_event);
> +	perf_trace_notify(TRACE_REG_PERF_CLOSE, p_event);
>  }
>  
>  static int perf_trace_event_init(struct trace_event_call *tp_event,