1 ==========================
2 Short users guide for SLUB
3 ==========================
4
5 The basic philosophy of SLUB is very different from SLAB. SLAB
6 requires rebuilding the kernel to activate debug options for all
7 slab caches. SLUB always includes full debugging but it is off by default.
8 SLUB can enable debugging only for selected slabs in order to avoid
9 an impact on overall system performance which may make a bug more
10 difficult to find.
11
12 In order to switch debugging on one can add an option ``slab_debug``
13 to the kernel command line. That will enable full debugging for
14 all slabs.
15
16 Typically one would then use the ``slabinfo`` command to get statistical
17 data and perform operation on the slabs. By default ``slabinfo`` only lists
18 slabs that have data in them. See "slabinfo -h" for more options when
19 running the command. ``slabinfo`` can be compiled with
20 ::
21
22 gcc -o slabinfo tools/mm/slabinfo.c
23
24 Some of the modes of operation of ``slabinfo`` require that slub debugging
25 be enabled on the command line. F.e. no tracking information will be
26 available without debugging on and validation can only partially
27 be performed if debugging was not switched on.
28
29 Some more sophisticated uses of slab_debug:
30 -------------------------------------------
31
32 Parameters may be given to ``slab_debug``. If none is specified then full
33 debugging is enabled. Format:
34
35 slab_debug=<Debug-Options>
36 Enable options for all slabs
37
38 slab_debug=<Debug-Options>,<slab name1>,<slab name2>,...
39 Enable options only for select slabs (no spaces
40 after a comma)
41
42 Multiple blocks of options for all slabs or selected slabs can be given, with
43 blocks of options delimited by ';'. The last of "all slabs" blocks is applied
44 to all slabs except those that match one of the "select slabs" block. Options
45 of the first "select slabs" blocks that matches the slab's name are applied.
46
47 Possible debug options are::
48
49 F Sanity checks on (enables SLAB_DEBUG_CONSISTENCY_CHECKS
50 Sorry SLAB legacy issues)
51 Z Red zoning
52 P Poisoning (object and padding)
53 U User tracking (free and alloc)
54 T Trace (please only use on single slabs)
55 A Enable failslab filter mark for the cache
56 O Switch debugging off for caches that would have
57 caused higher minimum slab orders
58 - Switch all debugging off (useful if the kernel is
59 configured with CONFIG_SLUB_DEBUG_ON)
60
61 F.e. in order to boot just with sanity checks and red zoning one would specify::
62
63 slab_debug=FZ
64
65 Trying to find an issue in the dentry cache? Try::
66
67 slab_debug=,dentry
68
69 to only enable debugging on the dentry cache. You may use an asterisk at the
70 end of the slab name, in order to cover all slabs with the same prefix. For
71 example, here's how you can poison the dentry cache as well as all kmalloc
72 slabs::
73
74 slab_debug=P,kmalloc-*,dentry
75
76 Red zoning and tracking may realign the slab. We can just apply sanity checks
77 to the dentry cache with::
78
79 slab_debug=F,dentry
80
81 Debugging options may require the minimum possible slab order to increase as
82 a result of storing the metadata (for example, caches with PAGE_SIZE object
83 sizes). This has a higher likelihood of resulting in slab allocation errors
84 in low memory situations or if there's high fragmentation of memory. To
85 switch off debugging for such caches by default, use::
86
87 slab_debug=O
88
89 You can apply different options to different list of slab names, using blocks
90 of options. This will enable red zoning for dentry and user tracking for
91 kmalloc. All other slabs will not get any debugging enabled::
92
93 slab_debug=Z,dentry;U,kmalloc-*
94
95 You can also enable options (e.g. sanity checks and poisoning) for all caches
96 except some that are deemed too performance critical and don't need to be
97 debugged by specifying global debug options followed by a list of slab names
98 with "-" as options::
99
100 slab_debug=FZ;-,zs_handle,zspage
101
102 The state of each debug option for a slab can be found in the respective files
103 under::
104
105 /sys/kernel/slab/<slab name>/
106
107 If the file contains 1, the option is enabled, 0 means disabled. The debug
108 options from the ``slab_debug`` parameter translate to the following files::
109
110 F sanity_checks
111 Z red_zone
112 P poison
113 U store_user
114 T trace
115 A failslab
116
117 failslab file is writable, so writing 1 or 0 will enable or disable
118 the option at runtime. Write returns -EINVAL if cache is an alias.
