]>
Commit | Line | Data |
---|---|---|
0c14398b MR |
1 | .. _slub: |
2 | ||
3 | ========================== | |
35243421 | 4 | Short users guide for SLUB |
0c14398b | 5 | ========================== |
35243421 | 6 | |
35243421 CL |
7 | The basic philosophy of SLUB is very different from SLAB. SLAB |
8 | requires rebuilding the kernel to activate debug options for all | |
c1aee215 | 9 | slab caches. SLUB always includes full debugging but it is off by default. |
35243421 CL |
10 | SLUB can enable debugging only for selected slabs in order to avoid |
11 | an impact on overall system performance which may make a bug more | |
12 | difficult to find. | |
13 | ||
0c14398b | 14 | In order to switch debugging on one can add an option ``slub_debug`` |
35243421 CL |
15 | to the kernel command line. That will enable full debugging for |
16 | all slabs. | |
17 | ||
0c14398b MR |
18 | Typically one would then use the ``slabinfo`` command to get statistical |
19 | data and perform operation on the slabs. By default ``slabinfo`` only lists | |
35243421 | 20 | slabs that have data in them. See "slabinfo -h" for more options when |
0c14398b MR |
21 | running the command. ``slabinfo`` can be compiled with |
22 | :: | |
35243421 | 23 | |
0c14398b | 24 | gcc -o slabinfo tools/vm/slabinfo.c |
35243421 | 25 | |
0c14398b | 26 | Some of the modes of operation of ``slabinfo`` require that slub debugging |
35243421 CL |
27 | be enabled on the command line. F.e. no tracking information will be |
28 | available without debugging on and validation can only partially | |
29 | be performed if debugging was not switched on. | |
30 | ||
31 | Some more sophisticated uses of slub_debug: | |
32 | ------------------------------------------- | |
33 | ||
0c14398b | 34 | Parameters may be given to ``slub_debug``. If none is specified then full |
35243421 CL |
35 | debugging is enabled. Format: |
36 | ||
0c14398b MR |
37 | slub_debug=<Debug-Options> |
38 | Enable options for all slabs | |
0c14398b | 39 | |
c5fd3ca0 AT |
40 | slub_debug=<Debug-Options>,<slab name1>,<slab name2>,... |
41 | Enable options only for select slabs (no spaces | |
42 | after a comma) | |
0c14398b MR |
43 | |
44 | Possible debug options are:: | |
35243421 | 45 | |
becfda68 LA |
46 | F Sanity checks on (enables SLAB_DEBUG_CONSISTENCY_CHECKS |
47 | Sorry SLAB legacy issues) | |
35243421 CL |
48 | Z Red zoning |
49 | P Poisoning (object and padding) | |
50 | U User tracking (free and alloc) | |
51 | T Trace (please only use on single slabs) | |
4c13dd3b | 52 | A Toggle failslab filter mark for the cache |
fa5ec8a1 DR |
53 | O Switch debugging off for caches that would have |
54 | caused higher minimum slab orders | |
f0630fff CL |
55 | - Switch all debugging off (useful if the kernel is |
56 | configured with CONFIG_SLUB_DEBUG_ON) | |
35243421 | 57 | |
0c14398b | 58 | F.e. in order to boot just with sanity checks and red zoning one would specify:: |
35243421 CL |
59 | |
60 | slub_debug=FZ | |
61 | ||
0c14398b | 62 | Trying to find an issue in the dentry cache? Try:: |
35243421 | 63 | |
989a7241 | 64 | slub_debug=,dentry |
35243421 | 65 | |
c5fd3ca0 AT |
66 | to only enable debugging on the dentry cache. You may use an asterisk at the |
67 | end of the slab name, in order to cover all slabs with the same prefix. For | |
68 | example, here's how you can poison the dentry cache as well as all kmalloc | |
11ede500 | 69 | slabs:: |
c5fd3ca0 AT |
70 | |
71 | slub_debug=P,kmalloc-*,dentry | |
35243421 CL |
72 | |
73 | Red zoning and tracking may realign the slab. We can just apply sanity checks | |
0c14398b | 74 | to the dentry cache with:: |
35243421 | 75 | |
