1============================ 2Transparent Hugepage Support 3============================ 4 5Objective 6========= 7 8Performance critical computing applications dealing with large memory 9working sets are already running on top of libhugetlbfs and in turn 10hugetlbfs. Transparent HugePage Support (THP) is an alternative mean of 11using huge pages for the backing of virtual memory with huge pages 12that supports the automatic promotion and demotion of page sizes and 13without the shortcomings of hugetlbfs. 14 15Currently THP only works for anonymous memory mappings and tmpfs/shmem. 16But in the future it can expand to other filesystems. 17 18.. note:: 19 in the examples below we presume that the basic page size is 4K and 20 the huge page size is 2M, although the actual numbers may vary 21 depending on the CPU architecture. 22 23The reason applications are running faster is because of two 24factors. The first factor is almost completely irrelevant and it's not 25of significant interest because it'll also have the downside of 26requiring larger clear-page copy-page in page faults which is a 27potentially negative effect. The first factor consists in taking a 28single page fault for each 2M virtual region touched by userland (so 29reducing the enter/exit kernel frequency by a 512 times factor). This 30only matters the first time the memory is accessed for the lifetime of 31a memory mapping. The second long lasting and much more important 32factor will affect all subsequent accesses to the memory for the whole 33runtime of the application. The second factor consist of two 34components: 35 361) the TLB miss will run faster (especially with virtualization using 37 nested pagetables but almost always also on bare metal without 38 virtualization) 39 402) a single TLB entry will be mapping a much larger amount of virtual 41 memory in turn reducing the number of TLB misses. With 42 virtualization and nested pagetables the TLB can be mapped of 43 larger size only if both KVM and the Linux guest are using 44 hugepages but a significant speedup already happens if only one of 45 the two is using hugepages just because of the fact the TLB miss is 46 going to run faster. 47 48Modern kernels support "multi-size THP" (mTHP), which introduces the 49ability to allocate memory in blocks that are bigger than a base page 50but smaller than traditional PMD-size (as described above), in 51increments of a power-of-2 number of pages. mTHP can back anonymous 52memory (for example 16K, 32K, 64K, etc). These THPs continue to be 53PTE-mapped, but in many cases can still provide similar benefits to 54those outlined above: Page faults are significantly reduced (by a 55factor of e.g. 4, 8, 16, etc), but latency spikes are much less 56prominent because the size of each page isn't as huge as the PMD-sized 57variant and there is less memory to clear in each page fault. Some 58architectures also employ TLB compression mechanisms to squeeze more 59entries in when a set of PTEs are virtually and physically contiguous 60and approporiately aligned. In this case, TLB misses will occur less 61often. 62 63THP can be enabled system wide or restricted to certain tasks or even 64memory ranges inside task's address space. Unless THP is completely 65disabled, there is ``khugepaged`` daemon that scans memory and 66collapses sequences of basic pages into PMD-sized huge pages. 67 68The THP behaviour is controlled via :ref:`sysfs <thp_sysfs>` 69interface and using madvise(2) and prctl(2) system calls. 70 71Transparent Hugepage Support maximizes the usefulness of free memory 72if compared to the reservation approach of hugetlbfs by allowing all 73unused memory to be used as cache or other movable (or even unmovable 74entities). It doesn't require reservation to prevent hugepage 75allocation failures to be noticeable from userland. It allows paging 76and all other advanced VM features to be available on the 77hugepages. It requires no modifications for applications to take 78advantage of it. 79 80Applications however can be further optimized to take advantage of 81this feature, like for example they've been optimized before to avoid 82a flood of mmap system calls for every malloc(4k). Optimizing userland 83is by far not mandatory and khugepaged already can take care of long 84lived page allocations even for hugepage unaware applications that 85deals with large amounts of memory. 86 87In certain cases when hugepages are enabled system wide, application 88may end up allocating more memory resources. An application may mmap a 89large region but only touch 1 byte of it, in that case a 2M page might 90be allocated instead of a 4k page for no good. This is why it's 91possible to disable hugepages system-wide and to only have them inside 92MADV_HUGEPAGE madvise regions. 