1======================================== 2Short users guide for the slab allocator 3======================================== 4 5The slab allocator includes full debugging support (when built with 6CONFIG_SLUB_DEBUG=y) but it is off by default (unless built with 7CONFIG_SLUB_DEBUG_ON=y). You can enable debugging only for selected 8slabs in order to avoid an impact on overall system performance which 9may make a bug more difficult to find. 10 11In order to switch debugging on one can add an option ``slab_debug`` 12to the kernel command line. That will enable full debugging for 13all slabs. 14 15Typically one would then use the ``slabinfo`` command to get statistical 16data and perform operation on the slabs. By default ``slabinfo`` only lists 17slabs that have data in them. See "slabinfo -h" for more options when 18running the command. ``slabinfo`` can be compiled with 19:: 20 21 gcc -o slabinfo tools/mm/slabinfo.c 22 23Some of the modes of operation of ``slabinfo`` require that slub debugging 24be enabled on the command line. F.e. no tracking information will be 25available without debugging on and validation can only partially 26be performed if debugging was not switched on. 27 28Some more sophisticated uses of slab_debug: 29------------------------------------------- 30 31Parameters may be given to ``slab_debug``. If none is specified then full 32debugging is enabled. Format: 33 34slab_debug=<Debug-Options> 35 Enable options for all slabs 36 37slab_debug=<Debug-Options>,<slab name1>,<slab name2>,... 38 Enable options only for select slabs (no spaces 39 after a comma) 40 41Multiple blocks of options for all slabs or selected slabs can be given, with 42blocks of options delimited by ';'. The last of "all slabs" blocks is applied 43to all slabs except those that match one of the "select slabs" block. Options 44of the first "select slabs" blocks that matches the slab's name are applied. 45 46Possible debug options are:: 47 48 F Sanity checks on (enables SLAB_DEBUG_CONSISTENCY_CHECKS 49 Sorry SLAB legacy issues) 50 Z Red zoning 51 P Poisoning (object and padding) 52 U User tracking (free and alloc) 53 T Trace (please only use on single slabs) 54 A Enable failslab filter mark for the cache 55 O Switch debugging off for caches that would have 56 caused higher minimum slab orders 57 - Switch all debugging off (useful if the kernel is 58 configured with CONFIG_SLUB_DEBUG_ON) 59 60F.e. in order to boot just with sanity checks and red zoning one would specify:: 61 62 slab_debug=FZ 63 64Trying to find an issue in the dentry cache? Try:: 65 66 slab_debug=,dentry 67 68to only enable debugging on the dentry cache. You may use an asterisk at the 69end of the slab name, in order to cover all slabs with the same prefix. For 70example, here's how you can poison the dentry cache as well as all kmalloc 71slabs:: 72 73 slab_debug=P,kmalloc-*,dentry 74 75Red zoning and tracking may realign the slab. We can just apply sanity checks 76to the dentry cache with:: 77 78 slab_debug=F,dentry 79 80Debugging options may require the minimum possible slab order to increase as 81a result of storing the metadata (for example, caches with PAGE_SIZE object 82sizes). This has a higher likelihood of resulting in slab allocation errors 83in low memory situations or if there's high fragmentation of memory. To 84switch off debugging for such caches by default, use:: 85 86 slab_debug=O 87 88You can apply different options to different list of slab names, using blocks 89of options. This will enable red zoning for dentry and user tracking for 90kmalloc. All other slabs will not get any debugging enabled:: 91 92 slab_debug=Z,dentry;U,kmalloc-* 93 94You can also enable options (e.g. sanity checks and poisoning) for all caches 