989a7241 | 76 | slub_debug=F,dentry |
35243421 | 77 | |
fa5ec8a1 DR |
78 | Debugging options may require the minimum possible slab order to increase as |
79 | a result of storing the metadata (for example, caches with PAGE_SIZE object | |
80 | sizes). This has a higher liklihood of resulting in slab allocation errors | |
81 | in low memory situations or if there's high fragmentation of memory. To | |
0c14398b | 82 | switch off debugging for such caches by default, use:: |
fa5ec8a1 DR |
83 | |
84 | slub_debug=O | |
85 | ||
35243421 CL |
86 | In case you forgot to enable debugging on the kernel command line: It is |
87 | possible to enable debugging manually when the kernel is up. Look at the | |
0c14398b | 88 | contents of:: |
35243421 | 89 | |
0c14398b | 90 | /sys/kernel/slab/<slab name>/ |
35243421 CL |
91 | |
92 | Look at the writable files. Writing 1 to them will enable the | |
93 | corresponding debug option. All options can be set on a slab that does | |
94 | not contain objects. If the slab already contains objects then sanity checks | |
95 | and tracing may only be enabled. The other options may cause the realignment | |
96 | of objects. | |
97 | ||
98 | Careful with tracing: It may spew out lots of information and never stop if | |
99 | used on the wrong slab. | |
100 | ||
c1aee215 | 101 | Slab merging |
0c14398b | 102 | ============ |
35243421 | 103 | |
c1aee215 | 104 | If no debug options are specified then SLUB may merge similar slabs together |
35243421 | 105 | in order to reduce overhead and increase cache hotness of objects. |
0c14398b | 106 | ``slabinfo -a`` displays which slabs were merged together. |
35243421 | 107 | |
c1aee215 | 108 | Slab validation |
0c14398b | 109 | =============== |
c1aee215 CL |
110 | |
111 | SLUB can validate all object if the kernel was booted with slub_debug. In | |
0c14398b MR |
112 | order to do so you must have the ``slabinfo`` tool. Then you can do |
113 | :: | |
c1aee215 | 114 | |
0c14398b | 115 | slabinfo -v |
c1aee215 CL |
116 | |
117 | which will test all objects. Output will be generated to the syslog. | |
118 | ||
119 | This also works in a more limited way if boot was without slab debug. | |
0c14398b | 120 | In that case ``slabinfo -v`` simply tests all reachable objects. Usually |
c1aee215 CL |
121 | these are in the cpu slabs and the partial slabs. Full slabs are not |
122 | tracked by SLUB in a non debug situation. | |
123 | ||
35243421 | 124 | Getting more performance |
0c14398b | 125 | ======================== |
35243421 CL |
126 | |
127 | To some degree SLUB's performance is limited by the need to take the | |
128 | list_lock once in a while to deal with partial slabs. That overhead is | |
129 | governed by the order of the allocation for each slab. The allocations | |
130 | can be influenced by kernel parameters: | |
131 | ||
0c14398b MR |
132 | .. slub_min_objects=x (default 4) |
133 | .. slub_min_order=x (default 0) | |
134 | .. slub_max_order=x (default 3 (PAGE_ALLOC_COSTLY_ORDER)) | |
135 | ||
136 | ``slub_min_objects`` | |
137 | allows to specify how many objects must at least fit into one | |
138 | slab in order for the allocation order to be acceptable. In | |
139 | general slub will be able to perform this number of | |
140 | allocations on a slab without consulting centralized resources | |
141 | (list_lock) where contention may occur. | |
142 | ||
143 | ``slub_min_order`` | |
358b6ba9 | 144 | specifies a minimum order of slabs. A similar effect like |
0c14398b MR |
145 | ``slub_min_objects``. |
146 | ||
147 | ``slub_max_order`` | |
148 | specified the order at which ``slub_min_objects`` should no | |
149 | longer be checked. This is useful to avoid SLUB trying to | |