119 Careful with tracing: It may spew out lots of information and never stop if
120 used on the wrong slab.
121
122 Slab merging
123 ============
124
125 If no debug options are specified then SLUB may merge similar slabs together
126 in order to reduce overhead and increase cache hotness of objects.
127 ``slabinfo -a`` displays which slabs were merged together.
128
129 Slab validation
130 ===============
131
132 SLUB can validate all object if the kernel was booted with slab_debug. In
133 order to do so you must have the ``slabinfo`` tool. Then you can do
134 ::
135
136 slabinfo -v
137
138 which will test all objects. Output will be generated to the syslog.
139
140 This also works in a more limited way if boot was without slab debug.
141 In that case ``slabinfo -v`` simply tests all reachable objects. Usually
142 these are in the cpu slabs and the partial slabs. Full slabs are not
143 tracked by SLUB in a non debug situation.
144
145 Getting more performance
146 ========================
147
148 To some degree SLUB's performance is limited by the need to take the
149 list_lock once in a while to deal with partial slabs. That overhead is
150 governed by the order of the allocation for each slab. The allocations
151 can be influenced by kernel parameters:
152
153 .. slab_min_objects=x (default: automatically scaled by number of cpus)
154 .. slab_min_order=x (default 0)
155 .. slab_max_order=x (default 3 (PAGE_ALLOC_COSTLY_ORDER))
156
157 ``slab_min_objects``
158 allows to specify how many objects must at least fit into one
159 slab in order for the allocation order to be acceptable. In
160 general slub will be able to perform this number of
161 allocations on a slab without consulting centralized resources
162 (list_lock) where contention may occur.
163
164 ``slab_min_order``
165 specifies a minimum order of slabs. A similar effect like
166 ``slab_min_objects``.
167
168 ``slab_max_order``
169 specified the order at which ``slab_min_objects`` should no
170 longer be checked. This is useful to avoid SLUB trying to
171 generate super large order pages to fit ``slab_min_objects``
172 of a slab cache with large object sizes into one high order
173 page. Setting command line parameter
174 ``debug_guardpage_minorder=N`` (N > 0), forces setting
175 ``slab_max_order`` to 0, what cause minimum possible order of
176 slabs allocation.
