1 /*
2 * Copyright (c) 1999, 2020, 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
25 // no precompiled headers
26 #include "jvm.h"
27 #include "classfile/classLoader.hpp"
28 #include "classfile/systemDictionary.hpp"
29 #include "classfile/vmSymbols.hpp"
30 #include "code/icBuffer.hpp"
31 #include "code/vtableStubs.hpp"
32 #include "compiler/compileBroker.hpp"
33 #include "compiler/disassembler.hpp"
34 #include "interpreter/interpreter.hpp"
35 #include "logging/log.hpp"
36 #include "logging/logStream.hpp"
37 #include "memory/allocation.inline.hpp"
38 #include "memory/filemap.hpp"
39 #include "oops/oop.inline.hpp"
40 #include "os_linux.inline.hpp"
41 #include "os_posix.inline.hpp"
42 #include "os_share_linux.hpp"
43 #include "osContainer_linux.hpp"
44 #include "prims/jniFastGetField.hpp"
45 #include "prims/jvm_misc.hpp"
46 #include "runtime/arguments.hpp"
47 #include "runtime/atomic.hpp"
48 #include "runtime/extendedPC.hpp"
49 #include "runtime/globals.hpp"
50 #include "runtime/interfaceSupport.inline.hpp"
51 #include "runtime/init.hpp"
52 #include "runtime/java.hpp"
53 #include "runtime/javaCalls.hpp"
54 #include "runtime/mutexLocker.hpp"
55 #include "runtime/objectMonitor.hpp"
56 #include "runtime/osThread.hpp"
57 #include "runtime/perfMemory.hpp"
58 #include "runtime/sharedRuntime.hpp"
59 #include "runtime/statSampler.hpp"
60 #include "runtime/stubRoutines.hpp"
61 #include "runtime/thread.inline.hpp"
62 #include "runtime/threadCritical.hpp"
63 #include "runtime/threadSMR.hpp"
64 #include "runtime/timer.hpp"
65 #include "runtime/vm_version.hpp"
66 #include "semaphore_posix.hpp"
67 #include "services/attachListener.hpp"
68 #include "services/memTracker.hpp"
69 #include "services/runtimeService.hpp"
70 #include "utilities/align.hpp"
71 #include "utilities/decoder.hpp"
72 #include "utilities/defaultStream.hpp"
73 #include "utilities/events.hpp"
74 #include "utilities/elfFile.hpp"
75 #include "utilities/growableArray.hpp"
76 #include "utilities/macros.hpp"
77 #include "utilities/powerOfTwo.hpp"
78 #include "utilities/vmError.hpp"
79
80 // put OS-includes here
81 # include <sys/types.h>
82 # include <sys/mman.h>
83 # include <sys/stat.h>
84 # include <sys/select.h>
85 # include <pthread.h>
86 # include <signal.h>
87 # include <endian.h>
88 # include <errno.h>
89 # include <dlfcn.h>
90 # include <stdio.h>
91 # include <unistd.h>
92 # include <sys/resource.h>
93 # include <pthread.h>
94 # include <sys/stat.h>
95 # include <sys/time.h>
96 # include <sys/times.h>
97 # include <sys/utsname.h>
98 # include <sys/socket.h>
99 # include <sys/wait.h>
100 # include <pwd.h>
101 # include <poll.h>
102 # include <fcntl.h>
103 # include <string.h>
104 # include <syscall.h>
105 # include <sys/sysinfo.h>
106 # include <gnu/libc-version.h>
107 # include <sys/ipc.h>
108 # include <sys/shm.h>
109 # include <link.h>
110 # include <stdint.h>
111 # include <inttypes.h>
112 # include <sys/ioctl.h>
113
114 #ifndef _GNU_SOURCE
115 #define _GNU_SOURCE
116 #include <sched.h>
117 #undef _GNU_SOURCE
118 #else
119 #include <sched.h>
120 #endif
121
122 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
123 // getrusage() is prepared to handle the associated failure.
124 #ifndef RUSAGE_THREAD
125 #define RUSAGE_THREAD (1) /* only the calling thread */
126 #endif
127
128 #define MAX_PATH (2 * K)
129
130 #define MAX_SECS 100000000
131
132 // for timer info max values which include all bits
133 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
134
135 enum CoredumpFilterBit {
136 FILE_BACKED_PVT_BIT = 1 << 2,
137 FILE_BACKED_SHARED_BIT = 1 << 3,
138 LARGEPAGES_BIT = 1 << 6,
139 DAX_SHARED_BIT = 1 << 8
140 };
141
142 ////////////////////////////////////////////////////////////////////////////////
143 // global variables
144 julong os::Linux::_physical_memory = 0;
145
146 address os::Linux::_initial_thread_stack_bottom = NULL;
147 uintptr_t os::Linux::_initial_thread_stack_size = 0;
148
149 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
150 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
151 pthread_t os::Linux::_main_thread;
152 int os::Linux::_page_size = -1;
153 bool os::Linux::_supports_fast_thread_cpu_time = false;
154 const char * os::Linux::_glibc_version = NULL;
155 const char * os::Linux::_libpthread_version = NULL;
156 size_t os::Linux::_default_large_page_size = 0;
157
158 static jlong initial_time_count=0;
159
160 static int clock_tics_per_sec = 100;
161
162 // If the VM might have been created on the primordial thread, we need to resolve the
163 // primordial thread stack bounds and check if the current thread might be the
164 // primordial thread in places. If we know that the primordial thread is never used,
165 // such as when the VM was created by one of the standard java launchers, we can
166 // avoid this
167 static bool suppress_primordial_thread_resolution = false;
168
169 // For diagnostics to print a message once. see run_periodic_checks
170 static sigset_t check_signal_done;
171 static bool check_signals = true;
172
173 // Signal number used to suspend/resume a thread
174
175 // do not use any signal number less than SIGSEGV, see 4355769
176 static int SR_signum = SIGUSR2;
177 sigset_t SR_sigset;
178
179 // utility functions
180
181 static int SR_initialize();
182
183 julong os::available_memory() {
184 return Linux::available_memory();
185 }
186
187 julong os::Linux::available_memory() {
188 // values in struct sysinfo are "unsigned long"
189 struct sysinfo si;
190 julong avail_mem;
191
192 if (OSContainer::is_containerized()) {
193 jlong mem_limit, mem_usage;
194 if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
195 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
196 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
197 }
198 if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
199 log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
200 }
201 if (mem_limit > 0 && mem_usage > 0 ) {
202 avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
203 log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
204 return avail_mem;
205 }
206 }
207
208 sysinfo(&si);
209 avail_mem = (julong)si.freeram * si.mem_unit;
210 log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
211 return avail_mem;
212 }
213
214 julong os::physical_memory() {
215 jlong phys_mem = 0;
216 if (OSContainer::is_containerized()) {
217 jlong mem_limit;
218 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
219 log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
220 return mem_limit;
221 }
222 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
223 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
224 }
225
226 phys_mem = Linux::physical_memory();
227 log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
228 return phys_mem;
229 }
230
231 static uint64_t initial_total_ticks = 0;
232 static uint64_t initial_steal_ticks = 0;
233 static bool has_initial_tick_info = false;
234
235 static void next_line(FILE *f) {
236 int c;
237 do {
238 c = fgetc(f);
239 } while (c != '\n' && c != EOF);
240 }
241
242 bool os::Linux::get_tick_information(CPUPerfTicks* pticks, int which_logical_cpu) {
243 FILE* fh;
244 uint64_t userTicks, niceTicks, systemTicks, idleTicks;
245 // since at least kernel 2.6 : iowait: time waiting for I/O to complete
246 // irq: time servicing interrupts; softirq: time servicing softirqs
247 uint64_t iowTicks = 0, irqTicks = 0, sirqTicks= 0;
248 // steal (since kernel 2.6.11): time spent in other OS when running in a virtualized environment
249 uint64_t stealTicks = 0;
250 // guest (since kernel 2.6.24): time spent running a virtual CPU for guest OS under the
251 // control of the Linux kernel
252 uint64_t guestNiceTicks = 0;
253 int logical_cpu = -1;
254 const int required_tickinfo_count = (which_logical_cpu == -1) ? 4 : 5;
255 int n;
256
257 memset(pticks, 0, sizeof(CPUPerfTicks));
258
259 if ((fh = fopen("/proc/stat", "r")) == NULL) {
260 return false;
261 }
262
263 if (which_logical_cpu == -1) {
264 n = fscanf(fh, "cpu " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
265 UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
266 UINT64_FORMAT " " UINT64_FORMAT " ",
267 &userTicks, &niceTicks, &systemTicks, &idleTicks,
268 &iowTicks, &irqTicks, &sirqTicks,
269 &stealTicks, &guestNiceTicks);
270 } else {
271 // Move to next line
272 next_line(fh);
273
274 // find the line for requested cpu faster to just iterate linefeeds?
275 for (int i = 0; i < which_logical_cpu; i++) {
276 next_line(fh);
277 }
278
279 n = fscanf(fh, "cpu%u " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
280 UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
281 UINT64_FORMAT " " UINT64_FORMAT " ",
282 &logical_cpu, &userTicks, &niceTicks,
283 &systemTicks, &idleTicks, &iowTicks, &irqTicks, &sirqTicks,
284 &stealTicks, &guestNiceTicks);
285 }
286
287 fclose(fh);
288 if (n < required_tickinfo_count || logical_cpu != which_logical_cpu) {
289 return false;
290 }
291 pticks->used = userTicks + niceTicks;
292 pticks->usedKernel = systemTicks + irqTicks + sirqTicks;
293 pticks->total = userTicks + niceTicks + systemTicks + idleTicks +
294 iowTicks + irqTicks + sirqTicks + stealTicks + guestNiceTicks;
295
296 if (n > required_tickinfo_count + 3) {
297 pticks->steal = stealTicks;
298 pticks->has_steal_ticks = true;
299 } else {
300 pticks->steal = 0;
301 pticks->has_steal_ticks = false;
302 }
303
304 return true;
305 }
306
307 // Return true if user is running as root.
308
309 bool os::have_special_privileges() {
310 static bool init = false;
311 static bool privileges = false;
312 if (!init) {
313 privileges = (getuid() != geteuid()) || (getgid() != getegid());
314 init = true;
315 }
316 return privileges;
317 }
318
319
320 #ifndef SYS_gettid
321 // i386: 224, ia64: 1105, amd64: 186, sparc 143
322 #ifdef __ia64__
323 #define SYS_gettid 1105
324 #else
325 #ifdef __i386__
326 #define SYS_gettid 224
327 #else
328 #ifdef __amd64__
329 #define SYS_gettid 186
330 #else
331 #ifdef __sparc__
332 #define SYS_gettid 143
333 #else
334 #error define gettid for the arch
335 #endif
336 #endif
337 #endif
338 #endif
339 #endif
340
341
342 // pid_t gettid()
343 //
344 // Returns the kernel thread id of the currently running thread. Kernel
345 // thread id is used to access /proc.
346 pid_t os::Linux::gettid() {
347 int rslt = syscall(SYS_gettid);
348 assert(rslt != -1, "must be."); // old linuxthreads implementation?
349 return (pid_t)rslt;
350 }
351
352 // Most versions of linux have a bug where the number of processors are
353 // determined by looking at the /proc file system. In a chroot environment,
354 // the system call returns 1.
355 static bool unsafe_chroot_detected = false;
356 static const char *unstable_chroot_error = "/proc file system not found.\n"
357 "Java may be unstable running multithreaded in a chroot "
358 "environment on Linux when /proc filesystem is not mounted.";
359
360 void os::Linux::initialize_system_info() {
361 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
362 if (processor_count() == 1) {
363 pid_t pid = os::Linux::gettid();
364 char fname[32];
365 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
366 FILE *fp = fopen(fname, "r");
367 if (fp == NULL) {
368 unsafe_chroot_detected = true;
369 } else {
370 fclose(fp);
371 }
372 }
373 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
374 assert(processor_count() > 0, "linux error");
375 }
376
377 void os::init_system_properties_values() {
378 // The next steps are taken in the product version:
379 //
380 // Obtain the JAVA_HOME value from the location of libjvm.so.
381 // This library should be located at:
382 // <JAVA_HOME>/lib/{client|server}/libjvm.so.
383 //
384 // If "/jre/lib/" appears at the right place in the path, then we
385 // assume libjvm.so is installed in a JDK and we use this path.
386 //
387 // Otherwise exit with message: "Could not create the Java virtual machine."
388 //
389 // The following extra steps are taken in the debugging version:
390 //
391 // If "/jre/lib/" does NOT appear at the right place in the path
392 // instead of exit check for $JAVA_HOME environment variable.
393 //
394 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
395 // then we append a fake suffix "hotspot/libjvm.so" to this path so
396 // it looks like libjvm.so is installed there
397 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
398 //
399 // Otherwise exit.
400 //
401 // Important note: if the location of libjvm.so changes this
402 // code needs to be changed accordingly.
403
404 // See ld(1):
405 // The linker uses the following search paths to locate required
406 // shared libraries:
407 // 1: ...
408 // ...
409 // 7: The default directories, normally /lib and /usr/lib.
410 #ifndef OVERRIDE_LIBPATH
411 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
412 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
413 #else
414 #define DEFAULT_LIBPATH "/lib:/usr/lib"
415 #endif
416 #else
417 #define DEFAULT_LIBPATH OVERRIDE_LIBPATH
418 #endif
419
420 // Base path of extensions installed on the system.
421 #define SYS_EXT_DIR "/usr/java/packages"
422 #define EXTENSIONS_DIR "/lib/ext"
423
424 // Buffer that fits several sprintfs.
425 // Note that the space for the colon and the trailing null are provided
426 // by the nulls included by the sizeof operator.
427 const size_t bufsize =
428 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
429 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
430 char *buf = NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
431
432 // sysclasspath, java_home, dll_dir
433 {
434 char *pslash;
435 os::jvm_path(buf, bufsize);
436
437 // Found the full path to libjvm.so.
438 // Now cut the path to <java_home>/jre if we can.
439 pslash = strrchr(buf, '/');
440 if (pslash != NULL) {
441 *pslash = '\0'; // Get rid of /libjvm.so.
442 }
443 pslash = strrchr(buf, '/');
444 if (pslash != NULL) {
445 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
446 }
447 Arguments::set_dll_dir(buf);
448
449 if (pslash != NULL) {
450 pslash = strrchr(buf, '/');
451 if (pslash != NULL) {
452 *pslash = '\0'; // Get rid of /lib.
453 }
454 }
455 Arguments::set_java_home(buf);
456 if (!set_boot_path('/', ':')) {
457 vm_exit_during_initialization("Failed setting boot class path.", NULL);
458 }
459 }
460
461 // Where to look for native libraries.
462 //
463 // Note: Due to a legacy implementation, most of the library path
464 // is set in the launcher. This was to accomodate linking restrictions
465 // on legacy Linux implementations (which are no longer supported).
466 // Eventually, all the library path setting will be done here.
467 //
468 // However, to prevent the proliferation of improperly built native
469 // libraries, the new path component /usr/java/packages is added here.
470 // Eventually, all the library path setting will be done here.
471 {
472 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
473 // should always exist (until the legacy problem cited above is
474 // addressed).
475 const char *v = ::getenv("LD_LIBRARY_PATH");
476 const char *v_colon = ":";
477 if (v == NULL) { v = ""; v_colon = ""; }
478 // That's +1 for the colon and +1 for the trailing '\0'.
479 char *ld_library_path = NEW_C_HEAP_ARRAY(char,
480 strlen(v) + 1 +
481 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
482 mtInternal);
483 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
484 Arguments::set_library_path(ld_library_path);
485 FREE_C_HEAP_ARRAY(char, ld_library_path);
486 }
487
488 // Extensions directories.
489 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
490 Arguments::set_ext_dirs(buf);
491
492 FREE_C_HEAP_ARRAY(char, buf);
493
494 #undef DEFAULT_LIBPATH
495 #undef SYS_EXT_DIR
496 #undef EXTENSIONS_DIR
497 }
498
499 ////////////////////////////////////////////////////////////////////////////////
500 // breakpoint support
501
502 void os::breakpoint() {
503 BREAKPOINT;
504 }
505
506 extern "C" void breakpoint() {
507 // use debugger to set breakpoint here
508 }
509
510 ////////////////////////////////////////////////////////////////////////////////
511 // signal support
512
513 debug_only(static bool signal_sets_initialized = false);
514 static sigset_t unblocked_sigs, vm_sigs;
515
516 void os::Linux::signal_sets_init() {
517 // Should also have an assertion stating we are still single-threaded.
518 assert(!signal_sets_initialized, "Already initialized");
519 // Fill in signals that are necessarily unblocked for all threads in
520 // the VM. Currently, we unblock the following signals:
521 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
522 // by -Xrs (=ReduceSignalUsage));
523 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
524 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
525 // the dispositions or masks wrt these signals.
526 // Programs embedding the VM that want to use the above signals for their
527 // own purposes must, at this time, use the "-Xrs" option to prevent
528 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
529 // (See bug 4345157, and other related bugs).
530 // In reality, though, unblocking these signals is really a nop, since
531 // these signals are not blocked by default.
532 sigemptyset(&unblocked_sigs);
533 sigaddset(&unblocked_sigs, SIGILL);
534 sigaddset(&unblocked_sigs, SIGSEGV);
535 sigaddset(&unblocked_sigs, SIGBUS);
536 sigaddset(&unblocked_sigs, SIGFPE);
537 #if defined(PPC64)
538 sigaddset(&unblocked_sigs, SIGTRAP);
539 #endif
540 sigaddset(&unblocked_sigs, SR_signum);
541
542 if (!ReduceSignalUsage) {
543 if (!os::Posix::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
544 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
545 }
546 if (!os::Posix::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
547 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
548 }
549 if (!os::Posix::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
550 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
551 }
552 }
553 // Fill in signals that are blocked by all but the VM thread.
554 sigemptyset(&vm_sigs);
555 if (!ReduceSignalUsage) {
556 sigaddset(&vm_sigs, BREAK_SIGNAL);
557 }
558 debug_only(signal_sets_initialized = true);
559
560 }
561
562 // These are signals that are unblocked while a thread is running Java.
563 // (For some reason, they get blocked by default.)
564 sigset_t* os::Linux::unblocked_signals() {
565 assert(signal_sets_initialized, "Not initialized");
566 return &unblocked_sigs;
567 }
568
569 // These are the signals that are blocked while a (non-VM) thread is
570 // running Java. Only the VM thread handles these signals.
571 sigset_t* os::Linux::vm_signals() {
572 assert(signal_sets_initialized, "Not initialized");
573 return &vm_sigs;
574 }
575
576 void os::Linux::hotspot_sigmask(Thread* thread) {
577
578 //Save caller's signal mask before setting VM signal mask
579 sigset_t caller_sigmask;
580 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
581
582 OSThread* osthread = thread->osthread();
583 osthread->set_caller_sigmask(caller_sigmask);
584
585 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
586
587 if (!ReduceSignalUsage) {
588 if (thread->is_VM_thread()) {
589 // Only the VM thread handles BREAK_SIGNAL ...
590 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
591 } else {
592 // ... all other threads block BREAK_SIGNAL
593 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
594 }
595 }
596 }
597
598 //////////////////////////////////////////////////////////////////////////////
599 // detecting pthread library
600
601 void os::Linux::libpthread_init() {
602 // Save glibc and pthread version strings.
603 #if !defined(_CS_GNU_LIBC_VERSION) || \
604 !defined(_CS_GNU_LIBPTHREAD_VERSION)
605 #error "glibc too old (< 2.3.2)"
606 #endif
607
608 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
609 assert(n > 0, "cannot retrieve glibc version");
610 char *str = (char *)malloc(n, mtInternal);
611 confstr(_CS_GNU_LIBC_VERSION, str, n);
612 os::Linux::set_glibc_version(str);
613
614 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
615 assert(n > 0, "cannot retrieve pthread version");
616 str = (char *)malloc(n, mtInternal);
617 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
618 os::Linux::set_libpthread_version(str);
619 }
620
621 /////////////////////////////////////////////////////////////////////////////
622 // thread stack expansion
623
624 // os::Linux::manually_expand_stack() takes care of expanding the thread
625 // stack. Note that this is normally not needed: pthread stacks allocate
626 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
627 // committed. Therefore it is not necessary to expand the stack manually.
628 //
629 // Manually expanding the stack was historically needed on LinuxThreads
630 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
631 // it is kept to deal with very rare corner cases:
632 //
633 // For one, user may run the VM on an own implementation of threads
634 // whose stacks are - like the old LinuxThreads - implemented using
635 // mmap(MAP_GROWSDOWN).
636 //
637 // Also, this coding may be needed if the VM is running on the primordial
638 // thread. Normally we avoid running on the primordial thread; however,
639 // user may still invoke the VM on the primordial thread.
640 //
641 // The following historical comment describes the details about running
642 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
643
644
645 // Force Linux kernel to expand current thread stack. If "bottom" is close
646 // to the stack guard, caller should block all signals.
647 //
648 // MAP_GROWSDOWN:
649 // A special mmap() flag that is used to implement thread stacks. It tells
650 // kernel that the memory region should extend downwards when needed. This
651 // allows early versions of LinuxThreads to only mmap the first few pages
652 // when creating a new thread. Linux kernel will automatically expand thread
653 // stack as needed (on page faults).
654 //
655 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
656 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
657 // region, it's hard to tell if the fault is due to a legitimate stack
658 // access or because of reading/writing non-exist memory (e.g. buffer
659 // overrun). As a rule, if the fault happens below current stack pointer,
660 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
661 // application (see Linux kernel fault.c).
662 //
663 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
664 // stack overflow detection.
665 //
666 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
667 // not use MAP_GROWSDOWN.
668 //
669 // To get around the problem and allow stack banging on Linux, we need to
670 // manually expand thread stack after receiving the SIGSEGV.
671 //
672 // There are two ways to expand thread stack to address "bottom", we used
673 // both of them in JVM before 1.5:
674 // 1. adjust stack pointer first so that it is below "bottom", and then
675 // touch "bottom"
676 // 2. mmap() the page in question
677 //
678 // Now alternate signal stack is gone, it's harder to use 2. For instance,
679 // if current sp is already near the lower end of page 101, and we need to
680 // call mmap() to map page 100, it is possible that part of the mmap() frame
681 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
682 // That will destroy the mmap() frame and cause VM to crash.
683 //
684 // The following code works by adjusting sp first, then accessing the "bottom"
685 // page to force a page fault. Linux kernel will then automatically expand the
686 // stack mapping.
