diff mbox series

[v8,04/12] user_events: Add basic perf and eBPF support

Message ID 20211216173511.10390-5-beaub@linux.microsoft.com (mailing list archive)
State Superseded
Headers show
Series user_events: Enable user processes to create and write to trace events | expand

Commit Message

Beau Belgrave Dec. 16, 2021, 5:35 p.m. UTC 
Adds support to write out user_event data to perf_probe/perf files as
well as to any attached eBPF program.

Signed-off-by: Beau Belgrave <beaub@linux.microsoft.com>
---
 kernel/trace/trace_events_user.c | 72 +++++++++++++++++++++++++++++++-
 1 file changed, 71 insertions(+), 1 deletion(-)

Comments

Masami Hiramatsu (Google) Dec. 22, 2021, 7:55 a.m. UTC | #1 
On Thu, 16 Dec 2021 09:35:03 -0800
Beau Belgrave <beaub@linux.microsoft.com> wrote:

> Adds support to write out user_event data to perf_probe/perf files as
> well as to any attached eBPF program.
> 

Looks good to me.

Reviewed-by: Masami Hiramatsu <mhiramat@kernel.org>

Thanks!

> Signed-off-by: Beau Belgrave <beaub@linux.microsoft.com>
> ---
>  kernel/trace/trace_events_user.c | 72 +++++++++++++++++++++++++++++++-
>  1 file changed, 71 insertions(+), 1 deletion(-)
> 
> diff --git a/kernel/trace/trace_events_user.c b/kernel/trace/trace_events_user.c
> index 9859e62b9838..cc30d1fcbb63 100644
> --- a/kernel/trace/trace_events_user.c
> +++ b/kernel/trace/trace_events_user.c
> @@ -550,6 +550,50 @@ static void user_event_ftrace(struct user_event *user, void *data, u32 datalen,
>  	trace_event_buffer_commit(&event_buffer);
>  }
>  
> +#ifdef CONFIG_PERF_EVENTS
> +/*
> + * Writes the user supplied payload out to perf ring buffer or eBPF program.
> + */
> +static void user_event_perf(struct user_event *user, void *data, u32 datalen,
> +			    void *tpdata)
> +{
> +	struct hlist_head *perf_head;
> +
> +	if (bpf_prog_array_valid(&user->call)) {
> +		struct user_bpf_context context = {0};
> +
> +		context.data_len = datalen;
> +		context.data_type = USER_BPF_DATA_KERNEL;
> +		context.kdata = data;
> +
> +		trace_call_bpf(&user->call, &context);
> +	}
> +
> +	perf_head = this_cpu_ptr(user->call.perf_events);
> +
> +	if (perf_head && !hlist_empty(perf_head)) {
> +		struct trace_entry *perf_entry;
> +		struct pt_regs *regs;
> +		size_t size = sizeof(*perf_entry) + datalen;
> +		int context;
> +
> +		perf_entry = perf_trace_buf_alloc(ALIGN(size, 8),
> +						  &regs, &context);
> +
> +		if (unlikely(!perf_entry))
> +			return;
> +
> +		perf_fetch_caller_regs(regs);
> +
> +		memcpy(perf_entry + 1, data, datalen);
> +
> +		perf_trace_buf_submit(perf_entry, size, context,
> +				      user->call.event.type, 1, regs,
> +				      perf_head, NULL);
> +	}
> +}
> +#endif
> +
>  /*
>   * Update the register page that is shared between user processes.
>   */
> @@ -572,6 +616,10 @@ static void update_reg_page_for(struct user_event *user)
>  
>  				if (probe_func == user_event_ftrace)
>  					status |= EVENT_STATUS_FTRACE;
> +#ifdef CONFIG_PERF_EVENTS
> +				else if (probe_func == user_event_perf)
> +					status |= EVENT_STATUS_PERF;
> +#endif
>  				else
>  					status |= EVENT_STATUS_OTHER;
>  			} while ((++probe_func_ptr)->func);
> @@ -611,8 +659,27 @@ static int user_event_reg(struct trace_event_call *call,
>  					    data);
>  		goto dec;
>  
> -	default:
> +#ifdef CONFIG_PERF_EVENTS
> +	case TRACE_REG_PERF_REGISTER:
> +		ret = tracepoint_probe_register(call->tp,
> +						call->class->perf_probe,
> +						data);
> +		if (!ret)
> +			goto inc;
> +		break;
> +
> +	case TRACE_REG_PERF_UNREGISTER:
> +		tracepoint_probe_unregister(call->tp,
> +					    call->class->perf_probe,
> +					    data);
> +		goto dec;
> +
> +	case TRACE_REG_PERF_OPEN:
> +	case TRACE_REG_PERF_CLOSE:
> +	case TRACE_REG_PERF_ADD:
> +	case TRACE_REG_PERF_DEL:
>  		break;
> +#endif
>  	}
>  
>  	return ret;
> @@ -864,6 +931,9 @@ static int user_event_parse(char *name, char *args, char *flags,
>  	user->class.get_fields = user_event_get_fields;
>  	user->class.reg = user_event_reg;
>  	user->class.probe = user_event_ftrace;
> +#ifdef CONFIG_PERF_EVENTS
> +	user->class.perf_probe = user_event_perf;
> +#endif
>  
>  	mutex_lock(&event_mutex);
>  	ret = user_event_trace_register(user);
> -- 
> 2.17.1
>
diff mbox series

