1 /*
   2  * Copyright (c) 2015, 2018, 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/zLargePages.inline.hpp"
  29 #include "gc/z/zMemory.hpp"
  30 #include "gc/z/zNUMA.hpp"
  31 #include "gc/z/zPhysicalMemory.inline.hpp"
  32 #include "gc/z/zPhysicalMemoryBacking_linux_x86.hpp"
  33 #include "logging/log.hpp"
  34 #include "runtime/os.hpp"
  35 #include "utilities/align.hpp"
  36 #include "utilities/debug.hpp"
  37 
  38 #include <stdio.h>
  39 #include <sys/mman.h>
  40 #include <sys/types.h>
  41 
  42 // Support for building on older Linux systems
  43 #ifndef MADV_HUGEPAGE
  44 #define MADV_HUGEPAGE                        14
  45 #endif
  46 
  47 // Proc file entry for max map mount
  48 #define ZFILENAME_PROC_MAX_MAP_COUNT         "/proc/sys/vm/max_map_count"
  49 
  50 ZPhysicalMemoryBacking::ZPhysicalMemoryBacking(size_t max_capacity, size_t granule_size) :
  51     _manager(),
  52     _file(),
  53     _granule_size(granule_size) {
  54 
  55   if (!_file.is_initialized()) {
  56     return;
  57   }
  58 
  59   // Check and warn if max map count is too low
  60   check_max_map_count(max_capacity, granule_size);
  61 
  62   // Check and warn if available space on filesystem is too low
  63   check_available_space_on_filesystem(max_capacity);
  64 }
  65 
  66 void ZPhysicalMemoryBacking::check_max_map_count(size_t max_capacity, size_t granule_size) const {
  67   const char* const filename = ZFILENAME_PROC_MAX_MAP_COUNT;
  68   FILE* const file = fopen(filename, "r");
  69   if (file == NULL) {
  70     // Failed to open file, skip check
  71     log_debug(gc, init)("Failed to open %s", filename);
  72     return;
  73   }
  74 
  75   size_t actual_max_map_count = 0;
  76   const int result = fscanf(file, SIZE_FORMAT, &actual_max_map_count);
  77   fclose(file);
  78   if (result != 1) {
  79     // Failed to read file, skip check
  80     log_debug(gc, init)("Failed to read %s", filename);
  81     return;
  82   }
  83 
  84   // The required max map count is impossible to calculate exactly since subsystems
  85   // other than ZGC are also creating memory mappings, and we have no control over that.
  86   // However, ZGC tends to create the most mappings and dominate the total count.
  87   // In the worst cases, ZGC will map each granule three times, i.e. once per heap view.
  88   // We speculate that we need another 20% to allow for non-ZGC subsystems to map memory.
  89   const size_t required_max_map_count = (max_capacity / granule_size) * 3 * 1.2;
  90   if (actual_max_map_count < required_max_map_count) {
  91     log_warning(gc, init)("***** WARNING! INCORRECT SYSTEM CONFIGURATION DETECTED! *****");
  92     log_warning(gc, init)("The system limit on number of memory mappings per process might be too low "
  93                           "for the given");
  94     log_warning(gc, init)("max Java heap size (" SIZE_FORMAT "M). Please adjust %s to allow for at",
  95                           max_capacity / M, filename);
  96     log_warning(gc, init)("least " SIZE_FORMAT " mappings (current limit is " SIZE_FORMAT "). Continuing "
  97                           "execution with the current", required_max_map_count, actual_max_map_count);
  98     log_warning(gc, init)("limit could lead to a fatal error, due to failure to map memory.");
  99   }
 100 }
 101 
 102 void ZPhysicalMemoryBacking::check_available_space_on_filesystem(size_t max_capacity) const {
 103   // Note that the available space on a tmpfs or a hugetlbfs filesystem
 104   // will be zero if no size limit was specified when it was mounted.