177
178 SLUB Debug output
179 =================
180
181 Here is a sample of slub debug output::
182
183 ====================================================================
184 BUG kmalloc-8: Right Redzone overwritten
185 --------------------------------------------------------------------
186
187 INFO: 0xc90f6d28-0xc90f6d2b. First byte 0x00 instead of 0xcc
188 INFO: Slab 0xc528c530 flags=0x400000c3 inuse=61 fp=0xc90f6d58
189 INFO: Object 0xc90f6d20 @offset=3360 fp=0xc90f6d58
190 INFO: Allocated in get_modalias+0x61/0xf5 age=53 cpu=1 pid=554
191
192 Bytes b4 (0xc90f6d10): 00 00 00 00 00 00 00 00 5a 5a 5a 5a 5a 5a 5a 5a ........ZZZZZZZZ
193 Object (0xc90f6d20): 31 30 31 39 2e 30 30 35 1019.005
194 Redzone (0xc90f6d28): 00 cc cc cc .
195 Padding (0xc90f6d50): 5a 5a 5a 5a 5a 5a 5a 5a ZZZZZZZZ
196
197 [<c010523d>] dump_trace+0x63/0x1eb
198 [<c01053df>] show_trace_log_lvl+0x1a/0x2f
199 [<c010601d>] show_trace+0x12/0x14
200 [<c0106035>] dump_stack+0x16/0x18
201 [<c017e0fa>] object_err+0x143/0x14b
202 [<c017e2cc>] check_object+0x66/0x234
203 [<c017eb43>] __slab_free+0x239/0x384
204 [<c017f446>] kfree+0xa6/0xc6
205 [<c02e2335>] get_modalias+0xb9/0xf5
206 [<c02e23b7>] dmi_dev_uevent+0x27/0x3c
207 [<c027866a>] dev_uevent+0x1ad/0x1da
208 [<c0205024>] kobject_uevent_env+0x20a/0x45b
209 [<c020527f>] kobject_uevent+0xa/0xf
210 [<c02779f1>] store_uevent+0x4f/0x58
211 [<c027758e>] dev_attr_store+0x29/0x2f
212 [<c01bec4f>] sysfs_write_file+0x16e/0x19c
213 [<c0183ba7>] vfs_write+0xd1/0x15a
214 [<c01841d7>] sys_write+0x3d/0x72
215 [<c0104112>] sysenter_past_esp+0x5f/0x99
216 [<b7f7b410>] 0xb7f7b410
217 =======================
218
219 FIX kmalloc-8: Restoring Redzone 0xc90f6d28-0xc90f6d2b=0xcc
220
221 If SLUB encounters a corrupted object (full detection requires the kernel
222 to be booted with slab_debug) then the following output will be dumped
223 into the syslog:
224
225 1. Description of the problem encountered
226
227 This will be a message in the system log starting with::
228
229 ===============================================
230 BUG <slab cache affected>: <What went wrong>
231 -----------------------------------------------
232
233 INFO: <corruption start>-<corruption_end> <more info>
234 INFO: Slab <address> <slab information>
235 INFO: Object <address> <object information>
236 INFO: Allocated in <kernel function> age=<jiffies since alloc> cpu=<allocated by
237 cpu> pid=<pid of the process>
238 INFO: Freed in <kernel function> age=<jiffies since free> cpu=<freed by cpu>
239 pid=<pid of the process>
240
241 (Object allocation / free information is only available if SLAB_STORE_USER is
242 set for the slab. slab_debug sets that option)
243
244 2. The object contents if an object was involved.
245
246 Various types of lines can follow the BUG SLUB line:
247
248 Bytes b4 <address> : <bytes>
249 Shows a few bytes before the object where the problem was detected.
250 Can be useful if the corruption does not stop with the start of the
251 object.
252
253 Object <address> : <bytes>
254 The bytes of the object. If the object is inactive then the bytes
255 typically contain poison values. Any non-poison value shows a
256 corruption by a write after free.
257
258 Redzone <address> : <bytes>
259 The Redzone following the object. The Redzone is used to detect
260 writes after the object. All bytes should always have the same
261 value. If there is any deviation then it is due to a write after
262 the object boundary.
263
264 (Redzone information is only available if SLAB_RED_ZONE is set.
265 slab_debug sets that option)
266
267 Padding <address> : <bytes>
268 Unused data to fill up the space in order to get the next object
269 properly aligned. In the debug case we make sure that there are
270 at least 4 bytes of padding. This allows the detection of writes
271 before the object.
272
273 3. A stackdump
274
275 The stackdump describes the location where the error was detected. The cause
276 of the corruption is may be more likely found by looking at the function that
277 allocated or freed the object.
278
279 4. Report on how the problem was dealt with in order to ensure the continued
280 operation of the system.
281
282 These are messages in the system log beginning with::
283
284 FIX <slab cache affected>: <corrective action taken>
285
286 In the above sample SLUB found that the Redzone of an active object has
287 been overwritten. Here a string of 8 characters was written into a slab that
288 has the length of 8 characters. However, a 8 character string needs a
289 terminating 0. That zero has overwritten the first byte of the Redzone field.
