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