lkml.org 
[lkml]   [2012]   [Jan]   [13]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v8 3.2.0-rc5 9/9] perf: perf interface for uprobes
(2012/01/13 14:14), Srikar Dronamraju wrote:
>>>>> +#define DEFAULT_FUNC_FILTER "!_*"
>>>>
>>>> This is a hidden rule for users ... please remove it.
>>>> (or, is there any reason why we need to have it?)
>>>>
>>>
>>> This is to be in sync with your commit
>>> 3c42258c9a4db70133fa6946a275b62a16792bb5
>>
>> I see, but that commit also provides filter option for changing
>> the function filter. Here, user can not change the filter rule.
>>
>> I think, currently, we don't need to filter any function by name
>> here, since the user obviously intends to probe given function :)
>
> Actually this was discussed in LKML here
> https://lkml.org/lkml/2010/7/20/5, please refer the sub-thread.
>
> Basically without this filter, the list of functions is too large
> including labels, weak, and local binding function which arent traced.

If you mean that this function is used for listing
function (perf probe -F), that's true. But it seems
this convert_name_to_addr() is used just for converting
given function.

As far as I can understand, this means that the user
specifies an actual and single function for the probe point.

If so, there is no need to list up all functions - just
find a function which has the given symbol. I guess, it
is enough to set given function name to
available_func_filter as below. :)

available_func_filter = function

then, map__load() loads only the function which has the
given function name, doesn't it? :)

>>>
>>> If the user provides a symbolic link, convert_name_to_addr would get the
>>> target executable for the given executable. This would handy if we were
>>> to compare existing probes registered on the same application using a
>>> different name (symbolic links). Since you seem to like that we register
>>> with the name the user has provided, I will just feed address here.
>>
>> Hmm, why do we need to compare the probe points? Of course, event-name
>> conflict should be solved, but I think it is acceptable that user puts
>> several probes on the same exec:vaddr. Since different users may want
>> to use it concurrently bit different ways.
>>
>
> The event-names themselves are generated from the probe points. There is
> no problem as such if two or more people use a different symlinks to
> create probes. I was just trying to see if we could solve the
> inconsitency where we warn a person if he is trying to place a probe on
> a existing probe but allow the same if he is trying to place a probe on
> a existing probe using a different symlink.
>
> This again I have changed as you suggested in the latest patches that I
> sent this week.

Yeah, I've checked out it. Thanks:)


Thank you,

--
Masami HIRAMATSU
Software Platform Research Dept. Linux Technology Center
Hitachi, Ltd., Yokohama Research Laboratory
E-mail: masami.hiramatsu.pt@hitachi.com


\
 
 \ /
  Last update: 2012-01-13 15:05    [W:0.101 / U:0.060 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site