| 1 | /* |
| 2 | * Copyright (c) 2015, 2019, Oracle and/or its affiliates. All rights reserved. |
| 3 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
| 4 | * |
| 5 | * This code is free software; you can redistribute it and/or modify it |
| 6 | * under the terms of the GNU General Public License version 2 only, as |
| 7 | * published by the Free Software Foundation. |
| 8 | * |
| 9 | * This code is distributed in the hope that it will be useful, but WITHOUT |
| 10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
| 11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
| 12 | * version 2 for more details (a copy is included in the LICENSE file that |
| 13 | * accompanied this code). |
| 14 | * |
| 15 | * You should have received a copy of the GNU General Public License version |
| 16 | * 2 along with this work; if not, write to the Free Software Foundation, |
| 17 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
| 18 | * |
| 19 | * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
| 20 | * or visit www.oracle.com if you need additional information or have any |
| 21 | * questions. |
| 22 | */ |
| 23 | |
| 24 | #include "precompiled.hpp" |
| 25 | #include "gc/z/zAddress.inline.hpp" |
| 26 | #include "gc/z/zBackingFile_linux_x86.hpp" |
| 27 | #include "gc/z/zErrno.hpp" |
| 28 | #include "gc/z/zGlobals.hpp" |
| 29 | #include "gc/z/zLargePages.inline.hpp" |
| 30 | #include "gc/z/zMemory.hpp" |
| 31 | #include "gc/z/zNUMA.hpp" |
| 32 | #include "gc/z/zPhysicalMemory.inline.hpp" |
| 33 | #include "gc/z/zPhysicalMemoryBacking_linux_x86.hpp" |
| 34 | #include "logging/log.hpp" |
| 35 | #include "runtime/init.hpp" |
| 36 | #include "runtime/os.hpp" |
| 37 | #include "utilities/align.hpp" |
| 38 | #include "utilities/debug.hpp" |
| 39 | |
| 40 | #include <stdio.h> |
| 41 | #include <sys/mman.h> |
| 42 | #include <sys/types.h> |
| 43 | |
| 44 | // |
| 45 | // Support for building on older Linux systems |
| 46 | // |
| 47 | |
| 48 | // madvise(2) flags |
| 49 | #ifndef MADV_HUGEPAGE |
| 50 | #define MADV_HUGEPAGE 14 |
| 51 | #endif |
| 52 | |
| 53 | // Proc file entry for max map mount |
| 54 | #define ZFILENAME_PROC_MAX_MAP_COUNT "/proc/sys/vm/max_map_count" |
| 55 | |
| 56 | bool ZPhysicalMemoryBacking::is_initialized() const { |
| 57 | return _file.is_initialized(); |
| 58 | } |
| 59 | |
| 60 | void ZPhysicalMemoryBacking::warn_available_space(size_t max) const { |
| 61 | // Note that the available space on a tmpfs or a hugetlbfs filesystem |
| 62 | // will be zero if no size limit was specified when it was mounted. |
| 63 | const size_t available = _file.available(); |
| 64 | if (available == 0) { |
| 65 | // No size limit set, skip check |
| 66 | log_info(gc, init)("Available space on backing filesystem: N/A" ); |
| 67 | return; |
| 68 | } |
| 69 | |
| 70 | log_info(gc, init)("Available space on backing filesystem: " SIZE_FORMAT "M" , available / M); |
| 71 | |
| 72 | // Warn if the filesystem doesn't currently have enough space available to hold |
| 73 | // the max heap size. The max heap size will be capped if we later hit this limit |
| 74 | // when trying to expand the heap. |
| 75 | if (available < max) { |
| 76 | log_warning(gc)("***** WARNING! INCORRECT SYSTEM CONFIGURATION DETECTED! *****" ); |
| 77 | log_warning(gc)("Not enough space available on the backing filesystem to hold the current max Java heap" ); |
| 78 | log_warning(gc)("size (" SIZE_FORMAT "M). Please adjust the size of the backing filesystem accordingly " |
| 79 | "(available" , max / M); |