687 //
688 // _expand_stack_to() assumes its frame size is less than page size, which
689 // should always be true if the function is not inlined.
690
691 static void NOINLINE _expand_stack_to(address bottom) {
692 address sp;
693 size_t size;
694 volatile char *p;
695
696 // Adjust bottom to point to the largest address within the same page, it
697 // gives us a one-page buffer if alloca() allocates slightly more memory.
698 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
699 bottom += os::Linux::page_size() - 1;
700
701 // sp might be slightly above current stack pointer; if that's the case, we
702 // will alloca() a little more space than necessary, which is OK. Don't use
703 // os::current_stack_pointer(), as its result can be slightly below current
704 // stack pointer, causing us to not alloca enough to reach "bottom".
705 sp = (address)&sp;
706
707 if (sp > bottom) {
708 size = sp - bottom;
709 p = (volatile char *)alloca(size);
710 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
711 p[0] = '\0';
712 }
713 }
714
715 void os::Linux::expand_stack_to(address bottom) {
716 _expand_stack_to(bottom);
717 }
718
719 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
720 assert(t!=NULL, "just checking");
721 assert(t->osthread()->expanding_stack(), "expand should be set");
722
723 if (t->is_in_usable_stack(addr)) {
724 sigset_t mask_all, old_sigset;
725 sigfillset(&mask_all);
726 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
727 _expand_stack_to(addr);
728 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
729 return true;
730 }
731 return false;
732 }
733
734 //////////////////////////////////////////////////////////////////////////////
735 // create new thread
736
737 // Thread start routine for all newly created threads
738 static void *thread_native_entry(Thread *thread) {
739
740 thread->record_stack_base_and_size();
741
742 // Try to randomize the cache line index of hot stack frames.
743 // This helps when threads of the same stack traces evict each other's
744 // cache lines. The threads can be either from the same JVM instance, or
745 // from different JVM instances. The benefit is especially true for
746 // processors with hyperthreading technology.
747 static int counter = 0;
748 int pid = os::current_process_id();
749 alloca(((pid ^ counter++) & 7) * 128);
750
751 thread->initialize_thread_current();
752
753 OSThread* osthread = thread->osthread();
754 Monitor* sync = osthread->startThread_lock();
755
756 osthread->set_thread_id(os::current_thread_id());
757
758 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
759 os::current_thread_id(), (uintx) pthread_self());
760
761 if (UseNUMA) {
762 int lgrp_id = os::numa_get_group_id();
763 if (lgrp_id != -1) {
764 thread->set_lgrp_id(lgrp_id);
765 }
766 }
767 // initialize signal mask for this thread
768 os::Linux::hotspot_sigmask(thread);
769
770 // initialize floating point control register
771 os::Linux::init_thread_fpu_state();
772
773 // handshaking with parent thread
774 {
775 MutexLocker ml(sync, Mutex::_no_safepoint_check_flag);
776
777 // notify parent thread
778 osthread->set_state(INITIALIZED);
779 sync->notify_all();
780
781 // wait until os::start_thread()
782 while (osthread->get_state() == INITIALIZED) {
783 sync->wait_without_safepoint_check();
784 }
785 }
786
787 assert(osthread->pthread_id() != 0, "pthread_id was not set as expected");
788
789 // call one more level start routine
790 thread->call_run();
791
792 // Note: at this point the thread object may already have deleted itself.
793 // Prevent dereferencing it from here on out.
794 thread = NULL;
795
796 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
797 os::current_thread_id(), (uintx) pthread_self());
798
799 return 0;
800 }
801
802 // On Linux, glibc places static TLS blocks (for __thread variables) on
803 // the thread stack. This decreases the stack size actually available
804 // to threads.
805 //
806 // For large static TLS sizes, this may cause threads to malfunction due
807 // to insufficient stack space. This is a well-known issue in glibc:
808 // http://sourceware.org/bugzilla/show_bug.cgi?id=11787.
809 //
810 // As a workaround, we call a private but assumed-stable glibc function,
811 // __pthread_get_minstack() to obtain the minstack size and derive the
812 // static TLS size from it. We then increase the user requested stack
813 // size by this TLS size.
814 //
815 // Due to compatibility concerns, this size adjustment is opt-in and
816 // controlled via AdjustStackSizeForTLS.
817 typedef size_t (*GetMinStack)(const pthread_attr_t *attr);
818
819 GetMinStack _get_minstack_func = NULL;
820
821 static void get_minstack_init() {
822 _get_minstack_func =
823 (GetMinStack)dlsym(RTLD_DEFAULT, "__pthread_get_minstack");
824 log_info(os, thread)("Lookup of __pthread_get_minstack %s",
825 _get_minstack_func == NULL ? "failed" : "succeeded");
826 }
827
828 // Returns the size of the static TLS area glibc puts on thread stacks.
829 // The value is cached on first use, which occurs when the first thread
830 // is created during VM initialization.
831 static size_t get_static_tls_area_size(const pthread_attr_t *attr) {
832 size_t tls_size = 0;
833 if (_get_minstack_func != NULL) {
834 // Obtain the pthread minstack size by calling __pthread_get_minstack.
835 size_t minstack_size = _get_minstack_func(attr);
836
837 // Remove non-TLS area size included in minstack size returned
838 // by __pthread_get_minstack() to get the static TLS size.
839 // In glibc before 2.27, minstack size includes guard_size.
840 // In glibc 2.27 and later, guard_size is automatically added
841 // to the stack size by pthread_create and is no longer included
842 // in minstack size. In both cases, the guard_size is taken into
843 // account, so there is no need to adjust the result for that.
844 //
845 // Although __pthread_get_minstack() is a private glibc function,
846 // it is expected to have a stable behavior across future glibc
847 // versions while glibc still allocates the static TLS blocks off
848 // the stack. Following is glibc 2.28 __pthread_get_minstack():
849 //
850 // size_t
851 // __pthread_get_minstack (const pthread_attr_t *attr)
852 // {
853 // return GLRO(dl_pagesize) + __static_tls_size + PTHREAD_STACK_MIN;
854 // }
855 //
856 //
857 // The following 'minstack_size > os::vm_page_size() + PTHREAD_STACK_MIN'
858 // if check is done for precaution.
859 if (minstack_size > (size_t)os::vm_page_size() + PTHREAD_STACK_MIN) {
860 tls_size = minstack_size - os::vm_page_size() - PTHREAD_STACK_MIN;
861 }
862 }
863
864 log_info(os, thread)("Stack size adjustment for TLS is " SIZE_FORMAT,
865 tls_size);
866 return tls_size;
867 }
868
869 bool os::create_thread(Thread* thread, ThreadType thr_type,
870 size_t req_stack_size) {
871 assert(thread->osthread() == NULL, "caller responsible");
872
873 // Allocate the OSThread object
874 OSThread* osthread = new OSThread(NULL, NULL);
875 if (osthread == NULL) {
876 return false;
877 }
878
879 // set the correct thread state
880 osthread->set_thread_type(thr_type);
881
882 // Initial state is ALLOCATED but not INITIALIZED
883 osthread->set_state(ALLOCATED);
884
885 thread->set_osthread(osthread);
886
887 // init thread attributes
888 pthread_attr_t attr;
889 pthread_attr_init(&attr);
890 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
891
892 // Calculate stack size if it's not specified by caller.
893 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
894 // In glibc versions prior to 2.7 the guard size mechanism
895 // is not implemented properly. The posix standard requires adding
896 // the size of the guard pages to the stack size, instead Linux
897 // takes the space out of 'stacksize'. Thus we adapt the requested
898 // stack_size by the size of the guard pages to mimick proper
899 // behaviour. However, be careful not to end up with a size
900 // of zero due to overflow. Don't add the guard page in that case.
901 size_t guard_size = os::Linux::default_guard_size(thr_type);
902 // Configure glibc guard page. Must happen before calling
903 // get_static_tls_area_size(), which uses the guard_size.
904 pthread_attr_setguardsize(&attr, guard_size);
905
906 size_t stack_adjust_size = 0;
907 if (AdjustStackSizeForTLS) {
908 // Adjust the stack_size for on-stack TLS - see get_static_tls_area_size().
909 stack_adjust_size += get_static_tls_area_size(&attr);
910 } else {
911 stack_adjust_size += guard_size;
912 }
913
914 stack_adjust_size = align_up(stack_adjust_size, os::vm_page_size());
915 if (stack_size <= SIZE_MAX - stack_adjust_size) {
916 stack_size += stack_adjust_size;
917 }
918 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
919
920 int status = pthread_attr_setstacksize(&attr, stack_size);
921 assert_status(status == 0, status, "pthread_attr_setstacksize");
922
923 ThreadState state;
924
925 {
926 pthread_t tid;
927 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
928
929 char buf[64];
930 if (ret == 0) {
931 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
932 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
933 } else {
934 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
935 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
936 // Log some OS information which might explain why creating the thread failed.
937 log_info(os, thread)("Number of threads approx. running in the VM: %d", Threads::number_of_threads());
938 LogStream st(Log(os, thread)::info());
939 os::Posix::print_rlimit_info(&st);
940 os::print_memory_info(&st);
941 os::Linux::print_proc_sys_info(&st);
942 os::Linux::print_container_info(&st);
943 }
944
945 pthread_attr_destroy(&attr);
946
947 if (ret != 0) {
948 // Need to clean up stuff we've allocated so far
949 thread->set_osthread(NULL);
950 delete osthread;
951 return false;
952 }
953
954 // Store pthread info into the OSThread
955 osthread->set_pthread_id(tid);
956
957 // Wait until child thread is either initialized or aborted
958 {
959 Monitor* sync_with_child = osthread->startThread_lock();
960 MutexLocker ml(sync_with_child, Mutex::_no_safepoint_check_flag);
961 while ((state = osthread->get_state()) == ALLOCATED) {
962 sync_with_child->wait_without_safepoint_check();
963 }
964 }
965 }
966
967 // Aborted due to thread limit being reached
968 if (state == ZOMBIE) {
969 thread->set_osthread(NULL);
970 delete osthread;
971 return false;
972 }
973
974 // The thread is returned suspended (in state INITIALIZED),
975 // and is started higher up in the call chain
976 assert(state == INITIALIZED, "race condition");
977 return true;
978 }
979
980 /////////////////////////////////////////////////////////////////////////////
981 // attach existing thread
982
983 // bootstrap the main thread
984 bool os::create_main_thread(JavaThread* thread) {
985 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
986 return create_attached_thread(thread);
987 }
988
989 bool os::create_attached_thread(JavaThread* thread) {
990 #ifdef ASSERT
991 thread->verify_not_published();
992 #endif
993
994 // Allocate the OSThread object
995 OSThread* osthread = new OSThread(NULL, NULL);
996
997 if (osthread == NULL) {
998 return false;
999 }
1000
1001 // Store pthread info into the OSThread
1002 osthread->set_thread_id(os::Linux::gettid());
1003 osthread->set_pthread_id(::pthread_self());
1004
1005 // initialize floating point control register
1006 os::Linux::init_thread_fpu_state();
1007
1008 // Initial thread state is RUNNABLE
1009 osthread->set_state(RUNNABLE);
1010
1011 thread->set_osthread(osthread);
1012
1013 if (UseNUMA) {
1014 int lgrp_id = os::numa_get_group_id();
1015 if (lgrp_id != -1) {
1016 thread->set_lgrp_id(lgrp_id);
1017 }
1018 }
1019
1020 if (os::is_primordial_thread()) {
1021 // If current thread is primordial thread, its stack is mapped on demand,
1022 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1023 // the entire stack region to avoid SEGV in stack banging.
1024 // It is also useful to get around the heap-stack-gap problem on SuSE
1025 // kernel (see 4821821 for details). We first expand stack to the top
1026 // of yellow zone, then enable stack yellow zone (order is significant,
1027 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1028 // is no gap between the last two virtual memory regions.
1029
1030 JavaThread *jt = (JavaThread *)thread;
1031 address addr = jt->stack_reserved_zone_base();
1032 assert(addr != NULL, "initialization problem?");
1033 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1034
1035 osthread->set_expanding_stack();
1036 os::Linux::manually_expand_stack(jt, addr);
1037 osthread->clear_expanding_stack();
1038 }
1039
1040 // initialize signal mask for this thread
1041 // and save the caller's signal mask
1042 os::Linux::hotspot_sigmask(thread);
1043
1044 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
1045 os::current_thread_id(), (uintx) pthread_self());
1046
1047 return true;
1048 }
1049
1050 void os::pd_start_thread(Thread* thread) {
1051 OSThread * osthread = thread->osthread();
1052 assert(osthread->get_state() != INITIALIZED, "just checking");
1053 Monitor* sync_with_child = osthread->startThread_lock();
1054 MutexLocker ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1055 sync_with_child->notify();
1056 }
1057
1058 // Free Linux resources related to the OSThread
1059 void os::free_thread(OSThread* osthread) {
1060 assert(osthread != NULL, "osthread not set");
1061
1062 // We are told to free resources of the argument thread,
1063 // but we can only really operate on the current thread.
1064 assert(Thread::current()->osthread() == osthread,
1065 "os::free_thread but not current thread");
1066
1067 #ifdef ASSERT
1068 sigset_t current;
1069 sigemptyset(¤t);
1070 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
1071 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
1072 #endif
1073
1074 // Restore caller's signal mask
1075 sigset_t sigmask = osthread->caller_sigmask();
1076 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1077
1078 delete osthread;
1079 }
1080
1081 //////////////////////////////////////////////////////////////////////////////
1082 // primordial thread
1083
1084 // Check if current thread is the primordial thread, similar to Solaris thr_main.
1085 bool os::is_primordial_thread(void) {
1086 if (suppress_primordial_thread_resolution) {
1087 return false;
1088 }
1089 char dummy;
1090 // If called before init complete, thread stack bottom will be null.
1091 // Can be called if fatal error occurs before initialization.
1092 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
1093 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
1094 os::Linux::initial_thread_stack_size() != 0,
1095 "os::init did not locate primordial thread's stack region");
1096 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
1097 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
1098 os::Linux::initial_thread_stack_size()) {
1099 return true;
1100 } else {
1101 return false;
1102 }
1103 }
1104
1105 // Find the virtual memory area that contains addr
1106 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1107 FILE *fp = fopen("/proc/self/maps", "r");
1108 if (fp) {
1109 address low, high;
1110 while (!feof(fp)) {
1111 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1112 if (low <= addr && addr < high) {
1113 if (vma_low) *vma_low = low;
1114 if (vma_high) *vma_high = high;
1115 fclose(fp);
1116 return true;
1117 }
1118 }
1119 for (;;) {
1120 int ch = fgetc(fp);
1121 if (ch == EOF || ch == (int)'\n') break;
1122 }
1123 }
1124 fclose(fp);
1125 }
1126 return false;
1127 }
1128
1129 // Locate primordial thread stack. This special handling of primordial thread stack
1130 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1131 // bogus value for the primordial process thread. While the launcher has created
1132 // the VM in a new thread since JDK 6, we still have to allow for the use of the
1133 // JNI invocation API from a primordial thread.
1134 void os::Linux::capture_initial_stack(size_t max_size) {
1135
1136 // max_size is either 0 (which means accept OS default for thread stacks) or
1137 // a user-specified value known to be at least the minimum needed. If we
1138 // are actually on the primordial thread we can make it appear that we have a
1139 // smaller max_size stack by inserting the guard pages at that location. But we
1140 // cannot do anything to emulate a larger stack than what has been provided by
1141 // the OS or threading library. In fact if we try to use a stack greater than
1142 // what is set by rlimit then we will crash the hosting process.
1143
1144 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
1145 // If this is "unlimited" then it will be a huge value.
1146 struct rlimit rlim;
1147 getrlimit(RLIMIT_STACK, &rlim);
1148 size_t stack_size = rlim.rlim_cur;
1149
1150 // 6308388: a bug in ld.so will relocate its own .data section to the
1151 // lower end of primordial stack; reduce ulimit -s value a little bit
1152 // so we won't install guard page on ld.so's data section.
1153 // But ensure we don't underflow the stack size - allow 1 page spare
1154 if (stack_size >= (size_t)(3 * page_size())) {
1155 stack_size -= 2 * page_size();
1156 }
1157
1158 // Try to figure out where the stack base (top) is. This is harder.
1159 //
1160 // When an application is started, glibc saves the initial stack pointer in
1161 // a global variable "__libc_stack_end", which is then used by system
1162 // libraries. __libc_stack_end should be pretty close to stack top. The
1163 // variable is available since the very early days. However, because it is
1164 // a private interface, it could disappear in the future.
1165 //
1166 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1167 // to __libc_stack_end, it is very close to stack top, but isn't the real
1168 // stack top. Note that /proc may not exist if VM is running as a chroot
1169 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1170 // /proc/<pid>/stat could change in the future (though unlikely).
1171 //
1172 // We try __libc_stack_end first. If that doesn't work, look for
1173 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1174 // as a hint, which should work well in most cases.
1175
1176 uintptr_t stack_start;
1177
1178 // try __libc_stack_end first
1179 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1180 if (p && *p) {
1181 stack_start = *p;
1182 } else {
1183 // see if we can get the start_stack field from /proc/self/stat
1184 FILE *fp;
1185 int pid;
1186 char state;
1187 int ppid;
1188 int pgrp;
1189 int session;
1190 int nr;
1191 int tpgrp;
1192 unsigned long flags;
1193 unsigned long minflt;
1194 unsigned long cminflt;
1195 unsigned long majflt;
1196 unsigned long cmajflt;
1197 unsigned long utime;
1198 unsigned long stime;
1199 long cutime;
1200 long cstime;
1201 long prio;
1202 long nice;
1203 long junk;
1204 long it_real;
1205 uintptr_t start;
1206 uintptr_t vsize;
1207 intptr_t rss;
1208 uintptr_t rsslim;
1209 uintptr_t scodes;
1210 uintptr_t ecode;
1211 int i;
1212
1213 // Figure what the primordial thread stack base is. Code is inspired
1214 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1215 // followed by command name surrounded by parentheses, state, etc.
1216 char stat[2048];
1217 int statlen;
1218
1219 fp = fopen("/proc/self/stat", "r");
1220 if (fp) {
1221 statlen = fread(stat, 1, 2047, fp);
1222 stat[statlen] = '\0';
1223 fclose(fp);
1224
1225 // Skip pid and the command string. Note that we could be dealing with
1226 // weird command names, e.g. user could decide to rename java launcher
1227 // to "java 1.4.2 :)", then the stat file would look like
1228 // 1234 (java 1.4.2 :)) R ... ...
1229 // We don't really need to know the command string, just find the last
1230 // occurrence of ")" and then start parsing from there. See bug 4726580.
1231 char * s = strrchr(stat, ')');
1232
1233 i = 0;
1234 if (s) {
1235 // Skip blank chars
1236 do { s++; } while (s && isspace(*s));
1237
1238 #define _UFM UINTX_FORMAT
1239 #define _DFM INTX_FORMAT
1240
1241 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1242 // 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8
1243 i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld " _UFM _UFM _DFM _UFM _UFM _UFM _UFM,
1244 &state, // 3 %c
1245 &ppid, // 4 %d
1246 &pgrp, // 5 %d
1247 &session, // 6 %d
1248 &nr, // 7 %d
1249 &tpgrp, // 8 %d
1250 &flags, // 9 %lu
1251 &minflt, // 10 %lu
1252 &cminflt, // 11 %lu
1253 &majflt, // 12 %lu
1254 &cmajflt, // 13 %lu
1255 &utime, // 14 %lu
1256 &stime, // 15 %lu
1257 &cutime, // 16 %ld
1258 &cstime, // 17 %ld
1259 &prio, // 18 %ld
1260 &nice, // 19 %ld
1261 &junk, // 20 %ld
1262 &it_real, // 21 %ld
1263 &start, // 22 UINTX_FORMAT
1264 &vsize, // 23 UINTX_FORMAT
1265 &rss, // 24 INTX_FORMAT
1266 &rsslim, // 25 UINTX_FORMAT
1267 &scodes, // 26 UINTX_FORMAT
1268 &ecode, // 27 UINTX_FORMAT
1269 &stack_start); // 28 UINTX_FORMAT
1270 }
1271
1272 #undef _UFM
1273 #undef _DFM
1274
1275 if (i != 28 - 2) {
1276 assert(false, "Bad conversion from /proc/self/stat");
1277 // product mode - assume we are the primordial thread, good luck in the
1278 // embedded case.
1279 warning("Can't detect primordial thread stack location - bad conversion");
1280 stack_start = (uintptr_t) &rlim;
1281 }
1282 } else {
1283 // For some reason we can't open /proc/self/stat (for example, running on
1284 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1285 // most cases, so don't abort:
1286 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1287 stack_start = (uintptr_t) &rlim;
1288 }
1289 }
1290
1291 // Now we have a pointer (stack_start) very close to the stack top, the
1292 // next thing to do is to figure out the exact location of stack top. We
1293 // can find out the virtual memory area that contains stack_start by
1294 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1295 // and its upper limit is the real stack top. (again, this would fail if
1296 // running inside chroot, because /proc may not exist.)
1297
1298 uintptr_t stack_top;
1299 address low, high;
1300 if (find_vma((address)stack_start, &low, &high)) {
1301 // success, "high" is the true stack top. (ignore "low", because initial
1302 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1303 stack_top = (uintptr_t)high;
1304 } else {
1305 // failed, likely because /proc/self/maps does not exist
1306 warning("Can't detect primordial thread stack location - find_vma failed");
1307 // best effort: stack_start is normally within a few pages below the real
1308 // stack top, use it as stack top, and reduce stack size so we won't put
1309 // guard page outside stack.
1310 stack_top = stack_start;
1311 stack_size -= 16 * page_size();
1312 }
1313
1314 // stack_top could be partially down the page so align it
1315 stack_top = align_up(stack_top, page_size());
1316
1317 // Allowed stack value is minimum of max_size and what we derived from rlimit
1318 if (max_size > 0) {
1319 _initial_thread_stack_size = MIN2(max_size, stack_size);
1320 } else {
1321 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1322 // clamp it at 8MB as we do on Solaris
1323 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1324 }
1325 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1326 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1327
1328 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1329
1330 if (log_is_enabled(Info, os, thread)) {
1331 // See if we seem to be on primordial process thread
1332 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1333 uintptr_t(&rlim) < stack_top;
1334
1335 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1336 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1337 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1338 stack_top, intptr_t(_initial_thread_stack_bottom));
1339 }
1340 }
1341
1342 ////////////////////////////////////////////////////////////////////////////////
1343 // time support
1344
1345 #ifndef SUPPORTS_CLOCK_MONOTONIC
1346 #error "Build platform doesn't support clock_gettime and related functionality"
1347 #endif
1348
1349 // Time since start-up in seconds to a fine granularity.
