Based on kernel version 4.2. Page generated on 2015-09-09 12:15 EST.
1 = Transparent Hugepage Support = 2 3 == Objective == 4 5 Performance critical computing applications dealing with large memory 6 working sets are already running on top of libhugetlbfs and in turn 7 hugetlbfs. Transparent Hugepage Support is an alternative means of 8 using huge pages for the backing of virtual memory with huge pages 9 that supports the automatic promotion and demotion of page sizes and 10 without the shortcomings of hugetlbfs. 11 12 Currently it only works for anonymous memory mappings but in the 13 future it can expand over the pagecache layer starting with tmpfs. 14 15 The reason applications are running faster is because of two 16 factors. The first factor is almost completely irrelevant and it's not 17 of significant interest because it'll also have the downside of 18 requiring larger clear-page copy-page in page faults which is a 19 potentially negative effect. The first factor consists in taking a 20 single page fault for each 2M virtual region touched by userland (so 21 reducing the enter/exit kernel frequency by a 512 times factor). This 22 only matters the first time the memory is accessed for the lifetime of 23 a memory mapping. The second long lasting and much more important 24 factor will affect all subsequent accesses to the memory for the whole 25 runtime of the application. The second factor consist of two 26 components: 1) the TLB miss will run faster (especially with 27 virtualization using nested pagetables but almost always also on bare 28 metal without virtualization) and 2) a single TLB entry will be 29 mapping a much larger amount of virtual memory in turn reducing the 30 number of TLB misses. With virtualization and nested pagetables the 31 TLB can be mapped of larger size only if both KVM and the Linux guest 32 are using hugepages but a significant speedup already happens if only 33 one of the two is using hugepages just because of the fact the TLB 34 miss is going to run faster. 35 36 == Design == 37 38 - "graceful fallback": mm components which don't have transparent 39 hugepage knowledge fall back to breaking a transparent hugepage and 40 working on the regular pages and their respective regular pmd/pte 41 mappings 42 43 - if a hugepage allocation fails because of memory fragmentation, 44 regular pages should be gracefully allocated instead and mixed in 45 the same vma without any failure or significant delay and without 46 userland noticing 47 48 - if some task quits and more hugepages become available (either 49 immediately in the buddy or through the VM), guest physical memory 50 backed by regular pages should be relocated on hugepages 51 automatically (with khugepaged) 52 53 - it doesn't require memory reservation and in turn it uses hugepages 54 whenever possible (the only possible reservation here is kernelcore= 55 to avoid unmovable pages to fragment all the memory but such a tweak 56 is not specific to transparent hugepage support and it's a generic 57 feature that applies to all dynamic high order allocations in the 58 kernel) 59 60 - this initial support only offers the feature in the anonymous memory 61 regions but it'd be ideal to move it to tmpfs and the pagecache 62 later 63 64 Transparent Hugepage Support maximizes the usefulness of free memory 65 if compared to the reservation approach of hugetlbfs by allowing all 66 unused memory to be used as cache or other movable (or even unmovable 67 entities). It doesn't require reservation to prevent hugepage 68 allocation failures to be noticeable from userland. It allows paging 69 and all other advanced VM features to be available on the 70 hugepages. It requires no modifications for applications to take 71 advantage of it. 72 73 Applications however can be further optimized to take advantage of 74 this feature, like for example they've been optimized before to avoid 75 a flood of mmap system calls for every malloc(4k). Optimizing userland 76 is by far not mandatory and khugepaged already can take care of long 77 lived page allocations even for hugepage unaware applications that 78 deals with large amounts of memory. 79 80 In certain cases when hugepages are enabled system wide, application 81 may end up allocating more memory resources. An application may mmap a 82 large region but only touch 1 byte of it, in that case a 2M page might 83 be allocated instead of a 4k page for no good. This is why it's 84 possible to disable hugepages system-wide and to only have them inside 85 MADV_HUGEPAGE madvise regions. 86 87 Embedded systems should enable hugepages only inside madvise regions 88 to eliminate any risk of wasting any precious byte of memory and to 89 only run faster. 90 91 Applications that gets a lot of benefit from hugepages and that don't 92 risk to lose memory by using hugepages, should use 93 madvise(MADV_HUGEPAGE) on their critical mmapped regions. 