93 94Embedded systems should enable hugepages only inside madvise regions 95to eliminate any risk of wasting any precious byte of memory and to 96only run faster. 97 98Applications that gets a lot of benefit from hugepages and that don't 99risk to lose memory by using hugepages, should use 100madvise(MADV_HUGEPAGE) on their critical mmapped regions. 101 102.. _thp_sysfs: 103 104sysfs 105===== 106 107Global THP controls 108------------------- 109 110Transparent Hugepage Support for anonymous memory can be entirely disabled 111(mostly for debugging purposes) or only enabled inside MADV_HUGEPAGE 112regions (to avoid the risk of consuming more memory resources) or enabled 113system wide. This can be achieved per-supported-THP-size with one of:: 114 115 echo always >/sys/kernel/mm/transparent_hugepage/hugepages-<size>kB/enabled 116 echo madvise >/sys/kernel/mm/transparent_hugepage/hugepages-<size>kB/enabled 117 echo never >/sys/kernel/mm/transparent_hugepage/hugepages-<size>kB/enabled 118 119where <size> is the hugepage size being addressed, the available sizes 120for which vary by system. 121 122For example:: 123 124 echo always >/sys/kernel/mm/transparent_hugepage/hugepages-2048kB/enabled 125 126Alternatively it is possible to specify that a given hugepage size 127will inherit the top-level "enabled" value:: 128 129 echo inherit >/sys/kernel/mm/transparent_hugepage/hugepages-<size>kB/enabled 130 131For example:: 132 133 echo inherit >/sys/kernel/mm/transparent_hugepage/hugepages-2048kB/enabled 134 135The top-level setting (for use with "inherit") can be set by issuing 136one of the following commands:: 137 138 echo always >/sys/kernel/mm/transparent_hugepage/enabled 139 echo madvise >/sys/kernel/mm/transparent_hugepage/enabled 140 echo never >/sys/kernel/mm/transparent_hugepage/enabled 141 142By default, PMD-sized hugepages have enabled="inherit" and all other 143hugepage sizes have enabled="never". If enabling multiple hugepage 144sizes, the kernel will select the most appropriate enabled size for a 145given allocation. 146 147It's also possible to limit defrag efforts in the VM to generate 148anonymous hugepages in case they're not immediately free to madvise 149regions or to never try to defrag memory and simply fallback to regular 150pages unless hugepages are immediately available. Clearly if we spend CPU 151time to defrag memory, we would expect to gain even more by the fact we 152use hugepages later instead of regular pages. This isn't always 153guaranteed, but it may be more likely in case the allocation is for a 154MADV_HUGEPAGE region. 155 156:: 157 158 echo always >/sys/kernel/mm/transparent_hugepage/defrag 159 echo defer >/sys/kernel/mm/transparent_hugepage/defrag 160 echo defer+madvise >/sys/kernel/mm/transparent_hugepage/defrag 161 echo madvise >/sys/kernel/mm/transparent_hugepage/defrag 162 echo never >/sys/kernel/mm/transparent_hugepage/defrag 163 164always 165 means that an application requesting THP will stall on 166 allocation failure and directly reclaim pages and compact 167 memory in an effort to allocate a THP immediately. This may be 168 desirable for virtual machines that benefit heavily from THP 169 use and are willing to delay the VM start to utilise them. 170 171defer 172 means that an application will wake kswapd in the background 173 to reclaim pages and wake kcompactd to compact memory so that 174 THP is available in the near future. It's the responsibility 175 of khugepaged to then install the THP pages later. 176 177defer+madvise 178 will enter direct reclaim and compaction like ``always``, but 179 only for regions that have used madvise(MADV_HUGEPAGE); all 180 other regions will wake kswapd in the background to reclaim 181 pages and wake kcompactd to compact memory so that THP is 182 available in the near future. 183 184madvise 185 will enter direct reclaim like ``always`` but only for regions 186 that are have used madvise(MADV_HUGEPAGE). This is the default 187 behaviour. 188 189never 190 should be self-explanatory. 191 192By default kernel tries to use huge, PMD-mappable zero page on read 193page fault to anonymous mapping. It's possible to disable huge zero 194page by writing 0 or enable it back by writing 1:: 195 196 echo 0 >/sys/kernel/mm/transparent_hugepage/use_zero_page 197 echo 1 >/sys/kernel/mm/transparent_hugepage/use_zero_page 198 199Some userspace (such as a test program, or an optimized memory 200allocation library) may want to know the size (in bytes) of a 201PMD-mappable transparent hugepage:: 202 203 cat /sys/kernel/mm/transparent_hugepage/hpage_pmd_size 204 205khugepaged will be automatically started when PMD-sized THP is enabled 206(either of the per-size anon control or the top-level control are set 207to "always" or "madvise"), and it'll be automatically shutdown when 208PMD-sized THP is disabled (when both the per-size anon control and the 209top-level control are "never") 210 211Khugepaged controls 212------------------- 213 214.. note:: 215 khugepaged currently only searches for opportunities to collapse to 216 PMD-sized THP and no attempt is made to collapse to other THP 217 sizes. 