| 80 | log_warning(gc)("space is currently " SIZE_FORMAT "M). Continuing execution with the current filesystem " |
| 81 | "size could" , available / M); |
| 82 | log_warning(gc)("lead to a premature OutOfMemoryError being thrown, due to failure to map memory." ); |
| 83 | } |
| 84 | } |
| 85 | |
| 86 | void ZPhysicalMemoryBacking::warn_max_map_count(size_t max) const { |
| 87 | const char* const filename = ZFILENAME_PROC_MAX_MAP_COUNT; |
| 88 | FILE* const file = fopen(filename, "r" ); |
| 89 | if (file == NULL) { |
| 90 | // Failed to open file, skip check |
| 91 | log_debug(gc, init)("Failed to open %s" , filename); |
| 92 | return; |
| 93 | } |
| 94 | |
| 95 | size_t actual_max_map_count = 0; |
| 96 | const int result = fscanf(file, SIZE_FORMAT, &actual_max_map_count); |
| 97 | fclose(file); |
| 98 | if (result != 1) { |
| 99 | // Failed to read file, skip check |
| 100 | log_debug(gc, init)("Failed to read %s" , filename); |
| 101 | return; |
| 102 | } |
| 103 | |
| 104 | // The required max map count is impossible to calculate exactly since subsystems |
| 105 | // other than ZGC are also creating memory mappings, and we have no control over that. |
| 106 | // However, ZGC tends to create the most mappings and dominate the total count. |
| 107 | // In the worst cases, ZGC will map each granule three times, i.e. once per heap view. |
| 108 | // We speculate that we need another 20% to allow for non-ZGC subsystems to map memory. |
| 109 | const size_t required_max_map_count = (max / ZGranuleSize) * 3 * 1.2; |
| 110 | if (actual_max_map_count < required_max_map_count) { |
| 111 | log_warning(gc)("***** WARNING! INCORRECT SYSTEM CONFIGURATION DETECTED! *****" ); |
| 112 | log_warning(gc)("The system limit on number of memory mappings per process might be too low for the given" ); |
| 113 | log_warning(gc)("max Java heap size (" SIZE_FORMAT "M). Please adjust %s to allow for at" , |
| 114 | max / M, filename); |
| 115 | log_warning(gc)("least " SIZE_FORMAT " mappings (current limit is " SIZE_FORMAT "). Continuing execution " |
| 116 | "with the current" , required_max_map_count, actual_max_map_count); |
| 117 | log_warning(gc)("limit could lead to a fatal error, due to failure to map memory." ); |
| 118 | } |
| 119 | } |
| 120 | |
| 121 | void ZPhysicalMemoryBacking::warn_commit_limits(size_t max) const { |
| 122 | // Warn if available space is too low |
| 123 | warn_available_space(max); |
| 124 | |
| 125 | // Warn if max map count is too low |
| 126 | warn_max_map_count(max); |
| 127 | } |
| 128 | |
| 129 | bool ZPhysicalMemoryBacking::supports_uncommit() { |
| 130 | assert(!is_init_completed(), "Invalid state" ); |
| 131 | assert(_file.size() >= ZGranuleSize, "Invalid size" ); |
| 132 | |
| 133 | // Test if uncommit is supported by uncommitting and then re-committing a granule |
| 134 | return commit(uncommit(ZGranuleSize)) == ZGranuleSize; |
| 135 | } |
| 136 | |
| 137 | size_t ZPhysicalMemoryBacking::commit(size_t size) { |
| 138 | size_t committed = 0; |
| 139 | |
| 140 | // Fill holes in the backing file |
| 141 | while (committed < size) { |
| 142 | size_t allocated = 0; |
| 143 | const size_t remaining = size - committed; |
| 144 | const uintptr_t start = _uncommitted.alloc_from_front_at_most(remaining, &allocated); |
| 145 | if (start == UINTPTR_MAX) { |
| 146 | // No holes to commit |
| 147 | break; |
| 148 | } |
| 149 | |
| 150 | // Try commit hole |
| 151 | const size_t filled = _file.commit(start, allocated); |
| 152 | if (filled > 0) { |
| 153 | // Successful or partialy successful |
| 154 | _committed.free(start, filled); |