95except some that are deemed too performance critical and don't need to be 96debugged by specifying global debug options followed by a list of slab names 97with "-" as options:: 98 99 slab_debug=FZ;-,zs_handle,zspage 100 101The state of each debug option for a slab can be found in the respective files 102under:: 103 104 /sys/kernel/slab/<slab name>/ 105 106If the file contains 1, the option is enabled, 0 means disabled. The debug 107options from the ``slab_debug`` parameter translate to the following files:: 108 109 F sanity_checks 110 Z red_zone 111 P poison 112 U store_user 113 T trace 114 A failslab 115 116failslab file is writable, so writing 1 or 0 will enable or disable 117the option at runtime. Write returns -EINVAL if cache is an alias. 118Careful with tracing: It may spew out lots of information and never stop if 119used on the wrong slab. 120 121Slab merging 122============ 123 124If no debug options are specified then SLUB may merge similar slabs together 125in order to reduce overhead and increase cache hotness of objects. 126``slabinfo -a`` displays which slabs were merged together. 127 128Slab validation 129=============== 130 131SLUB can validate all object if the kernel was booted with slab_debug. In 132order to do so you must have the ``slabinfo`` tool. Then you can do 133:: 134 135 slabinfo -v 136 137which will test all objects. Output will be generated to the syslog. 138 139This also works in a more limited way if boot was without slab debug. 140In that case ``slabinfo -v`` simply tests all reachable objects. Usually 141these are in the cpu slabs and the partial slabs. Full slabs are not 142tracked by SLUB in a non debug situation. 143 144Getting more performance 145======================== 146 147To some degree SLUB's performance is limited by the need to take the 148list_lock once in a while to deal with partial slabs. That overhead is 149governed by the order of the allocation for each slab. The allocations 150can be influenced by kernel parameters: 151 152.. slab_min_objects=x (default: automatically scaled by number of cpus) 153.. slab_min_order=x (default 0) 154.. slab_max_order=x (default 3 (PAGE_ALLOC_COSTLY_ORDER)) 155 156``slab_min_objects`` 157 allows to specify how many objects must at least fit into one 158 slab in order for the allocation order to be acceptable. In 159 general slub will be able to perform this number of 160 allocations on a slab without consulting centralized resources 161 (list_lock) where contention may occur. 162 163``slab_min_order`` 164 specifies a minimum order of slabs. A similar effect like 165 ``slab_min_objects``. 166 167``slab_max_order`` 168 specified the order at which ``slab_min_objects`` should no 169 longer be checked. This is useful to avoid SLUB trying to 170 generate super large order pages to fit ``slab_min_objects`` 171 of a slab cache with large object sizes into one high order 172 page. Setting command line parameter 173 ``debug_guardpage_minorder=N`` (N > 0), forces setting 174 ``slab_max_order`` to 0, what cause minimum possible order of 175 slabs allocation. 176 177``slab_strict_numa`` 178 Enables the application of memory policies on each 179 allocation. This results in more accurate placement of 180 objects which may result in the reduction of accesses 181 to remote nodes. The default is to only apply memory 182 policies at the folio level when a new folio is acquired 183 or a folio is retrieved from the lists. Enabling this 184 option reduces the fastpath performance of the slab allocator. 