218 219khugepaged runs usually at low frequency so while one may not want to 220invoke defrag algorithms synchronously during the page faults, it 221should be worth invoking defrag at least in khugepaged. However it's 222also possible to disable defrag in khugepaged by writing 0 or enable 223defrag in khugepaged by writing 1:: 224 225 echo 0 >/sys/kernel/mm/transparent_hugepage/khugepaged/defrag 226 echo 1 >/sys/kernel/mm/transparent_hugepage/khugepaged/defrag 227 228You can also control how many pages khugepaged should scan at each 229pass:: 230 231 /sys/kernel/mm/transparent_hugepage/khugepaged/pages_to_scan 232 233and how many milliseconds to wait in khugepaged between each pass (you 234can set this to 0 to run khugepaged at 100% utilization of one core):: 235 236 /sys/kernel/mm/transparent_hugepage/khugepaged/scan_sleep_millisecs 237 238and how many milliseconds to wait in khugepaged if there's an hugepage 239allocation failure to throttle the next allocation attempt:: 240 241 /sys/kernel/mm/transparent_hugepage/khugepaged/alloc_sleep_millisecs 242 243The khugepaged progress can be seen in the number of pages collapsed (note 244that this counter may not be an exact count of the number of pages 245collapsed, since "collapsed" could mean multiple things: (1) A PTE mapping 246being replaced by a PMD mapping, or (2) All 4K physical pages replaced by 247one 2M hugepage. Each may happen independently, or together, depending on 248the type of memory and the failures that occur. As such, this value should 249be interpreted roughly as a sign of progress, and counters in /proc/vmstat 250consulted for more accurate accounting):: 251 252 /sys/kernel/mm/transparent_hugepage/khugepaged/pages_collapsed 253 254for each pass:: 255 256 /sys/kernel/mm/transparent_hugepage/khugepaged/full_scans 257 258``max_ptes_none`` specifies how many extra small pages (that are 259not already mapped) can be allocated when collapsing a group 260of small pages into one large page:: 261 262 /sys/kernel/mm/transparent_hugepage/khugepaged/max_ptes_none 263 264A higher value leads to use additional memory for programs. 265A lower value leads to gain less thp performance. Value of 266max_ptes_none can waste cpu time very little, you can 267ignore it. 268 269``max_ptes_swap`` specifies how many pages can be brought in from 270swap when collapsing a group of pages into a transparent huge page:: 271 272 /sys/kernel/mm/transparent_hugepage/khugepaged/max_ptes_swap 273 274A higher value can cause excessive swap IO and waste 275memory. A lower value can prevent THPs from being 276collapsed, resulting fewer pages being collapsed into 277THPs, and lower memory access performance. 278 279``max_ptes_shared`` specifies how many pages can be shared across multiple 280processes. khugepaged might treat pages of THPs as shared if any page of 281that THP is shared. Exceeding the number would block the collapse:: 282 283 /sys/kernel/mm/transparent_hugepage/khugepaged/max_ptes_shared 284 285A higher value may increase memory footprint for some workloads. 286 287Boot parameter 288============== 289 290You can change the sysfs boot time defaults of Transparent Hugepage 291Support by passing the parameter ``transparent_hugepage=always`` or 292``transparent_hugepage=madvise`` or ``transparent_hugepage=never`` 293to the kernel command line. 294 295Hugepages in tmpfs/shmem 296======================== 297 298You can control hugepage allocation policy in tmpfs with mount option 299``huge=``. It can have following values: 300 301always 302 Attempt to allocate huge pages every time we need a new page; 303 304never 305 Do not allocate huge pages; 306 307within_size 308 Only allocate huge page if it will be fully within i_size. 309 Also respect fadvise()/madvise() hints; 310 311advise 312 Only allocate huge pages if requested with fadvise()/madvise(); 313 314The default policy is ``never``. 315 316``mount -o remount,huge= /mountpoint`` works fine after mount: remounting 317``huge=never`` will not attempt to break up huge pages at all, just stop more 318from being allocated. 319 320There's also sysfs knob to control hugepage allocation policy for internal 321shmem mount: /sys/kernel/mm/transparent_hugepage/shmem_enabled. The mount 322is used for SysV SHM, memfds, shared anonymous mmaps (of /dev/zero or 323MAP_ANONYMOUS), GPU drivers' DRM objects, Ashmem. 