150 | generate super large order pages to fit ``slub_min_objects`` | |
151 | of a slab cache with large object sizes into one high order | |
152 | page. Setting command line parameter | |
153 | ``debug_guardpage_minorder=N`` (N > 0), forces setting | |
154 | ``slub_max_order`` to 0, what cause minimum possible order of | |
155 | slabs allocation. | |
35243421 | 156 | |
c1aee215 | 157 | SLUB Debug output |
0c14398b MR |
158 | ================= |
159 | ||
160 | Here is a sample of slub debug output:: | |
161 | ||
162 | ==================================================================== | |
163 | BUG kmalloc-8: Redzone overwritten | |
164 | -------------------------------------------------------------------- | |
165 | ||
166 | INFO: 0xc90f6d28-0xc90f6d2b. First byte 0x00 instead of 0xcc | |
167 | INFO: Slab 0xc528c530 flags=0x400000c3 inuse=61 fp=0xc90f6d58 | |
168 | INFO: Object 0xc90f6d20 @offset=3360 fp=0xc90f6d58 | |
169 | INFO: Allocated in get_modalias+0x61/0xf5 age=53 cpu=1 pid=554 | |
170 | ||
171 | Bytes b4 0xc90f6d10: 00 00 00 00 00 00 00 00 5a 5a 5a 5a 5a 5a 5a 5a ........ZZZZZZZZ | |
172 | Object 0xc90f6d20: 31 30 31 39 2e 30 30 35 1019.005 | |
173 | Redzone 0xc90f6d28: 00 cc cc cc . | |
174 | Padding 0xc90f6d50: 5a 5a 5a 5a 5a 5a 5a 5a ZZZZZZZZ | |
175 | ||
176 | [<c010523d>] dump_trace+0x63/0x1eb | |
177 | [<c01053df>] show_trace_log_lvl+0x1a/0x2f | |
178 | [<c010601d>] show_trace+0x12/0x14 | |
179 | [<c0106035>] dump_stack+0x16/0x18 | |
180 | [<c017e0fa>] object_err+0x143/0x14b | |
181 | [<c017e2cc>] check_object+0x66/0x234 | |
182 | [<c017eb43>] __slab_free+0x239/0x384 | |
183 | [<c017f446>] kfree+0xa6/0xc6 | |
184 | [<c02e2335>] get_modalias+0xb9/0xf5 | |
185 | [<c02e23b7>] dmi_dev_uevent+0x27/0x3c | |
186 | [<c027866a>] dev_uevent+0x1ad/0x1da | |
187 | [<c0205024>] kobject_uevent_env+0x20a/0x45b | |
188 | [<c020527f>] kobject_uevent+0xa/0xf | |
189 | [<c02779f1>] store_uevent+0x4f/0x58 | |
190 | [<c027758e>] dev_attr_store+0x29/0x2f | |
191 | [<c01bec4f>] sysfs_write_file+0x16e/0x19c | |
192 | [<c0183ba7>] vfs_write+0xd1/0x15a | |
193 | [<c01841d7>] sys_write+0x3d/0x72 | |
194 | [<c0104112>] sysenter_past_esp+0x5f/0x99 | |
195 | [<b7f7b410>] 0xb7f7b410 | |
196 | ======================= | |
197 | ||
198 | FIX kmalloc-8: Restoring Redzone 0xc90f6d28-0xc90f6d2b=0xcc | |
c1aee215 | 199 | |
24922684 CL |
200 | If SLUB encounters a corrupted object (full detection requires the kernel |
201 | to be booted with slub_debug) then the following output will be dumped | |
202 | into the syslog: | |
c1aee215 | 203 | |
24922684 | 204 | 1. Description of the problem encountered |
c1aee215 | 205 | |
0c14398b | 206 | This will be a message in the system log starting with:: |
c1aee215 | 207 | |
0c14398b MR |
208 | =============================================== |
209 | BUG <slab cache affected>: <What went wrong> | |
210 | ----------------------------------------------- | |
c1aee215 | 211 | |
0c14398b MR |
212 | INFO: <corruption start>-<corruption_end> <more info> |
213 | INFO: Slab <address> <slab information> | |
214 | INFO: Object <address> <object information> | |
215 | INFO: Allocated in <kernel function> age=<jiffies since alloc> cpu=<allocated by | |
24922684 | 216 | cpu> pid=<pid of the process> |
0c14398b MR |
217 | INFO: Freed in <kernel function> age=<jiffies since free> cpu=<freed by cpu> |
218 | pid=<pid of the process> | |
c1aee215 | 219 | |
0c14398b MR |
220 | (Object allocation / free information is only available if SLAB_STORE_USER is |
221 | set for the slab. slub_debug sets that option) | |
c1aee215 | 222 | |
24922684 | 223 | 2. The object contents if an object was involved. |