185 186SLUB Debug output 187================= 188 189Here is a sample of slub debug output:: 190 191 ==================================================================== 192 BUG kmalloc-8: Right Redzone overwritten 193 -------------------------------------------------------------------- 194 195 INFO: 0xc90f6d28-0xc90f6d2b. First byte 0x00 instead of 0xcc 196 INFO: Slab 0xc528c530 flags=0x400000c3 inuse=61 fp=0xc90f6d58 197 INFO: Object 0xc90f6d20 @offset=3360 fp=0xc90f6d58 198 INFO: Allocated in get_modalias+0x61/0xf5 age=53 cpu=1 pid=554 199 200 Bytes b4 (0xc90f6d10): 00 00 00 00 00 00 00 00 5a 5a 5a 5a 5a 5a 5a 5a ........ZZZZZZZZ 201 Object (0xc90f6d20): 31 30 31 39 2e 30 30 35 1019.005 202 Redzone (0xc90f6d28): 00 cc cc cc . 203 Padding (0xc90f6d50): 5a 5a 5a 5a 5a 5a 5a 5a ZZZZZZZZ 204 205 [<c010523d>] dump_trace+0x63/0x1eb 206 [<c01053df>] show_trace_log_lvl+0x1a/0x2f 207 [<c010601d>] show_trace+0x12/0x14 208 [<c0106035>] dump_stack+0x16/0x18 209 [<c017e0fa>] object_err+0x143/0x14b 210 [<c017e2cc>] check_object+0x66/0x234 211 [<c017eb43>] __slab_free+0x239/0x384 212 [<c017f446>] kfree+0xa6/0xc6 213 [<c02e2335>] get_modalias+0xb9/0xf5 214 [<c02e23b7>] dmi_dev_uevent+0x27/0x3c 215 [<c027866a>] dev_uevent+0x1ad/0x1da 216 [<c0205024>] kobject_uevent_env+0x20a/0x45b 217 [<c020527f>] kobject_uevent+0xa/0xf 218 [<c02779f1>] store_uevent+0x4f/0x58 219 [<c027758e>] dev_attr_store+0x29/0x2f 220 [<c01bec4f>] sysfs_write_file+0x16e/0x19c 221 [<c0183ba7>] vfs_write+0xd1/0x15a 222 [<c01841d7>] sys_write+0x3d/0x72 223 [<c0104112>] sysenter_past_esp+0x5f/0x99 224 [<b7f7b410>] 0xb7f7b410 225 ======================= 226 227 FIX kmalloc-8: Restoring Redzone 0xc90f6d28-0xc90f6d2b=0xcc 228 229If SLUB encounters a corrupted object (full detection requires the kernel 230to be booted with slab_debug) then the following output will be dumped 231into the syslog: 232 2331. Description of the problem encountered 234 235 This will be a message in the system log starting with:: 236 237 =============================================== 238 BUG <slab cache affected>: <What went wrong> 239 ----------------------------------------------- 240 241 INFO: <corruption start>-<corruption_end> <more info> 242 INFO: Slab <address> <slab information> 243 INFO: Object <address> <object information> 244 INFO: Allocated in <kernel function> age=<jiffies since alloc> cpu=<allocated by 245 cpu> pid=<pid of the process> 246 INFO: Freed in <kernel function> age=<jiffies since free> cpu=<freed by cpu> 247 pid=<pid of the process> 248 249 (Object allocation / free information is only available if SLAB_STORE_USER is 250 set for the slab. slab_debug sets that option) 251 2522. The object contents if an object was involved. 253 254 Various types of lines can follow the BUG SLUB line: 255 256 Bytes b4 <address> : <bytes> 257 Shows a few bytes before the object where the problem was detected. 258 Can be useful if the corruption does not stop with the start of the 259 object. 260 261 Object <address> : <bytes> 262 The bytes of the object. If the object is inactive then the bytes 263 typically contain poison values. Any non-poison value shows a 264 corruption by a write after free. 265 266 Redzone <address> : <bytes> 267 The Redzone following the object. The Redzone is used to detect 268 writes after the object. All bytes should always have the same 269 value. If there is any deviation then it is due to a write after 270 the object boundary. 271 272 (Redzone information is only available if SLAB_RED_ZONE is set. 273 slab_debug sets that option) 274 275 Padding <address> : <bytes> 276 Unused data to fill up the space in order to get the next object 277 properly aligned. In the debug case we make sure that there are 278 at least 4 bytes of padding. This allows the detection of writes 279 before the object. 280 2813. A stackdump 282 283 The stackdump describes the location where the error was detected. The cause 284 of the corruption is may be more likely found by looking at the function that 285 allocated or freed the object. 