290 After reporting the details of the issue encountered the FIX SLUB message
291 tells us that SLUB has restored the Redzone to its proper value and then
292 system operations continue.
293
294 Emergency operations
295 ====================
296
297 Minimal debugging (sanity checks alone) can be enabled by booting with::
298
299 slab_debug=F
300
301 This will be generally be enough to enable the resiliency features of slub
302 which will keep the system running even if a bad kernel component will
303 keep corrupting objects. This may be important for production systems.
304 Performance will be impacted by the sanity checks and there will be a
305 continual stream of error messages to the syslog but no additional memory
306 will be used (unlike full debugging).
307
308 No guarantees. The kernel component still needs to be fixed. Performance
309 may be optimized further by locating the slab that experiences corruption
310 and enabling debugging only for that cache
311
312 I.e.::
313
314 slab_debug=F,dentry
315
316 If the corruption occurs by writing after the end of the object then it
317 may be advisable to enable a Redzone to avoid corrupting the beginning
318 of other objects::
319
320 slab_debug=FZ,dentry
321
322 Extended slabinfo mode and plotting
323 ===================================
324
325 The ``slabinfo`` tool has a special 'extended' ('-X') mode that includes:
326 - Slabcache Totals
327 - Slabs sorted by size (up to -N <num> slabs, default 1)
328 - Slabs sorted by loss (up to -N <num> slabs, default 1)
329
330 Additionally, in this mode ``slabinfo`` does not dynamically scale
331 sizes (G/M/K) and reports everything in bytes (this functionality is
332 also available to other slabinfo modes via '-B' option) which makes
333 reporting more precise and accurate. Moreover, in some sense the `-X'
334 mode also simplifies the analysis of slabs' behaviour, because its
335 output can be plotted using the ``slabinfo-gnuplot.sh`` script. So it
336 pushes the analysis from looking through the numbers (tons of numbers)
337 to something easier -- visual analysis.
338
339 To generate plots:
340
341 a) collect slabinfo extended records, for example::
342
343 while [ 1 ]; do slabinfo -X >> FOO_STATS; sleep 1; done
344
345 b) pass stats file(-s) to ``slabinfo-gnuplot.sh`` script::
346
347 slabinfo-gnuplot.sh FOO_STATS [FOO_STATS2 .. FOO_STATSN]
348
349 The ``slabinfo-gnuplot.sh`` script will pre-processes the collected records
350 and generates 3 png files (and 3 pre-processing cache files) per STATS
351 file:
352 - Slabcache Totals: FOO_STATS-totals.png
353 - Slabs sorted by size: FOO_STATS-slabs-by-size.png
354 - Slabs sorted by loss: FOO_STATS-slabs-by-loss.png
355
356 Another use case, when ``slabinfo-gnuplot.sh`` can be useful, is when you
357 need to compare slabs' behaviour "prior to" and "after" some code
358 modification. To help you out there, ``slabinfo-gnuplot.sh`` script
359 can 'merge' the `Slabcache Totals` sections from different
360 measurements. To visually compare N plots:
361
362 a) Collect as many STATS1, STATS2, .. STATSN files as you need::
363
364 while [ 1 ]; do slabinfo -X >> STATS<X>; sleep 1; done
365
366 b) Pre-process those STATS files::
367
368 slabinfo-gnuplot.sh STATS1 STATS2 .. STATSN
369
370 c) Execute ``slabinfo-gnuplot.sh`` in '-t' mode, passing all of the
371 generated pre-processed \*-totals::
372
373 slabinfo-gnuplot.sh -t STATS1-totals STATS2-totals .. STATSN-totals
374
375 This will produce a single plot (png file).
376
377 Plots, expectedly, can be large so some fluctuations or small spikes
378 can go unnoticed. To deal with that, ``slabinfo-gnuplot.sh`` has two
379 options to 'zoom-in'/'zoom-out':
380
381 a) ``-s %d,%d`` -- overwrites the default image width and height
382 b) ``-r %d,%d`` -- specifies a range of samples to use (for example,
383 in ``slabinfo -X >> FOO_STATS; sleep 1;`` case, using a ``-r
384 40,60`` range will plot only samples collected between 40th and
385 60th seconds).