94 95 == sysfs == 96 97 Transparent Hugepage Support can be entirely disabled (mostly for 98 debugging purposes) or only enabled inside MADV_HUGEPAGE regions (to 99 avoid the risk of consuming more memory resources) or enabled system 100 wide. This can be achieved with one of: 101 102 echo always >/sys/kernel/mm/transparent_hugepage/enabled 103 echo madvise >/sys/kernel/mm/transparent_hugepage/enabled 104 echo never >/sys/kernel/mm/transparent_hugepage/enabled 105 106 It's also possible to limit defrag efforts in the VM to generate 107 hugepages in case they're not immediately free to madvise regions or 108 to never try to defrag memory and simply fallback to regular pages 109 unless hugepages are immediately available. Clearly if we spend CPU 110 time to defrag memory, we would expect to gain even more by the fact 111 we use hugepages later instead of regular pages. This isn't always 112 guaranteed, but it may be more likely in case the allocation is for a 113 MADV_HUGEPAGE region. 114 115 echo always >/sys/kernel/mm/transparent_hugepage/defrag 116 echo madvise >/sys/kernel/mm/transparent_hugepage/defrag 117 echo never >/sys/kernel/mm/transparent_hugepage/defrag 118 119 By default kernel tries to use huge zero page on read page fault. 120 It's possible to disable huge zero page by writing 0 or enable it 121 back by writing 1: 122 123 echo 0 >/sys/kernel/mm/transparent_hugepage/use_zero_page 124 echo 1 >/sys/kernel/mm/transparent_hugepage/use_zero_page 125 126 khugepaged will be automatically started when 127 transparent_hugepage/enabled is set to "always" or "madvise, and it'll 128 be automatically shutdown if it's set to "never". 129 130 khugepaged runs usually at low frequency so while one may not want to 131 invoke defrag algorithms synchronously during the page faults, it 132 should be worth invoking defrag at least in khugepaged. However it's 133 also possible to disable defrag in khugepaged by writing 0 or enable 134 defrag in khugepaged by writing 1: 135 136 echo 0 >/sys/kernel/mm/transparent_hugepage/khugepaged/defrag 137 echo 1 >/sys/kernel/mm/transparent_hugepage/khugepaged/defrag 138 139 You can also control how many pages khugepaged should scan at each 140 pass: 141 142 /sys/kernel/mm/transparent_hugepage/khugepaged/pages_to_scan 143 144 and how many milliseconds to wait in khugepaged between each pass (you 145 can set this to 0 to run khugepaged at 100% utilization of one core): 146 147 /sys/kernel/mm/transparent_hugepage/khugepaged/scan_sleep_millisecs 148 149 and how many milliseconds to wait in khugepaged if there's an hugepage 150 allocation failure to throttle the next allocation attempt. 151 152 /sys/kernel/mm/transparent_hugepage/khugepaged/alloc_sleep_millisecs 153 154 The khugepaged progress can be seen in the number of pages collapsed: 155 156 /sys/kernel/mm/transparent_hugepage/khugepaged/pages_collapsed 157 158 for each pass: 159 160 /sys/kernel/mm/transparent_hugepage/khugepaged/full_scans 161 162 max_ptes_none specifies how many extra small pages (that are 163 not already mapped) can be allocated when collapsing a group 164 of small pages into one large page. 165 166 /sys/kernel/mm/transparent_hugepage/khugepaged/max_ptes_none 167 168 A higher value leads to use additional memory for programs. 169 A lower value leads to gain less thp performance. Value of 170 max_ptes_none can waste cpu time very little, you can 171 ignore it. 172 173 == Boot parameter == 174 175 You can change the sysfs boot time defaults of Transparent Hugepage 176 Support by passing the parameter "transparent_hugepage=always" or 177 "transparent_hugepage=madvise" or "transparent_hugepage=never" 178 (without "") to the kernel command line. 179 180 == Need of application restart == 181 182 The transparent_hugepage/enabled values only affect future 183 behavior. So to make them effective you need to restart any 184 application that could have been using hugepages. This also applies to 185 the regions registered in khugepaged. 186 187 == Monitoring usage == 188 189 The number of transparent huge pages currently used by the system is 190 available by reading the AnonHugePages field in /proc/meminfo. To 191 identify what applications are using transparent huge pages, it is 192 necessary to read /proc/PID/smaps and count the AnonHugePages fields 193 for each mapping. Note that reading the smaps file is expensive and 194 reading it frequently will incur overhead. 195 196 There are a number of counters in /proc/vmstat that may be used to 197 monitor how successfully the system is providing huge pages for use. 198 199 thp_fault_alloc is incremented every time a huge page is successfully 200 allocated to handle a page fault. This applies to both the 201 first time a page is faulted and for COW faults. 