HELGRIND TIPS

• Adhere to pthread best practices to minimize error occurrences.
• When crafting your thread functions or alloc functions that recycle a buffer pool, label them using macros from helgrind.h.

HELGRIND

Detect common threading errors in ./program.

valgrind --tool=helgrind --free-is-write=yes ./program

Also detect potential deadlocks. Can be verbose for many programs.

valgrind --tool=helgrind --track-lockorders=yes ./program

HELGRIND ERRORS

Lock was unlocked without first being locked.
Helgrind shows where it originated.

Thread #1 unlocked a not-locked lock at 0x7FEFFFA90
    at 0x4C2408D: pthread_mutex_unlock
    by 0x40073A: nearly_main (tc09_bad_unlock.c:27)
    by 0x40079B: main (tc09_bad_unlock.c:50)
   Lock at 0x7FEFFFA90 was first observed
    at 0x4C25D01: pthread_mutex_init
    by 0x40071F: nearly_main (tc09_bad_unlock.c:23)
    by 0x40079B: main (tc09_bad_unlock.c:50)

Possible race.
Helgrind shows the backtrace for both threads’ accesses.

Thread #1 is the program's root thread

Thread #2 was created
...
Possible data race during read of size 4 at 0x601038 by thread #1
Locks held: none
    at 0x400606: main (simple_race.c:13)

This conflicts with a previous write of size 4 by thread #2
...
Location 0x601038 is 0 bytes inside global var "var" declared at simple_race.c:3

Helgrind also detects pthreads API misuses, as well as races and deadlocks.