324 325In addition to policies listed above, shmem_enabled allows two further 326values: 327 328deny 329 For use in emergencies, to force the huge option off from 330 all mounts; 331force 332 Force the huge option on for all - very useful for testing; 333 334Shmem can also use "multi-size THP" (mTHP) by adding a new sysfs knob to 335control mTHP allocation: 336'/sys/kernel/mm/transparent_hugepage/hugepages-<size>kB/shmem_enabled', 337and its value for each mTHP is essentially consistent with the global 338setting. An 'inherit' option is added to ensure compatibility with these 339global settings. Conversely, the options 'force' and 'deny' are dropped, 340which are rather testing artifacts from the old ages. 341 342always 343 Attempt to allocate <size> huge pages every time we need a new page; 344 345inherit 346 Inherit the top-level "shmem_enabled" value. By default, PMD-sized hugepages 347 have enabled="inherit" and all other hugepage sizes have enabled="never"; 348 349never 350 Do not allocate <size> huge pages; 351 352within_size 353 Only allocate <size> huge page if it will be fully within i_size. 354 Also respect fadvise()/madvise() hints; 355 356advise 357 Only allocate <size> huge pages if requested with fadvise()/madvise(); 358 359Need of application restart 360=========================== 361 362The transparent_hugepage/enabled and 363transparent_hugepage/hugepages-<size>kB/enabled values and tmpfs mount 364option only affect future behavior. So to make them effective you need 365to restart any application that could have been using hugepages. This 366also applies to the regions registered in khugepaged. 367 368Monitoring usage 369================ 370 371The number of PMD-sized anonymous transparent huge pages currently used by the 372system is available by reading the AnonHugePages field in ``/proc/meminfo``. 373To identify what applications are using PMD-sized anonymous transparent huge 374pages, it is necessary to read ``/proc/PID/smaps`` and count the AnonHugePages 375fields for each mapping. (Note that AnonHugePages only applies to traditional 376PMD-sized THP for historical reasons and should have been called 377AnonHugePmdMapped). 378 379The number of file transparent huge pages mapped to userspace is available 380by reading ShmemPmdMapped and ShmemHugePages fields in ``/proc/meminfo``. 381To identify what applications are mapping file transparent huge pages, it 382is necessary to read ``/proc/PID/smaps`` and count the FileHugeMapped fields 383for each mapping. 384 385Note that reading the smaps file is expensive and reading it 386frequently will incur overhead. 387 388There are a number of counters in ``/proc/vmstat`` that may be used to 389monitor how successfully the system is providing huge pages for use. 390 391thp_fault_alloc 392 is incremented every time a huge page is successfully 393 allocated and charged to handle a page fault. 394 395thp_collapse_alloc 396 is incremented by khugepaged when it has found 397 a range of pages to collapse into one huge page and has 398 successfully allocated a new huge page to store the data. 399 400thp_fault_fallback 401 is incremented if a page fault fails to allocate or charge 402 a huge page and instead falls back to using small pages. 403 404thp_fault_fallback_charge 405 is incremented if a page fault fails to charge a huge page and 406 instead falls back to using small pages even though the 407 allocation was successful. 408 409thp_collapse_alloc_failed 410 is incremented if khugepaged found a range 411 of pages that should be collapsed into one huge page but failed 412 the allocation. 413 414thp_file_alloc 415 is incremented every time a shmem huge page is successfully 416 allocated (Note that despite being named after "file", the counter 417 measures only shmem). 418 419thp_file_fallback 420 is incremented if a shmem huge page is attempted to be allocated 421 but fails and instead falls back to using small pages. (Note that 422 despite being named after "file", the counter measures only shmem). 423 424thp_file_fallback_charge 425 is incremented if a shmem huge page cannot be charged and instead 426 falls back to using small pages even though the allocation was 427 successful. (Note that despite being named after "file", the 428 counter measures only shmem). 429 430thp_file_mapped 431 is incremented every time a file or shmem huge page is mapped into 432 user address space. 433 434thp_split_page 435 is incremented every time a huge page is split into base 436 pages. This can happen for a variety of reasons but a common 437 reason is that a huge page is old and is being reclaimed. 438 This action implies splitting all PMD the page mapped with. 439 440thp_split_page_failed 441 is incremented if kernel fails to split huge 442 page. This can happen if the page was pinned by somebody. 