| 155 | committed += filled; |
| 156 | } |
| 157 | if (filled < allocated) { |
| 158 | // Failed or partialy failed |
| 159 | _uncommitted.free(start + filled, allocated - filled); |
| 160 | return committed; |
| 161 | } |
| 162 | } |
| 163 | |
| 164 | // Expand backing file |
| 165 | if (committed < size) { |
| 166 | const size_t remaining = size - committed; |
| 167 | const uintptr_t start = _file.size(); |
| 168 | const size_t expanded = _file.commit(start, remaining); |
| 169 | if (expanded > 0) { |
| 170 | // Successful or partialy successful |
| 171 | _committed.free(start, expanded); |
| 172 | committed += expanded; |
| 173 | } |
| 174 | } |
| 175 | |
| 176 | return committed; |
| 177 | } |
| 178 | |
| 179 | size_t ZPhysicalMemoryBacking::uncommit(size_t size) { |
| 180 | size_t uncommitted = 0; |
| 181 | |
| 182 | // Punch holes in backing file |
| 183 | while (uncommitted < size) { |
| 184 | size_t allocated = 0; |
| 185 | const size_t remaining = size - uncommitted; |
| 186 | const uintptr_t start = _committed.alloc_from_back_at_most(remaining, &allocated); |
| 187 | assert(start != UINTPTR_MAX, "Allocation should never fail" ); |
| 188 | |
| 189 | // Try punch hole |
| 190 | const size_t punched = _file.uncommit(start, allocated); |
| 191 | if (punched > 0) { |
| 192 | // Successful or partialy successful |
| 193 | _uncommitted.free(start, punched); |
| 194 | uncommitted += punched; |
| 195 | } |
| 196 | if (punched < allocated) { |
| 197 | // Failed or partialy failed |
| 198 | _committed.free(start + punched, allocated - punched); |
| 199 | return uncommitted; |
| 200 | } |
| 201 | } |
| 202 | |
| 203 | return uncommitted; |
| 204 | } |
| 205 | |
| 206 | ZPhysicalMemory ZPhysicalMemoryBacking::alloc(size_t size) { |
| 207 | assert(is_aligned(size, ZGranuleSize), "Invalid size" ); |
| 208 | |
| 209 | ZPhysicalMemory pmem; |
| 210 | |
| 211 | // Allocate segments |
| 212 | for (size_t allocated = 0; allocated < size; allocated += ZGranuleSize) { |
| 213 | const uintptr_t start = _committed.alloc_from_front(ZGranuleSize); |
| 214 | assert(start != UINTPTR_MAX, "Allocation should never fail" ); |
| 215 | pmem.add_segment(ZPhysicalMemorySegment(start, ZGranuleSize)); |
| 216 | } |
| 217 | |
| 218 | return pmem; |
| 219 | } |
| 220 | |
| 221 | void ZPhysicalMemoryBacking::free(const ZPhysicalMemory& pmem) { |
| 222 | const size_t nsegments = pmem.nsegments(); |
| 223 | |
| 224 | // Free segments |
| 225 | for (size_t i = 0; i < nsegments; i++) { |
| 226 | const ZPhysicalMemorySegment& segment = pmem.segment(i); |
| 227 | _committed.free(segment.start(), segment.size()); |
| 228 | } |
| 229 | } |
| 230 | |
| 231 | void ZPhysicalMemoryBacking::map_failed(ZErrno err) const { |
| 232 | if (err == ENOMEM) { |
| 233 | fatal("Failed to map memory. Please check the system limit on number of " |
| 234 | "memory mappings allowed per process (see %s)" , ZFILENAME_PROC_MAX_MAP_COUNT); |
| 235 | } else { |
| 236 | fatal("Failed to map memory (%s)" , err.to_string()); |
| 237 | } |
| 238 | } |
| 239 | |
| 240 | void ZPhysicalMemoryBacking::advise_view(uintptr_t addr, size_t size, int advice) const { |
| 241 | if (madvise((void*)addr, size, advice) == -1) { |
| 242 | ZErrno err; |
| 243 | log_error(gc)("Failed to advise on memory (advice %d, %s)" , advice, err.to_string()); |
| 244 | } |
| 245 | } |
| 246 | |
| 247 | void ZPhysicalMemoryBacking::pretouch_view(uintptr_t addr, size_t size) const { |
| 248 | const size_t page_size = ZLargePages::is_explicit() ? os::large_page_size() : os::vm_page_size(); |
