libbpf-tools: add futexctn

[ethercflow]

Although systemtap already has a futexes.stp tool, it doesn't feel very easy to use. For example, there are only maximum, minimum and average values, but no delay distribution information. In addition, although the mutex address is behind, there is no call stack information, which is inconvenient to check when you cannot use gdb to attach the process.

This tool identifies lock contention information by filtering futex FUTEX_WAIT operations and retrieves the call stack information when entering lock contention. The lock contention counter is partitioned based on the call stack, enabling users to locate all path information involved in this lock contention using the printed lock address, which serves as the basis for optimization.

➜ ✗ sudo ./futexctn -p 381676 1 1 -mT

15:07:13


fmu[381694] lock 0x55bb40a44060 contended 61 times, 16 avg msecs [max: 21 msecs, min 16 msecs]
    -
    __lll_lock_wait_private
    bar
    foo
    thread_func
    pthread_condattr_setpshared
    -
     msecs               : count    distribution
         0 -> 1          : 0        |                                        |
         2 -> 3          : 0        |                                        |
         4 -> 7          : 0        |                                        |
         8 -> 15         : 0        |                                        |
        16 -> 31         : 61       |****************************************|


fmu[381689] lock 0x55bb40a44060 contended 61 times, 16 avg msecs [max: 21 msecs, min 16 msecs]
    -
    __lll_lock_wait_private
    bar
    foo
    thread_func
    pthread_condattr_setpshared
    -
     msecs               : count    distribution
         0 -> 1          : 0        |                                        |
         2 -> 3          : 0        |                                        |
         4 -> 7          : 0        |                                        |
         8 -> 15         : 0        |                                        |
        16 -> 31         : 61       |****************************************|

[chenhengqi]

Could you explain how does this tool work, especially the contended counter ?

[ethercflow]

Could you explain how does this tool work, especially the contended counter ?

Done

[chenhengqi]

• Since this tool is working on a single process basis, FUTEX_WAIT_PRIVATE should be used instead of FUTEX_WAIT.

When executing a futex operation that requests to block a thread,
the kernel will block only if the futex word has the value that
the calling thread supplied (as one of the arguments of the
futex() call) as the expected value of the futex word.

• According to the manpage, I guess more work should be done to check whether there is a contention or not.

[ethercflow]

Since this tool is working on a single process basis, FUTEX_WAIT_PRIVATE should be used instead of FUTEX_WAIT.

@chenhengqi Glibc (pthread) only use FUTEX_WAIT, so we need to add comments to notice this?

[ethercflow]

Hi @chenhengqi @yonghong-song, do you have any other suggestions for modification?

[yonghong-song]

Similar to other tools, please guard your usage of blazesym with macro USE_BLAZESYM as it may not be available in all systems.

[ethercflow]

Similar to other tools, please guard your usage of blazesym with macro USE_BLAZESYM as it may not be available in all systems.

@yonghong-song Fixed, thanks!

[ethercflow]

Hi @yonghong-song @chenhengqi , sorry to bother you. May I ask if this PR be merged? Is there anything else I need to do?