1350 // Used by VMSelfDestructTimer and the MemProfiler.
1351 double os::elapsedTime() {
1352
1353 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1354 }
1355
1356 jlong os::elapsed_counter() {
1357 return javaTimeNanos() - initial_time_count;
1358 }
1359
1360 jlong os::elapsed_frequency() {
1361 return NANOSECS_PER_SEC; // nanosecond resolution
1362 }
1363
1364 bool os::supports_vtime() { return true; }
1365
1366 double os::elapsedVTime() {
1367 struct rusage usage;
1368 int retval = getrusage(RUSAGE_THREAD, &usage);
1369 if (retval == 0) {
1370 return (double) (usage.ru_utime.tv_sec + usage.ru_stime.tv_sec) + (double) (usage.ru_utime.tv_usec + usage.ru_stime.tv_usec) / (1000 * 1000);
1371 } else {
1372 // better than nothing, but not much
1373 return elapsedTime();
1374 }
1375 }
1376
1377 jlong os::javaTimeMillis() {
1378 timeval time;
1379 int status = gettimeofday(&time, NULL);
1380 assert(status != -1, "linux error");
1381 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1382 }
1383
1384 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1385 timeval time;
1386 int status = gettimeofday(&time, NULL);
1387 assert(status != -1, "linux error");
1388 seconds = jlong(time.tv_sec);
1389 nanos = jlong(time.tv_usec) * 1000;
1390 }
1391
1392 void os::Linux::fast_thread_clock_init() {
1393 if (!UseLinuxPosixThreadCPUClocks) {
1394 return;
1395 }
1396 clockid_t clockid;
1397 struct timespec tp;
1398 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1399 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1400
1401 // Switch to using fast clocks for thread cpu time if
1402 // the clock_getres() returns 0 error code.
1403 // Note, that some kernels may support the current thread
1404 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1405 // returned by the pthread_getcpuclockid().
1406 // If the fast Posix clocks are supported then the clock_getres()
1407 // must return at least tp.tv_sec == 0 which means a resolution
1408 // better than 1 sec. This is extra check for reliability.
1409
1410 if (pthread_getcpuclockid_func &&
1411 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1412 os::Posix::clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1413 _supports_fast_thread_cpu_time = true;
1414 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1415 }
1416 }
1417
1418 jlong os::javaTimeNanos() {
1419 if (os::supports_monotonic_clock()) {
1420 struct timespec tp;
1421 int status = os::Posix::clock_gettime(CLOCK_MONOTONIC, &tp);
1422 assert(status == 0, "gettime error");
1423 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1424 return result;
1425 } else {
1426 timeval time;
1427 int status = gettimeofday(&time, NULL);
1428 assert(status != -1, "linux error");
1429 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1430 return 1000 * usecs;
1431 }
1432 }
1433
1434 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1435 if (os::supports_monotonic_clock()) {
1436 info_ptr->max_value = ALL_64_BITS;
1437
1438 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1439 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1440 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1441 } else {
1442 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1443 info_ptr->max_value = ALL_64_BITS;
1444
1445 // gettimeofday is a real time clock so it skips
1446 info_ptr->may_skip_backward = true;
1447 info_ptr->may_skip_forward = true;
1448 }
1449
1450 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1451 }
1452
1453 // Return the real, user, and system times in seconds from an
1454 // arbitrary fixed point in the past.
1455 bool os::getTimesSecs(double* process_real_time,
1456 double* process_user_time,
1457 double* process_system_time) {
1458 struct tms ticks;
1459 clock_t real_ticks = times(&ticks);
1460
1461 if (real_ticks == (clock_t) (-1)) {
1462 return false;
1463 } else {
1464 double ticks_per_second = (double) clock_tics_per_sec;
1465 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1466 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1467 *process_real_time = ((double) real_ticks) / ticks_per_second;
1468
1469 return true;
1470 }
1471 }
1472
1473
1474 char * os::local_time_string(char *buf, size_t buflen) {
1475 struct tm t;
1476 time_t long_time;
1477 time(&long_time);
1478 localtime_r(&long_time, &t);
1479 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1480 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1481 t.tm_hour, t.tm_min, t.tm_sec);
1482 return buf;
1483 }
1484
1485 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1486 return localtime_r(clock, res);
1487 }
1488
1489 ////////////////////////////////////////////////////////////////////////////////
1490 // runtime exit support
1491
1492 // Note: os::shutdown() might be called very early during initialization, or
1493 // called from signal handler. Before adding something to os::shutdown(), make
1494 // sure it is async-safe and can handle partially initialized VM.
1495 void os::shutdown() {
1496
1497 // allow PerfMemory to attempt cleanup of any persistent resources
1498 perfMemory_exit();
1499
1500 // needs to remove object in file system
1501 AttachListener::abort();
1502
1503 // flush buffered output, finish log files
1504 ostream_abort();
1505
1506 // Check for abort hook
1507 abort_hook_t abort_hook = Arguments::abort_hook();
1508 if (abort_hook != NULL) {
1509 abort_hook();
1510 }
1511
1512 }
1513
1514 // Note: os::abort() might be called very early during initialization, or
1515 // called from signal handler. Before adding something to os::abort(), make
1516 // sure it is async-safe and can handle partially initialized VM.
1517 void os::abort(bool dump_core, void* siginfo, const void* context) {
1518 os::shutdown();
1519 if (dump_core) {
1520 if (DumpPrivateMappingsInCore) {
1521 ClassLoader::close_jrt_image();
1522 }
1523 #ifndef PRODUCT
1524 fdStream out(defaultStream::output_fd());
1525 out.print_raw("Current thread is ");
1526 char buf[16];
1527 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1528 out.print_raw_cr(buf);
1529 out.print_raw_cr("Dumping core ...");
1530 #endif
1531 ::abort(); // dump core
1532 }
1533
1534 ::exit(1);
1535 }
1536
1537 // Die immediately, no exit hook, no abort hook, no cleanup.
1538 // Dump a core file, if possible, for debugging.
1539 void os::die() {
1540 if (TestUnresponsiveErrorHandler && !CreateCoredumpOnCrash) {
1541 // For TimeoutInErrorHandlingTest.java, we just kill the VM
1542 // and don't take the time to generate a core file.
1543 os::signal_raise(SIGKILL);
1544 } else {
1545 ::abort();
1546 }
1547 }
1548
1549 // thread_id is kernel thread id (similar to Solaris LWP id)
1550 intx os::current_thread_id() { return os::Linux::gettid(); }
1551 int os::current_process_id() {
1552 return ::getpid();
1553 }
1554
1555 // DLL functions
1556
1557 const char* os::dll_file_extension() { return ".so"; }
1558
1559 // This must be hard coded because it's the system's temporary
1560 // directory not the java application's temp directory, ala java.io.tmpdir.
1561 const char* os::get_temp_directory() { return "/tmp"; }
1562
1563 static bool file_exists(const char* filename) {
1564 struct stat statbuf;
1565 if (filename == NULL || strlen(filename) == 0) {
1566 return false;
1567 }
1568 return os::stat(filename, &statbuf) == 0;
1569 }
1570
1571 // check if addr is inside libjvm.so
1572 bool os::address_is_in_vm(address addr) {
1573 static address libjvm_base_addr;
1574 Dl_info dlinfo;
1575
1576 if (libjvm_base_addr == NULL) {
1577 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1578 libjvm_base_addr = (address)dlinfo.dli_fbase;
1579 }
1580 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1581 }
1582
1583 if (dladdr((void *)addr, &dlinfo) != 0) {
1584 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1585 }
1586
1587 return false;
1588 }
1589
1590 bool os::dll_address_to_function_name(address addr, char *buf,
1591 int buflen, int *offset,
1592 bool demangle) {
1593 // buf is not optional, but offset is optional
1594 assert(buf != NULL, "sanity check");
1595
1596 Dl_info dlinfo;
1597
1598 if (dladdr((void*)addr, &dlinfo) != 0) {
1599 // see if we have a matching symbol
1600 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1601 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1602 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1603 }
1604 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1605 return true;
1606 }
1607 // no matching symbol so try for just file info
1608 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1609 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1610 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1611 return true;
1612 }
1613 }
1614 }
1615
1616 buf[0] = '\0';
1617 if (offset != NULL) *offset = -1;
1618 return false;
1619 }
1620
1621 struct _address_to_library_name {
1622 address addr; // input : memory address
1623 size_t buflen; // size of fname
1624 char* fname; // output: library name
1625 address base; // library base addr
1626 };
1627
1628 static int address_to_library_name_callback(struct dl_phdr_info *info,
1629 size_t size, void *data) {
1630 int i;
1631 bool found = false;
1632 address libbase = NULL;
1633 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1634
1635 // iterate through all loadable segments
1636 for (i = 0; i < info->dlpi_phnum; i++) {
1637 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1638 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1639 // base address of a library is the lowest address of its loaded
1640 // segments.
1641 if (libbase == NULL || libbase > segbase) {
1642 libbase = segbase;
1643 }
1644 // see if 'addr' is within current segment
1645 if (segbase <= d->addr &&
1646 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1647 found = true;
1648 }
1649 }
1650 }
1651
1652 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1653 // so dll_address_to_library_name() can fall through to use dladdr() which
1654 // can figure out executable name from argv[0].
1655 if (found && info->dlpi_name && info->dlpi_name[0]) {
1656 d->base = libbase;
1657 if (d->fname) {
1658 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1659 }
1660 return 1;
1661 }
1662 return 0;
1663 }
1664
1665 bool os::dll_address_to_library_name(address addr, char* buf,
1666 int buflen, int* offset) {
1667 // buf is not optional, but offset is optional
1668 assert(buf != NULL, "sanity check");
1669
1670 Dl_info dlinfo;
1671 struct _address_to_library_name data;
1672
1673 // There is a bug in old glibc dladdr() implementation that it could resolve
1674 // to wrong library name if the .so file has a base address != NULL. Here
1675 // we iterate through the program headers of all loaded libraries to find
1676 // out which library 'addr' really belongs to. This workaround can be
1677 // removed once the minimum requirement for glibc is moved to 2.3.x.
1678 data.addr = addr;
1679 data.fname = buf;
1680 data.buflen = buflen;
1681 data.base = NULL;
1682 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1683
1684 if (rslt) {
1685 // buf already contains library name
1686 if (offset) *offset = addr - data.base;
1687 return true;
1688 }
1689 if (dladdr((void*)addr, &dlinfo) != 0) {
1690 if (dlinfo.dli_fname != NULL) {
1691 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1692 }
1693 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1694 *offset = addr - (address)dlinfo.dli_fbase;
1695 }
1696 return true;
1697 }
1698
1699 buf[0] = '\0';
1700 if (offset) *offset = -1;
1701 return false;
1702 }
1703
1704 // Loads .dll/.so and
1705 // in case of error it checks if .dll/.so was built for the
1706 // same architecture as Hotspot is running on
1707
1708
1709 // Remember the stack's state. The Linux dynamic linker will change
1710 // the stack to 'executable' at most once, so we must safepoint only once.
1711 bool os::Linux::_stack_is_executable = false;
1712
1713 // VM operation that loads a library. This is necessary if stack protection
1714 // of the Java stacks can be lost during loading the library. If we
1715 // do not stop the Java threads, they can stack overflow before the stacks
1716 // are protected again.
1717 class VM_LinuxDllLoad: public VM_Operation {
1718 private:
1719 const char *_filename;
1720 char *_ebuf;
1721 int _ebuflen;
1722 void *_lib;
1723 public:
1724 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1725 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1726 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1727 void doit() {
1728 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1729 os::Linux::_stack_is_executable = true;
1730 }
1731 void* loaded_library() { return _lib; }
1732 };
1733
1734 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1735 void * result = NULL;
1736 bool load_attempted = false;
1737
1738 log_info(os)("attempting shared library load of %s", filename);
1739
1740 // Check whether the library to load might change execution rights
1741 // of the stack. If they are changed, the protection of the stack
1742 // guard pages will be lost. We need a safepoint to fix this.
1743 //
1744 // See Linux man page execstack(8) for more info.
1745 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1746 if (!ElfFile::specifies_noexecstack(filename)) {
1747 if (!is_init_completed()) {
1748 os::Linux::_stack_is_executable = true;
1749 // This is OK - No Java threads have been created yet, and hence no
1750 // stack guard pages to fix.
1751 //
1752 // Dynamic loader will make all stacks executable after
1753 // this function returns, and will not do that again.
1754 assert(Threads::number_of_threads() == 0, "no Java threads should exist yet.");
1755 } else {
1756 warning("You have loaded library %s which might have disabled stack guard. "
1757 "The VM will try to fix the stack guard now.\n"
1758 "It's highly recommended that you fix the library with "
1759 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1760 filename);
1761
1762 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1763 JavaThread *jt = JavaThread::current();
1764 if (jt->thread_state() != _thread_in_native) {
1765 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1766 // that requires ExecStack. Cannot enter safe point. Let's give up.
1767 warning("Unable to fix stack guard. Giving up.");
1768 } else {
1769 if (!LoadExecStackDllInVMThread) {
1770 // This is for the case where the DLL has an static
1771 // constructor function that executes JNI code. We cannot
1772 // load such DLLs in the VMThread.
1773 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1774 }
1775
1776 ThreadInVMfromNative tiv(jt);
1777 debug_only(VMNativeEntryWrapper vew;)
1778
1779 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1780 VMThread::execute(&op);
1781 if (LoadExecStackDllInVMThread) {
1782 result = op.loaded_library();
1783 }
1784 load_attempted = true;
1785 }
1786 }
1787 }
1788 }
1789
1790 if (!load_attempted) {
1791 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1792 }
1793
1794 if (result != NULL) {
1795 // Successful loading
1796 return result;
1797 }
1798
1799 Elf32_Ehdr elf_head;
1800 int diag_msg_max_length=ebuflen-strlen(ebuf);
1801 char* diag_msg_buf=ebuf+strlen(ebuf);
1802
1803 if (diag_msg_max_length==0) {
1804 // No more space in ebuf for additional diagnostics message
1805 return NULL;
1806 }
1807
1808
1809 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1810
1811 if (file_descriptor < 0) {
1812 // Can't open library, report dlerror() message
1813 return NULL;
1814 }
1815
1816 bool failed_to_read_elf_head=
1817 (sizeof(elf_head)!=
1818 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1819
1820 ::close(file_descriptor);
1821 if (failed_to_read_elf_head) {
1822 // file i/o error - report dlerror() msg
1823 return NULL;
1824 }
1825
1826 if (elf_head.e_ident[EI_DATA] != LITTLE_ENDIAN_ONLY(ELFDATA2LSB) BIG_ENDIAN_ONLY(ELFDATA2MSB)) {
1827 // handle invalid/out of range endianness values
1828 if (elf_head.e_ident[EI_DATA] == 0 || elf_head.e_ident[EI_DATA] > 2) {
1829 return NULL;
1830 }
1831
1832 #if defined(VM_LITTLE_ENDIAN)
1833 // VM is LE, shared object BE
1834 elf_head.e_machine = be16toh(elf_head.e_machine);
1835 #else
1836 // VM is BE, shared object LE
1837 elf_head.e_machine = le16toh(elf_head.e_machine);
1838 #endif
1839 }
1840
1841 typedef struct {
1842 Elf32_Half code; // Actual value as defined in elf.h
1843 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1844 unsigned char elf_class; // 32 or 64 bit
1845 unsigned char endianness; // MSB or LSB
1846 char* name; // String representation
1847 } arch_t;
1848
1849 #ifndef EM_486
1850 #define EM_486 6 /* Intel 80486 */
1851 #endif
1852 #ifndef EM_AARCH64
1853 #define EM_AARCH64 183 /* ARM AARCH64 */
1854 #endif
1855 #ifndef EM_RISCV
1856 #define EM_RISCV 243 /* RISC-V */
1857 #endif
1858
1859 static const arch_t arch_array[]={
1860 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1861 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1862 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1863 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1864 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1865 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1866 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1867 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1868 #if defined(VM_LITTLE_ENDIAN)
1869 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1870 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1871 #else
1872 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1873 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1874 #endif
1875 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1876 // we only support 64 bit z architecture
1877 {EM_S390, EM_S390, ELFCLASS64, ELFDATA2MSB, (char*)"IBM System/390"},
1878 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1879 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1880 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1881 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1882 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1883 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1884 {EM_RISCV, EM_RISCV, ELFCLASS64, ELFDATA2LSB, (char*)"RISC-V"},
1885 };
1886
1887 #if (defined IA32)
1888 static Elf32_Half running_arch_code=EM_386;
1889 #elif (defined AMD64) || (defined X32)
1890 static Elf32_Half running_arch_code=EM_X86_64;
1891 #elif (defined IA64)
1892 static Elf32_Half running_arch_code=EM_IA_64;
1893 #elif (defined __sparc) && (defined _LP64)
1894 static Elf32_Half running_arch_code=EM_SPARCV9;
1895 #elif (defined __sparc) && (!defined _LP64)
1896 static Elf32_Half running_arch_code=EM_SPARC;
1897 #elif (defined __powerpc64__)
1898 static Elf32_Half running_arch_code=EM_PPC64;
1899 #elif (defined __powerpc__)
1900 static Elf32_Half running_arch_code=EM_PPC;
1901 #elif (defined AARCH64)
1902 static Elf32_Half running_arch_code=EM_AARCH64;
1903 #elif (defined ARM)
1904 static Elf32_Half running_arch_code=EM_ARM;
1905 #elif (defined S390)
1906 static Elf32_Half running_arch_code=EM_S390;
1907 #elif (defined ALPHA)
1908 static Elf32_Half running_arch_code=EM_ALPHA;
1909 #elif (defined MIPSEL)
1910 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1911 #elif (defined PARISC)
1912 static Elf32_Half running_arch_code=EM_PARISC;
1913 #elif (defined MIPS)
1914 static Elf32_Half running_arch_code=EM_MIPS;
1915 #elif (defined M68K)
1916 static Elf32_Half running_arch_code=EM_68K;
1917 #elif (defined SH)
1918 static Elf32_Half running_arch_code=EM_SH;
1919 #elif (defined RISCV)
1920 static Elf32_Half running_arch_code=EM_RISCV;
1921 #else
1922 #error Method os::dll_load requires that one of following is defined:\
1923 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, RISCV, S390, SH, __sparc
1924 #endif
1925
1926 // Identify compatibility class for VM's architecture and library's architecture
1927 // Obtain string descriptions for architectures
1928
1929 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1930 int running_arch_index=-1;
1931
1932 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1933 if (running_arch_code == arch_array[i].code) {
1934 running_arch_index = i;
1935 }
1936 if (lib_arch.code == arch_array[i].code) {
1937 lib_arch.compat_class = arch_array[i].compat_class;
1938 lib_arch.name = arch_array[i].name;
1939 }
1940 }
1941
1942 assert(running_arch_index != -1,
1943 "Didn't find running architecture code (running_arch_code) in arch_array");
1944 if (running_arch_index == -1) {
1945 // Even though running architecture detection failed
1946 // we may still continue with reporting dlerror() message
1947 return NULL;
1948 }
1949
1950 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1951 if (lib_arch.name != NULL) {
1952 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1953 " (Possible cause: can't load %s .so on a %s platform)",
1954 lib_arch.name, arch_array[running_arch_index].name);
1955 } else {
1956 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1957 " (Possible cause: can't load this .so (machine code=0x%x) on a %s platform)",
1958 lib_arch.code, arch_array[running_arch_index].name);
1959 }
1960 return NULL;
1961 }
1962
1963 if (lib_arch.endianness != arch_array[running_arch_index].endianness) {
1964 ::snprintf(diag_msg_buf, diag_msg_max_length-1, " (Possible cause: endianness mismatch)");
1965 return NULL;
1966 }
1967
1968 // ELF file class/capacity : 0 - invalid, 1 - 32bit, 2 - 64bit
1969 if (lib_arch.elf_class > 2 || lib_arch.elf_class < 1) {
1970 ::snprintf(diag_msg_buf, diag_msg_max_length-1, " (Possible cause: invalid ELF file class)");
1971 return NULL;
1972 }
1973
1974 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1975 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1976 " (Possible cause: architecture word width mismatch, can't load %d-bit .so on a %d-bit platform)",
1977 (int) lib_arch.elf_class * 32, arch_array[running_arch_index].elf_class * 32);
1978 return NULL;
1979 }
1980
1981 return NULL;
1982 }
1983
1984 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1985 int ebuflen) {
1986 void * result = ::dlopen(filename, RTLD_LAZY);
1987 if (result == NULL) {
1988 const char* error_report = ::dlerror();
1989 if (error_report == NULL) {
1990 error_report = "dlerror returned no error description";
1991 }
1992 if (ebuf != NULL && ebuflen > 0) {
1993 ::strncpy(ebuf, error_report, ebuflen-1);
1994 ebuf[ebuflen-1]='\0';
1995 }
1996 Events::log(NULL, "Loading shared library %s failed, %s", filename, error_report);
1997 log_info(os)("shared library load of %s failed, %s", filename, error_report);
1998 } else {
1999 Events::log(NULL, "Loaded shared library %s", filename);
2000 log_info(os)("shared library load of %s was successful", filename);
2001 }
2002 return result;
2003 }
2004
2005 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
2006 int ebuflen) {
2007 void * result = NULL;
2008 if (LoadExecStackDllInVMThread) {
2009 result = dlopen_helper(filename, ebuf, ebuflen);
2010 }
2011
2012 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2013 // library that requires an executable stack, or which does not have this
2014 // stack attribute set, dlopen changes the stack attribute to executable. The
2015 // read protection of the guard pages gets lost.
2016 //
2017 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2018 // may have been queued at the same time.