386
387
388 DebugFS files for SLUB
389 ======================
390
391 For more information about current state of SLUB caches with the user tracking
392 debug option enabled, debugfs files are available, typically under
393 /sys/kernel/debug/slab/<cache>/ (created only for caches with enabled user
394 tracking). There are 2 types of these files with the following debug
395 information:
396
397 1. alloc_traces::
398
399 Prints information about unique allocation traces of the currently
400 allocated objects. The output is sorted by frequency of each trace.
401
402 Information in the output:
403 Number of objects, allocating function, possible memory wastage of
404 kmalloc objects(total/per-object), minimal/average/maximal jiffies
405 since alloc, pid range of the allocating processes, cpu mask of
406 allocating cpus, numa node mask of origins of memory, and stack trace.
407
408 Example:::
409
410 338 pci_alloc_dev+0x2c/0xa0 waste=521872/1544 age=290837/291891/293509 pid=1 cpus=106 nodes=0-1
411 __kmem_cache_alloc_node+0x11f/0x4e0
412 kmalloc_trace+0x26/0xa0
413 pci_alloc_dev+0x2c/0xa0
414 pci_scan_single_device+0xd2/0x150
415 pci_scan_slot+0xf7/0x2d0
416 pci_scan_child_bus_extend+0x4e/0x360
417 acpi_pci_root_create+0x32e/0x3b0
418 pci_acpi_scan_root+0x2b9/0x2d0
419 acpi_pci_root_add.cold.11+0x110/0xb0a
420 acpi_bus_attach+0x262/0x3f0
421 device_for_each_child+0xb7/0x110
422 acpi_dev_for_each_child+0x77/0xa0
423 acpi_bus_attach+0x108/0x3f0
424 device_for_each_child+0xb7/0x110
425 acpi_dev_for_each_child+0x77/0xa0
426 acpi_bus_attach+0x108/0x3f0
427
428 2. free_traces::
429
430 Prints information about unique freeing traces of the currently allocated
431 objects. The freeing traces thus come from the previous life-cycle of the
432 objects and are reported as not available for objects allocated for the first
433 time. The output is sorted by frequency of each trace.
434
435 Information in the output:
436 Number of objects, freeing function, minimal/average/maximal jiffies since free,
437 pid range of the freeing processes, cpu mask of freeing cpus, and stack trace.
438
439 Example:::
440
441 1980 <not-available> age=4294912290 pid=0 cpus=0
442 51 acpi_ut_update_ref_count+0x6a6/0x782 age=236886/237027/237772 pid=1 cpus=1
443 kfree+0x2db/0x420
444 acpi_ut_update_ref_count+0x6a6/0x782
445 acpi_ut_update_object_reference+0x1ad/0x234
446 acpi_ut_remove_reference+0x7d/0x84
447 acpi_rs_get_prt_method_data+0x97/0xd6
448 acpi_get_irq_routing_table+0x82/0xc4
449 acpi_pci_irq_find_prt_entry+0x8e/0x2e0
450 acpi_pci_irq_lookup+0x3a/0x1e0
451 acpi_pci_irq_enable+0x77/0x240
452 pcibios_enable_device+0x39/0x40
453 do_pci_enable_device.part.0+0x5d/0xe0
454 pci_enable_device_flags+0xfc/0x120
455 pci_enable_device+0x13/0x20
456 virtio_pci_probe+0x9e/0x170
457 local_pci_probe+0x48/0x80
458 pci_device_probe+0x105/0x1c0
459
460 Christoph Lameter, May 30, 2007
461 Sergey Senozhatsky, October 23, 2015
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.