202 203 thp_collapse_alloc is incremented by khugepaged when it has found 204 a range of pages to collapse into one huge page and has 205 successfully allocated a new huge page to store the data. 206 207 thp_fault_fallback is incremented if a page fault fails to allocate 208 a huge page and instead falls back to using small pages. 209 210 thp_collapse_alloc_failed is incremented if khugepaged found a range 211 of pages that should be collapsed into one huge page but failed 212 the allocation. 213 214 thp_split is incremented every time a huge page is split into base 215 pages. This can happen for a variety of reasons but a common 216 reason is that a huge page is old and is being reclaimed. 217 218 thp_zero_page_alloc is incremented every time a huge zero page is 219 successfully allocated. It includes allocations which where 220 dropped due race with other allocation. Note, it doesn't count 221 every map of the huge zero page, only its allocation. 222 223 thp_zero_page_alloc_failed is incremented if kernel fails to allocate 224 huge zero page and falls back to using small pages. 225 226 As the system ages, allocating huge pages may be expensive as the 227 system uses memory compaction to copy data around memory to free a 228 huge page for use. There are some counters in /proc/vmstat to help 229 monitor this overhead. 230 231 compact_stall is incremented every time a process stalls to run 232 memory compaction so that a huge page is free for use. 233 234 compact_success is incremented if the system compacted memory and 235 freed a huge page for use. 236 237 compact_fail is incremented if the system tries to compact memory 238 but failed. 239 240 compact_pages_moved is incremented each time a page is moved. If 241 this value is increasing rapidly, it implies that the system 242 is copying a lot of data to satisfy the huge page allocation. 243 It is possible that the cost of copying exceeds any savings 244 from reduced TLB misses. 245 246 compact_pagemigrate_failed is incremented when the underlying mechanism 247 for moving a page failed. 248 249 compact_blocks_moved is incremented each time memory compaction examines 250 a huge page aligned range of pages. 251 252 It is possible to establish how long the stalls were using the function 253 tracer to record how long was spent in __alloc_pages_nodemask and 254 using the mm_page_alloc tracepoint to identify which allocations were 255 for huge pages. 256 257 == get_user_pages and follow_page == 258 259 get_user_pages and follow_page if run on a hugepage, will return the 260 head or tail pages as usual (exactly as they would do on 261 hugetlbfs). Most gup users will only care about the actual physical 262 address of the page and its temporary pinning to release after the I/O 263 is complete, so they won't ever notice the fact the page is huge. But 264 if any driver is going to mangle over the page structure of the tail 265 page (like for checking page->mapping or other bits that are relevant 266 for the head page and not the tail page), it should be updated to jump 267 to check head page instead (while serializing properly against 268 split_huge_page() to avoid the head and tail pages to disappear from 269 under it, see the futex code to see an example of that, hugetlbfs also 270 needed special handling in futex code for similar reasons). 271 272 NOTE: these aren't new constraints to the GUP API, and they match the 273 same constrains that applies to hugetlbfs too, so any driver capable 274 of handling GUP on hugetlbfs will also work fine on transparent 275 hugepage backed mappings. 276 277 In case you can't handle compound pages if they're returned by 278 follow_page, the FOLL_SPLIT bit can be specified as parameter to 279 follow_page, so that it will split the hugepages before returning 280 them. Migration for example passes FOLL_SPLIT as parameter to 281 follow_page because it's not hugepage aware and in fact it can't work 282 at all on hugetlbfs (but it instead works fine on transparent 283 hugepages thanks to FOLL_SPLIT). migration simply can't deal with 284 hugepages being returned (as it's not only checking the pfn of the 285 page and pinning it during the copy but it pretends to migrate the 286 memory in regular page sizes and with regular pte/pmd mappings). 287 288 == Optimizing the applications == 289 290 To be guaranteed that the kernel will map a 2M page immediately in any 291 memory region, the mmap region has to be hugepage naturally 292 aligned. posix_memalign() can provide that guarantee. 293 294 == Hugetlbfs == 295 296 You can use hugetlbfs on a kernel that has transparent hugepage 297 support enabled just fine as always. No difference can be noted in 298 hugetlbfs other than there will be less overall fragmentation. All 299 usual features belonging to hugetlbfs are preserved and 300 unaffected. libhugetlbfs will also work fine as usual. 