Message-ID: <CA+wEVJax6owyYJVfVn+=uoFXsy8qgc_sTHHwJBMU0fiAXDtUfA@mail.gmail.com>
Date: Mon, 30 Jun 2025 16:23:38 +0200
From: Gabriele Paoloni <gpaoloni@...hat.com>
To: rostedt@...dmis.org, mhiramat@...nel.org, mathieu.desnoyers@...icios.com, 
	linux-kernel@...r.kernel.org, linux-trace-kernel@...r.kernel.org
Subject: Re: [RFC PATCH v2] tracing: add testable specifications for event_enable_write/read

On Mon, Jun 30, 2025 at 3:55 PM Gabriele Paoloni <gpaoloni@...hat.com> wrote:
>
> This patch implements the documentation of event_enable_write and
> event_enable_read in the form of testable function's expectations.
>
> Signed-off-by: Gabriele Paoloni <gpaoloni@...hat.com>
> ---
> Following the discussion in v1 (see [1]), v2 adds improvements with
> the "Context" specifications and the definition of the assumptions
> of use for the correct invocation of the documented functions.
>
> [1] https://lore.kernel.org/linux-trace-kernel/20250612104349.5047-1-gpaoloni@redhat.com/
> ---
>  kernel/trace/trace_events.c | 81 +++++++++++++++++++++++++++++++++++++
>  1 file changed, 81 insertions(+)
>
> diff --git a/kernel/trace/trace_events.c b/kernel/trace/trace_events.c
> index 120531268abf..57284f540566 100644
> --- a/kernel/trace/trace_events.c
> +++ b/kernel/trace/trace_events.c
> @@ -1771,6 +1771,52 @@ static void p_stop(struct seq_file *m, void *p)
>         mutex_unlock(&event_mutex);
>  }
>
> +/**
> + * event_enable_read - read from a trace event file to retrieve its status.
> + * @filp: file pointer associated with the target trace event;
> + * @ubuf: user space buffer where the event status is copied to;
> + * @cnt: number of bytes to be copied to the user space buffer;
> + * @ppos: the current position in the buffer.
> + *
> + * This is a way for user space executables to retrieve the status of a
> + * specific event
> + *
> + * Function's expectations:
> + * - This function shall lock the global event_mutex before performing any
> + *   operation on the target event file and unlock it after all operations on
> + *   the target event file have completed;
> + * - This function shall retrieve the status flags from the file associated
> + *   with the target event;
> + * - This function shall format the string to report the event status to user
> + *   space:
> + *   - The first character of the string to be copied to user space shall be
> + *     set to "1" if the enabled flag is set AND the soft_disabled flag is not
> + *     set, else it shall be set to "0";
> + *   - The second character of the string to be copied to user space shall be
> + *     set to "*" if either the soft_disabled flag or the soft_mode flag is
> + *     set, else it shall be set to "\n";
> + *   - The third character of the string to be copied to user space shall b
> + *     set to "\n" if either the soft_disabled flag or the soft_mode flag is
> + *     set, else it shall be set to "0";
> + *   - Any other character in the string to be copied to user space shall be
> + *     set to "0";
> + * - This function shall invoke simple_read_from_buffer() to perform the copy
> + *   of the kernel space string to ubuf.
> + *
> + * Assumptions of Use:
> + * - The caller shall  pass cnt equal or greater than the length of the string
> + *   to be copied to user space;
> + * - Any read operation on a file descriptor, unless it is the first operation
> + *   following a trace event file open, shall be preceded by an lseek
> + *   invocation to reposition the file offset to zero.
> + *
> + * Context: process context, locks and unlocks event_mutex.
> + *
> + * Return: the number of copied bytes on success, -ENODEV if the event file
> + * cannot be retrieved from the input filp, -EINVAL if cnt is less than the
> + * length of the string to be copied to ubuf, any error returned by
> + * simple_read_from_buffer

Please neglect this RFC since the above return value is not correct.
I have fixed the issue in v3.

Apologies for the noise
Gab

> + */
>  static ssize_t
>  event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
>                   loff_t *ppos)
> @@ -1801,6 +1847,41 @@ event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
>         return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
>  }
>
> +/**
> + * event_enable_write - write to a trace event file to enable/disable it.
> + * @filp: file pointer associated with the target trace event;
> + * @ubuf: user space buffer where the enable/disable value is copied from;
> + * @cnt: number of bytes to be copied from the user space buffer;
> + * @ppos: the current position in the buffer.
> + *
> + * This is a way for user space executables to enable or disable event
> + * recording.
> + *
> + * Function's expectations:
> + * - This function shall copy cnt bytes from the input ubuf buffer to a kernel
> + *   space buffer and shall convert the string within the kernel space buffer
> + *   into a decimal base format number;
> + * - This function shall lock the global event_mutex before performing any
> + *   operation on the target event file and unlock it after all operations on
> + *   the target event file have completed;
> + * - This function shall check the size of the per-cpu ring-buffers used for
> + *   the event trace data and, if smaller than TRACE_BUF_SIZE_DEFAULT, expand
> + *   them to TRACE_BUF_SIZE_DEFAULT bytes (sizes larger than
> + *   TRACE_BUF_SIZE_DEFAULT are not allowed);
> + * - This function shall invoke ftrace_event_enable_disable to enable or
> + *   disable the target trace event according to the value read from user space
> + *   (0 - disable, 1 - enable);
> + * - This function shall increase the file position pointed by ppos by the
> + *   number of bytes speicified by cnt;
> + *
> + * Context: process context, locks and unlocks event_mutex.
> + *
> + * Return: the number of written bytes on success, any error returned by
> + * kstrtoul_from_user, -ENODEV if the event file cannot be retrieved from the
> + * input filp, any error returned by tracing_update_buffers, any error returned
> + * by ftrace_event_enable_disable, -EINVAL if the value copied from the user
> + * space ubuf is different from 0 or 1.
> + */
>  static ssize_t
>  event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
>                    loff_t *ppos)
> --
> 2.48.1
>

Powered by blists - more mailing lists