Anton Korobeynikov
765fc9c0ca
Unify statistic collection:
...
- For MI_STAT == 0 no allocation stats are collected
- For MI_STAT == 1 only aggregated values (across normal, large and huge heaps) are collected
- For MI_STAT == 1 separate per-bin collection for normal heap is done as well
2020-11-11 11:41:39 +03:00
Anton Korobeynikov
00fb89f771
Rename the field
2020-11-11 11:41:33 +03:00
daan
14b8d27386
track pinned memory separately from large os pages
2020-09-08 16:46:03 -07:00
daan
2e311f341b
fix msvc compilation in C mode
2020-09-05 09:37:09 -07:00
daan
900c97664a
merge from dev-atomic
2020-09-03 09:47:01 -07:00
daan
8033b62979
allow overriding MI_MAX_ALIGN_SIZE
2020-08-29 09:59:15 -07:00
daan
ef8e5d18a6
replace atomics with C11/C++ atomics with explicit memory order; passes tsan. Issue #130
2020-07-26 18:01:33 -07:00
daan
95afd0509f
make segment abandoned_next atomic; tsan passes without warnings now (issue #130 )
2020-07-25 23:50:22 -07:00
daan
09ade02429
bring inline with C11 atomics; no volatile and cas order of expected/desired
2020-07-25 22:52:27 -07:00
Anton Korobeynikov
079b886feb
Add cmake option to specify whether warnings / errors are enabled by default.
...
Currently warnings / errors are enabled by default in debug build.
Otherwise they could be enabled only via environmental variable or
API option call. Add possibility to specify the default during the
build time. This simplifies e.g. integration of the library into
bigger projects as no source changes would be required.
2020-04-28 16:44:44 +03:00
daan
69a0846478
add MI_PADDING flag to cmake to supress use of padding in debug mode
2020-04-07 10:01:18 -07:00
daan
7745dde8d2
allow retirement for all object sizes (issue #212 )
2020-03-16 15:31:37 -07:00
daan
8a2a52843d
delete all thread owned heaps when a thread is terminated (issue #202 )
2020-02-13 12:15:23 -08:00
daan
aa68b8cbc7
improve encoding of padding canary and buffer overflow detection
2020-02-01 12:16:37 -08:00
daan
40f1e1e07b
byte-precise heap block overflow checking with encoded padding
2020-01-31 23:39:51 -08:00
daan
68112a2751
better padding implementation, more precise statistics
2020-01-31 20:34:24 -08:00
daan
4531367de2
fix padding check for aligned allocation; improve perf for small aligned allocations
2020-01-31 13:20:02 -08:00
daan
5d212d688f
add MI_PADDING build option to add padding to each block to detect heap block overflows
2020-01-29 17:10:57 -08:00
daan
5e32d00aab
add visit count to abandoned to limit list length
2020-01-25 12:26:08 -08:00
daan
7a1e86fc20
merge from dev
2020-01-15 18:03:37 -08:00
daan
0099707af9
use delayed free for all pages; reduce size of the page structure for improved address calculation
2020-01-15 17:19:01 -08:00
daan
c9b5ac80b3
update page reset queue to just do delayed page resets
2020-01-15 12:00:44 -08:00
daan
941c55ee42
wip: first implementation of page free list on segments for effecient delayed page reset
2020-01-14 21:47:18 -08:00
daan
683d8998d4
fix potential A-B-A problem with segment abandonment; noticed by Manual Poeter and Sam Gross
2020-01-08 17:45:38 -08:00
daan
eeb623e6af
increase retire limit, collect retired pages
2020-01-03 17:06:41 -08:00
daan
f9ca88f71c
set secure default to 0 again
2020-01-02 17:57:41 -08:00
daan
fc3e537bd4
improve double free detection with faster same page check
2019-12-28 15:28:13 -08:00
daan
e3391d9a53
stronger encoding of free lists using two keys per page
2019-12-28 00:57:42 -08:00
daan
ba87a39d9f
updated random cookie generation using OS primitives and chacha20
2019-12-22 19:37:49 -08:00
daan
50575b12c0
Merge branch 'dev' into dev-exp
2019-11-21 16:29:59 -08:00
daan
41caf6d0f8
set secure default to 0
2019-11-21 16:29:46 -08:00
daan
6568059cc6
merge from dev
2019-11-21 16:28:28 -08:00
daan
74dbfc30be
improved security by encoding NULL values; double free mitigation on by default; more precise free list corruption detection
2019-11-21 15:21:23 -08:00
Daan Leijen
211f1aa519
remove reset delay slots; add reset tracking per page and segment
2019-11-20 14:55:12 -08:00
daan
a0958b2da6
enable more reset delay slots
2019-11-11 17:06:16 -08:00
daan
181bef382c
merge dev-exp
2019-11-10 11:39:53 -08:00
daan
27f1a8b3d2
fix avg display; set secure default to 0`
2019-11-07 10:35:30 -08:00
daan
36090dee5c
Merge branch 'dev' into dev-exp
2019-11-07 10:34:06 -08:00
daan
31d11f64d5
fix secure free list extension where a non-empty initial free list was discarded
2019-11-07 10:33:45 -08:00
daan
378716c467
refactor and improve atomic bitmap usage
2019-11-07 10:26:52 -08:00
daan
829fd872f4
initial delay slots
2019-11-04 11:48:41 -08:00
daan
fd9d8c85ae
change numa support on linux to use getcpu
2019-11-02 11:55:03 -07:00
daan
2c12d7f223
optimized numa calls; better Linux support
2019-11-01 22:01:52 -07:00
daan
b052d3b731
enable double free and heap corruption detection in debug mode
2019-10-28 15:54:33 -07:00
daan
081e2d1eb6
fix statistics display
2019-10-28 13:43:42 -07:00
daan
2affdbbd2e
stronger secure mode when defining MI_SECURE=4: checks for double free, corrupted free list, and invalid pointer frees. Performance is impacted but not too much -- more perf testing is needed
2019-10-18 18:11:04 -07:00
daan
5de851a84d
update page_flags to have more portable definition
2019-10-17 16:48:16 -07:00
daan
26c27fbf58
use uint8_t bit fields, and improve portability of page_flags type
2019-10-17 12:07:26 -07:00
Johannes Schindelin
559688ec64
Suppress warning about unnamed struct
...
This prevents MSVC complaining with
warning C4201: nonstandard extension used: nameless struct/union
The struct might seem unnecessary to the occasional reader (it did seem
so to this commit's author), but it is not! It is required to align the
fields to a boundary, which is verified by the test suite. Removing that
"unnecessary" `struct` results in this failure:
1: Test command: mimalloc-test-api
[...]
1: test: malloc-zero... mimalloc: assertion failed: at src/page.c:591, mi_page_init
1: assertion: "!mi_page_has_aligned(page)"
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
2019-10-17 20:43:18 +02:00
Johannes Schindelin
e747a6f3a6
Use unsigned for bit-field variables
...
It is actually non-standard to use `bool` with a bit-field quantifier,
and VS 2019 complains about this.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
2019-10-17 20:43:18 +02:00