443 444thp_deferred_split_page 445 is incremented when a huge page is put onto split 446 queue. This happens when a huge page is partially unmapped and 447 splitting it would free up some memory. Pages on split queue are 448 going to be split under memory pressure. 449 450thp_split_pmd 451 is incremented every time a PMD split into table of PTEs. 452 This can happen, for instance, when application calls mprotect() or 453 munmap() on part of huge page. It doesn't split huge page, only 454 page table entry. 455 456thp_zero_page_alloc 457 is incremented every time a huge zero page used for thp is 458 successfully allocated. Note, it doesn't count every map of 459 the huge zero page, only its allocation. 460 461thp_zero_page_alloc_failed 462 is incremented if kernel fails to allocate 463 huge zero page and falls back to using small pages. 464 465thp_swpout 466 is incremented every time a huge page is swapout in one 467 piece without splitting. 468 469thp_swpout_fallback 470 is incremented if a huge page has to be split before swapout. 471 Usually because failed to allocate some continuous swap space 472 for the huge page. 473 474In /sys/kernel/mm/transparent_hugepage/hugepages-<size>kB/stats, There are 475also individual counters for each huge page size, which can be utilized to 476monitor the system's effectiveness in providing huge pages for usage. Each 477counter has its own corresponding file. 478 479anon_fault_alloc 480 is incremented every time a huge page is successfully 481 allocated and charged to handle a page fault. 482 483anon_fault_fallback 484 is incremented if a page fault fails to allocate or charge 485 a huge page and instead falls back to using huge pages with 486 lower orders or small pages. 487 488anon_fault_fallback_charge 489 is incremented if a page fault fails to charge a huge page and 490 instead falls back to using huge pages with lower orders or 491 small pages even though the allocation was successful. 492 493swpout 494 is incremented every time a huge page is swapped out in one 495 piece without splitting. 496 497swpout_fallback 498 is incremented if a huge page has to be split before swapout. 499 Usually because failed to allocate some continuous swap space 500 for the huge page. 501 502shmem_alloc 503 is incremented every time a shmem huge page is successfully 504 allocated. 505 506shmem_fallback 507 is incremented if a shmem huge page is attempted to be allocated 508 but fails and instead falls back to using small pages. 509 510shmem_fallback_charge 511 is incremented if a shmem huge page cannot be charged and instead 512 falls back to using small pages even though the allocation was 513 successful. 514 515split 516 is incremented every time a huge page is successfully split into 517 smaller orders. This can happen for a variety of reasons but a 518 common reason is that a huge page is old and is being reclaimed. 519 520split_failed 521 is incremented if kernel fails to split huge 522 page. This can happen if the page was pinned by somebody. 523 524split_deferred 525 is incremented when a huge page is put onto split queue. 526 This happens when a huge page is partially unmapped and splitting 527 it would free up some memory. Pages on split queue are going to 528 be split under memory pressure, if splitting is possible. 529 530As the system ages, allocating huge pages may be expensive as the 531system uses memory compaction to copy data around memory to free a 532huge page for use. There are some counters in ``/proc/vmstat`` to help 533monitor this overhead. 534 535compact_stall 536 is incremented every time a process stalls to run 537 memory compaction so that a huge page is free for use. 538 539compact_success 540 is incremented if the system compacted memory and 541 freed a huge page for use. 542 543compact_fail 544 is incremented if the system tries to compact memory 545 but failed. 546 547It is possible to establish how long the stalls were using the function 548tracer to record how long was spent in __alloc_pages() and 549using the mm_page_alloc tracepoint to identify which allocations were 550for huge pages. 551 552Optimizing the applications 553=========================== 554 555To be guaranteed that the kernel will map a THP immediately in any 556memory region, the mmap region has to be hugepage naturally 557aligned. posix_memalign() can provide that guarantee. 558 559Hugetlbfs 560========= 561 562You can use hugetlbfs on a kernel that has transparent hugepage 563support enabled just fine as always. No difference can be noted in 564hugetlbfs other than there will be less overall fragmentation. All 565usual features belonging to hugetlbfs are preserved and 566unaffected. libhugetlbfs will also work fine as usual. 567