Re: [PATCH 2/5] tracing_bprintk: don't increment @pos in t_start()

[Li Zefan <lizf@cn.fujitsu.com>]

Wang Liming wrote:
> Li Zefan wrote:
>> Wang Liming wrote:
>>> Hi,
>>>
>>> Li Zefan wrote:
>>>> It's wrong to increment @pos in t_start(), otherwise we'll lose
>>>> some entries when reading printk_formats, if the output is large
>>>> than PAGE_SIZE.
>>>>
>>>> [ Impact: fix missing entries when reading printk_formats ]
>>>>
>>>> Reported-by: Lai Jiangshan <laijs@cn.fujitsu.com>
>>>> Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
>>>> ---
>>>>  kernel/trace/trace_printk.c |   26 ++++++--------------------
>>>>  1 files changed, 6 insertions(+), 20 deletions(-)
>>>>
>>>> diff --git a/kernel/trace/trace_printk.c b/kernel/trace/trace_printk.c
>>>> index 9bece96..7b62781 100644
>>>> --- a/kernel/trace/trace_printk.c
>>>> +++ b/kernel/trace/trace_printk.c
>>>> @@ -155,25 +155,19 @@ int __ftrace_vprintk(unsigned long ip, const
>>>> char *fmt, va_list ap)
>>>>  EXPORT_SYMBOL_GPL(__ftrace_vprintk);
>>>>  
>>>>  static void *
>>>> -t_next(struct seq_file *m, void *v, loff_t *pos)
>>>> +t_start(struct seq_file *m, loff_t *pos)
>>>>  {
>>>> -    const char **fmt = m->private;
>>>> -    const char **next = fmt;
>>>> -
>>>> -    (*pos)++;
>>>> +    const char **fmt = __start___trace_bprintk_fmt + *pos;
>>>>  
>>>>      if ((unsigned long)fmt >= (unsigned
>>>> long)__stop___trace_bprintk_fmt)
>>>>          return NULL;
>>>> -
>>>> -    next = fmt;
>>>> -    m->private = ++next;
>>>> -
>>>>      return fmt;
>>>>  }
>>>>  
>>>> -static void *t_start(struct seq_file *m, loff_t *pos)
>>>> +static void *t_next(struct seq_file *m, void * v, loff_t *pos)
>>>>  {
>>>> -    return t_next(m, NULL, pos);
>>>> +    (*pos)++;
>>>> +    return t_start(m, pos);
>>>>  }
>>>>  
>>> I prefer to .start to call .next, so I rewrite it to following:
>>>
>>
>> Thanks for the comment, but I don't think .next calls .start is bad,
>> and I'm not the only one doing this. Grep c_start() to see some of
>> them.
> Yes, you are not the only one, but it's the only one in the tracing
> code. :)
> I just think we should make the seq_* uniform so that we can understand
> them more clearly.
> 

I don't see how this make seq_* un-uniform..

And I don't want to add extra checking for this kind of uniform.

And if some next()s check (v == NULL) while others don't, do
you think it's uniform or not?