2019
2020 if (!_stack_is_executable) {
2021 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
2022 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2023 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
2024 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
2025 warning("Attempt to reguard stack yellow zone failed.");
2026 }
2027 }
2028 }
2029 }
2030
2031 return result;
2032 }
2033
2034 void* os::dll_lookup(void* handle, const char* name) {
2035 void* res = dlsym(handle, name);
2036 return res;
2037 }
2038
2039 void* os::get_default_process_handle() {
2040 return (void*)::dlopen(NULL, RTLD_LAZY);
2041 }
2042
2043 static bool _print_ascii_file(const char* filename, outputStream* st, const char* hdr = NULL) {
2044 int fd = ::open(filename, O_RDONLY);
2045 if (fd == -1) {
2046 return false;
2047 }
2048
2049 if (hdr != NULL) {
2050 st->print_cr("%s", hdr);
2051 }
2052
2053 char buf[33];
2054 int bytes;
2055 buf[32] = '\0';
2056 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
2057 st->print_raw(buf, bytes);
2058 }
2059
2060 ::close(fd);
2061
2062 return true;
2063 }
2064
2065 void os::print_dll_info(outputStream *st) {
2066 st->print_cr("Dynamic libraries:");
2067
2068 char fname[32];
2069 pid_t pid = os::Linux::gettid();
2070
2071 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2072
2073 if (!_print_ascii_file(fname, st)) {
2074 st->print("Can not get library information for pid = %d\n", pid);
2075 }
2076 }
2077
2078 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2079 FILE *procmapsFile = NULL;
2080
2081 // Open the procfs maps file for the current process
2082 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2083 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2084 char line[PATH_MAX + 100];
2085
2086 // Read line by line from 'file'
2087 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2088 u8 base, top, inode;
2089 char name[sizeof(line)];
2090
2091 // Parse fields from line, discard perms, offset and device
2092 int matches = sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %*s %*s %*s " INT64_FORMAT " %s",
2093 &base, &top, &inode, name);
2094 // the last entry 'name' is empty for some entries, so we might have 3 matches instead of 4 for some lines
2095 if (matches < 3) continue;
2096 if (matches == 3) name[0] = '\0';
2097
2098 // Filter by inode 0 so that we only get file system mapped files.
2099 if (inode != 0) {
2100
2101 // Call callback with the fields of interest
2102 if(callback(name, (address)base, (address)top, param)) {
2103 // Oops abort, callback aborted
2104 fclose(procmapsFile);
2105 return 1;
2106 }
2107 }
2108 }
2109 fclose(procmapsFile);
2110 }
2111 return 0;
2112 }
2113
2114 void os::print_os_info_brief(outputStream* st) {
2115 os::Linux::print_distro_info(st);
2116
2117 os::Posix::print_uname_info(st);
2118
2119 os::Linux::print_libversion_info(st);
2120
2121 }
2122
2123 void os::print_os_info(outputStream* st) {
2124 st->print("OS:");
2125
2126 os::Linux::print_distro_info(st);
2127
2128 os::Posix::print_uname_info(st);
2129
2130 os::Linux::print_uptime_info(st);
2131
2132 // Print warning if unsafe chroot environment detected
2133 if (unsafe_chroot_detected) {
2134 st->print("WARNING!! ");
2135 st->print_cr("%s", unstable_chroot_error);
2136 }
2137
2138 os::Linux::print_libversion_info(st);
2139
2140 os::Posix::print_rlimit_info(st);
2141
2142 os::Posix::print_load_average(st);
2143
2144 os::Linux::print_full_memory_info(st);
2145
2146 os::Linux::print_proc_sys_info(st);
2147
2148 os::Linux::print_ld_preload_file(st);
2149
2150 os::Linux::print_container_info(st);
2151
2152 VM_Version::print_platform_virtualization_info(st);
2153
2154 os::Linux::print_steal_info(st);
2155 }
2156
2157 // Try to identify popular distros.
2158 // Most Linux distributions have a /etc/XXX-release file, which contains
2159 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2160 // file that also contains the OS version string. Some have more than one
2161 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2162 // /etc/redhat-release.), so the order is important.
2163 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2164 // their own specific XXX-release file as well as a redhat-release file.
2165 // Because of this the XXX-release file needs to be searched for before the
2166 // redhat-release file.
2167 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2168 // search for redhat-release / SuSE-release needs to be before lsb-release.
2169 // Since the lsb-release file is the new standard it needs to be searched
2170 // before the older style release files.
2171 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2172 // next to last resort. The os-release file is a new standard that contains
2173 // distribution information and the system-release file seems to be an old
2174 // standard that has been replaced by the lsb-release and os-release files.
2175 // Searching for the debian_version file is the last resort. It contains
2176 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2177 // "Debian " is printed before the contents of the debian_version file.
2178
2179 const char* distro_files[] = {
2180 "/etc/oracle-release",
2181 "/etc/mandriva-release",
2182 "/etc/mandrake-release",
2183 "/etc/sun-release",
2184 "/etc/redhat-release",
2185 "/etc/SuSE-release",
2186 "/etc/lsb-release",
2187 "/etc/turbolinux-release",
2188 "/etc/gentoo-release",
2189 "/etc/ltib-release",
2190 "/etc/angstrom-version",
2191 "/etc/system-release",
2192 "/etc/os-release",
2193 NULL };
2194
2195 void os::Linux::print_distro_info(outputStream* st) {
2196 for (int i = 0;; i++) {
2197 const char* file = distro_files[i];
2198 if (file == NULL) {
2199 break; // done
2200 }
2201 // If file prints, we found it.
2202 if (_print_ascii_file(file, st)) {
2203 return;
2204 }
2205 }
2206
2207 if (file_exists("/etc/debian_version")) {
2208 st->print("Debian ");
2209 _print_ascii_file("/etc/debian_version", st);
2210 } else {
2211 st->print("Linux");
2212 }
2213 st->cr();
2214 }
2215
2216 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2217 char buf[256];
2218 while (fgets(buf, sizeof(buf), fp)) {
2219 // Edit out extra stuff in expected format
2220 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2221 char* ptr = strstr(buf, "\""); // the name is in quotes
2222 if (ptr != NULL) {
2223 ptr++; // go beyond first quote
2224 char* nl = strchr(ptr, '\"');
2225 if (nl != NULL) *nl = '\0';
2226 strncpy(distro, ptr, length);
2227 } else {
2228 ptr = strstr(buf, "=");
2229 ptr++; // go beyond equals then
2230 char* nl = strchr(ptr, '\n');
2231 if (nl != NULL) *nl = '\0';
2232 strncpy(distro, ptr, length);
2233 }
2234 return;
2235 } else if (get_first_line) {
2236 char* nl = strchr(buf, '\n');
2237 if (nl != NULL) *nl = '\0';
2238 strncpy(distro, buf, length);
2239 return;
2240 }
2241 }
2242 // print last line and close
2243 char* nl = strchr(buf, '\n');
2244 if (nl != NULL) *nl = '\0';
2245 strncpy(distro, buf, length);
2246 }
2247
2248 static void parse_os_info(char* distro, size_t length, const char* file) {
2249 FILE* fp = fopen(file, "r");
2250 if (fp != NULL) {
2251 // if suse format, print out first line
2252 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2253 parse_os_info_helper(fp, distro, length, get_first_line);
2254 fclose(fp);
2255 }
2256 }
2257
2258 void os::get_summary_os_info(char* buf, size_t buflen) {
2259 for (int i = 0;; i++) {
2260 const char* file = distro_files[i];
2261 if (file == NULL) {
2262 break; // ran out of distro_files
2263 }
2264 if (file_exists(file)) {
2265 parse_os_info(buf, buflen, file);
2266 return;
2267 }
2268 }
2269 // special case for debian
2270 if (file_exists("/etc/debian_version")) {
2271 strncpy(buf, "Debian ", buflen);
2272 if (buflen > 7) {
2273 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2274 }
2275 } else {
2276 strncpy(buf, "Linux", buflen);
2277 }
2278 }
2279
2280 void os::Linux::print_libversion_info(outputStream* st) {
2281 // libc, pthread
2282 st->print("libc:");
2283 st->print("%s ", os::Linux::glibc_version());
2284 st->print("%s ", os::Linux::libpthread_version());
2285 st->cr();
2286 }
2287
2288 void os::Linux::print_proc_sys_info(outputStream* st) {
2289 st->cr();
2290 st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):");
2291 _print_ascii_file("/proc/sys/kernel/threads-max", st);
2292 st->cr();
2293 st->cr();
2294
2295 st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):");
2296 _print_ascii_file("/proc/sys/vm/max_map_count", st);
2297 st->cr();
2298 st->cr();
2299
2300 st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):");
2301 _print_ascii_file("/proc/sys/kernel/pid_max", st);
2302 st->cr();
2303 st->cr();
2304 }
2305
2306 void os::Linux::print_full_memory_info(outputStream* st) {
2307 st->print("\n/proc/meminfo:\n");
2308 _print_ascii_file("/proc/meminfo", st);
2309 st->cr();
2310
2311 // some information regarding THPs; for details see
2312 // https://www.kernel.org/doc/Documentation/vm/transhuge.txt
2313 st->print_cr("/sys/kernel/mm/transparent_hugepage/enabled:");
2314 if (!_print_ascii_file("/sys/kernel/mm/transparent_hugepage/enabled", st)) {
2315 st->print_cr(" <Not Available>");
2316 }
2317 st->cr();
2318 st->print_cr("/sys/kernel/mm/transparent_hugepage/defrag (defrag/compaction efforts parameter):");
2319 if (!_print_ascii_file("/sys/kernel/mm/transparent_hugepage/defrag", st)) {
2320 st->print_cr(" <Not Available>");
2321 }
2322 st->cr();
2323 }
2324
2325 void os::Linux::print_ld_preload_file(outputStream* st) {
2326 _print_ascii_file("/etc/ld.so.preload", st, "\n/etc/ld.so.preload:");
2327 st->cr();
2328 }
2329
2330 void os::Linux::print_uptime_info(outputStream* st) {
2331 struct sysinfo sinfo;
2332 int ret = sysinfo(&sinfo);
2333 if (ret == 0) {
2334 os::print_dhm(st, "OS uptime:", (long) sinfo.uptime);
2335 }
2336 }
2337
2338
2339 void os::Linux::print_container_info(outputStream* st) {
2340 if (!OSContainer::is_containerized()) {
2341 return;
2342 }
2343
2344 st->print("container (cgroup) information:\n");
2345
2346 const char *p_ct = OSContainer::container_type();
2347 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "not supported");
2348
2349 char *p = OSContainer::cpu_cpuset_cpus();
2350 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "not supported");
2351 free(p);
2352
2353 p = OSContainer::cpu_cpuset_memory_nodes();
2354 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "not supported");
2355 free(p);
2356
2357 int i = OSContainer::active_processor_count();
2358 st->print("active_processor_count: ");
2359 if (i > 0) {
2360 st->print("%d\n", i);
2361 } else {
2362 st->print("not supported\n");
2363 }
2364
2365 i = OSContainer::cpu_quota();
2366 st->print("cpu_quota: ");
2367 if (i > 0) {
2368 st->print("%d\n", i);
2369 } else {
2370 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no quota");
2371 }
2372
2373 i = OSContainer::cpu_period();
2374 st->print("cpu_period: ");
2375 if (i > 0) {
2376 st->print("%d\n", i);
2377 } else {
2378 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no period");
2379 }
2380
2381 i = OSContainer::cpu_shares();
2382 st->print("cpu_shares: ");
2383 if (i > 0) {
2384 st->print("%d\n", i);
2385 } else {
2386 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no shares");
2387 }
2388
2389 jlong j = OSContainer::memory_limit_in_bytes();
2390 st->print("memory_limit_in_bytes: ");
2391 if (j > 0) {
2392 st->print(JLONG_FORMAT "\n", j);
2393 } else {
2394 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2395 }
2396
2397 j = OSContainer::memory_and_swap_limit_in_bytes();
2398 st->print("memory_and_swap_limit_in_bytes: ");
2399 if (j > 0) {
2400 st->print(JLONG_FORMAT "\n", j);
2401 } else {
2402 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2403 }
2404
2405 j = OSContainer::memory_soft_limit_in_bytes();
2406 st->print("memory_soft_limit_in_bytes: ");
2407 if (j > 0) {
2408 st->print(JLONG_FORMAT "\n", j);
2409 } else {
2410 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2411 }
2412
2413 j = OSContainer::OSContainer::memory_usage_in_bytes();
2414 st->print("memory_usage_in_bytes: ");
2415 if (j > 0) {
2416 st->print(JLONG_FORMAT "\n", j);
2417 } else {
2418 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2419 }
2420
2421 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2422 st->print("memory_max_usage_in_bytes: ");
2423 if (j > 0) {
2424 st->print(JLONG_FORMAT "\n", j);
2425 } else {
2426 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2427 }
2428 st->cr();
2429 }
2430
2431 void os::Linux::print_steal_info(outputStream* st) {
2432 if (has_initial_tick_info) {
2433 CPUPerfTicks pticks;
2434 bool res = os::Linux::get_tick_information(&pticks, -1);
2435
2436 if (res && pticks.has_steal_ticks) {
2437 uint64_t steal_ticks_difference = pticks.steal - initial_steal_ticks;
2438 uint64_t total_ticks_difference = pticks.total - initial_total_ticks;
2439 double steal_ticks_perc = 0.0;
2440 if (total_ticks_difference != 0) {
2441 steal_ticks_perc = (double) steal_ticks_difference / total_ticks_difference;
2442 }
2443 st->print_cr("Steal ticks since vm start: " UINT64_FORMAT, steal_ticks_difference);
2444 st->print_cr("Steal ticks percentage since vm start:%7.3f", steal_ticks_perc);
2445 }
2446 }
2447 }
2448
2449 void os::print_memory_info(outputStream* st) {
2450
2451 st->print("Memory:");
2452 st->print(" %dk page", os::vm_page_size()>>10);
2453
2454 // values in struct sysinfo are "unsigned long"
2455 struct sysinfo si;
2456 sysinfo(&si);
2457
2458 st->print(", physical " UINT64_FORMAT "k",
2459 os::physical_memory() >> 10);
2460 st->print("(" UINT64_FORMAT "k free)",
2461 os::available_memory() >> 10);
2462 st->print(", swap " UINT64_FORMAT "k",
2463 ((jlong)si.totalswap * si.mem_unit) >> 10);
2464 st->print("(" UINT64_FORMAT "k free)",
2465 ((jlong)si.freeswap * si.mem_unit) >> 10);
2466 st->cr();
2467 }
2468
2469 // Print the first "model name" line and the first "flags" line
2470 // that we find and nothing more. We assume "model name" comes
2471 // before "flags" so if we find a second "model name", then the
2472 // "flags" field is considered missing.
2473 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2474 #if defined(IA32) || defined(AMD64)
2475 // Other platforms have less repetitive cpuinfo files
2476 FILE *fp = fopen("/proc/cpuinfo", "r");
2477 if (fp) {
2478 while (!feof(fp)) {
2479 if (fgets(buf, buflen, fp)) {
2480 // Assume model name comes before flags
2481 bool model_name_printed = false;
2482 if (strstr(buf, "model name") != NULL) {
2483 if (!model_name_printed) {
2484 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2485 st->print_raw(buf);
2486 model_name_printed = true;
2487 } else {
2488 // model name printed but not flags? Odd, just return
2489 fclose(fp);
2490 return true;
2491 }
2492 }
2493 // print the flags line too
2494 if (strstr(buf, "flags") != NULL) {
2495 st->print_raw(buf);
2496 fclose(fp);
2497 return true;
2498 }
2499 }
2500 }
2501 fclose(fp);
2502 }
2503 #endif // x86 platforms
2504 return false;
2505 }
2506
2507 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2508 // Only print the model name if the platform provides this as a summary
2509 if (!print_model_name_and_flags(st, buf, buflen)) {
2510 st->print("\n/proc/cpuinfo:\n");
2511 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2512 st->print_cr(" <Not Available>");
2513 }
2514 }
2515 }
2516
2517 #if defined(AMD64) || defined(IA32) || defined(X32)
2518 const char* search_string = "model name";
2519 #elif defined(M68K)
2520 const char* search_string = "CPU";
2521 #elif defined(PPC64)
2522 const char* search_string = "cpu";
2523 #elif defined(S390)
2524 const char* search_string = "machine =";
2525 #elif defined(SPARC)
2526 const char* search_string = "cpu";
2527 #else
2528 const char* search_string = "Processor";
2529 #endif
2530
2531 // Parses the cpuinfo file for string representing the model name.
2532 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2533 FILE* fp = fopen("/proc/cpuinfo", "r");
2534 if (fp != NULL) {
2535 while (!feof(fp)) {
2536 char buf[256];
2537 if (fgets(buf, sizeof(buf), fp)) {
2538 char* start = strstr(buf, search_string);
2539 if (start != NULL) {
2540 char *ptr = start + strlen(search_string);
2541 char *end = buf + strlen(buf);
2542 while (ptr != end) {
2543 // skip whitespace and colon for the rest of the name.
2544 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2545 break;
2546 }
2547 ptr++;
2548 }
2549 if (ptr != end) {
2550 // reasonable string, get rid of newline and keep the rest
2551 char* nl = strchr(buf, '\n');
2552 if (nl != NULL) *nl = '\0';
2553 strncpy(cpuinfo, ptr, length);
2554 fclose(fp);
2555 return;
2556 }
2557 }
2558 }
2559 }
2560 fclose(fp);
2561 }
2562 // cpuinfo not found or parsing failed, just print generic string. The entire
2563 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2564 #if defined(AARCH64)
2565 strncpy(cpuinfo, "AArch64", length);
2566 #elif defined(AMD64)
2567 strncpy(cpuinfo, "x86_64", length);
2568 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2569 strncpy(cpuinfo, "ARM", length);
2570 #elif defined(IA32)
2571 strncpy(cpuinfo, "x86_32", length);
2572 #elif defined(IA64)
2573 strncpy(cpuinfo, "IA64", length);
2574 #elif defined(PPC)
2575 strncpy(cpuinfo, "PPC64", length);
2576 #elif defined(S390)
2577 strncpy(cpuinfo, "S390", length);
2578 #elif defined(SPARC)
2579 strncpy(cpuinfo, "sparcv9", length);
2580 #elif defined(ZERO_LIBARCH)
2581 strncpy(cpuinfo, ZERO_LIBARCH, length);
2582 #else
2583 strncpy(cpuinfo, "unknown", length);
2584 #endif
2585 }
2586
2587 static void print_signal_handler(outputStream* st, int sig,
2588 char* buf, size_t buflen);
2589
2590 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2591 st->print_cr("Signal Handlers:");
2592 print_signal_handler(st, SIGSEGV, buf, buflen);
2593 print_signal_handler(st, SIGBUS , buf, buflen);
2594 print_signal_handler(st, SIGFPE , buf, buflen);
2595 print_signal_handler(st, SIGPIPE, buf, buflen);
2596 print_signal_handler(st, SIGXFSZ, buf, buflen);
2597 print_signal_handler(st, SIGILL , buf, buflen);
2598 print_signal_handler(st, SR_signum, buf, buflen);
2599 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2600 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2601 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2602 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2603 #if defined(PPC64)
2604 print_signal_handler(st, SIGTRAP, buf, buflen);
2605 #endif
2606 }
2607
2608 static char saved_jvm_path[MAXPATHLEN] = {0};
2609
2610 // Find the full path to the current module, libjvm.so
2611 void os::jvm_path(char *buf, jint buflen) {
2612 // Error checking.
2613 if (buflen < MAXPATHLEN) {
2614 assert(false, "must use a large-enough buffer");
2615 buf[0] = '\0';
2616 return;
2617 }
2618 // Lazy resolve the path to current module.
2619 if (saved_jvm_path[0] != 0) {
2620 strcpy(buf, saved_jvm_path);
2621 return;
2622 }
2623
2624 char dli_fname[MAXPATHLEN];
2625 bool ret = dll_address_to_library_name(
2626 CAST_FROM_FN_PTR(address, os::jvm_path),
2627 dli_fname, sizeof(dli_fname), NULL);
2628 assert(ret, "cannot locate libjvm");
2629 char *rp = NULL;
2630 if (ret && dli_fname[0] != '\0') {
2631 rp = os::Posix::realpath(dli_fname, buf, buflen);
2632 }
2633 if (rp == NULL) {
2634 return;
2635 }
2636
2637 if (Arguments::sun_java_launcher_is_altjvm()) {
2638 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2639 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2640 // If "/jre/lib/" appears at the right place in the string, then
2641 // assume we are installed in a JDK and we're done. Otherwise, check
2642 // for a JAVA_HOME environment variable and fix up the path so it
2643 // looks like libjvm.so is installed there (append a fake suffix
2644 // hotspot/libjvm.so).
2645 const char *p = buf + strlen(buf) - 1;
2646 for (int count = 0; p > buf && count < 5; ++count) {
2647 for (--p; p > buf && *p != '/'; --p)
2648 /* empty */ ;
2649 }
2650
2651 if (strncmp(p, "/jre/lib/", 9) != 0) {
2652 // Look for JAVA_HOME in the environment.
2653 char* java_home_var = ::getenv("JAVA_HOME");
2654 if (java_home_var != NULL && java_home_var[0] != 0) {
2655 char* jrelib_p;
2656 int len;
2657
2658 // Check the current module name "libjvm.so".
2659 p = strrchr(buf, '/');
2660 if (p == NULL) {
2661 return;
2662 }
2663 assert(strstr(p, "/libjvm") == p, "invalid library name");
2664
2665 rp = os::Posix::realpath(java_home_var, buf, buflen);
2666 if (rp == NULL) {
2667 return;
2668 }
2669
2670 // determine if this is a legacy image or modules image
2671 // modules image doesn't have "jre" subdirectory
2672 len = strlen(buf);
2673 assert(len < buflen, "Ran out of buffer room");
2674 jrelib_p = buf + len;
2675 snprintf(jrelib_p, buflen-len, "/jre/lib");
2676 if (0 != access(buf, F_OK)) {
2677 snprintf(jrelib_p, buflen-len, "/lib");
2678 }
2679
2680 if (0 == access(buf, F_OK)) {
2681 // Use current module name "libjvm.so"
2682 len = strlen(buf);
2683 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2684 } else {
2685 // Go back to path of .so
2686 rp = os::Posix::realpath(dli_fname, buf, buflen);
2687 if (rp == NULL) {
2688 return;
2689 }
2690 }
2691 }
2692 }
2693 }
2694
2695 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2696 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2697 }
2698
2699 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2700 // no prefix required, not even "_"
2701 }
2702
2703 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2704 // no suffix required
2705 }
2706
2707 ////////////////////////////////////////////////////////////////////////////////
2708 // sun.misc.Signal support
2709
2710 static void UserHandler(int sig, void *siginfo, void *context) {
2711 // Ctrl-C is pressed during error reporting, likely because the error
2712 // handler fails to abort. Let VM die immediately.
