Based on kernel version 3.9. Page generated on 2013-05-02 23:17 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/khugepaged/use_zero_page 124 echo 1 >/sys/kernel/mm/transparent_hugepage/khugepaged/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 == Boot parameter == 163 164 You can change the sysfs boot time defaults of Transparent Hugepage 165 Support by passing the parameter "transparent_hugepage=always" or 166 "transparent_hugepage=madvise" or "transparent_hugepage=never" 167 (without "") to the kernel command line. 168 169 == Need of application restart == 170 171 The transparent_hugepage/enabled values only affect future 172 behavior. So to make them effective you need to restart any 173 application that could have been using hugepages. This also applies to 174 the regions registered in khugepaged. 175 176 == Monitoring usage == 177 178 The number of transparent huge pages currently used by the system is 179 available by reading the AnonHugePages field in /proc/meminfo. To 180 identify what applications are using transparent huge pages, it is 181 necessary to read /proc/PID/smaps and count the AnonHugePages fields 182 for each mapping. Note that reading the smaps file is expensive and 183 reading it frequently will incur overhead. 184 185 There are a number of counters in /proc/vmstat that may be used to 186 monitor how successfully the system is providing huge pages for use. 187 188 thp_fault_alloc is incremented every time a huge page is successfully 189 allocated to handle a page fault. This applies to both the 190 first time a page is faulted and for COW faults. 191 192 thp_collapse_alloc is incremented by khugepaged when it has found 193 a range of pages to collapse into one huge page and has 194 successfully allocated a new huge page to store the data. 195 196 thp_fault_fallback is incremented if a page fault fails to allocate 197 a huge page and instead falls back to using small pages. 198 199 thp_collapse_alloc_failed is incremented if khugepaged found a range 200 of pages that should be collapsed into one huge page but failed 201 the allocation. 202 203 thp_split is incremented every time a huge page is split into base 204 pages. This can happen for a variety of reasons but a common 205 reason is that a huge page is old and is being reclaimed. 206 207 thp_zero_page_alloc is incremented every time a huge zero page is 208 successfully allocated. It includes allocations which where 209 dropped due race with other allocation. Note, it doesn't count 210 every map of the huge zero page, only its allocation. 211 212 thp_zero_page_alloc_failed is incremented if kernel fails to allocate 213 huge zero page and falls back to using small pages. 214 215 As the system ages, allocating huge pages may be expensive as the 216 system uses memory compaction to copy data around memory to free a 217 huge page for use. There are some counters in /proc/vmstat to help 218 monitor this overhead. 219 220 compact_stall is incremented every time a process stalls to run 221 memory compaction so that a huge page is free for use. 222 223 compact_success is incremented if the system compacted memory and 224 freed a huge page for use. 225 226 compact_fail is incremented if the system tries to compact memory 227 but failed. 228 229 compact_pages_moved is incremented each time a page is moved. If 230 this value is increasing rapidly, it implies that the system 231 is copying a lot of data to satisfy the huge page allocation. 232 It is possible that the cost of copying exceeds any savings 233 from reduced TLB misses. 234 235 compact_pagemigrate_failed is incremented when the underlying mechanism 236 for moving a page failed. 237 238 compact_blocks_moved is incremented each time memory compaction examines 239 a huge page aligned range of pages. 240 241 It is possible to establish how long the stalls were using the function 242 tracer to record how long was spent in __alloc_pages_nodemask and 243 using the mm_page_alloc tracepoint to identify which allocations were 244 for huge pages. 245 246 == get_user_pages and follow_page == 247 248 get_user_pages and follow_page if run on a hugepage, will return the 249 head or tail pages as usual (exactly as they would do on 250 hugetlbfs). Most gup users will only care about the actual physical 251 address of the page and its temporary pinning to release after the I/O 252 is complete, so they won't ever notice the fact the page is huge. But 253 if any driver is going to mangle over the page structure of the tail 254 page (like for checking page->mapping or other bits that are relevant 255 for the head page and not the tail page), it should be updated to jump 256 to check head page instead (while serializing properly against 257 split_huge_page() to avoid the head and tail pages to disappear from 258 under it, see the futex code to see an example of that, hugetlbfs also 259 needed special handling in futex code for similar reasons). 260 261 NOTE: these aren't new constraints to the GUP API, and they match the 262 same constrains that applies to hugetlbfs too, so any driver capable 263 of handling GUP on hugetlbfs will also work fine on transparent 264 hugepage backed mappings. 265 266 In case you can't handle compound pages if they're returned by 267 follow_page, the FOLL_SPLIT bit can be specified as parameter to 268 follow_page, so that it will split the hugepages before returning 269 them. Migration for example passes FOLL_SPLIT as parameter to 270 follow_page because it's not hugepage aware and in fact it can't work 271 at all on hugetlbfs (but it instead works fine on transparent 272 hugepages thanks to FOLL_SPLIT). migration simply can't deal with 273 hugepages being returned (as it's not only checking the pfn of the 274 page and pinning it during the copy but it pretends to migrate the 275 memory in regular page sizes and with regular pte/pmd mappings). 