Patch

diff --git a/kernel/trace/trace_events_user.c b/kernel/trace/trace_events_user.c
index 9859e62b9838..cc30d1fcbb63 100644
--- a/kernel/trace/trace_events_user.c
+++ b/kernel/trace/trace_events_user.c
@@ -550,6 +550,50 @@  static void user_event_ftrace(struct user_event *user, void *data, u32 datalen,
 	trace_event_buffer_commit(&event_buffer);
 }
 
+#ifdef CONFIG_PERF_EVENTS
+/*
+ * Writes the user supplied payload out to perf ring buffer or eBPF program.
+ */
+static void user_event_perf(struct user_event *user, void *data, u32 datalen,
+			    void *tpdata)
+{
+	struct hlist_head *perf_head;
+
+	if (bpf_prog_array_valid(&user->call)) {
+		struct user_bpf_context context = {0};
+
+		context.data_len = datalen;
+		context.data_type = USER_BPF_DATA_KERNEL;
+		context.kdata = data;
+
+		trace_call_bpf(&user->call, &context);
+	}
+
+	perf_head = this_cpu_ptr(user->call.perf_events);
+
+	if (perf_head && !hlist_empty(perf_head)) {
+		struct trace_entry *perf_entry;
+		struct pt_regs *regs;
+		size_t size = sizeof(*perf_entry) + datalen;
+		int context;
+
+		perf_entry = perf_trace_buf_alloc(ALIGN(size, 8),
+						  &regs, &context);
+
+		if (unlikely(!perf_entry))
+			return;
+
+		perf_fetch_caller_regs(regs);
+
+		memcpy(perf_entry + 1, data, datalen);
+
+		perf_trace_buf_submit(perf_entry, size, context,
+				      user->call.event.type, 1, regs,
+				      perf_head, NULL);
+	}
+}
+#endif
+
 /*
  * Update the register page that is shared between user processes.
  */
@@ -572,6 +616,10 @@  static void update_reg_page_for(struct user_event *user)
 
 				if (probe_func == user_event_ftrace)
 					status |= EVENT_STATUS_FTRACE;
+#ifdef CONFIG_PERF_EVENTS
+				else if (probe_func == user_event_perf)
+					status |= EVENT_STATUS_PERF;
+#endif
 				else
 					status |= EVENT_STATUS_OTHER;
 			} while ((++probe_func_ptr)->func);
@@ -611,8 +659,27 @@  static int user_event_reg(struct trace_event_call *call,
 					    data);
 		goto dec;
 
-	default:
+#ifdef CONFIG_PERF_EVENTS
+	case TRACE_REG_PERF_REGISTER:
+		ret = tracepoint_probe_register(call->tp,
+						call->class->perf_probe,
+						data);
+		if (!ret)
+			goto inc;
+		break;
+
+	case TRACE_REG_PERF_UNREGISTER:
+		tracepoint_probe_unregister(call->tp,
+					    call->class->perf_probe,
+					    data);
+		goto dec;
+
+	case TRACE_REG_PERF_OPEN:
+	case TRACE_REG_PERF_CLOSE:
+	case TRACE_REG_PERF_ADD:
+	case TRACE_REG_PERF_DEL:
 		break;
+#endif
 	}
 
 	return ret;
@@ -864,6 +931,9 @@  static int user_event_parse(char *name, char *args, char *flags,
 	user->class.get_fields = user_event_get_fields;
 	user->class.reg = user_event_reg;
 	user->class.probe = user_event_ftrace;
+#ifdef CONFIG_PERF_EVENTS
+	user->class.perf_probe = user_event_perf;
+#endif
 
 	mutex_lock(&event_mutex);
 	ret = user_event_trace_register(user);