2713 if (sig == SIGINT && VMError::is_error_reported()) {
2714 os::die();
2715 }
2716
2717 os::signal_notify(sig);
2718 }
2719
2720 void* os::user_handler() {
2721 return CAST_FROM_FN_PTR(void*, UserHandler);
2722 }
2723
2724 extern "C" {
2725 typedef void (*sa_handler_t)(int);
2726 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2727 }
2728
2729 void* os::signal(int signal_number, void* handler) {
2730 struct sigaction sigAct, oldSigAct;
2731
2732 sigfillset(&(sigAct.sa_mask));
2733 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2734 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2735
2736 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2737 // -1 means registration failed
2738 return (void *)-1;
2739 }
2740
2741 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2742 }
2743
2744 void os::signal_raise(int signal_number) {
2745 ::raise(signal_number);
2746 }
2747
2748 // The following code is moved from os.cpp for making this
2749 // code platform specific, which it is by its very nature.
2750
2751 // Will be modified when max signal is changed to be dynamic
2752 int os::sigexitnum_pd() {
2753 return NSIG;
2754 }
2755
2756 // a counter for each possible signal value
2757 static volatile jint pending_signals[NSIG+1] = { 0 };
2758
2759 // Linux(POSIX) specific hand shaking semaphore.
2760 static Semaphore* sig_sem = NULL;
2761 static PosixSemaphore sr_semaphore;
2762
2763 static void jdk_misc_signal_init() {
2764 // Initialize signal structures
2765 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2766
2767 // Initialize signal semaphore
2768 sig_sem = new Semaphore();
2769 }
2770
2771 void os::signal_notify(int sig) {
2772 if (sig_sem != NULL) {
2773 Atomic::inc(&pending_signals[sig]);
2774 sig_sem->signal();
2775 } else {
2776 // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2777 // initialization isn't called.
2778 assert(ReduceSignalUsage, "signal semaphore should be created");
2779 }
2780 }
2781
2782 static int check_pending_signals() {
2783 for (;;) {
2784 for (int i = 0; i < NSIG + 1; i++) {
2785 jint n = pending_signals[i];
2786 if (n > 0 && n == Atomic::cmpxchg(&pending_signals[i], n, n - 1)) {
2787 return i;
2788 }
2789 }
2790 JavaThread *thread = JavaThread::current();
2791 ThreadBlockInVM tbivm(thread);
2792
2793 bool threadIsSuspended;
2794 do {
2795 thread->set_suspend_equivalent();
2796 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2797 sig_sem->wait();
2798
2799 // were we externally suspended while we were waiting?
2800 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2801 if (threadIsSuspended) {
2802 // The semaphore has been incremented, but while we were waiting
2803 // another thread suspended us. We don't want to continue running
2804 // while suspended because that would surprise the thread that
2805 // suspended us.
2806 sig_sem->signal();
2807
2808 thread->java_suspend_self();
2809 }
2810 } while (threadIsSuspended);
2811 }
2812 }
2813
2814 int os::signal_wait() {
2815 return check_pending_signals();
2816 }
2817
2818 ////////////////////////////////////////////////////////////////////////////////
2819 // Virtual Memory
2820
2821 int os::vm_page_size() {
2822 // Seems redundant as all get out
2823 assert(os::Linux::page_size() != -1, "must call os::init");
2824 return os::Linux::page_size();
2825 }
2826
2827 // Solaris allocates memory by pages.
2828 int os::vm_allocation_granularity() {
2829 assert(os::Linux::page_size() != -1, "must call os::init");
2830 return os::Linux::page_size();
2831 }
2832
2833 // Rationale behind this function:
2834 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2835 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2836 // samples for JITted code. Here we create private executable mapping over the code cache
2837 // and then we can use standard (well, almost, as mapping can change) way to provide
2838 // info for the reporting script by storing timestamp and location of symbol
2839 void linux_wrap_code(char* base, size_t size) {
2840 static volatile jint cnt = 0;
2841
2842 if (!UseOprofile) {
2843 return;
2844 }
2845
2846 char buf[PATH_MAX+1];
2847 int num = Atomic::add(&cnt, 1);
2848
2849 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2850 os::get_temp_directory(), os::current_process_id(), num);
2851 unlink(buf);
2852
2853 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2854
2855 if (fd != -1) {
2856 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2857 if (rv != (off_t)-1) {
2858 if (::write(fd, "", 1) == 1) {
2859 mmap(base, size,
2860 PROT_READ|PROT_WRITE|PROT_EXEC,
2861 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2862 }
2863 }
2864 ::close(fd);
2865 unlink(buf);
2866 }
2867 }
2868
2869 static bool recoverable_mmap_error(int err) {
2870 // See if the error is one we can let the caller handle. This
2871 // list of errno values comes from JBS-6843484. I can't find a
2872 // Linux man page that documents this specific set of errno
2873 // values so while this list currently matches Solaris, it may
2874 // change as we gain experience with this failure mode.
2875 switch (err) {
2876 case EBADF:
2877 case EINVAL:
2878 case ENOTSUP:
2879 // let the caller deal with these errors
2880 return true;
2881
2882 default:
2883 // Any remaining errors on this OS can cause our reserved mapping
2884 // to be lost. That can cause confusion where different data
2885 // structures think they have the same memory mapped. The worst
2886 // scenario is if both the VM and a library think they have the
2887 // same memory mapped.
2888 return false;
2889 }
2890 }
2891
2892 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2893 int err) {
2894 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2895 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2896 os::strerror(err), err);
2897 }
2898
2899 static void warn_fail_commit_memory(char* addr, size_t size,
2900 size_t alignment_hint, bool exec,
2901 int err) {
2902 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2903 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2904 alignment_hint, exec, os::strerror(err), err);
2905 }
2906
2907 // NOTE: Linux kernel does not really reserve the pages for us.
2908 // All it does is to check if there are enough free pages
2909 // left at the time of mmap(). This could be a potential
2910 // problem.
2911 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2912 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2913 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2914 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2915 if (res != (uintptr_t) MAP_FAILED) {
2916 if (UseNUMAInterleaving) {
2917 numa_make_global(addr, size);
2918 }
2919 return 0;
2920 }
2921
2922 int err = errno; // save errno from mmap() call above
2923
2924 if (!recoverable_mmap_error(err)) {
2925 warn_fail_commit_memory(addr, size, exec, err);
2926 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2927 }
2928
2929 return err;
2930 }
2931
2932 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2933 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2934 }
2935
2936 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2937 const char* mesg) {
2938 assert(mesg != NULL, "mesg must be specified");
2939 int err = os::Linux::commit_memory_impl(addr, size, exec);
2940 if (err != 0) {
2941 // the caller wants all commit errors to exit with the specified mesg:
2942 warn_fail_commit_memory(addr, size, exec, err);
2943 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2944 }
2945 }
2946
2947 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2948 #ifndef MAP_HUGETLB
2949 #define MAP_HUGETLB 0x40000
2950 #endif
2951
2952 // If MAP_HUGETLB is set, and the system supports multiple huge page sizes,
2953 // flag bits [26:31] can be used to encode the log2 of the desired huge page size.
2954 // Otherwise the system's default huge page size will be used.
2955 // See mmap(2) man page for more info (since Linux 3.8).
2956 // https://lwn.net/Articles/533499/
2957 #ifndef MAP_HUGE_SHIFT
2958 #define MAP_HUGE_SHIFT 26
2959 #endif
2960
2961 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2962 #ifndef MADV_HUGEPAGE
2963 #define MADV_HUGEPAGE 14
2964 #endif
2965
2966 int os::Linux::commit_memory_impl(char* addr, size_t size,
2967 size_t alignment_hint, bool exec) {
2968 int err = os::Linux::commit_memory_impl(addr, size, exec);
2969 if (err == 0) {
2970 realign_memory(addr, size, alignment_hint);
2971 }
2972 return err;
2973 }
2974
2975 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2976 bool exec) {
2977 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2978 }
2979
2980 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2981 size_t alignment_hint, bool exec,
2982 const char* mesg) {
2983 assert(mesg != NULL, "mesg must be specified");
2984 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2985 if (err != 0) {
2986 // the caller wants all commit errors to exit with the specified mesg:
2987 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2988 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2989 }
2990 }
2991
2992 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2993 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2994 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2995 // be supported or the memory may already be backed by huge pages.
2996 ::madvise(addr, bytes, MADV_HUGEPAGE);
2997 }
2998 }
2999
3000 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
3001 // This method works by doing an mmap over an existing mmaping and effectively discarding
3002 // the existing pages. However it won't work for SHM-based large pages that cannot be
3003 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
3004 // small pages on top of the SHM segment. This method always works for small pages, so we
3005 // allow that in any case.
3006 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
3007 commit_memory(addr, bytes, alignment_hint, !ExecMem);
3008 }
3009 }
3010
3011 void os::numa_make_global(char *addr, size_t bytes) {
3012 Linux::numa_interleave_memory(addr, bytes);
3013 }
3014
3015 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
3016 // bind policy to MPOL_PREFERRED for the current thread.
3017 #define USE_MPOL_PREFERRED 0
3018
3019 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
3020 // To make NUMA and large pages more robust when both enabled, we need to ease
3021 // the requirements on where the memory should be allocated. MPOL_BIND is the
3022 // default policy and it will force memory to be allocated on the specified
3023 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
3024 // the specified node, but will not force it. Using this policy will prevent
3025 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
3026 // free large pages.
3027 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
3028 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
3029 }
3030
3031 bool os::numa_topology_changed() { return false; }
3032
3033 size_t os::numa_get_groups_num() {
3034 // Return just the number of nodes in which it's possible to allocate memory
3035 // (in numa terminology, configured nodes).
3036 return Linux::numa_num_configured_nodes();
3037 }
3038
3039 int os::numa_get_group_id() {
3040 int cpu_id = Linux::sched_getcpu();
3041 if (cpu_id != -1) {
3042 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
3043 if (lgrp_id != -1) {
3044 return lgrp_id;
3045 }
3046 }
3047 return 0;
3048 }
3049
3050 int os::numa_get_group_id_for_address(const void* address) {
3051 void** pages = const_cast<void**>(&address);
3052 int id = -1;
3053
3054 if (os::Linux::numa_move_pages(0, 1, pages, NULL, &id, 0) == -1) {
3055 return -1;
3056 }
3057 if (id < 0) {
3058 return -1;
3059 }
3060 return id;
3061 }
3062
3063 int os::Linux::get_existing_num_nodes() {
3064 int node;
3065 int highest_node_number = Linux::numa_max_node();
3066 int num_nodes = 0;
3067
3068 // Get the total number of nodes in the system including nodes without memory.
3069 for (node = 0; node <= highest_node_number; node++) {
3070 if (is_node_in_existing_nodes(node)) {
3071 num_nodes++;
3072 }
3073 }
3074 return num_nodes;
3075 }
3076
3077 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
3078 int highest_node_number = Linux::numa_max_node();
3079 size_t i = 0;
3080
3081 // Map all node ids in which it is possible to allocate memory. Also nodes are
3082 // not always consecutively available, i.e. available from 0 to the highest
3083 // node number. If the nodes have been bound explicitly using numactl membind,
3084 // then allocate memory from those nodes only.
3085 for (int node = 0; node <= highest_node_number; node++) {
3086 if (Linux::is_node_in_bound_nodes((unsigned int)node)) {
3087 ids[i++] = node;
3088 }
3089 }
3090 return i;
3091 }
3092
3093 bool os::get_page_info(char *start, page_info* info) {
3094 return false;
3095 }
3096
3097 char *os::scan_pages(char *start, char* end, page_info* page_expected,
3098 page_info* page_found) {
3099 return end;
3100 }
3101
3102
3103 int os::Linux::sched_getcpu_syscall(void) {
3104 unsigned int cpu = 0;
3105 int retval = -1;
3106
3107 #if defined(IA32)
3108 #ifndef SYS_getcpu
3109 #define SYS_getcpu 318
3110 #endif
3111 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
3112 #elif defined(AMD64)
3113 // Unfortunately we have to bring all these macros here from vsyscall.h
3114 // to be able to compile on old linuxes.
3115 #define __NR_vgetcpu 2
3116 #define VSYSCALL_START (-10UL << 20)
3117 #define VSYSCALL_SIZE 1024
3118 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
3119 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
3120 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
3121 retval = vgetcpu(&cpu, NULL, NULL);
3122 #endif
3123
3124 return (retval == -1) ? retval : cpu;
3125 }
3126
3127 void os::Linux::sched_getcpu_init() {
3128 // sched_getcpu() should be in libc.
3129 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
3130 dlsym(RTLD_DEFAULT, "sched_getcpu")));
3131
3132 // If it's not, try a direct syscall.
3133 if (sched_getcpu() == -1) {
3134 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
3135 (void*)&sched_getcpu_syscall));
3136 }
3137
3138 if (sched_getcpu() == -1) {
3139 vm_exit_during_initialization("getcpu(2) system call not supported by kernel");
3140 }
3141 }
3142
3143 // Something to do with the numa-aware allocator needs these symbols
3144 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
3145 extern "C" JNIEXPORT void numa_error(char *where) { }
3146
3147 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
3148 // load symbol from base version instead.
3149 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
3150 void *f = dlvsym(handle, name, "libnuma_1.1");
3151 if (f == NULL) {
3152 f = dlsym(handle, name);
3153 }
3154 return f;
3155 }
3156
3157 // Handle request to load libnuma symbol version 1.2 (API v2) only.
3158 // Return NULL if the symbol is not defined in this particular version.
3159 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
3160 return dlvsym(handle, name, "libnuma_1.2");
3161 }
3162
3163 bool os::Linux::libnuma_init() {
3164 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
3165 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
3166 if (handle != NULL) {
3167 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
3168 libnuma_dlsym(handle, "numa_node_to_cpus")));
3169 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
3170 libnuma_dlsym(handle, "numa_max_node")));
3171 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
3172 libnuma_dlsym(handle, "numa_num_configured_nodes")));
3173 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
3174 libnuma_dlsym(handle, "numa_available")));
3175 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
3176 libnuma_dlsym(handle, "numa_tonode_memory")));
3177 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
3178 libnuma_dlsym(handle, "numa_interleave_memory")));
3179 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
3180 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
3181 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
3182 libnuma_dlsym(handle, "numa_set_bind_policy")));
3183 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
3184 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
3185 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
3186 libnuma_dlsym(handle, "numa_distance")));
3187 set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t,
3188 libnuma_v2_dlsym(handle, "numa_get_membind")));
3189 set_numa_get_interleave_mask(CAST_TO_FN_PTR(numa_get_interleave_mask_func_t,
3190 libnuma_v2_dlsym(handle, "numa_get_interleave_mask")));
3191 set_numa_move_pages(CAST_TO_FN_PTR(numa_move_pages_func_t,
3192 libnuma_dlsym(handle, "numa_move_pages")));
3193 set_numa_set_preferred(CAST_TO_FN_PTR(numa_set_preferred_func_t,
3194 libnuma_dlsym(handle, "numa_set_preferred")));
3195
3196 if (numa_available() != -1) {
3197 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
3198 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
3199 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
3200 set_numa_interleave_bitmask(_numa_get_interleave_mask());
3201 set_numa_membind_bitmask(_numa_get_membind());
3202 // Create an index -> node mapping, since nodes are not always consecutive
3203 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3204 rebuild_nindex_to_node_map();
3205 // Create a cpu -> node mapping
3206 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3207 rebuild_cpu_to_node_map();
3208 return true;
3209 }
3210 }
3211 }
3212 return false;
3213 }
3214
3215 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
3216 // Creating guard page is very expensive. Java thread has HotSpot
3217 // guard pages, only enable glibc guard page for non-Java threads.
3218 // (Remember: compiler thread is a Java thread, too!)
3219 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
3220 }
3221
3222 void os::Linux::rebuild_nindex_to_node_map() {
3223 int highest_node_number = Linux::numa_max_node();
3224
3225 nindex_to_node()->clear();
3226 for (int node = 0; node <= highest_node_number; node++) {
3227 if (Linux::is_node_in_existing_nodes(node)) {
3228 nindex_to_node()->append(node);
3229 }
3230 }
3231 }
3232
3233 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
3234 // The table is later used in get_node_by_cpu().
3235 void os::Linux::rebuild_cpu_to_node_map() {
3236 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
3237 // in libnuma (possible values are starting from 16,
3238 // and continuing up with every other power of 2, but less
3239 // than the maximum number of CPUs supported by kernel), and
3240 // is a subject to change (in libnuma version 2 the requirements
3241 // are more reasonable) we'll just hardcode the number they use
3242 // in the library.
3243 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
3244
3245 size_t cpu_num = processor_count();
3246 size_t cpu_map_size = NCPUS / BitsPerCLong;
3247 size_t cpu_map_valid_size =
3248 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
3249
3250 cpu_to_node()->clear();
3251 cpu_to_node()->at_grow(cpu_num - 1);
3252
3253 size_t node_num = get_existing_num_nodes();
3254
3255 int distance = 0;
3256 int closest_distance = INT_MAX;
3257 int closest_node = 0;
3258 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
3259 for (size_t i = 0; i < node_num; i++) {
3260 // Check if node is configured (not a memory-less node). If it is not, find
3261 // the closest configured node. Check also if node is bound, i.e. it's allowed
3262 // to allocate memory from the node. If it's not allowed, map cpus in that node
3263 // to the closest node from which memory allocation is allowed.
3264 if (!is_node_in_configured_nodes(nindex_to_node()->at(i)) ||
3265 !is_node_in_bound_nodes(nindex_to_node()->at(i))) {
3266 closest_distance = INT_MAX;
3267 // Check distance from all remaining nodes in the system. Ignore distance
3268 // from itself, from another non-configured node, and from another non-bound
3269 // node.
3270 for (size_t m = 0; m < node_num; m++) {
3271 if (m != i &&
3272 is_node_in_configured_nodes(nindex_to_node()->at(m)) &&
3273 is_node_in_bound_nodes(nindex_to_node()->at(m))) {
3274 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3275 // If a closest node is found, update. There is always at least one
3276 // configured and bound node in the system so there is always at least
3277 // one node close.
3278 if (distance != 0 && distance < closest_distance) {
3279 closest_distance = distance;
3280 closest_node = nindex_to_node()->at(m);
3281 }
3282 }
3283 }
3284 } else {
3285 // Current node is already a configured node.
3286 closest_node = nindex_to_node()->at(i);
3287 }
3288
3289 // Get cpus from the original node and map them to the closest node. If node
3290 // is a configured node (not a memory-less node), then original node and
3291 // closest node are the same.
3292 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3293 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3294 if (cpu_map[j] != 0) {
3295 for (size_t k = 0; k < BitsPerCLong; k++) {
3296 if (cpu_map[j] & (1UL << k)) {
3297 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3298 }
3299 }
3300 }
3301 }
3302 }
3303 }
3304 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3305 }
3306
3307 int os::Linux::get_node_by_cpu(int cpu_id) {
3308 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3309 return cpu_to_node()->at(cpu_id);
3310 }
3311 return -1;
3312 }
3313
3314 GrowableArray<int>* os::Linux::_cpu_to_node;
3315 GrowableArray<int>* os::Linux::_nindex_to_node;
3316 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3317 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3318 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3319 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3320 os::Linux::numa_available_func_t os::Linux::_numa_available;
3321 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3322 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3323 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3324 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3325 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3326 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3327 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind;
3328 os::Linux::numa_get_interleave_mask_func_t os::Linux::_numa_get_interleave_mask;
3329 os::Linux::numa_move_pages_func_t os::Linux::_numa_move_pages;
3330 os::Linux::numa_set_preferred_func_t os::Linux::_numa_set_preferred;
3331 os::Linux::NumaAllocationPolicy os::Linux::_current_numa_policy;
3332 unsigned long* os::Linux::_numa_all_nodes;
3333 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3334 struct bitmask* os::Linux::_numa_nodes_ptr;
3335 struct bitmask* os::Linux::_numa_interleave_bitmask;
3336 struct bitmask* os::Linux::_numa_membind_bitmask;
3337
3338 bool os::pd_uncommit_memory(char* addr, size_t size) {
3339 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3340 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3341 return res != (uintptr_t) MAP_FAILED;
3342 }
3343
3344 static address get_stack_commited_bottom(address bottom, size_t size) {
3345 address nbot = bottom;
3346 address ntop = bottom + size;
3347
3348 size_t page_sz = os::vm_page_size();
3349 unsigned pages = size / page_sz;
3350
3351 unsigned char vec[1];
3352 unsigned imin = 1, imax = pages + 1, imid;
3353 int mincore_return_value = 0;
3354
3355 assert(imin <= imax, "Unexpected page size");
3356
3357 while (imin < imax) {
3358 imid = (imax + imin) / 2;
3359 nbot = ntop - (imid * page_sz);
3360
3361 // Use a trick with mincore to check whether the page is mapped or not.
3362 // mincore sets vec to 1 if page resides in memory and to 0 if page
3363 // is swapped output but if page we are asking for is unmapped
3364 // it returns -1,ENOMEM
3365 mincore_return_value = mincore(nbot, page_sz, vec);
3366
3367 if (mincore_return_value == -1) {
3368 // Page is not mapped go up
3369 // to find first mapped page
3370 if (errno != EAGAIN) {
3371 assert(errno == ENOMEM, "Unexpected mincore errno");
3372 imax = imid;
3373 }
3374 } else {
3375 // Page is mapped go down
3376 // to find first not mapped page
3377 imin = imid + 1;
3378 }
3379 }
3380
3381 nbot = nbot + page_sz;
3382
3383 // Adjust stack bottom one page up if last checked page is not mapped
3384 if (mincore_return_value == -1) {
3385 nbot = nbot + page_sz;
3386 }
3387
3388 return nbot;
3389 }
3390
3391 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3392 int mincore_return_value;
3393 const size_t stripe = 1024; // query this many pages each time
3394 unsigned char vec[stripe + 1];
3395 // set a guard
3396 vec[stripe] = 'X';
3397
3398 const size_t page_sz = os::vm_page_size();
3399 size_t pages = size / page_sz;
3400
3401 assert(is_aligned(start, page_sz), "Start address must be page aligned");
3402 assert(is_aligned(size, page_sz), "Size must be page aligned");
3403
3404 committed_start = NULL;
3405
3406 int loops = (pages + stripe - 1) / stripe;
3407 int committed_pages = 0;
3408 address loop_base = start;
3409 bool found_range = false;
3410
3411 for (int index = 0; index < loops && !found_range; index ++) {
3412 assert(pages > 0, "Nothing to do");
3413 int pages_to_query = (pages >= stripe) ? stripe : pages;
3414 pages -= pages_to_query;
3415
3416 // Get stable read
3417 while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3418
3419 // During shutdown, some memory goes away without properly notifying NMT,
3420 // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3421 // Bailout and return as not committed for now.
