Data race in compact_gc
While looking at !1603 (closed) I found the following data race triggered by the compact_gc test in the threaded2 way:
==================
WARNING: ThreadSanitizer: data race (pid=29004)
Atomic read of size 2 at 0x7ee5c5700168 by main thread (mutexes: write M16, write M28):
#0 __tsan_atomic16_load <null> (libtsan.so.0+0x000000060eec)
#1 evacuate rts/sm/Evac.c:637 (compact_gc+0x000000411982)
#2 evacuate_hash_entry rts/sm/Scav.c:165 (compact_gc+0x00000087453e)
#3 mapHashTable rts/Hash.c:394 (compact_gc+0x000000845e27)
#4 scavenge_compact rts/sm/Scav.c:182 (compact_gc+0x000000875037)
#5 scavenge_one rts/sm/Scav.c:1551 (compact_gc+0x000000875037)
#6 scavenge_large rts/sm/Scav.c:2008 (compact_gc+0x000000876afa)
#7 scavenge_find_work rts/sm/Scav.c:2068 (compact_gc+0x000000876afa)
#8 scavenge_loop rts/sm/Scav.c:2137 (compact_gc+0x000000876afa)
#9 scavenge_until_all_done rts/sm/GC.c:1112 (compact_gc+0x00000086a4fb)
#10 GarbageCollect rts/sm/GC.c:438 (compact_gc+0x00000086bc31)
#11 scheduleDoGC rts/Schedule.c:1814 (compact_gc+0x00000084db15)
#12 schedule rts/Schedule.c:552 (compact_gc+0x00000084eea7)
#13 scheduleWaitThread rts/Schedule.c:2559 (compact_gc+0x000000851b8d)
#14 rts_evalLazyIO rts/RtsAPI.c:530 (compact_gc+0x0000008991d1)
#15 hs_main rts/RtsMain.c:72 (compact_gc+0x000000849215)
#16 main <null> (compact_gc+0x000000414a41)
Previous write of size 8 at 0x7ee5c5700168 by thread T6:
#0 mmap <null> (libtsan.so.0+0x00000005d920)
#1 my_mmap rts/posix/OSMem.c:240 (compact_gc+0x00000087cae2)
#2 osCommitMemory rts/posix/OSMem.c:599 (compact_gc+0x00000087d345)
#3 getFreshMBlocks rts/sm/MBlock.c:205 (compact_gc+0x00000086fef3)
#4 getCommittedMBlocks rts/sm/MBlock.c:216 (compact_gc+0x00000086fef3)
#5 getMBlocks rts/sm/MBlock.c:580 (compact_gc+0x000000870218)
#6 alloc_mega_group rts/sm/BlockAlloc.c:378 (compact_gc+0x000000868be1)
#7 allocGroupOnNode rts/sm/BlockAlloc.c:429 (compact_gc+0x00000086970d)
#8 allocLargeChunkOnNode rts/sm/BlockAlloc.c:506 (compact_gc+0x000000869e32)
#9 allocBlocks_sync rts/sm/GCUtils.c:59 (compact_gc+0x00000086f8ee)
#10 alloc_todo_block rts/sm/GCUtils.c:329 (compact_gc+0x00000086f8ee)
#11 todo_block_full rts/sm/GCUtils.c:298 (compact_gc+0x00000086fc5d)
#12 alloc_for_copy rts/sm/Evac.c:80 (compact_gc+0x00000040e3f5)
#13 copy_tag_nolock rts/sm/Evac.c:157 (compact_gc+0x00000040e3f5)
#14 evacuate rts/sm/Evac.c:706 (compact_gc+0x00000040e3f5)
#15 scavenge_block rts/sm/Scav.c:496 (compact_gc+0x000000409024)
#16 scavenge_find_work rts/sm/Scav.c:2061 (compact_gc+0x0000008768e7)
#17 scavenge_loop rts/sm/Scav.c:2137 (compact_gc+0x0000008768e7)
#18 scavenge_until_all_done rts/sm/GC.c:1112 (compact_gc+0x00000086a4fb)
#19 gcWorkerThread rts/sm/GC.c:1185 (compact_gc+0x00000086e368)
#20 yieldCapability rts/Capability.c:904 (compact_gc+0x000000843a65)
#21 scheduleYield rts/Schedule.c:681 (compact_gc+0x00000084f452)
#22 schedule rts/Schedule.c:295 (compact_gc+0x00000084f452)
#23 scheduleWorker rts/Schedule.c:2576 (compact_gc+0x000000851bfe)
#24 workerStart rts/Task.c:445 (compact_gc+0x000000858815)
#25 <null> <null> (libtsan.so.0+0x000000028d5b)
Mutex M16 (0x000000a530c0) created at:
#0 pthread_mutex_init <null> (libtsan.so.0+0x00000002c81e)
#1 initMutex rts/posix/OSThreads.c:170 (compact_gc+0x00000087d777)
#2 initStorage rts/sm/Storage.c:148 (compact_gc+0x00000087791b)
#3 hs_init_ghc rts/RtsStartup.c:245 (compact_gc+0x00000084a040)
#4 hs_main rts/RtsMain.c:57 (compact_gc+0x0000008491eb)
#5 main <null> (compact_gc+0x000000414a41)
Mutex M28 (0x000000a52320) created at:
#0 pthread_mutex_init <null> (libtsan.so.0+0x00000002c81e)
#1 initMutex rts/posix/OSThreads.c:170 (compact_gc+0x00000087d777)
#2 initStablePtrTable rts/StablePtr.c:162 (compact_gc+0x0000008538c1)
#3 initStablePtrTable rts/StablePtr.c:155 (compact_gc+0x0000008539b4)
#4 hs_init_ghc rts/RtsStartup.c:248 (compact_gc+0x00000084a045)
#5 hs_main rts/RtsMain.c:57 (compact_gc+0x0000008491eb)
#6 main <null> (compact_gc+0x000000414a41)
Thread T6 (tid=29011, running) created by thread T4 at:
#0 pthread_create <null> (libtsan.so.0+0x00000002c010)
#1 createOSThread rts/posix/OSThreads.c:137 (compact_gc+0x00000087d6ef)
#2 startWorkerTask rts/Task.c:497 (compact_gc+0x00000085922a)
#3 releaseCapability_ rts/Capability.c:567 (compact_gc+0x000000843197)
#4 suspendThread rts/Schedule.c:2424 (compact_gc+0x0000008513e4)
#5 <null> <null> (compact_gc+0x0000007ad139)
#6 scheduleWorker rts/Schedule.c:2576 (compact_gc+0x000000851bfe)
#7 workerStart rts/Task.c:445 (compact_gc+0x000000858815)
#8 <null> <null> (libtsan.so.0+0x000000028d5b)
SUMMARY: ThreadSanitizer: data race (/nix/store/c7hj2bk4aqgpb3q0h5xhq7lag0lq3jm7-gcc-7.4.0-lib/lib/libtsan.so.0+0x60eec) in __tsan_atomic16_loadThe load in question in evacuate is:
StgClosure *e = (StgClosure*)UN_FORWARDING_PTR(info);
RELAXED_STORE(p, TAG_CLOSURE(tag,e));
if (gen_no < gct->evac_gen_no) { // optimisation
if (RELAXED_LOAD(&Bdescr((P_)e)->gen_no) // <===== this, I think
< gct->evac_gen_no) {
gct->failed_to_evac = true;
TICK_GC_FAILED_PROMOTION();
}
}
return; Given that I only see this in compact_gc, I suspect this is a bug in the CNF compactor which fails to initialize block descriptors correctly.
Edited by Ben Gamari