Concerning TSAN failures in STM

@angerman mentioned that his testing of the memory ordering work on the Cardano codebase resulted in some concerning TSAN failures in STM.

Log: $1695

GDB Backtrace: $1696

stm.c: $1699

changed milestone to %9.0.2

added Phigh STM Tbug labels

changed the description

changed the description

@angerman, are you certain you applied all of the memory ordering patches (specifically, all of the branches listed in the "Branch structure" section of the preparatory MR)? Specifically, this race:

  Atomic write of size 8 at 0x558007ab5378 by thread T34:
    #0 __tsan_atomic64_compare_exchange_strong <null> (libtsan.so.0+0x6e8ee)
    #1 cas includes/stg/SMP.h:142 (cardano-node+0x1643058)
    #2 cond_lock_tvar rts/STM.c:313 (cardano-node+0x1643058)
    #3 validate_and_acquire_ownership rts/STM.c:772 (cardano-node+0x1643058)
    #4 stmReWait rts/STM.c:1247 (cardano-node+0x1644a2f)
    #5 stg_atomically_waiting_frame_info <null> (cardano-node+0x1639154)
    #6 scheduleWorker rts/Schedule.c:2573 (cardano-node+0x1614a1f)
    #7 workerStart rts/Task.c:445 (cardano-node+0x1646093)
    #8 <null> <null> (libtsan.so.0+0x2e0b6)

  Previous write of size 8 at 0x558007ab5378 by thread T46:
    #0 unlock_tvar rts/STM.c:303 (cardano-node+0x16440ee)
    #1 stmCommitTransaction rts/STM.c:1105 (cardano-node+0x16440ee)
    #2 stg_atomically_frame_info <null> (cardano-node+0x16390d3)
    #3 scheduleWorker rts/Schedule.c:2573 (cardano-node+0x1614a1f)
    #4 workerStart rts/Task.c:445 (cardano-node+0x1646093)
    #5 <null> <null> (libtsan.so.0+0x2e0b6)

is fixed by !2648 (closed) yet I don't see any of the STM changes present in your tree.

@angerman and I concluded that his backport of the memory ordering work to 8.6 is missing some STM-relevant changes.

closed