3422 if (mincore_return_value == -1 && errno == ENOMEM) {
3423 return false;
3424 }
3425
3426 assert(vec[stripe] == 'X', "overflow guard");
3427 assert(mincore_return_value == 0, "Range must be valid");
3428 // Process this stripe
3429 for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3430 if ((vec[vecIdx] & 0x01) == 0) { // not committed
3431 // End of current contiguous region
3432 if (committed_start != NULL) {
3433 found_range = true;
3434 break;
3435 }
3436 } else { // committed
3437 // Start of region
3438 if (committed_start == NULL) {
3439 committed_start = loop_base + page_sz * vecIdx;
3440 }
3441 committed_pages ++;
3442 }
3443 }
3444
3445 loop_base += pages_to_query * page_sz;
3446 }
3447
3448 if (committed_start != NULL) {
3449 assert(committed_pages > 0, "Must have committed region");
3450 assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3451 assert(committed_start >= start && committed_start < start + size, "Out of range");
3452 committed_size = page_sz * committed_pages;
3453 return true;
3454 } else {
3455 assert(committed_pages == 0, "Should not have committed region");
3456 return false;
3457 }
3458 }
3459
3460
3461 // Linux uses a growable mapping for the stack, and if the mapping for
3462 // the stack guard pages is not removed when we detach a thread the
3463 // stack cannot grow beyond the pages where the stack guard was
3464 // mapped. If at some point later in the process the stack expands to
3465 // that point, the Linux kernel cannot expand the stack any further
3466 // because the guard pages are in the way, and a segfault occurs.
3467 //
3468 // However, it's essential not to split the stack region by unmapping
3469 // a region (leaving a hole) that's already part of the stack mapping,
3470 // so if the stack mapping has already grown beyond the guard pages at
3471 // the time we create them, we have to truncate the stack mapping.
3472 // So, we need to know the extent of the stack mapping when
3473 // create_stack_guard_pages() is called.
3474
3475 // We only need this for stacks that are growable: at the time of
3476 // writing thread stacks don't use growable mappings (i.e. those
3477 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3478 // only applies to the main thread.
3479
3480 // If the (growable) stack mapping already extends beyond the point
3481 // where we're going to put our guard pages, truncate the mapping at
3482 // that point by munmap()ping it. This ensures that when we later
3483 // munmap() the guard pages we don't leave a hole in the stack
3484 // mapping. This only affects the main/primordial thread
3485
3486 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3487 if (os::is_primordial_thread()) {
3488 // As we manually grow stack up to bottom inside create_attached_thread(),
3489 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3490 // we don't need to do anything special.
3491 // Check it first, before calling heavy function.
3492 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3493 unsigned char vec[1];
3494
3495 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3496 // Fallback to slow path on all errors, including EAGAIN
3497 stack_extent = (uintptr_t) get_stack_commited_bottom(
3498 os::Linux::initial_thread_stack_bottom(),
3499 (size_t)addr - stack_extent);
3500 }
3501
3502 if (stack_extent < (uintptr_t)addr) {
3503 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3504 }
3505 }
3506
3507 return os::commit_memory(addr, size, !ExecMem);
3508 }
3509
3510 // If this is a growable mapping, remove the guard pages entirely by
3511 // munmap()ping them. If not, just call uncommit_memory(). This only
3512 // affects the main/primordial thread, but guard against future OS changes.
3513 // It's safe to always unmap guard pages for primordial thread because we
3514 // always place it right after end of the mapped region.
3515
3516 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3517 uintptr_t stack_extent, stack_base;
3518
3519 if (os::is_primordial_thread()) {
3520 return ::munmap(addr, size) == 0;
3521 }
3522
3523 return os::uncommit_memory(addr, size);
3524 }
3525
3526 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3527 // at 'requested_addr'. If there are existing memory mappings at the same
3528 // location, however, they will be overwritten. If 'fixed' is false,
3529 // 'requested_addr' is only treated as a hint, the return value may or
3530 // may not start from the requested address. Unlike Linux mmap(), this
3531 // function returns NULL to indicate failure.
3532 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3533 char * addr;
3534 int flags;
3535
3536 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3537 if (fixed) {
3538 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3539 flags |= MAP_FIXED;
3540 }
3541
3542 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3543 // touch an uncommitted page. Otherwise, the read/write might
3544 // succeed if we have enough swap space to back the physical page.
3545 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3546 flags, -1, 0);
3547
3548 return addr == MAP_FAILED ? NULL : addr;
3549 }
3550
3551 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3552 // (req_addr != NULL) or with a given alignment.
3553 // - bytes shall be a multiple of alignment.
3554 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3555 // - alignment sets the alignment at which memory shall be allocated.
3556 // It must be a multiple of allocation granularity.
3557 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3558 // req_addr or NULL.
3559 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3560
3561 size_t extra_size = bytes;
3562 if (req_addr == NULL && alignment > 0) {
3563 extra_size += alignment;
3564 }
3565
3566 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3567 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3568 -1, 0);
3569 if (start == MAP_FAILED) {
3570 start = NULL;
3571 } else {
3572 if (req_addr != NULL) {
3573 if (start != req_addr) {
3574 ::munmap(start, extra_size);
3575 start = NULL;
3576 }
3577 } else {
3578 char* const start_aligned = align_up(start, alignment);
3579 char* const end_aligned = start_aligned + bytes;
3580 char* const end = start + extra_size;
3581 if (start_aligned > start) {
3582 ::munmap(start, start_aligned - start);
3583 }
3584 if (end_aligned < end) {
3585 ::munmap(end_aligned, end - end_aligned);
3586 }
3587 start = start_aligned;
3588 }
3589 }
3590 return start;
3591 }
3592
3593 static int anon_munmap(char * addr, size_t size) {
3594 return ::munmap(addr, size) == 0;
3595 }
3596
3597 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3598 size_t alignment_hint) {
3599 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3600 }
3601
3602 bool os::pd_release_memory(char* addr, size_t size) {
3603 return anon_munmap(addr, size);
3604 }
3605
3606 static bool linux_mprotect(char* addr, size_t size, int prot) {
3607 // Linux wants the mprotect address argument to be page aligned.
3608 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3609
3610 // According to SUSv3, mprotect() should only be used with mappings
3611 // established by mmap(), and mmap() always maps whole pages. Unaligned
3612 // 'addr' likely indicates problem in the VM (e.g. trying to change
3613 // protection of malloc'ed or statically allocated memory). Check the
3614 // caller if you hit this assert.
3615 assert(addr == bottom, "sanity check");
3616
3617 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3618 Events::log(NULL, "Protecting memory [" INTPTR_FORMAT "," INTPTR_FORMAT "] with protection modes %x", p2i(bottom), p2i(bottom+size), prot);
3619 return ::mprotect(bottom, size, prot) == 0;
3620 }
3621
3622 // Set protections specified
3623 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3624 bool is_committed) {
3625 unsigned int p = 0;
3626 switch (prot) {
3627 case MEM_PROT_NONE: p = PROT_NONE; break;
3628 case MEM_PROT_READ: p = PROT_READ; break;
3629 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3630 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3631 default:
3632 ShouldNotReachHere();
3633 }
3634 // is_committed is unused.
3635 return linux_mprotect(addr, bytes, p);
3636 }
3637
3638 bool os::guard_memory(char* addr, size_t size) {
3639 return linux_mprotect(addr, size, PROT_NONE);
3640 }
3641
3642 bool os::unguard_memory(char* addr, size_t size) {
3643 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3644 }
3645
3646 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3647 size_t page_size) {
3648 bool result = false;
3649 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3650 MAP_ANONYMOUS|MAP_PRIVATE,
3651 -1, 0);
3652 if (p != MAP_FAILED) {
3653 void *aligned_p = align_up(p, page_size);
3654
3655 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3656
3657 munmap(p, page_size * 2);
3658 }
3659
3660 if (warn && !result) {
3661 warning("TransparentHugePages is not supported by the operating system.");
3662 }
3663
3664 return result;
3665 }
3666
3667 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3668 bool result = false;
3669 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3670 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3671 -1, 0);
3672
3673 if (p != MAP_FAILED) {
3674 // We don't know if this really is a huge page or not.
3675 FILE *fp = fopen("/proc/self/maps", "r");
3676 if (fp) {
3677 while (!feof(fp)) {
3678 char chars[257];
3679 long x = 0;
3680 if (fgets(chars, sizeof(chars), fp)) {
3681 if (sscanf(chars, "%lx-%*x", &x) == 1
3682 && x == (long)p) {
3683 if (strstr (chars, "hugepage")) {
3684 result = true;
3685 break;
3686 }
3687 }
3688 }
3689 }
3690 fclose(fp);
3691 }
3692 munmap(p, page_size);
3693 }
3694
3695 if (warn && !result) {
3696 warning("HugeTLBFS is not supported by the operating system.");
3697 }
3698
3699 return result;
3700 }
3701
3702 // From the coredump_filter documentation:
3703 //
3704 // - (bit 0) anonymous private memory
3705 // - (bit 1) anonymous shared memory
3706 // - (bit 2) file-backed private memory
3707 // - (bit 3) file-backed shared memory
3708 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3709 // effective only if the bit 2 is cleared)
3710 // - (bit 5) hugetlb private memory
3711 // - (bit 6) hugetlb shared memory
3712 // - (bit 7) dax private memory
3713 // - (bit 8) dax shared memory
3714 //
3715 static void set_coredump_filter(CoredumpFilterBit bit) {
3716 FILE *f;
3717 long cdm;
3718
3719 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3720 return;
3721 }
3722
3723 if (fscanf(f, "%lx", &cdm) != 1) {
3724 fclose(f);
3725 return;
3726 }
3727
3728 long saved_cdm = cdm;
3729 rewind(f);
3730 cdm |= bit;
3731
3732 if (cdm != saved_cdm) {
3733 fprintf(f, "%#lx", cdm);
3734 }
3735
3736 fclose(f);
3737 }
3738
3739 // Large page support
3740
3741 static size_t _large_page_size = 0;
3742
3743 size_t os::Linux::find_default_large_page_size() {
3744 size_t large_page_size = 0;
3745
3746 // large_page_size on Linux is used to round up heap size. x86 uses either
3747 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3748 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3749 // page as large as 256M.
3750 //
3751 // Here we try to figure out page size by parsing /proc/meminfo and looking
3752 // for a line with the following format:
3753 // Hugepagesize: 2048 kB
3754 //
3755 // If we can't determine the value (e.g. /proc is not mounted, or the text
3756 // format has been changed), we'll use the largest page size supported by
3757 // the processor.
3758
3759 #ifndef ZERO
3760 large_page_size =
3761 AARCH64_ONLY(2 * M)
3762 AMD64_ONLY(2 * M)
3763 ARM32_ONLY(2 * M)
3764 IA32_ONLY(4 * M)
3765 IA64_ONLY(256 * M)
3766 PPC_ONLY(4 * M)
3767 S390_ONLY(1 * M)
3768 SPARC_ONLY(4 * M);
3769 #endif // ZERO
3770
3771 FILE *fp = fopen("/proc/meminfo", "r");
3772 if (fp) {
3773 while (!feof(fp)) {
3774 int x = 0;
3775 char buf[16];
3776 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3777 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3778 large_page_size = x * K;
3779 break;
3780 }
3781 } else {
3782 // skip to next line
3783 for (;;) {
3784 int ch = fgetc(fp);
3785 if (ch == EOF || ch == (int)'\n') break;
3786 }
3787 }
3788 }
3789 fclose(fp);
3790 }
3791 return large_page_size;
3792 }
3793
3794 bool os::Linux::is_valid_large_page_size(size_t large_page_size) {
3795 // We need to scan /sys/kernel/mm/hugepages
3796 // to discover the available page sizes
3797 const char* sys_hugepages = "/sys/kernel/mm/hugepages";
3798 if (dir_is_empty(sys_hugepages)) {
3799 return false;
3800 }
3801
3802 DIR *dir = opendir(sys_hugepages);
3803 if (dir == NULL) {
3804 return false;
3805 }
3806
3807 struct dirent *entry;
3808 size_t page_size;
3809 bool is_valid = false;
3810 while ( (entry = readdir(dir)) != NULL) {
3811 if (entry->d_type == DT_DIR &&
3812 sscanf(entry->d_name, "hugepages-%zukB", &page_size) == 1) {
3813 // The kernel is using kB, hotspot uses bytes
3814 if (large_page_size == page_size * K) {
3815 is_valid = true;
3816 break;
3817 }
3818 }
3819 }
3820 closedir(dir);
3821 return is_valid;
3822 }
3823
3824 size_t os::Linux::setup_large_page_size() {
3825 _default_large_page_size = Linux::find_default_large_page_size();
3826 _large_page_size = _default_large_page_size;
3827
3828 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != _large_page_size ) {
3829 if (is_valid_large_page_size(LargePageSizeInBytes)) {
3830 _large_page_size = LargePageSizeInBytes;
3831 } else {
3832 warning("Setting LargePageSizeInBytes=" SIZE_FORMAT " has no effect on this OS. Default large page size is "
3833 SIZE_FORMAT "%s.",
3834 LargePageSizeInBytes,
3835 byte_size_in_proper_unit(_large_page_size), proper_unit_for_byte_size(_large_page_size));
3836 }
3837 }
3838
3839 const size_t default_page_size = (size_t)Linux::page_size();
3840 if (_large_page_size > default_page_size) {
3841 _page_sizes[0] = _large_page_size;
3842 _page_sizes[1] = default_page_size;
3843 _page_sizes[2] = 0;
3844 }
3845
3846 return _large_page_size;
3847 }
3848
3849 size_t os::Linux::default_large_page_size() {
3850 return _default_large_page_size;
3851 }
3852
3853 bool os::Linux::setup_large_page_type(size_t page_size) {
3854 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3855 FLAG_IS_DEFAULT(UseSHM) &&
3856 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3857
3858 // The type of large pages has not been specified by the user.
3859
3860 // Try UseHugeTLBFS and then UseSHM.
3861 UseHugeTLBFS = UseSHM = true;
3862
3863 // Don't try UseTransparentHugePages since there are known
3864 // performance issues with it turned on. This might change in the future.
3865 UseTransparentHugePages = false;
3866 }
3867
3868 if (UseTransparentHugePages) {
3869 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3870 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3871 UseHugeTLBFS = false;
3872 UseSHM = false;
3873 return true;
3874 }
3875 UseTransparentHugePages = false;
3876 }
3877
3878 if (UseHugeTLBFS) {
3879 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3880 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3881 UseSHM = false;
3882 return true;
3883 }
3884 UseHugeTLBFS = false;
3885 }
3886
3887 return UseSHM;
3888 }
3889
3890 void os::large_page_init() {
3891 if (!UseLargePages &&
3892 !UseTransparentHugePages &&
3893 !UseHugeTLBFS &&
3894 !UseSHM) {
3895 // Not using large pages.
3896 return;
3897 }
3898
3899 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3900 // The user explicitly turned off large pages.
3901 // Ignore the rest of the large pages flags.
3902 UseTransparentHugePages = false;
3903 UseHugeTLBFS = false;
3904 UseSHM = false;
3905 return;
3906 }
3907
3908 size_t large_page_size = Linux::setup_large_page_size();
3909 UseLargePages = Linux::setup_large_page_type(large_page_size);
3910
3911 set_coredump_filter(LARGEPAGES_BIT);
3912 }
3913
3914 #ifndef SHM_HUGETLB
3915 #define SHM_HUGETLB 04000
3916 #endif
3917
3918 #define shm_warning_format(format, ...) \
3919 do { \
3920 if (UseLargePages && \
3921 (!FLAG_IS_DEFAULT(UseLargePages) || \
3922 !FLAG_IS_DEFAULT(UseSHM) || \
3923 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3924 warning(format, __VA_ARGS__); \
3925 } \
3926 } while (0)
3927
3928 #define shm_warning(str) shm_warning_format("%s", str)
3929
3930 #define shm_warning_with_errno(str) \
3931 do { \
3932 int err = errno; \
3933 shm_warning_format(str " (error = %d)", err); \
3934 } while (0)
3935
3936 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3937 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3938
3939 if (!is_aligned(alignment, SHMLBA)) {
3940 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3941 return NULL;
3942 }
3943
3944 // To ensure that we get 'alignment' aligned memory from shmat,
3945 // we pre-reserve aligned virtual memory and then attach to that.
3946
3947 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3948 if (pre_reserved_addr == NULL) {
3949 // Couldn't pre-reserve aligned memory.
3950 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3951 return NULL;
3952 }
3953
3954 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3955 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3956
3957 if ((intptr_t)addr == -1) {
3958 int err = errno;
3959 shm_warning_with_errno("Failed to attach shared memory.");
3960
3961 assert(err != EACCES, "Unexpected error");
3962 assert(err != EIDRM, "Unexpected error");
3963 assert(err != EINVAL, "Unexpected error");
3964
3965 // Since we don't know if the kernel unmapped the pre-reserved memory area
3966 // we can't unmap it, since that would potentially unmap memory that was
3967 // mapped from other threads.
3968 return NULL;
3969 }
3970
3971 return addr;
3972 }
3973
3974 static char* shmat_at_address(int shmid, char* req_addr) {
3975 if (!is_aligned(req_addr, SHMLBA)) {
3976 assert(false, "Requested address needs to be SHMLBA aligned");
3977 return NULL;
3978 }
3979
3980 char* addr = (char*)shmat(shmid, req_addr, 0);
3981
3982 if ((intptr_t)addr == -1) {
3983 shm_warning_with_errno("Failed to attach shared memory.");
3984 return NULL;
3985 }
3986
3987 return addr;
3988 }
3989
3990 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3991 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3992 if (req_addr != NULL) {
3993 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3994 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3995 return shmat_at_address(shmid, req_addr);
3996 }
3997
3998 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3999 // return large page size aligned memory addresses when req_addr == NULL.
4000 // However, if the alignment is larger than the large page size, we have
4001 // to manually ensure that the memory returned is 'alignment' aligned.
4002 if (alignment > os::large_page_size()) {
4003 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
4004 return shmat_with_alignment(shmid, bytes, alignment);
4005 } else {
4006 return shmat_at_address(shmid, NULL);
4007 }
4008 }
4009
4010 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
4011 char* req_addr, bool exec) {
4012 // "exec" is passed in but not used. Creating the shared image for
4013 // the code cache doesn't have an SHM_X executable permission to check.
4014 assert(UseLargePages && UseSHM, "only for SHM large pages");
4015 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
4016 assert(is_aligned(req_addr, alignment), "Unaligned address");
4017
4018 if (!is_aligned(bytes, os::large_page_size())) {
4019 return NULL; // Fallback to small pages.
4020 }
4021
4022 // Create a large shared memory region to attach to based on size.
4023 // Currently, size is the total size of the heap.
4024 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
4025 if (shmid == -1) {
4026 // Possible reasons for shmget failure:
4027 // 1. shmmax is too small for Java heap.
4028 // > check shmmax value: cat /proc/sys/kernel/shmmax
4029 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
4030 // 2. not enough large page memory.
4031 // > check available large pages: cat /proc/meminfo
4032 // > increase amount of large pages:
4033 // echo new_value > /proc/sys/vm/nr_hugepages
4034 // Note 1: different Linux may use different name for this property,
4035 // e.g. on Redhat AS-3 it is "hugetlb_pool".
4036 // Note 2: it's possible there's enough physical memory available but
4037 // they are so fragmented after a long run that they can't
4038 // coalesce into large pages. Try to reserve large pages when
4039 // the system is still "fresh".
4040 shm_warning_with_errno("Failed to reserve shared memory.");
4041 return NULL;
4042 }
4043
4044 // Attach to the region.
4045 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
4046
4047 // Remove shmid. If shmat() is successful, the actual shared memory segment
4048 // will be deleted when it's detached by shmdt() or when the process
4049 // terminates. If shmat() is not successful this will remove the shared
4050 // segment immediately.
4051 shmctl(shmid, IPC_RMID, NULL);
4052
4053 return addr;
4054 }
4055
4056 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
4057 int error) {
4058 assert(error == ENOMEM, "Only expect to fail if no memory is available");
4059
4060 bool warn_on_failure = UseLargePages &&
4061 (!FLAG_IS_DEFAULT(UseLargePages) ||
4062 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
4063 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
4064
4065 if (warn_on_failure) {
4066 char msg[128];
4067 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
4068 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
4069 warning("%s", msg);
4070 }
4071 }
4072
4073 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
4074 char* req_addr,
4075 bool exec) {
4076 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
4077 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
4078 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
4079
4080 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
4081 int flags = MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB;
4082
4083 if (os::large_page_size() != default_large_page_size()) {
4084 flags |= (exact_log2(os::large_page_size()) << MAP_HUGE_SHIFT);
4085 }
4086 char* addr = (char*)::mmap(req_addr, bytes, prot, flags, -1, 0);
4087
4088 if (addr == MAP_FAILED) {
4089 warn_on_large_pages_failure(req_addr, bytes, errno);
4090 return NULL;
4091 }
4092
4093 assert(is_aligned(addr, os::large_page_size()), "Must be");
4094
4095 return addr;
4096 }
4097
4098 // Reserve memory using mmap(MAP_HUGETLB).
4099 // - bytes shall be a multiple of alignment.
4100 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
4101 // - alignment sets the alignment at which memory shall be allocated.
4102 // It must be a multiple of allocation granularity.
4103 // Returns address of memory or NULL. If req_addr was not NULL, will only return
4104 // req_addr or NULL.
4105 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
4106 size_t alignment,
4107 char* req_addr,
4108 bool exec) {
4109 size_t large_page_size = os::large_page_size();
4110 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
4111
4112 assert(is_aligned(req_addr, alignment), "Must be");
4113 assert(is_aligned(bytes, alignment), "Must be");
4114
4115 // First reserve - but not commit - the address range in small pages.
4116 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
4117
4118 if (start == NULL) {
4119 return NULL;
4120 }
4121
4122 assert(is_aligned(start, alignment), "Must be");
4123
4124 char* end = start + bytes;
4125
4126 // Find the regions of the allocated chunk that can be promoted to large pages.