286 2874. Report on how the problem was dealt with in order to ensure the continued 288 operation of the system. 289 290 These are messages in the system log beginning with:: 291 292 FIX <slab cache affected>: <corrective action taken> 293 294 In the above sample SLUB found that the Redzone of an active object has 295 been overwritten. Here a string of 8 characters was written into a slab that 296 has the length of 8 characters. However, a 8 character string needs a 297 terminating 0. That zero has overwritten the first byte of the Redzone field. 298 After reporting the details of the issue encountered the FIX SLUB message 299 tells us that SLUB has restored the Redzone to its proper value and then 300 system operations continue. 301 302Emergency operations 303==================== 304 305Minimal debugging (sanity checks alone) can be enabled by booting with:: 306 307 slab_debug=F 308 309This will be generally be enough to enable the resiliency features of slub 310which will keep the system running even if a bad kernel component will 311keep corrupting objects. This may be important for production systems. 312Performance will be impacted by the sanity checks and there will be a 313continual stream of error messages to the syslog but no additional memory 314will be used (unlike full debugging). 315 316No guarantees. The kernel component still needs to be fixed. Performance 317may be optimized further by locating the slab that experiences corruption 318and enabling debugging only for that cache 319 320I.e.:: 321 322 slab_debug=F,dentry 323 324If the corruption occurs by writing after the end of the object then it 325may be advisable to enable a Redzone to avoid corrupting the beginning 326of other objects:: 327 328 slab_debug=FZ,dentry 329 330Extended slabinfo mode and plotting 331=================================== 332 333The ``slabinfo`` tool has a special 'extended' ('-X') mode that includes: 334 - Slabcache Totals 335 - Slabs sorted by size (up to -N <num> slabs, default 1) 336 - Slabs sorted by loss (up to -N <num> slabs, default 1) 337 338Additionally, in this mode ``slabinfo`` does not dynamically scale 339sizes (G/M/K) and reports everything in bytes (this functionality is 340also available to other slabinfo modes via '-B' option) which makes 341reporting more precise and accurate. Moreover, in some sense the `-X' 342mode also simplifies the analysis of slabs' behaviour, because its 343output can be plotted using the ``slabinfo-gnuplot.sh`` script. So it 344pushes the analysis from looking through the numbers (tons of numbers) 345to something easier -- visual analysis. 346 347To generate plots: 348 349a) collect slabinfo extended records, for example:: 350 351 while [ 1 ]; do slabinfo -X >> FOO_STATS; sleep 1; done 352 353b) pass stats file(-s) to ``slabinfo-gnuplot.sh`` script:: 354 355 slabinfo-gnuplot.sh FOO_STATS [FOO_STATS2 .. FOO_STATSN] 356 357 The ``slabinfo-gnuplot.sh`` script will pre-processes the collected records 358 and generates 3 png files (and 3 pre-processing cache files) per STATS 359 file: 360 - Slabcache Totals: FOO_STATS-totals.png 361 - Slabs sorted by size: FOO_STATS-slabs-by-size.png 362 - Slabs sorted by loss: FOO_STATS-slabs-by-loss.png 363 364Another use case, when ``slabinfo-gnuplot.sh`` can be useful, is when you 365need to compare slabs' behaviour "prior to" and "after" some code 366modification. To help you out there, ``slabinfo-gnuplot.sh`` script 367can 'merge' the `Slabcache Totals` sections from different 368measurements. To visually compare N plots: 369 370a) Collect as many STATS1, STATS2, .. STATSN files as you need:: 371 372 while [ 1 ]; do slabinfo -X >> STATS<X>; sleep 1; done 373 374b) Pre-process those STATS files:: 375 376 slabinfo-gnuplot.sh STATS1 STATS2 .. STATSN 377 378c) Execute ``slabinfo-gnuplot.sh`` in '-t' mode, passing all of the 379 generated pre-processed \*-totals:: 380 381 slabinfo-gnuplot.sh -t STATS1-totals STATS2-totals .. STATSN-totals 382 383 This will produce a single plot (png file). 