c1aee215 | 224 | |
0c14398b | 225 | Various types of lines can follow the BUG SLUB line: |
c1aee215 | 226 | |
0c14398b | 227 | Bytes b4 <address> : <bytes> |
24922684 | 228 | Shows a few bytes before the object where the problem was detected. |
c1aee215 CL |
229 | Can be useful if the corruption does not stop with the start of the |
230 | object. | |
231 | ||
0c14398b | 232 | Object <address> : <bytes> |
c1aee215 | 233 | The bytes of the object. If the object is inactive then the bytes |
24922684 | 234 | typically contain poison values. Any non-poison value shows a |
c1aee215 CL |
235 | corruption by a write after free. |
236 | ||
0c14398b | 237 | Redzone <address> : <bytes> |
24922684 | 238 | The Redzone following the object. The Redzone is used to detect |
c1aee215 CL |
239 | writes after the object. All bytes should always have the same |
240 | value. If there is any deviation then it is due to a write after | |
241 | the object boundary. | |
242 | ||
24922684 CL |
243 | (Redzone information is only available if SLAB_RED_ZONE is set. |
244 | slub_debug sets that option) | |
c1aee215 | 245 | |
0c14398b | 246 | Padding <address> : <bytes> |
c1aee215 CL |
247 | Unused data to fill up the space in order to get the next object |
248 | properly aligned. In the debug case we make sure that there are | |
24922684 | 249 | at least 4 bytes of padding. This allows the detection of writes |
c1aee215 CL |
250 | before the object. |
251 | ||
24922684 CL |
252 | 3. A stackdump |
253 | ||
0c14398b MR |
254 | The stackdump describes the location where the error was detected. The cause |
255 | of the corruption is may be more likely found by looking at the function that | |
256 | allocated or freed the object. | |
24922684 CL |
257 | |
258 | 4. Report on how the problem was dealt with in order to ensure the continued | |
0c14398b | 259 | operation of the system. |
24922684 | 260 | |
0c14398b | 261 | These are messages in the system log beginning with:: |
24922684 | 262 | |
0c14398b | 263 | FIX <slab cache affected>: <corrective action taken> |
24922684 | 264 | |
0c14398b MR |
265 | In the above sample SLUB found that the Redzone of an active object has |
266 | been overwritten. Here a string of 8 characters was written into a slab that | |
267 | has the length of 8 characters. However, a 8 character string needs a | |
268 | terminating 0. That zero has overwritten the first byte of the Redzone field. | |
269 | After reporting the details of the issue encountered the FIX SLUB message | |
270 | tells us that SLUB has restored the Redzone to its proper value and then | |
271 | system operations continue. | |
24922684 | 272 | |
0c14398b MR |
273 | Emergency operations |
274 | ==================== | |
24922684 | 275 | |
0c14398b | 276 | Minimal debugging (sanity checks alone) can be enabled by booting with:: |
24922684 CL |
277 | |
278 | slub_debug=F | |
279 | ||
280 | This will be generally be enough to enable the resiliency features of slub | |
281 | which will keep the system running even if a bad kernel component will | |
282 | keep corrupting objects. This may be important for production systems. | |
283 | Performance will be impacted by the sanity checks and there will be a | |
284 | continual stream of error messages to the syslog but no additional memory | |
285 | will be used (unlike full debugging). | |
286 | ||
287 | No guarantees. The kernel component still needs to be fixed. Performance | |
288 | may be optimized further by locating the slab that experiences corruption | |
289 | and enabling debugging only for that cache | |
290 | ||
0c14398b | 291 | I.e.:: |
24922684 CL |