 105   const size_t available = _file.available();
 106   if (available == 0) {
 107     // No size limit set, skip check
 108     log_info(gc, init)("Available space on backing filesystem: N/A");
 109     return;
 110   }
 111 
 112   log_info(gc, init)("Available space on backing filesystem: " SIZE_FORMAT "M",
 113                      available / M);
 114 
 115   // Warn if the filesystem doesn't currently have enough space available to hold
 116   // the max heap size. The max heap size will be capped if we later hit this limit
 117   // when trying to expand the heap.
 118   if (available < max_capacity) {
 119     log_warning(gc, init)("***** WARNING! INCORRECT SYSTEM CONFIGURATION DETECTED! *****");
 120     log_warning(gc, init)("Not enough space available on the backing filesystem to hold the current "
 121                           "max Java heap");
 122     log_warning(gc, init)("size (" SIZE_FORMAT "M). Please adjust the size of the backing filesystem "
 123                           "accordingly (available", max_capacity / M);
 124     log_warning(gc, init)("space is currently " SIZE_FORMAT "M). Continuing execution with the current "
 125                           "filesystem size could", available / M);
 126     log_warning(gc, init)("lead to a premature OutOfMemoryError being thrown, due to failure to map "
 127                           "memory.");
 128   }
 129 }
 130 
 131 bool ZPhysicalMemoryBacking::is_initialized() const {
 132   return _file.is_initialized();
 133 }
 134 
 135 size_t ZPhysicalMemoryBacking::try_expand(size_t old_capacity, size_t new_capacity) {
 136   assert(old_capacity < new_capacity, "Invalid old/new capacity");
 137 
 138   const size_t capacity = _file.try_expand(old_capacity, new_capacity - old_capacity, _granule_size);
 139   if (capacity > old_capacity) {
 140     // Add expanded capacity to free list
 141     _manager.free(old_capacity, capacity - old_capacity);
 142   }
 143 
 144   return capacity;
 145 }
 146 
 147 ZPhysicalMemory ZPhysicalMemoryBacking::alloc(size_t size) {
 148   assert(is_aligned(size, _granule_size), "Invalid size");
 149 
 150   ZPhysicalMemory pmem;
 151 
 152   // Allocate segments
 153   for (size_t allocated = 0; allocated < size; allocated += _granule_size) {
 154     const uintptr_t start = _manager.alloc_from_front(_granule_size);
 155     assert(start != UINTPTR_MAX, "Allocation should never fail");
 156     pmem.add_segment(ZPhysicalMemorySegment(start, _granule_size));
 157   }
 158 
 159   return pmem;
 160 }
 161 
 162 void ZPhysicalMemoryBacking::free(ZPhysicalMemory pmem) {
 163   const size_t nsegments = pmem.nsegments();
 164 
 165   // Free segments
 166   for (size_t i = 0; i < nsegments; i++) {
 167     const ZPhysicalMemorySegment segment = pmem.segment(i);
 168     _manager.free(segment.start(), segment.size());
 169   }
 170 }
 171 
 172 void ZPhysicalMemoryBacking::map_failed(ZErrno err) const {
 173   if (err == ENOMEM) {
 174     fatal("Failed to map memory. Please check the system limit on number of "
 175           "memory mappings allowed per process (see %s)", ZFILENAME_PROC_MAX_MAP_COUNT);
 176   } else {
 177     fatal("Failed to map memory (%s)", err.to_string());
 178   }
 179 }
 180 
 181 void ZPhysicalMemoryBacking::advise_view(uintptr_t addr, size_t size) const {
 182   if (madvise((void*)addr, size, MADV_HUGEPAGE) == -1) {
 183     ZErrno err;
 184     log_error(gc)("Failed to advise use of transparent huge pages (%s)", err.to_string());
