Re: [PATCH tracing/kprobes 0/7] tracing/kprobes: kprobe-based event tracer update and perf support

[Frederic Weisbecker <fweisbec@gmail.com>]

On Mon, Sep 14, 2009 at 05:10:33PM -0400, Masami Hiramatsu wrote:
> Frederic Weisbecker wrote:
>> On Mon, Sep 14, 2009 at 01:16:13PM -0400, Masami Hiramatsu wrote:
>>> Frederic Weisbecker wrote:
>>>> On Fri, Sep 11, 2009 at 12:03:30PM -0400, Masami Hiramatsu wrote:
>>>>> Frederic Weisbecker wrote:
>>>>>> May be another step in the todo-list that would be nice: define the format
>>>>>> for a type. Like it's done from ftrace events.
>>>>>
>>>>> Thanks!
>>>>>
>>>>> BTW, I'm not sure what the type means. Each event already has its own
>>>>> event ID and event_call. Could you tell me which part of ftrace I should
>>>>> refer to ?
>>>>>
>>>>
>>>>
>>>> Actually I meant the format for a field.
>>>> Say you define filename=arg1, it would be nice to have
>>>>
>>>> 	print "%s", filename
>>>>
>>>> in the format file.
>>>
>>> Ah, indeed. It is better to support 'type' casting for each argument.
>>> I think type casting can be done as below syntax.
>>>
>>> NAME=ARG:TYPE
>>> e.g.
>>> jiffies64=@jiffies64:u64
>>> message=%ax:str
>>>
>>
>>
>> Yeah looks good!
>>
>>
>>
>>>> Hmm, now that I think about it, we can't dereference an array...for now :-)
>>>
>>> :-)
>>> BTW, currently, an entry of the array can be shown, e.g. +10(sa).
>>> Hmm, for more complex dereference(e.g. accessing a->b[a->c]), it might need
>>> another dereferencing syntax(e.g. "sa[16][sa[8]]"), or
>>> just allow to use braces(e.g. "+(+8(sa))(+16(sa))").
>>
>>
>> Well, that may be too much complexity.
>> I guess if we want multi level dereference, say you want a->b->c
>> it should be sufficient to probe the point where b->c gets it's
>> value (if any).
>>
>> But it would be nice to fetch a range: sa[begin:end]
>> Or at least just giving the length of the range.
>
> Hmm, I think it is better to treat the range as a type, because
> we don't know how to express each entry without the type.
>
> NAME=ARG:TYPE[LEN]
> e.g.
> regs=sa:u64[16]
>


Yeah indeed, and that fits well in the NAME=ARG:TYPE logic
that you have suggested.