292 | |
293 | slub_debug=F,dentry | |
294 | ||
295 | If the corruption occurs by writing after the end of the object then it | |
296 | may be advisable to enable a Redzone to avoid corrupting the beginning | |
0c14398b | 297 | of other objects:: |
24922684 CL |
298 | |
299 | slub_debug=FZ,dentry | |
c1aee215 | 300 | |
05be9617 | 301 | Extended slabinfo mode and plotting |
0c14398b | 302 | =================================== |
05be9617 | 303 | |
0c14398b | 304 | The ``slabinfo`` tool has a special 'extended' ('-X') mode that includes: |
05be9617 SS |
305 | - Slabcache Totals |
306 | - Slabs sorted by size (up to -N <num> slabs, default 1) | |
307 | - Slabs sorted by loss (up to -N <num> slabs, default 1) | |
308 | ||
0c14398b MR |
309 | Additionally, in this mode ``slabinfo`` does not dynamically scale |
310 | sizes (G/M/K) and reports everything in bytes (this functionality is | |
311 | also available to other slabinfo modes via '-B' option) which makes | |
312 | reporting more precise and accurate. Moreover, in some sense the `-X' | |
313 | mode also simplifies the analysis of slabs' behaviour, because its | |
314 | output can be plotted using the ``slabinfo-gnuplot.sh`` script. So it | |
315 | pushes the analysis from looking through the numbers (tons of numbers) | |
316 | to something easier -- visual analysis. | |
05be9617 SS |
317 | |
318 | To generate plots: | |
0c14398b MR |
319 | |
320 | a) collect slabinfo extended records, for example:: | |
321 | ||
322 | while [ 1 ]; do slabinfo -X >> FOO_STATS; sleep 1; done | |
323 | ||
324 | b) pass stats file(-s) to ``slabinfo-gnuplot.sh`` script:: | |
325 | ||
326 | slabinfo-gnuplot.sh FOO_STATS [FOO_STATS2 .. FOO_STATSN] | |
327 | ||
328 | The ``slabinfo-gnuplot.sh`` script will pre-processes the collected records | |
329 | and generates 3 png files (and 3 pre-processing cache files) per STATS | |
330 | file: | |
331 | - Slabcache Totals: FOO_STATS-totals.png | |
332 | - Slabs sorted by size: FOO_STATS-slabs-by-size.png | |
333 | - Slabs sorted by loss: FOO_STATS-slabs-by-loss.png | |
334 | ||
335 | Another use case, when ``slabinfo-gnuplot.sh`` can be useful, is when you | |
336 | need to compare slabs' behaviour "prior to" and "after" some code | |
337 | modification. To help you out there, ``slabinfo-gnuplot.sh`` script | |
338 | can 'merge' the `Slabcache Totals` sections from different | |
339 | measurements. To visually compare N plots: | |
340 | ||
341 | a) Collect as many STATS1, STATS2, .. STATSN files as you need:: | |
342 | ||
343 | while [ 1 ]; do slabinfo -X >> STATS<X>; sleep 1; done | |
344 | ||
345 | b) Pre-process those STATS files:: | |
346 | ||
347 | slabinfo-gnuplot.sh STATS1 STATS2 .. STATSN | |
348 | ||
349 | c) Execute ``slabinfo-gnuplot.sh`` in '-t' mode, passing all of the | |
350 | generated pre-processed \*-totals:: | |
351 | ||
352 | slabinfo-gnuplot.sh -t STATS1-totals STATS2-totals .. STATSN-totals | |
353 | ||
354 | This will produce a single plot (png file). | |
355 | ||
356 | Plots, expectedly, can be large so some fluctuations or small spikes | |
357 | can go unnoticed. To deal with that, ``slabinfo-gnuplot.sh`` has two | |
358 | options to 'zoom-in'/'zoom-out': | |
359 | ||
360 | a) ``-s %d,%d`` -- overwrites the default image width and heigh | |
361 | b) ``-r %d,%d`` -- specifies a range of samples to use (for example, | |
362 | in ``slabinfo -X >> FOO_STATS; sleep 1;`` case, using a ``-r | |
363 | 40,60`` range will plot only samples collected between 40th and | |
364 | 60th seconds). | |
05be9617 | 365 | |
cde53535 | 366 | Christoph Lameter, May 30, 2007 |
05be9617 | 367 | Sergey Senozhatsky, October 23, 2015 |