4127 char* lp_start = align_up(start, large_page_size);
4128 char* lp_end = align_down(end, large_page_size);
4129
4130 size_t lp_bytes = lp_end - lp_start;
4131
4132 assert(is_aligned(lp_bytes, large_page_size), "Must be");
4133
4134 if (lp_bytes == 0) {
4135 // The mapped region doesn't even span the start and the end of a large page.
4136 // Fall back to allocate a non-special area.
4137 ::munmap(start, end - start);
4138 return NULL;
4139 }
4140
4141 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
4142 int flags = MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED;
4143 void* result;
4144
4145 // Commit small-paged leading area.
4146 if (start != lp_start) {
4147 result = ::mmap(start, lp_start - start, prot, flags, -1, 0);
4148 if (result == MAP_FAILED) {
4149 ::munmap(lp_start, end - lp_start);
4150 return NULL;
4151 }
4152 }
4153
4154 // Commit large-paged area.
4155 flags |= MAP_HUGETLB;
4156
4157 if (os::large_page_size() != default_large_page_size()) {
4158 flags |= (exact_log2(os::large_page_size()) << MAP_HUGE_SHIFT);
4159 }
4160
4161 result = ::mmap(lp_start, lp_bytes, prot, flags, -1, 0);
4162 if (result == MAP_FAILED) {
4163 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
4164 // If the mmap above fails, the large pages region will be unmapped and we
4165 // have regions before and after with small pages. Release these regions.
4166 //
4167 // | mapped | unmapped | mapped |
4168 // ^ ^ ^ ^
4169 // start lp_start lp_end end
4170 //
4171 ::munmap(start, lp_start - start);
4172 ::munmap(lp_end, end - lp_end);
4173 return NULL;
4174 }
4175
4176 // Commit small-paged trailing area.
4177 if (lp_end != end) {
4178 result = ::mmap(lp_end, end - lp_end, prot,
4179 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
4180 -1, 0);
4181 if (result == MAP_FAILED) {
4182 ::munmap(start, lp_end - start);
4183 return NULL;
4184 }
4185 }
4186
4187 return start;
4188 }
4189
4190 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
4191 size_t alignment,
4192 char* req_addr,
4193 bool exec) {
4194 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
4195 assert(is_aligned(req_addr, alignment), "Must be");
4196 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
4197 assert(is_power_of_2(os::large_page_size()), "Must be");
4198 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
4199
4200 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
4201 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
4202 } else {
4203 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
4204 }
4205 }
4206
4207 char* os::pd_reserve_memory_special(size_t bytes, size_t alignment,
4208 char* req_addr, bool exec) {
4209 assert(UseLargePages, "only for large pages");
4210
4211 char* addr;
4212 if (UseSHM) {
4213 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
4214 } else {
4215 assert(UseHugeTLBFS, "must be");
4216 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
4217 }
4218
4219 if (addr != NULL) {
4220 if (UseNUMAInterleaving) {
4221 numa_make_global(addr, bytes);
4222 }
4223 }
4224
4225 return addr;
4226 }
4227
4228 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
4229 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
4230 return shmdt(base) == 0;
4231 }
4232
4233 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
4234 return pd_release_memory(base, bytes);
4235 }
4236
4237 bool os::pd_release_memory_special(char* base, size_t bytes) {
4238 assert(UseLargePages, "only for large pages");
4239 bool res;
4240
4241 if (UseSHM) {
4242 res = os::Linux::release_memory_special_shm(base, bytes);
4243 } else {
4244 assert(UseHugeTLBFS, "must be");
4245 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
4246 }
4247 return res;
4248 }
4249
4250 size_t os::large_page_size() {
4251 return _large_page_size;
4252 }
4253
4254 // With SysV SHM the entire memory region must be allocated as shared
4255 // memory.
4256 // HugeTLBFS allows application to commit large page memory on demand.
4257 // However, when committing memory with HugeTLBFS fails, the region
4258 // that was supposed to be committed will lose the old reservation
4259 // and allow other threads to steal that memory region. Because of this
4260 // behavior we can't commit HugeTLBFS memory.
4261 bool os::can_commit_large_page_memory() {
4262 return UseTransparentHugePages;
4263 }
4264
4265 bool os::can_execute_large_page_memory() {
4266 return UseTransparentHugePages || UseHugeTLBFS;
4267 }
4268
4269 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
4270 assert(file_desc >= 0, "file_desc is not valid");
4271 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
4272 if (result != NULL) {
4273 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
4274 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
4275 }
4276 }
4277 return result;
4278 }
4279
4280 // Reserve memory at an arbitrary address, only if that area is
4281 // available (and not reserved for something else).
4282
4283 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
4284 // Assert only that the size is a multiple of the page size, since
4285 // that's all that mmap requires, and since that's all we really know
4286 // about at this low abstraction level. If we need higher alignment,
4287 // we can either pass an alignment to this method or verify alignment
4288 // in one of the methods further up the call chain. See bug 5044738.
4289 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
4290
4291 // Repeatedly allocate blocks until the block is allocated at the
4292 // right spot.
4293
4294 // Linux mmap allows caller to pass an address as hint; give it a try first,
4295 // if kernel honors the hint then we can return immediately.
4296 char * addr = anon_mmap(requested_addr, bytes, false);
4297 if (addr == requested_addr) {
4298 return requested_addr;
4299 }
4300
4301 if (addr != NULL) {
4302 // mmap() is successful but it fails to reserve at the requested address
4303 anon_munmap(addr, bytes);
4304 }
4305
4306 return NULL;
4307 }
4308
4309 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4310 void os::infinite_sleep() {
4311 while (true) { // sleep forever ...
4312 ::sleep(100); // ... 100 seconds at a time
4313 }
4314 }
4315
4316 // Used to convert frequent JVM_Yield() to nops
4317 bool os::dont_yield() {
4318 return DontYieldALot;
4319 }
4320
4321 // Linux CFS scheduler (since 2.6.23) does not guarantee sched_yield(2) will
4322 // actually give up the CPU. Since skip buddy (v2.6.28):
4323 //
4324 // * Sets the yielding task as skip buddy for current CPU's run queue.
4325 // * Picks next from run queue, if empty, picks a skip buddy (can be the yielding task).
4326 // * Clears skip buddies for this run queue (yielding task no longer a skip buddy).
4327 //
4328 // An alternative is calling os::naked_short_nanosleep with a small number to avoid
4329 // getting re-scheduled immediately.
4330 //
4331 void os::naked_yield() {
4332 sched_yield();
4333 }
4334
4335 ////////////////////////////////////////////////////////////////////////////////
4336 // thread priority support
4337
4338 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4339 // only supports dynamic priority, static priority must be zero. For real-time
4340 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4341 // However, for large multi-threaded applications, SCHED_RR is not only slower
4342 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4343 // of 5 runs - Sep 2005).
4344 //
4345 // The following code actually changes the niceness of kernel-thread/LWP. It
4346 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4347 // not the entire user process, and user level threads are 1:1 mapped to kernel
4348 // threads. It has always been the case, but could change in the future. For
4349 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4350 // It is only used when ThreadPriorityPolicy=1 and may require system level permission
4351 // (e.g., root privilege or CAP_SYS_NICE capability).
4352
4353 int os::java_to_os_priority[CriticalPriority + 1] = {
4354 19, // 0 Entry should never be used
4355
4356 4, // 1 MinPriority
4357 3, // 2
4358 2, // 3
4359
4360 1, // 4
4361 0, // 5 NormPriority
4362 -1, // 6
4363
4364 -2, // 7
4365 -3, // 8
4366 -4, // 9 NearMaxPriority
4367
4368 -5, // 10 MaxPriority
4369
4370 -5 // 11 CriticalPriority
4371 };
4372
4373 static int prio_init() {
4374 if (ThreadPriorityPolicy == 1) {
4375 if (geteuid() != 0) {
4376 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy) && !FLAG_IS_JIMAGE_RESOURCE(ThreadPriorityPolicy)) {
4377 warning("-XX:ThreadPriorityPolicy=1 may require system level permission, " \
4378 "e.g., being the root user. If the necessary permission is not " \
4379 "possessed, changes to priority will be silently ignored.");
4380 }
4381 }
4382 }
4383 if (UseCriticalJavaThreadPriority) {
4384 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4385 }
4386 return 0;
4387 }
4388
4389 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4390 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4391
4392 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4393 return (ret == 0) ? OS_OK : OS_ERR;
4394 }
4395
4396 OSReturn os::get_native_priority(const Thread* const thread,
4397 int *priority_ptr) {
4398 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4399 *priority_ptr = java_to_os_priority[NormPriority];
4400 return OS_OK;
4401 }
4402
4403 errno = 0;
4404 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4405 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4406 }
4407
4408 ////////////////////////////////////////////////////////////////////////////////
4409 // suspend/resume support
4410
4411 // The low-level signal-based suspend/resume support is a remnant from the
4412 // old VM-suspension that used to be for java-suspension, safepoints etc,
4413 // within hotspot. Currently used by JFR's OSThreadSampler
4414 //
4415 // The remaining code is greatly simplified from the more general suspension
4416 // code that used to be used.
4417 //
4418 // The protocol is quite simple:
4419 // - suspend:
4420 // - sends a signal to the target thread
4421 // - polls the suspend state of the osthread using a yield loop
4422 // - target thread signal handler (SR_handler) sets suspend state
4423 // and blocks in sigsuspend until continued
4424 // - resume:
4425 // - sets target osthread state to continue
4426 // - sends signal to end the sigsuspend loop in the SR_handler
4427 //
4428 // Note that the SR_lock plays no role in this suspend/resume protocol,
4429 // but is checked for NULL in SR_handler as a thread termination indicator.
4430 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4431 //
4432 // Note that resume_clear_context() and suspend_save_context() are needed
4433 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4434 // which in part is used by:
4435 // - Forte Analyzer: AsyncGetCallTrace()
4436 // - StackBanging: get_frame_at_stack_banging_point()
4437
4438 static void resume_clear_context(OSThread *osthread) {
4439 osthread->set_ucontext(NULL);
4440 osthread->set_siginfo(NULL);
4441 }
4442
4443 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4444 ucontext_t* context) {
4445 osthread->set_ucontext(context);
4446 osthread->set_siginfo(siginfo);
4447 }
4448
4449 // Handler function invoked when a thread's execution is suspended or
4450 // resumed. We have to be careful that only async-safe functions are
4451 // called here (Note: most pthread functions are not async safe and
4452 // should be avoided.)
4453 //
4454 // Note: sigwait() is a more natural fit than sigsuspend() from an
4455 // interface point of view, but sigwait() prevents the signal hander
4456 // from being run. libpthread would get very confused by not having
4457 // its signal handlers run and prevents sigwait()'s use with the
4458 // mutex granting granting signal.
4459 //
4460 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4461 //
4462 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4463 // Save and restore errno to avoid confusing native code with EINTR
4464 // after sigsuspend.
4465 int old_errno = errno;
4466
4467 Thread* thread = Thread::current_or_null_safe();
4468 assert(thread != NULL, "Missing current thread in SR_handler");
4469
4470 // On some systems we have seen signal delivery get "stuck" until the signal
4471 // mask is changed as part of thread termination. Check that the current thread
4472 // has not already terminated (via SR_lock()) - else the following assertion
4473 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4474 // destructor has completed.
4475
4476 if (thread->SR_lock() == NULL) {
4477 return;
4478 }
4479
4480 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4481
4482 OSThread* osthread = thread->osthread();
4483
4484 os::SuspendResume::State current = osthread->sr.state();
4485 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4486 suspend_save_context(osthread, siginfo, context);
4487
4488 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4489 os::SuspendResume::State state = osthread->sr.suspended();
4490 if (state == os::SuspendResume::SR_SUSPENDED) {
4491 sigset_t suspend_set; // signals for sigsuspend()
4492 sigemptyset(&suspend_set);
4493 // get current set of blocked signals and unblock resume signal
4494 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4495 sigdelset(&suspend_set, SR_signum);
4496
4497 sr_semaphore.signal();
4498 // wait here until we are resumed
4499 while (1) {
4500 sigsuspend(&suspend_set);
4501
4502 os::SuspendResume::State result = osthread->sr.running();
4503 if (result == os::SuspendResume::SR_RUNNING) {
4504 sr_semaphore.signal();
4505 break;
4506 }
4507 }
4508
4509 } else if (state == os::SuspendResume::SR_RUNNING) {
4510 // request was cancelled, continue
4511 } else {
4512 ShouldNotReachHere();
4513 }
4514
4515 resume_clear_context(osthread);
4516 } else if (current == os::SuspendResume::SR_RUNNING) {
4517 // request was cancelled, continue
4518 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4519 // ignore
4520 } else {
4521 // ignore
4522 }
4523
4524 errno = old_errno;
4525 }
4526
4527 static int SR_initialize() {
4528 struct sigaction act;
4529 char *s;
4530
4531 // Get signal number to use for suspend/resume
4532 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4533 int sig = ::strtol(s, 0, 10);
4534 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4535 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4536 SR_signum = sig;
4537 } else {
4538 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4539 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4540 }
4541 }
4542
4543 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4544 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4545
4546 sigemptyset(&SR_sigset);
4547 sigaddset(&SR_sigset, SR_signum);
4548
4549 // Set up signal handler for suspend/resume
4550 act.sa_flags = SA_RESTART|SA_SIGINFO;
4551 act.sa_handler = (void (*)(int)) SR_handler;
4552
4553 // SR_signum is blocked by default.
4554 // 4528190 - We also need to block pthread restart signal (32 on all
4555 // supported Linux platforms). Note that LinuxThreads need to block
4556 // this signal for all threads to work properly. So we don't have
4557 // to use hard-coded signal number when setting up the mask.
4558 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4559
4560 if (sigaction(SR_signum, &act, 0) == -1) {
4561 return -1;
4562 }
4563
4564 // Save signal flag
4565 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4566 return 0;
4567 }
4568
4569 static int sr_notify(OSThread* osthread) {
4570 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4571 assert_status(status == 0, status, "pthread_kill");
4572 return status;
4573 }
4574
4575 // "Randomly" selected value for how long we want to spin
4576 // before bailing out on suspending a thread, also how often
4577 // we send a signal to a thread we want to resume
4578 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4579 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4580
4581 // returns true on success and false on error - really an error is fatal
4582 // but this seems the normal response to library errors
4583 static bool do_suspend(OSThread* osthread) {
4584 assert(osthread->sr.is_running(), "thread should be running");
4585 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4586
4587 // mark as suspended and send signal
4588 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4589 // failed to switch, state wasn't running?
4590 ShouldNotReachHere();
4591 return false;
4592 }
4593
4594 if (sr_notify(osthread) != 0) {
4595 ShouldNotReachHere();
4596 }
4597
4598 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4599 while (true) {
4600 if (sr_semaphore.timedwait(2)) {
4601 break;
4602 } else {
4603 // timeout
4604 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4605 if (cancelled == os::SuspendResume::SR_RUNNING) {
4606 return false;
4607 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4608 // make sure that we consume the signal on the semaphore as well
4609 sr_semaphore.wait();
4610 break;
4611 } else {
4612 ShouldNotReachHere();
4613 return false;
4614 }
4615 }
4616 }
4617
4618 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4619 return true;
4620 }
4621
4622 static void do_resume(OSThread* osthread) {
4623 assert(osthread->sr.is_suspended(), "thread should be suspended");
4624 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4625
4626 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4627 // failed to switch to WAKEUP_REQUEST
4628 ShouldNotReachHere();
4629 return;
4630 }
4631
4632 while (true) {
4633 if (sr_notify(osthread) == 0) {
4634 if (sr_semaphore.timedwait(2)) {
4635 if (osthread->sr.is_running()) {
4636 return;
4637 }
4638 }
4639 } else {
4640 ShouldNotReachHere();
4641 }
4642 }
4643
4644 guarantee(osthread->sr.is_running(), "Must be running!");
4645 }
4646
4647 ///////////////////////////////////////////////////////////////////////////////////
4648 // signal handling (except suspend/resume)
4649
4650 // This routine may be used by user applications as a "hook" to catch signals.
4651 // The user-defined signal handler must pass unrecognized signals to this
4652 // routine, and if it returns true (non-zero), then the signal handler must
4653 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4654 // routine will never retun false (zero), but instead will execute a VM panic
4655 // routine kill the process.
4656 //
4657 // If this routine returns false, it is OK to call it again. This allows
4658 // the user-defined signal handler to perform checks either before or after
4659 // the VM performs its own checks. Naturally, the user code would be making
4660 // a serious error if it tried to handle an exception (such as a null check
4661 // or breakpoint) that the VM was generating for its own correct operation.
4662 //
4663 // This routine may recognize any of the following kinds of signals:
4664 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4665 // It should be consulted by handlers for any of those signals.
4666 //
4667 // The caller of this routine must pass in the three arguments supplied
4668 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4669 // field of the structure passed to sigaction(). This routine assumes that
4670 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4671 //
4672 // Note that the VM will print warnings if it detects conflicting signal
4673 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4674 //
4675 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4676 siginfo_t* siginfo,
4677 void* ucontext,
4678 int abort_if_unrecognized);
4679
4680 static void signalHandler(int sig, siginfo_t* info, void* uc) {
4681 assert(info != NULL && uc != NULL, "it must be old kernel");
4682 int orig_errno = errno; // Preserve errno value over signal handler.
4683 JVM_handle_linux_signal(sig, info, uc, true);
4684 errno = orig_errno;
4685 }
4686
4687
4688 // This boolean allows users to forward their own non-matching signals
4689 // to JVM_handle_linux_signal, harmlessly.
4690 bool os::Linux::signal_handlers_are_installed = false;
4691
4692 // For signal-chaining
4693 bool os::Linux::libjsig_is_loaded = false;
4694 typedef struct sigaction *(*get_signal_t)(int);
4695 get_signal_t os::Linux::get_signal_action = NULL;
4696
4697 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4698 struct sigaction *actp = NULL;
4699
4700 if (libjsig_is_loaded) {
4701 // Retrieve the old signal handler from libjsig
4702 actp = (*get_signal_action)(sig);
4703 }
4704 if (actp == NULL) {
4705 // Retrieve the preinstalled signal handler from jvm
4706 actp = os::Posix::get_preinstalled_handler(sig);
4707 }
4708
4709 return actp;
4710 }
4711
4712 static bool call_chained_handler(struct sigaction *actp, int sig,
4713 siginfo_t *siginfo, void *context) {
4714 // Call the old signal handler
4715 if (actp->sa_handler == SIG_DFL) {
4716 // It's more reasonable to let jvm treat it as an unexpected exception
4717 // instead of taking the default action.
4718 return false;
4719 } else if (actp->sa_handler != SIG_IGN) {
4720 if ((actp->sa_flags & SA_NODEFER) == 0) {
4721 // automaticlly block the signal
4722 sigaddset(&(actp->sa_mask), sig);
4723 }
4724
4725 sa_handler_t hand = NULL;
4726 sa_sigaction_t sa = NULL;
4727 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4728 // retrieve the chained handler
4729 if (siginfo_flag_set) {
4730 sa = actp->sa_sigaction;
4731 } else {
4732 hand = actp->sa_handler;
4733 }
4734
4735 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4736 actp->sa_handler = SIG_DFL;
4737 }
4738
4739 // try to honor the signal mask
4740 sigset_t oset;
4741 sigemptyset(&oset);
4742 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4743
4744 // call into the chained handler
4745 if (siginfo_flag_set) {
4746 (*sa)(sig, siginfo, context);
4747 } else {
4748 (*hand)(sig);
4749 }
4750
4751 // restore the signal mask
4752 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4753 }
4754 // Tell jvm's signal handler the signal is taken care of.
4755 return true;
4756 }
4757
4758 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4759 bool chained = false;
4760 // signal-chaining
4761 if (UseSignalChaining) {
4762 struct sigaction *actp = get_chained_signal_action(sig);
4763 if (actp != NULL) {
4764 chained = call_chained_handler(actp, sig, siginfo, context);
4765 }
4766 }
4767 return chained;
4768 }
4769
4770 // for diagnostic
4771 int sigflags[NSIG];
4772
4773 int os::Linux::get_our_sigflags(int sig) {
4774 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4775 return sigflags[sig];
4776 }
4777
4778 void os::Linux::set_our_sigflags(int sig, int flags) {
4779 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4780 if (sig > 0 && sig < NSIG) {
4781 sigflags[sig] = flags;
4782 }
4783 }
4784
4785 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4786 // Check for overwrite.
4787 struct sigaction oldAct;
4788 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4789
4790 void* oldhand = oldAct.sa_sigaction
4791 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4792 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4793 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4794 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4795 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4796 if (AllowUserSignalHandlers || !set_installed) {
4797 // Do not overwrite; user takes responsibility to forward to us.
4798 return;
4799 } else if (UseSignalChaining) {
4800 // save the old handler in jvm
4801 os::Posix::save_preinstalled_handler(sig, oldAct);
4802 // libjsig also interposes the sigaction() call below and saves the
4803 // old sigaction on it own.
4804 } else {
4805 fatal("Encountered unexpected pre-existing sigaction handler "
4806 "%#lx for signal %d.", (long)oldhand, sig);
4807 }
4808 }
4809
4810 struct sigaction sigAct;
4811 sigfillset(&(sigAct.sa_mask));
4812 sigAct.sa_handler = SIG_DFL;
4813 if (!set_installed) {
4814 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4815 } else {
4816 sigAct.sa_sigaction = signalHandler;
4817 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4818 }
4819 // Save flags, which are set by ours
4820 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4821 sigflags[sig] = sigAct.sa_flags;
4822
4823 int ret = sigaction(sig, &sigAct, &oldAct);
4824 assert(ret == 0, "check");
4825
4826 void* oldhand2 = oldAct.sa_sigaction
4827 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4828 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4829 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4830 }
4831
4832 // install signal handlers for signals that HotSpot needs to
4833 // handle in order to support Java-level exception handling.
