Scrub deallocations before reallocation.

[mjp41]

This adds an additional mitigation to scrub allocations before they are reallocated. We should do perf analysis before we decide if this is an always on mitigation.

[mjp41]

So I ran a single pass of mimalloc-bench on this

#------------------------------------------------------------------
# test    alloc   time  rss    user  sys  page-faults page-reclaims
cfrac       sn-m0 06.36 3312 6.35 0.00 0 434
cfrac       sn-m12 06.61 3560 6.61 0.00 0 510
cfrac       sn-m13 06.87 3648 6.86 0.00 0 511
cfrac       sn-chk 06.84 3736 6.84 0.00 0 511
espresso    sn-m0 05.18 6248 5.13 0.02 5 656
espresso    sn-m12 05.26 14708 5.24 0.02 0 722
espresso    sn-m13 05.46 14816 5.45 0.01 0 725
espresso    sn-chk 05.50 14672 5.48 0.01 0 717
barnes      sn-m0 02.87 65552 2.84 0.02 2 2839
barnes      sn-m12 02.85 65868 2.84 0.01 0 2864
barnes      sn-m13 02.88 65872 2.86 0.02 0 2865
barnes      sn-chk 02.86 65676 2.85 0.01 0 2848
redis       sn-m0 4.299 8328 0.20 0.02 18 1273
redis       sn-m12 4.380 11868 0.19 0.03 0 1684
redis       sn-m13 4.519 13680 0.17 0.06 0 2133
redis       sn-chk 4.540 13636 0.18 0.05 0 2129
leanN       sn-m0 25.37 554224 97.48 0.72 166 13964
leanN       sn-m12 26.70 729428 103.30 0.90 0 14205
leanN       sn-m13 26.78 701400 99.95 1.68 0 372164
leanN       sn-chk 28.45 681480 108.82 1.71 0 370780
larsonN-sized sn-m0 1.224 921164 357.46 1.17 0 75204
larsonN-sized sn-m12 1.448 948324 355.82 1.40 0 69590
larsonN-sized sn-m13 7.780 880144 350.75 1.28 0 44900
larsonN-sized sn-chk 8.642 851820 352.24 1.29 0 44474
mstressN    sn-m0 01.92 1575824 44.34 2.23 0 30869
mstressN    sn-m12 02.46 1620004 44.79 3.21 0 39147
mstressN    sn-m13 09.60 1362696 51.99 61.56 0 7511285
mstressN    sn-chk 09.63 1381832 50.47 65.04 0 7505614
rptestN     sn-m0 2.755 1042792 31.96 5.00 5 36736
rptestN     sn-m12 3.049 1026724 37.92 4.92 5 36158
rptestN     sn-m13 4.507 964600 53.84 5.08 0 31708
rptestN     sn-chk 4.813 846072 57.91 4.91 2 35348
gs          sn-m0 01.94 48208 1.12 0.07 213 1835
gs          sn-m12 01.16 54212 1.13 0.02 0 1940
gs          sn-m13 01.41 50948 1.23 0.17 0 73796
gs          sn-chk 01.44 50920 1.26 0.17 0 73672
lua         sn-m0 06.37 97692 4.92 0.54 300 147891
lua         sn-m12 05.54 103520 4.97 0.56 0 154032
lua         sn-m13 05.76 104196 5.13 0.62 0 174304
lua         sn-chk 05.72 103868 5.10 0.61 0 175901
alloc-test1 sn-m0 03.64 13968 3.62 0.01 0 1352
alloc-test1 sn-m12 03.74 16128 3.73 0.00 0 1405
alloc-test1 sn-m13 03.92 15068 3.92 0.00 0 1406
alloc-test1 sn-chk 03.92 15192 3.92 0.00 0 1403
alloc-testN sn-m0 03.66 21408 56.72 0.02 0 7061
alloc-testN sn-m12 03.73 21700 58.48 0.03 0 9359
alloc-testN sn-m13 03.83 21384 61.08 0.02 0 9312
alloc-testN sn-chk 03.91 21328 61.33 0.01 0 9370
sh6benchN   sn-m0 00.13 299380 1.67 0.30 0 39842
sh6benchN   sn-m12 00.55 492468 30.33 0.29 0 47556
sh6benchN   sn-m13 00.33 511348 12.85 0.34 0 51693
sh6benchN   sn-chk 00.32 523556 11.77 0.34 0 54782
sh8benchN   sn-m0 00.20 300464 4.86 0.18 0 16619
sh8benchN   sn-m12 00.43 459880 8.68 0.63 0 20091
sh8benchN   sn-m13 01.89 787908 47.79 4.51 0 228719
sh8benchN   sn-chk 01.63 818524 31.11 4.19 0 216240
xmalloc-testN sn-m0 0.202 262404 187.86 45.13 0 39989
xmalloc-testN sn-m12 0.268 299488 195.74 12.09 4 41547
xmalloc-testN sn-m13 0.282 314124 217.25 9.41 0 44752
xmalloc-testN sn-chk 0.282 305172 218.47 9.60 0 46864
cache-scratch1 sn-m0 01.42 3920 1.42 0.00 0 280
cache-scratch1 sn-m12 01.41 3876 1.41 0.00 0 283
cache-scratch1 sn-m13 01.41 3900 1.41 0.00 0 278
cache-scratch1 sn-chk 01.42 3984 1.42 0.00 0 282
cache-scratchN sn-m0 00.05 4412 3.31 0.00 0 588
cache-scratchN sn-m12 00.05 4664 3.30 0.01 0 653
cache-scratchN sn-m13 00.05 4428 3.26 0.01 0 650
cache-scratchN sn-chk 00.05 4764 3.26 0.01 0 655
glibc-simple sn-m0 01.64 2460 1.63 0.00 0 291
glibc-simple sn-m12 02.41 2616 2.41 0.00 0 316
glibc-simple sn-m13 02.27 2604 2.27 0.00 0 329
glibc-simple sn-chk 02.27 2620 2.27 0.00 0 325
glibc-thread sn-m0 .339 63960 141.38 0.09 69 20390
glibc-thread sn-m12 .394 282480 138.76 0.35 0 43582
glibc-thread sn-m13 .456 282524 138.78 0.24 0 42587
glibc-thread sn-chk .457 282576 137.92 0.27 0 41156
rocksdb     sn-m0 05.29 102108 3.91 1.53 86 2398
rocksdb     sn-m12 05.21 107144 3.87 1.67 0 2580
rocksdb     sn-m13 05.26 100960 4.05 1.59 0 20659
rocksdb     sn-chk 05.38 99392 4.15 1.63 0 20808

So m0 is not mitigations, m12 is everything except free scrub and guarded mempcy, m13 has free scrub, and sn-chk is all the checks. This looks a bit pricey in places to turn on always. But worth having as an option.

[nwf]

WesZero. Hah! ❤️

[mjp41]

WesZero. Hah! ❤️

Sadly for Larson it goes slower. But I think Larson is mostly untouched allocations below the size of a page.