276 277 == Optimizing the applications == 278 279 To be guaranteed that the kernel will map a 2M page immediately in any 280 memory region, the mmap region has to be hugepage naturally 281 aligned. posix_memalign() can provide that guarantee. 282 283 == Hugetlbfs == 284 285 You can use hugetlbfs on a kernel that has transparent hugepage 286 support enabled just fine as always. No difference can be noted in 287 hugetlbfs other than there will be less overall fragmentation. All 288 usual features belonging to hugetlbfs are preserved and 289 unaffected. libhugetlbfs will also work fine as usual. 290 291 == Graceful fallback == 292 293 Code walking pagetables but unware about huge pmds can simply call 294 split_huge_page_pmd(vma, addr, pmd) where the pmd is the one returned by 295 pmd_offset. It's trivial to make the code transparent hugepage aware 296 by just grepping for "pmd_offset" and adding split_huge_page_pmd where 297 missing after pmd_offset returns the pmd. Thanks to the graceful 298 fallback design, with a one liner change, you can avoid to write 299 hundred if not thousand of lines of complex code to make your code 300 hugepage aware. 301 302 If you're not walking pagetables but you run into a physical hugepage 303 but you can't handle it natively in your code, you can split it by 304 calling split_huge_page(page). This is what the Linux VM does before 305 it tries to swapout the hugepage for example. 306 307 Example to make mremap.c transparent hugepage aware with a one liner 308 change: 309 310 diff --git a/mm/mremap.c b/mm/mremap.c 311 --- a/mm/mremap.c 312 +++ b/mm/mremap.c 313 @@ -41,6 +41,7 @@ static pmd_t *get_old_pmd(struct mm_stru 314 return NULL; 315 316 pmd = pmd_offset(pud, addr); 317 + split_huge_page_pmd(vma, addr, pmd); 318 if (pmd_none_or_clear_bad(pmd)) 319 return NULL; 320 321 == Locking in hugepage aware code == 322 323 We want as much code as possible hugepage aware, as calling 324 split_huge_page() or split_huge_page_pmd() has a cost. 325 326 To make pagetable walks huge pmd aware, all you need to do is to call 327 pmd_trans_huge() on the pmd returned by pmd_offset. You must hold the 328 mmap_sem in read (or write) mode to be sure an huge pmd cannot be 329 created from under you by khugepaged (khugepaged collapse_huge_page 330 takes the mmap_sem in write mode in addition to the anon_vma lock). If 331 pmd_trans_huge returns false, you just fallback in the old code 332 paths. If instead pmd_trans_huge returns true, you have to take the 333 mm->page_table_lock and re-run pmd_trans_huge. Taking the 334 page_table_lock will prevent the huge pmd to be converted into a 335 regular pmd from under you (split_huge_page can run in parallel to the 336 pagetable walk). If the second pmd_trans_huge returns false, you 337 should just drop the page_table_lock and fallback to the old code as 338 before. Otherwise you should run pmd_trans_splitting on the pmd. In 339 case pmd_trans_splitting returns true, it means split_huge_page is 340 already in the middle of splitting the page. So if pmd_trans_splitting 341 returns true it's enough to drop the page_table_lock and call 342 wait_split_huge_page and then fallback the old code paths. You are 343 guaranteed by the time wait_split_huge_page returns, the pmd isn't 344 huge anymore. If pmd_trans_splitting returns false, you can proceed to 345 process the huge pmd and the hugepage natively. Once finished you can 346 drop the page_table_lock. 347 348 == compound_lock, get_user_pages and put_page == 349 350 split_huge_page internally has to distribute the refcounts in the head 351 page to the tail pages before clearing all PG_head/tail bits from the 352 page structures. It can do that easily for refcounts taken by huge pmd 353 mappings. But the GUI API as created by hugetlbfs (that returns head 354 and tail pages if running get_user_pages on an address backed by any 355 hugepage), requires the refcount to be accounted on the tail pages and 356 not only in the head pages, if we want to be able to run 357 split_huge_page while there are gup pins established on any tail 358 page. Failure to be able to run split_huge_page if there's any gup pin 359 on any tail page, would mean having to split all hugepages upfront in 360 get_user_pages which is unacceptable as too many gup users are 361 performance critical and they must work natively on hugepages like 362 they work natively on hugetlbfs already (hugetlbfs is simpler because 363 hugetlbfs pages cannot be splitted so there wouldn't be requirement of 364 accounting the pins on the tail pages for hugetlbfs). If we wouldn't 365 account the gup refcounts on the tail pages during gup, we won't know 366 anymore which tail page is pinned by gup and which is not while we run 367 split_huge_page. But we still have to add the gup pin to the head page 368 too, to know when we can free the compound page in case it's never 369 splitted during its lifetime. That requires changing not just 370 get_page, but put_page as well so that when put_page runs on a tail 371 page (and only on a tail page) it will find its respective head page, 372 and then it will decrease the head page refcount in addition to the 373 tail page refcount. To obtain a head page reliably and to decrease its 374 refcount without race conditions, put_page has to serialize against 375 __split_huge_page_refcount using a special per-page lock called 376 compound_lock.