384 385 Plots, expectedly, can be large so some fluctuations or small spikes 386 can go unnoticed. To deal with that, ``slabinfo-gnuplot.sh`` has two 387 options to 'zoom-in'/'zoom-out': 388 389 a) ``-s %d,%d`` -- overwrites the default image width and height 390 b) ``-r %d,%d`` -- specifies a range of samples to use (for example, 391 in ``slabinfo -X >> FOO_STATS; sleep 1;`` case, using a ``-r 392 40,60`` range will plot only samples collected between 40th and 393 60th seconds). 394 395 396DebugFS files for SLUB 397====================== 398 399For more information about current state of SLUB caches with the user tracking 400debug option enabled, debugfs files are available, typically under 401/sys/kernel/debug/slab/<cache>/ (created only for caches with enabled user 402tracking). There are 2 types of these files with the following debug 403information: 404 4051. alloc_traces:: 406 407 Prints information about unique allocation traces of the currently 408 allocated objects. The output is sorted by frequency of each trace. 409 410 Information in the output: 411 Number of objects, allocating function, possible memory wastage of 412 kmalloc objects(total/per-object), minimal/average/maximal jiffies 413 since alloc, pid range of the allocating processes, cpu mask of 414 allocating cpus, numa node mask of origins of memory, and stack trace. 415 416 Example::: 417 418 338 pci_alloc_dev+0x2c/0xa0 waste=521872/1544 age=290837/291891/293509 pid=1 cpus=106 nodes=0-1 419 __kmem_cache_alloc_node+0x11f/0x4e0 420 kmalloc_trace+0x26/0xa0 421 pci_alloc_dev+0x2c/0xa0 422 pci_scan_single_device+0xd2/0x150 423 pci_scan_slot+0xf7/0x2d0 424 pci_scan_child_bus_extend+0x4e/0x360 425 acpi_pci_root_create+0x32e/0x3b0 426 pci_acpi_scan_root+0x2b9/0x2d0 427 acpi_pci_root_add.cold.11+0x110/0xb0a 428 acpi_bus_attach+0x262/0x3f0 429 device_for_each_child+0xb7/0x110 430 acpi_dev_for_each_child+0x77/0xa0 431 acpi_bus_attach+0x108/0x3f0 432 device_for_each_child+0xb7/0x110 433 acpi_dev_for_each_child+0x77/0xa0 434 acpi_bus_attach+0x108/0x3f0 435 4362. free_traces:: 437 438 Prints information about unique freeing traces of the currently allocated 439 objects. The freeing traces thus come from the previous life-cycle of the 440 objects and are reported as not available for objects allocated for the first 441 time. The output is sorted by frequency of each trace. 442 443 Information in the output: 444 Number of objects, freeing function, minimal/average/maximal jiffies since free, 445 pid range of the freeing processes, cpu mask of freeing cpus, and stack trace. 446 447 Example::: 448 449 1980 <not-available> age=4294912290 pid=0 cpus=0 450 51 acpi_ut_update_ref_count+0x6a6/0x782 age=236886/237027/237772 pid=1 cpus=1 451 kfree+0x2db/0x420 452 acpi_ut_update_ref_count+0x6a6/0x782 453 acpi_ut_update_object_reference+0x1ad/0x234 454 acpi_ut_remove_reference+0x7d/0x84 455 acpi_rs_get_prt_method_data+0x97/0xd6 456 acpi_get_irq_routing_table+0x82/0xc4 457 acpi_pci_irq_find_prt_entry+0x8e/0x2e0 458 acpi_pci_irq_lookup+0x3a/0x1e0 459 acpi_pci_irq_enable+0x77/0x240 460 pcibios_enable_device+0x39/0x40 461 do_pci_enable_device.part.0+0x5d/0xe0 462 pci_enable_device_flags+0xfc/0x120 463 pci_enable_device+0x13/0x20 464 virtio_pci_probe+0x9e/0x170 465 local_pci_probe+0x48/0x80 466 pci_device_probe+0x105/0x1c0 467 468Christoph Lameter, May 30, 2007 469Sergey Senozhatsky, October 23, 2015 470
