Prev: [ANNOUNCE] TCM/ConfigFS v3.4.0-rc1
Next: oprofile: remove double ring buffering
From: Masami Hiramatsu on 20 Mar 2010 00:30
Frederic Weisbecker wrote:
> On Wed, Mar 17, 2010 at 03:14:43PM -0400, Masami Hiramatsu wrote:
>> Mark Wielaard wrote:
>>> On Tue, 2010-03-16 at 18:06 -0400, Masami Hiramatsu wrote:
>>>> Support accessing members in the data structures. With this,
>>>> perf-probe accepts data-structure members(IOW, it now accepts
>>>> dot '.' and arrow '->' operators) as probe arguemnts.
>>>>
>>>> e.g.
>>>>
>>>> ./perf probe --add 'schedule:44 rq->curr'
>>>>
>>>> ./perf probe --add 'vfs_read file->f_op->read file->f_path.dentry'
>>>>
>>>> Note that '>' can be interpreted as redirection in command-line.
>>>
>>> If you find that a problem then you can do like SystemTap does and allow
>>> '.' in place of '->'. In the code you already use the
>>> perf_probe_arg_field ref flag only to check that the DIE gives you the
>>> same information. So you could just drop that and use any separator.
>>> Then you decide based on whether you see a DW_TAG_pointer_type. This
>>> gives the user some extra flexibility by letting them not having to care
>>> about specifying extra type information already available elsewhere.
>>
>> Thanks, when designing this feature, I considered it too.
>>
>> Since perf probe already support displaying source code by --line option,
>> users will read the probed code itself and try to probe it. In that case,
>> I think they naturally use '.' and '->' as they read (they might try to
>> copy & paste it).
>>
>> So, I think that it would be good to support both of '.' and '->' as
>> they are used in the code, because it will not confuse users.
>>
>> Thank you,
>
>
> Agreed.
>
> And lets people use what is common for them: expressions that follow
> C rules in the context.
>
> And those who will be more familiar with perf probe will know they can
> use the simplified "." based scheme.

Hi Frederic, Steven,

BTW, currently perf trace can't parse these 'data structure' type
field yet.

e.g.
execute following commands;

../perf probe vfs_read 'file->f_mode'
../perf record -e probe:vfs_read -afR ls -l
../perf trace
Warning: Error: expected ';' but read '->'
Warning: failed to read event format for vfs_read

Because perf trace expects that the field name should be a
C-variable name. Moreover, if perf probe supports array element,
it becomes more problematic, e.g. 'field:u8 array[8]'! oops!
what does the field contain? an array with 8 elements? or
8th element of 'array' array? :(

Should I suppose that trace event format only accept C-style
variable name for field?


Thank you,

--
Masami Hiramatsu
e-mail: mhiramat(a)redhat.com
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
 | 
Pages: 1
Prev: [ANNOUNCE] TCM/ConfigFS v3.4.0-rc1
Next: oprofile: remove double ring buffering