[lttng-dev] Problem in Tracing C Program

[Julien Desfossez]

On 2017-11-30 12:51 PM, MMM wrote:
> Dear Julien,
> 
> Thanks for your response. Is the mentioned task easier in Java or C program?
> indeed, I think it can be easier in Java program. If it is true, I can
> switch to Java language.

No, your test program should be in C because it will be easy to link the
open()/read() calls from your C program to the open/read system calls.

Julien

> 
> On Thursday, November 30, 2017, 7:15:21 PM GMT+3:30, Julien Desfossez
> <jdesfossez at efficios.com> wrote:
> 
> 
> On 2017-11-30 10:13 AM, MMM via lttng-dev wrote:
> 
>> Dear Users,
>>
>> I have posted a couple of email about my problem in the past, while
>> nobody response those. It may be true that my question is amateurish,
>> but due to the fact that I am a beginner, I can not find the sufficient
>> document for solving my problem. I am going to trace a C program that
>> read from a file and show the ratio of reads from page cache vs reads
>> from disk. I read document of LTTng website, but I can not find
>> appropriate explanation for this task. I just need a simple guidance to
>> show me the right way for doing it. Could you please help me.
>>
>> Regards,
>> Mehdi.
> 
> 
> Hi,
> 
> This is a really non-trivial task to do. Basically, you need to
> distinguish the read() syscalls that lead to FS/disk access with the
> ones that don't (for the same FD).
> The main difficulty is to make the link between the syscall and the
> lower-level events and I am not sure we have all the data needed to
> establish this link from the payload of the events we have (something
> like FD -> FS inode -> block).
> 
> I added Houssem to this thread because I think I remember he did some
> work around that, he may have more information.
> 
> Julien