| 249 | os::pretouch_memory((void*)addr, (void*)(addr + size), page_size); |
| 250 | } |
| 251 | |
| 252 | void ZPhysicalMemoryBacking::map_view(const ZPhysicalMemory& pmem, uintptr_t addr, bool pretouch) const { |
| 253 | const size_t nsegments = pmem.nsegments(); |
| 254 | size_t size = 0; |
| 255 | |
| 256 | // Map segments |
| 257 | for (size_t i = 0; i < nsegments; i++) { |
| 258 | const ZPhysicalMemorySegment& segment = pmem.segment(i); |
| 259 | const uintptr_t segment_addr = addr + size; |
| 260 | const void* const res = mmap((void*)segment_addr, segment.size(), PROT_READ|PROT_WRITE, MAP_FIXED|MAP_SHARED, _file.fd(), segment.start()); |
| 261 | if (res == MAP_FAILED) { |
| 262 | ZErrno err; |
| 263 | map_failed(err); |
| 264 | } |
| 265 | |
| 266 | size += segment.size(); |
| 267 | } |
| 268 | |
| 269 | // Advise on use of transparent huge pages before touching it |
| 270 | if (ZLargePages::is_transparent()) { |
| 271 | advise_view(addr, size, MADV_HUGEPAGE); |
| 272 | } |
| 273 | |
| 274 | // NUMA interleave memory before touching it |
| 275 | ZNUMA::memory_interleave(addr, size); |
| 276 | |
| 277 | // Pre-touch memory |
| 278 | if (pretouch) { |
| 279 | pretouch_view(addr, size); |
| 280 | } |
| 281 | } |
| 282 | |
| 283 | void ZPhysicalMemoryBacking::unmap_view(const ZPhysicalMemory& pmem, uintptr_t addr) const { |
| 284 | // Note that we must keep the address space reservation intact and just detach |
| 285 | // the backing memory. For this reason we map a new anonymous, non-accessible |
| 286 | // and non-reserved page over the mapping instead of actually unmapping. |
| 287 | const void* const res = mmap((void*)addr, pmem.size(), PROT_NONE, MAP_FIXED|MAP_ANONYMOUS|MAP_PRIVATE|MAP_NORESERVE, -1, 0); |
| 288 | if (res == MAP_FAILED) { |
| 289 | ZErrno err; |
| 290 | map_failed(err); |
| 291 | } |
| 292 | } |
| 293 | |
| 294 | uintptr_t ZPhysicalMemoryBacking::nmt_address(uintptr_t offset) const { |
| 295 | // From an NMT point of view we treat the first heap view (marked0) as committed |
| 296 | return ZAddress::marked0(offset); |
| 297 | } |
| 298 | |
| 299 | void ZPhysicalMemoryBacking::map(const ZPhysicalMemory& pmem, uintptr_t offset) const { |
| 300 | if (ZVerifyViews) { |
| 301 | // Map good view |
| 302 | map_view(pmem, ZAddress::good(offset), AlwaysPreTouch); |
| 303 | } else { |
| 304 | // Map all views |
| 305 | map_view(pmem, ZAddress::marked0(offset), AlwaysPreTouch); |
| 306 | map_view(pmem, ZAddress::marked1(offset), AlwaysPreTouch); |
| 307 | map_view(pmem, ZAddress::remapped(offset), AlwaysPreTouch); |
| 308 | } |
| 309 | } |
| 310 | |
| 311 | void ZPhysicalMemoryBacking::unmap(const ZPhysicalMemory& pmem, uintptr_t offset) const { |
| 312 | if (ZVerifyViews) { |
| 313 | // Unmap good view |
| 314 | unmap_view(pmem, ZAddress::good(offset)); |
| 315 | } else { |
| 316 | // Unmap all views |
| 317 | unmap_view(pmem, ZAddress::marked0(offset)); |
| 318 | unmap_view(pmem, ZAddress::marked1(offset)); |
| 319 | unmap_view(pmem, ZAddress::remapped(offset)); |
| 320 | } |
| 321 | } |
| 322 | |
| 323 | void ZPhysicalMemoryBacking::debug_map(const ZPhysicalMemory& pmem, uintptr_t offset) const { |
| 324 | // Map good view |
| 325 | assert(ZVerifyViews, "Should be enabled" ); |
| 326 | map_view(pmem, ZAddress::good(offset), false /* pretouch */); |
| 327 | } |
| 328 | |
| 329 | void ZPhysicalMemoryBacking::debug_unmap(const ZPhysicalMemory& pmem, uintptr_t offset) const { |
| 330 | // Unmap good view |
| 331 | assert(ZVerifyViews, "Should be enabled" ); |
| 332 | unmap_view(pmem, ZAddress::good(offset)); |
| 333 | } |
| 334 | |