301 302 == Graceful fallback == 303 304 Code walking pagetables but unware about huge pmds can simply call 305 split_huge_page_pmd(vma, addr, pmd) where the pmd is the one returned by 306 pmd_offset. It's trivial to make the code transparent hugepage aware 307 by just grepping for "pmd_offset" and adding split_huge_page_pmd where 308 missing after pmd_offset returns the pmd. Thanks to the graceful 309 fallback design, with a one liner change, you can avoid to write 310 hundred if not thousand of lines of complex code to make your code 311 hugepage aware. 312 313 If you're not walking pagetables but you run into a physical hugepage 314 but you can't handle it natively in your code, you can split it by 315 calling split_huge_page(page). This is what the Linux VM does before 316 it tries to swapout the hugepage for example. 317 318 Example to make mremap.c transparent hugepage aware with a one liner 319 change: 320 321 diff --git a/mm/mremap.c b/mm/mremap.c 322 --- a/mm/mremap.c 323 +++ b/mm/mremap.c 324 @@ -41,6 +41,7 @@ static pmd_t *get_old_pmd(struct mm_stru 325 return NULL; 326 327 pmd = pmd_offset(pud, addr); 328 + split_huge_page_pmd(vma, addr, pmd); 329 if (pmd_none_or_clear_bad(pmd)) 330 return NULL; 331 332 == Locking in hugepage aware code == 333 334 We want as much code as possible hugepage aware, as calling 335 split_huge_page() or split_huge_page_pmd() has a cost. 336 337 To make pagetable walks huge pmd aware, all you need to do is to call 338 pmd_trans_huge() on the pmd returned by pmd_offset. You must hold the 339 mmap_sem in read (or write) mode to be sure an huge pmd cannot be 340 created from under you by khugepaged (khugepaged collapse_huge_page 341 takes the mmap_sem in write mode in addition to the anon_vma lock). If 342 pmd_trans_huge returns false, you just fallback in the old code 343 paths. If instead pmd_trans_huge returns true, you have to take the 344 mm->page_table_lock and re-run pmd_trans_huge. Taking the 345 page_table_lock will prevent the huge pmd to be converted into a 346 regular pmd from under you (split_huge_page can run in parallel to the 347 pagetable walk). If the second pmd_trans_huge returns false, you 348 should just drop the page_table_lock and fallback to the old code as 349 before. Otherwise you should run pmd_trans_splitting on the pmd. In 350 case pmd_trans_splitting returns true, it means split_huge_page is 351 already in the middle of splitting the page. So if pmd_trans_splitting 352 returns true it's enough to drop the page_table_lock and call 353 wait_split_huge_page and then fallback the old code paths. You are 354 guaranteed by the time wait_split_huge_page returns, the pmd isn't 355 huge anymore. If pmd_trans_splitting returns false, you can proceed to 356 process the huge pmd and the hugepage natively. Once finished you can 357 drop the page_table_lock. 358 359 == compound_lock, get_user_pages and put_page == 360 361 split_huge_page internally has to distribute the refcounts in the head 362 page to the tail pages before clearing all PG_head/tail bits from the 363 page structures. It can do that easily for refcounts taken by huge pmd 364 mappings. But the GUI API as created by hugetlbfs (that returns head 365 and tail pages if running get_user_pages on an address backed by any 366 hugepage), requires the refcount to be accounted on the tail pages and 367 not only in the head pages, if we want to be able to run 368 split_huge_page while there are gup pins established on any tail 369 page. Failure to be able to run split_huge_page if there's any gup pin 370 on any tail page, would mean having to split all hugepages upfront in 371 get_user_pages which is unacceptable as too many gup users are 372 performance critical and they must work natively on hugepages like 373 they work natively on hugetlbfs already (hugetlbfs is simpler because 374 hugetlbfs pages cannot be split so there wouldn't be requirement of 375 accounting the pins on the tail pages for hugetlbfs). If we wouldn't 376 account the gup refcounts on the tail pages during gup, we won't know 377 anymore which tail page is pinned by gup and which is not while we run 378 split_huge_page. But we still have to add the gup pin to the head page 379 too, to know when we can free the compound page in case it's never 380 split during its lifetime. That requires changing not just 381 get_page, but put_page as well so that when put_page runs on a tail 382 page (and only on a tail page) it will find its respective head page, 383 and then it will decrease the head page refcount in addition to the 384 tail page refcount. To obtain a head page reliably and to decrease its 385 refcount without race conditions, put_page has to serialize against 386 __split_huge_page_refcount using a special per-page lock called 387 compound_lock.