 185   }
 186 }
 187 
 188 void ZPhysicalMemoryBacking::pretouch_view(uintptr_t addr, size_t size) const {
 189   const size_t page_size = ZLargePages::is_explicit() ? os::large_page_size() : os::vm_page_size();
 190   os::pretouch_memory((void*)addr, (void*)(addr + size), page_size);
 191 }
 192 
 193 void ZPhysicalMemoryBacking::map_view(ZPhysicalMemory pmem, uintptr_t addr, bool pretouch) const {
 194   const size_t nsegments = pmem.nsegments();
 195 
 196   // Map segments
 197   for (size_t i = 0; i < nsegments; i++) {
 198     const ZPhysicalMemorySegment segment = pmem.segment(i);
 199     const size_t size = segment.size();
 200     const void* const res = mmap((void*)addr, size, PROT_READ|PROT_WRITE, MAP_FIXED|MAP_SHARED, _file.fd(), segment.start());
 201     if (res == MAP_FAILED) {
 202       ZErrno err;
 203       map_failed(err);
 204     }
 205 
 206     // Advise on use of transparent huge pages before touching it
 207     if (ZLargePages::is_transparent()) {
 208       advise_view(addr, size);
 209     }
 210 
 211     // NUMA interleave memory before touching it
 212     ZNUMA::memory_interleave(addr, size);
 213 
 214     if (pretouch) {
 215       pretouch_view(addr, size);
 216     }
 217 
 218     addr += size;
 219   }
 220 }
 221 
 222 void ZPhysicalMemoryBacking::unmap_view(ZPhysicalMemory pmem, uintptr_t addr) const {
 223   // Note that we must keep the address space reservation intact and just detach
 224   // the backing memory. For this reason we map a new anonymous, non-accessible
 225   // and non-reserved page over the mapping instead of actually unmapping.
 226   const size_t size = pmem.size();
 227   const void* const res = mmap((void*)addr, size, PROT_NONE, MAP_FIXED|MAP_ANONYMOUS|MAP_PRIVATE|MAP_NORESERVE, -1, 0);
 228   if (res == MAP_FAILED) {
 229     ZErrno err;
 230     map_failed(err);
 231   }
 232 }
 233 
 234 uintptr_t ZPhysicalMemoryBacking::nmt_address(uintptr_t offset) const {
 235   // From an NMT point of view we treat the first heap mapping (marked0) as committed
 236   return ZAddress::marked0(offset);
 237 }
 238 
 239 void ZPhysicalMemoryBacking::map(ZPhysicalMemory pmem, uintptr_t offset) const {
 240   if (ZUnmapBadViews) {
 241     // Only map the good view, for debugging only
 242     map_view(pmem, ZAddress::good(offset), AlwaysPreTouch);
 243   } else {
 244     // Map all views
 245     map_view(pmem, ZAddress::marked0(offset), AlwaysPreTouch);
 246     map_view(pmem, ZAddress::marked1(offset), AlwaysPreTouch);
 247     map_view(pmem, ZAddress::remapped(offset), AlwaysPreTouch);
 248   }
 249 }
 250 
 251 void ZPhysicalMemoryBacking::unmap(ZPhysicalMemory pmem, uintptr_t offset) const {
 252   if (ZUnmapBadViews) {
 253     // Only map the good view, for debugging only
 254     unmap_view(pmem, ZAddress::good(offset));
 255   } else {
 256     // Unmap all views
 257     unmap_view(pmem, ZAddress::marked0(offset));
 258     unmap_view(pmem, ZAddress::marked1(offset));
 259     unmap_view(pmem, ZAddress::remapped(offset));
 260   }
 261 }
 262 
 263 void ZPhysicalMemoryBacking::flip(ZPhysicalMemory pmem, uintptr_t offset) const {
 264   assert(ZUnmapBadViews, "Should be enabled");
 265   const uintptr_t addr_good = ZAddress::good(offset);
 266   const uintptr_t addr_bad = ZAddress::is_marked(ZAddressGoodMask) ? ZAddress::remapped(offset) : ZAddress::marked(offset);
 267   // Map/Unmap views
 268   map_view(pmem, addr_good, false /* pretouch */);
 269   unmap_view(pmem, addr_bad);
 270 }