4834
4835 void os::Linux::install_signal_handlers() {
4836 if (!signal_handlers_are_installed) {
4837 signal_handlers_are_installed = true;
4838
4839 // signal-chaining
4840 typedef void (*signal_setting_t)();
4841 signal_setting_t begin_signal_setting = NULL;
4842 signal_setting_t end_signal_setting = NULL;
4843 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4844 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4845 if (begin_signal_setting != NULL) {
4846 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4847 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4848 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4849 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4850 libjsig_is_loaded = true;
4851 assert(UseSignalChaining, "should enable signal-chaining");
4852 }
4853 if (libjsig_is_loaded) {
4854 // Tell libjsig jvm is setting signal handlers
4855 (*begin_signal_setting)();
4856 }
4857
4858 set_signal_handler(SIGSEGV, true);
4859 set_signal_handler(SIGPIPE, true);
4860 set_signal_handler(SIGBUS, true);
4861 set_signal_handler(SIGILL, true);
4862 set_signal_handler(SIGFPE, true);
4863 #if defined(PPC64)
4864 set_signal_handler(SIGTRAP, true);
4865 #endif
4866 set_signal_handler(SIGXFSZ, true);
4867
4868 if (libjsig_is_loaded) {
4869 // Tell libjsig jvm finishes setting signal handlers
4870 (*end_signal_setting)();
4871 }
4872
4873 // We don't activate signal checker if libjsig is in place, we trust ourselves
4874 // and if UserSignalHandler is installed all bets are off.
4875 // Log that signal checking is off only if -verbose:jni is specified.
4876 if (CheckJNICalls) {
4877 if (libjsig_is_loaded) {
4878 log_debug(jni, resolve)("Info: libjsig is activated, all active signal checking is disabled");
4879 check_signals = false;
4880 }
4881 if (AllowUserSignalHandlers) {
4882 log_debug(jni, resolve)("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4883 check_signals = false;
4884 }
4885 }
4886 }
4887 }
4888
4889 // This is the fastest way to get thread cpu time on Linux.
4890 // Returns cpu time (user+sys) for any thread, not only for current.
4891 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4892 // It might work on 2.6.10+ with a special kernel/glibc patch.
4893 // For reference, please, see IEEE Std 1003.1-2004:
4894 // http://www.unix.org/single_unix_specification
4895
4896 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4897 struct timespec tp;
4898 int rc = os::Posix::clock_gettime(clockid, &tp);
4899 assert(rc == 0, "clock_gettime is expected to return 0 code");
4900
4901 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4902 }
4903
4904 /////
4905 // glibc on Linux platform uses non-documented flag
4906 // to indicate, that some special sort of signal
4907 // trampoline is used.
4908 // We will never set this flag, and we should
4909 // ignore this flag in our diagnostic
4910 #ifdef SIGNIFICANT_SIGNAL_MASK
4911 #undef SIGNIFICANT_SIGNAL_MASK
4912 #endif
4913 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4914
4915 static const char* get_signal_handler_name(address handler,
4916 char* buf, int buflen) {
4917 int offset = 0;
4918 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4919 if (found) {
4920 // skip directory names
4921 const char *p1, *p2;
4922 p1 = buf;
4923 size_t len = strlen(os::file_separator());
4924 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4925 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4926 } else {
4927 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4928 }
4929 return buf;
4930 }
4931
4932 static void print_signal_handler(outputStream* st, int sig,
4933 char* buf, size_t buflen) {
4934 struct sigaction sa;
4935
4936 sigaction(sig, NULL, &sa);
4937
4938 // See comment for SIGNIFICANT_SIGNAL_MASK define
4939 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4940
4941 st->print("%s: ", os::exception_name(sig, buf, buflen));
4942
4943 address handler = (sa.sa_flags & SA_SIGINFO)
4944 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4945 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4946
4947 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4948 st->print("SIG_DFL");
4949 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4950 st->print("SIG_IGN");
4951 } else {
4952 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4953 }
4954
4955 st->print(", sa_mask[0]=");
4956 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4957
4958 address rh = VMError::get_resetted_sighandler(sig);
4959 // May be, handler was resetted by VMError?
4960 if (rh != NULL) {
4961 handler = rh;
4962 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4963 }
4964
4965 st->print(", sa_flags=");
4966 os::Posix::print_sa_flags(st, sa.sa_flags);
4967
4968 // Check: is it our handler?
4969 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4970 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4971 // It is our signal handler
4972 // check for flags, reset system-used one!
4973 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4974 st->print(
4975 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4976 os::Linux::get_our_sigflags(sig));
4977 }
4978 }
4979 st->cr();
4980 }
4981
4982
4983 #define DO_SIGNAL_CHECK(sig) \
4984 do { \
4985 if (!sigismember(&check_signal_done, sig)) { \
4986 os::Linux::check_signal_handler(sig); \
4987 } \
4988 } while (0)
4989
4990 // This method is a periodic task to check for misbehaving JNI applications
4991 // under CheckJNI, we can add any periodic checks here
4992
4993 void os::run_periodic_checks() {
4994 if (check_signals == false) return;
4995
4996 // SEGV and BUS if overridden could potentially prevent
4997 // generation of hs*.log in the event of a crash, debugging
4998 // such a case can be very challenging, so we absolutely
4999 // check the following for a good measure:
5000 DO_SIGNAL_CHECK(SIGSEGV);
5001 DO_SIGNAL_CHECK(SIGILL);
5002 DO_SIGNAL_CHECK(SIGFPE);
5003 DO_SIGNAL_CHECK(SIGBUS);
5004 DO_SIGNAL_CHECK(SIGPIPE);
5005 DO_SIGNAL_CHECK(SIGXFSZ);
5006 #if defined(PPC64)
5007 DO_SIGNAL_CHECK(SIGTRAP);
5008 #endif
5009
5010 // ReduceSignalUsage allows the user to override these handlers
5011 // see comments at the very top and jvm_md.h
5012 if (!ReduceSignalUsage) {
5013 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
5014 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
5015 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
5016 DO_SIGNAL_CHECK(BREAK_SIGNAL);
5017 }
5018
5019 DO_SIGNAL_CHECK(SR_signum);
5020 }
5021
5022 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
5023
5024 static os_sigaction_t os_sigaction = NULL;
5025
5026 void os::Linux::check_signal_handler(int sig) {
5027 char buf[O_BUFLEN];
5028 address jvmHandler = NULL;
5029
5030
5031 struct sigaction act;
5032 if (os_sigaction == NULL) {
5033 // only trust the default sigaction, in case it has been interposed
5034 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
5035 if (os_sigaction == NULL) return;
5036 }
5037
5038 os_sigaction(sig, (struct sigaction*)NULL, &act);
5039
5040
5041 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
5042
5043 address thisHandler = (act.sa_flags & SA_SIGINFO)
5044 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
5045 : CAST_FROM_FN_PTR(address, act.sa_handler);
5046
5047
5048 switch (sig) {
5049 case SIGSEGV:
5050 case SIGBUS:
5051 case SIGFPE:
5052 case SIGPIPE:
5053 case SIGILL:
5054 case SIGXFSZ:
5055 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
5056 break;
5057
5058 case SHUTDOWN1_SIGNAL:
5059 case SHUTDOWN2_SIGNAL:
5060 case SHUTDOWN3_SIGNAL:
5061 case BREAK_SIGNAL:
5062 jvmHandler = (address)user_handler();
5063 break;
5064
5065 default:
5066 if (sig == SR_signum) {
5067 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
5068 } else {
5069 return;
5070 }
5071 break;
5072 }
5073
5074 if (thisHandler != jvmHandler) {
5075 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
5076 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
5077 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
5078 // No need to check this sig any longer
5079 sigaddset(&check_signal_done, sig);
5080 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
5081 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
5082 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
5083 exception_name(sig, buf, O_BUFLEN));
5084 }
5085 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
5086 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
5087 tty->print("expected:");
5088 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
5089 tty->cr();
5090 tty->print(" found:");
5091 os::Posix::print_sa_flags(tty, act.sa_flags);
5092 tty->cr();
5093 // No need to check this sig any longer
5094 sigaddset(&check_signal_done, sig);
5095 }
5096
5097 // Dump all the signal
5098 if (sigismember(&check_signal_done, sig)) {
5099 print_signal_handlers(tty, buf, O_BUFLEN);
5100 }
5101 }
5102
5103 extern void report_error(char* file_name, int line_no, char* title,
5104 char* format, ...);
5105
5106 // this is called _before_ most of the global arguments have been parsed
5107 void os::init(void) {
5108 char dummy; // used to get a guess on initial stack address
5109
5110 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
5111
5112 init_random(1234567);
5113
5114 Linux::set_page_size(sysconf(_SC_PAGESIZE));
5115 if (Linux::page_size() == -1) {
5116 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
5117 os::strerror(errno));
5118 }
5119 init_page_sizes((size_t) Linux::page_size());
5120
5121 Linux::initialize_system_info();
5122
5123 os::Linux::CPUPerfTicks pticks;
5124 bool res = os::Linux::get_tick_information(&pticks, -1);
5125
5126 if (res && pticks.has_steal_ticks) {
5127 has_initial_tick_info = true;
5128 initial_total_ticks = pticks.total;
5129 initial_steal_ticks = pticks.steal;
5130 }
5131
5132 // _main_thread points to the thread that created/loaded the JVM.
5133 Linux::_main_thread = pthread_self();
5134
5135 // retrieve entry point for pthread_setname_np
5136 Linux::_pthread_setname_np =
5137 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
5138
5139 os::Posix::init();
5140
5141 initial_time_count = javaTimeNanos();
5142
5143 // Always warn if no monotonic clock available
5144 if (!os::Posix::supports_monotonic_clock()) {
5145 warning("No monotonic clock was available - timed services may " \
5146 "be adversely affected if the time-of-day clock changes");
5147 }
5148 }
5149
5150 // To install functions for atexit system call
5151 extern "C" {
5152 static void perfMemory_exit_helper() {
5153 perfMemory_exit();
5154 }
5155 }
5156
5157 void os::pd_init_container_support() {
5158 OSContainer::init();
5159 }
5160
5161 void os::Linux::numa_init() {
5162
5163 // Java can be invoked as
5164 // 1. Without numactl and heap will be allocated/configured on all nodes as
5165 // per the system policy.
5166 // 2. With numactl --interleave:
5167 // Use numa_get_interleave_mask(v2) API to get nodes bitmask. The same
5168 // API for membind case bitmask is reset.
5169 // Interleave is only hint and Kernel can fallback to other nodes if
5170 // no memory is available on the target nodes.
5171 // 3. With numactl --membind:
5172 // Use numa_get_membind(v2) API to get nodes bitmask. The same API for
5173 // interleave case returns bitmask of all nodes.
5174 // numa_all_nodes_ptr holds bitmask of all nodes.
5175 // numa_get_interleave_mask(v2) and numa_get_membind(v2) APIs returns correct
5176 // bitmask when externally configured to run on all or fewer nodes.
5177
5178 if (!Linux::libnuma_init()) {
5179 UseNUMA = false;
5180 } else {
5181 if ((Linux::numa_max_node() < 1) || Linux::is_bound_to_single_node()) {
5182 // If there's only one node (they start from 0) or if the process
5183 // is bound explicitly to a single node using membind, disable NUMA unless
5184 // user explicilty forces NUMA optimizations on single-node/UMA systems
5185 UseNUMA = ForceNUMA;
5186 } else {
5187
5188 LogTarget(Info,os) log;
5189 LogStream ls(log);
5190
5191 Linux::set_configured_numa_policy(Linux::identify_numa_policy());
5192
5193 struct bitmask* bmp = Linux::_numa_membind_bitmask;
5194 const char* numa_mode = "membind";
5195
5196 if (Linux::is_running_in_interleave_mode()) {
5197 bmp = Linux::_numa_interleave_bitmask;
5198 numa_mode = "interleave";
5199 }
5200
5201 ls.print("UseNUMA is enabled and invoked in '%s' mode."
5202 " Heap will be configured using NUMA memory nodes:", numa_mode);
5203
5204 for (int node = 0; node <= Linux::numa_max_node(); node++) {
5205 if (Linux::_numa_bitmask_isbitset(bmp, node)) {
5206 ls.print(" %d", node);
5207 }
5208 }
5209 }
5210 }
5211
5212 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5213 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5214 // we can make the adaptive lgrp chunk resizing work. If the user specified both
5215 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5216 // and disable adaptive resizing.
5217 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5218 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5219 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5220 UseAdaptiveSizePolicy = false;
5221 UseAdaptiveNUMAChunkSizing = false;
5222 }
5223 }
5224 }
5225
5226 // this is called _after_ the global arguments have been parsed
5227 jint os::init_2(void) {
5228
5229 // This could be set after os::Posix::init() but all platforms
5230 // have to set it the same so we have to mirror Solaris.
5231 DEBUG_ONLY(os::set_mutex_init_done();)
5232
5233 os::Posix::init_2();
5234
5235 Linux::fast_thread_clock_init();
5236
5237 // initialize suspend/resume support - must do this before signal_sets_init()
5238 if (SR_initialize() != 0) {
5239 perror("SR_initialize failed");
5240 return JNI_ERR;
5241 }
5242
5243 Linux::signal_sets_init();
5244 Linux::install_signal_handlers();
5245 // Initialize data for jdk.internal.misc.Signal
5246 if (!ReduceSignalUsage) {
5247 jdk_misc_signal_init();
5248 }
5249
5250 if (AdjustStackSizeForTLS) {
5251 get_minstack_init();
5252 }
5253
5254 // Check and sets minimum stack sizes against command line options
5255 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
5256 return JNI_ERR;
5257 }
5258
5259 #if defined(IA32)
5260 // Need to ensure we've determined the process's initial stack to
5261 // perform the workaround
5262 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5263 workaround_expand_exec_shield_cs_limit();
5264 #else
5265 suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
5266 if (!suppress_primordial_thread_resolution) {
5267 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5268 }
5269 #endif
5270
5271 Linux::libpthread_init();
5272 Linux::sched_getcpu_init();
5273 log_info(os)("HotSpot is running with %s, %s",
5274 Linux::glibc_version(), Linux::libpthread_version());
5275
5276 if (UseNUMA) {
5277 Linux::numa_init();
5278 }
5279
5280 if (MaxFDLimit) {
5281 // set the number of file descriptors to max. print out error
5282 // if getrlimit/setrlimit fails but continue regardless.
5283 struct rlimit nbr_files;
5284 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5285 if (status != 0) {
5286 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5287 } else {
5288 nbr_files.rlim_cur = nbr_files.rlim_max;
5289 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5290 if (status != 0) {
5291 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5292 }
5293 }
5294 }
5295
5296 // at-exit methods are called in the reverse order of their registration.
5297 // atexit functions are called on return from main or as a result of a
5298 // call to exit(3C). There can be only 32 of these functions registered
5299 // and atexit() does not set errno.
5300
5301 if (PerfAllowAtExitRegistration) {
5302 // only register atexit functions if PerfAllowAtExitRegistration is set.
5303 // atexit functions can be delayed until process exit time, which
5304 // can be problematic for embedded VM situations. Embedded VMs should
5305 // call DestroyJavaVM() to assure that VM resources are released.
5306
5307 // note: perfMemory_exit_helper atexit function may be removed in
5308 // the future if the appropriate cleanup code can be added to the
5309 // VM_Exit VMOperation's doit method.
5310 if (atexit(perfMemory_exit_helper) != 0) {
5311 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5312 }
5313 }
5314
5315 // initialize thread priority policy
5316 prio_init();
5317
5318 if (!FLAG_IS_DEFAULT(AllocateHeapAt) || !FLAG_IS_DEFAULT(AllocateOldGenAt)) {
5319 set_coredump_filter(DAX_SHARED_BIT);
5320 }
5321
5322 if (DumpPrivateMappingsInCore) {
5323 set_coredump_filter(FILE_BACKED_PVT_BIT);
5324 }
5325
5326 if (DumpSharedMappingsInCore) {
5327 set_coredump_filter(FILE_BACKED_SHARED_BIT);
5328 }
5329
5330 return JNI_OK;
5331 }
5332
5333 // older glibc versions don't have this macro (which expands to
5334 // an optimized bit-counting function) so we have to roll our own
5335 #ifndef CPU_COUNT
5336
5337 static int _cpu_count(const cpu_set_t* cpus) {
5338 int count = 0;
5339 // only look up to the number of configured processors
5340 for (int i = 0; i < os::processor_count(); i++) {
5341 if (CPU_ISSET(i, cpus)) {
5342 count++;
5343 }
5344 }
5345 return count;
5346 }
5347
5348 #define CPU_COUNT(cpus) _cpu_count(cpus)
5349
5350 #endif // CPU_COUNT
5351
5352 // Get the current number of available processors for this process.
5353 // This value can change at any time during a process's lifetime.
5354 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5355 // If it appears there may be more than 1024 processors then we do a
5356 // dynamic check - see 6515172 for details.
5357 // If anything goes wrong we fallback to returning the number of online
5358 // processors - which can be greater than the number available to the process.
5359 int os::Linux::active_processor_count() {
5360 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5361 cpu_set_t* cpus_p = &cpus;
5362 int cpus_size = sizeof(cpu_set_t);
5363
5364 int configured_cpus = os::processor_count(); // upper bound on available cpus
5365 int cpu_count = 0;
5366
5367 // old build platforms may not support dynamic cpu sets
5368 #ifdef CPU_ALLOC
5369
5370 // To enable easy testing of the dynamic path on different platforms we
5371 // introduce a diagnostic flag: UseCpuAllocPath
5372 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5373 // kernel may use a mask bigger than cpu_set_t
5374 log_trace(os)("active_processor_count: using dynamic path %s"
5375 "- configured processors: %d",
5376 UseCpuAllocPath ? "(forced) " : "",
5377 configured_cpus);
5378 cpus_p = CPU_ALLOC(configured_cpus);
5379 if (cpus_p != NULL) {
5380 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5381 // zero it just to be safe
5382 CPU_ZERO_S(cpus_size, cpus_p);
5383 }
5384 else {
5385 // failed to allocate so fallback to online cpus
5386 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5387 log_trace(os)("active_processor_count: "
5388 "CPU_ALLOC failed (%s) - using "
5389 "online processor count: %d",
5390 os::strerror(errno), online_cpus);
5391 return online_cpus;
5392 }
5393 }
5394 else {
5395 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5396 configured_cpus);
5397 }
5398 #else // CPU_ALLOC
5399 // these stubs won't be executed
5400 #define CPU_COUNT_S(size, cpus) -1
5401 #define CPU_FREE(cpus)
5402
5403 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5404 configured_cpus);
5405 #endif // CPU_ALLOC
5406
5407 // pid 0 means the current thread - which we have to assume represents the process
5408 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5409 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5410 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5411 }
5412 else {
5413 cpu_count = CPU_COUNT(cpus_p);
5414 }
5415 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5416 }
5417 else {
5418 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5419 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5420 "which may exceed available processors", os::strerror(errno), cpu_count);
5421 }
5422
5423 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5424 CPU_FREE(cpus_p);
5425 }
5426
5427 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5428 return cpu_count;
5429 }
5430
5431 // Determine the active processor count from one of
5432 // three different sources:
5433 //
5434 // 1. User option -XX:ActiveProcessorCount
5435 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5436 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5437 //
5438 // Option 1, if specified, will always override.
5439 // If the cgroup subsystem is active and configured, we
5440 // will return the min of the cgroup and option 2 results.
5441 // This is required since tools, such as numactl, that
5442 // alter cpu affinity do not update cgroup subsystem
5443 // cpuset configuration files.
5444 int os::active_processor_count() {
5445 // User has overridden the number of active processors
5446 if (ActiveProcessorCount > 0) {
5447 log_trace(os)("active_processor_count: "
5448 "active processor count set by user : %d",
5449 ActiveProcessorCount);
5450 return ActiveProcessorCount;
5451 }
5452
5453 int active_cpus;
5454 if (OSContainer::is_containerized()) {
5455 active_cpus = OSContainer::active_processor_count();
5456 log_trace(os)("active_processor_count: determined by OSContainer: %d",
5457 active_cpus);
5458 } else {
5459 active_cpus = os::Linux::active_processor_count();
5460 }
5461
5462 return active_cpus;
5463 }
5464
5465 uint os::processor_id() {
5466 const int id = Linux::sched_getcpu();
5467 assert(id >= 0 && id < _processor_count, "Invalid processor id");
5468 return (uint)id;
5469 }
5470
5471 void os::set_native_thread_name(const char *name) {
5472 if (Linux::_pthread_setname_np) {
5473 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5474 snprintf(buf, sizeof(buf), "%s", name);
5475 buf[sizeof(buf) - 1] = '\0';
5476 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5477 // ERANGE should not happen; all other errors should just be ignored.
5478 assert(rc != ERANGE, "pthread_setname_np failed");
5479 }
5480 }
5481
5482 bool os::bind_to_processor(uint processor_id) {
5483 // Not yet implemented.
5484 return false;
5485 }
5486
5487 ///
5488
5489 void os::SuspendedThreadTask::internal_do_task() {
5490 if (do_suspend(_thread->osthread())) {
5491 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5492 do_task(context);
5493 do_resume(_thread->osthread());
5494 }
5495 }
5496
5497 ////////////////////////////////////////////////////////////////////////////////
5498 // debug support
5499
5500 bool os::find(address addr, outputStream* st) {
5501 Dl_info dlinfo;
5502 memset(&dlinfo, 0, sizeof(dlinfo));
5503 if (dladdr(addr, &dlinfo) != 0) {
5504 st->print(PTR_FORMAT ": ", p2i(addr));
5505 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5506 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5507 p2i(addr) - p2i(dlinfo.dli_saddr));
5508 } else if (dlinfo.dli_fbase != NULL) {
5509 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5510 } else {
5511 st->print("<absolute address>");
5512 }
5513 if (dlinfo.dli_fname != NULL) {
5514 st->print(" in %s", dlinfo.dli_fname);
5515 }
5516 if (dlinfo.dli_fbase != NULL) {
5517 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5518 }
5519 st->cr();
5520
5521 if (Verbose) {
5522 // decode some bytes around the PC
5523 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5524 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5525 address lowest = (address) dlinfo.dli_sname;
5526 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5527 if (begin < lowest) begin = lowest;
5528 Dl_info dlinfo2;
5529 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5530 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5531 end = (address) dlinfo2.dli_saddr;
5532 }
5533 Disassembler::decode(begin, end, st);
5534 }
5535 return true;
5536 }
5537 return false;
5538 }
5539
5540 ////////////////////////////////////////////////////////////////////////////////
5541 // misc
5542
5543 // This does not do anything on Linux. This is basically a hook for being
5544 // able to use structured exception handling (thread-local exception filters)
5545 // on, e.g., Win32.
5546 void
5547 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5548 JavaCallArguments* args, Thread* thread) {
5549 f(value, method, args, thread);
5550 }
5551
5552 void os::print_statistics() {
5553 }
5554
5555 bool os::message_box(const char* title, const char* message) {
5556 int i;
5557 fdStream err(defaultStream::error_fd());
5558 for (i = 0; i < 78; i++) err.print_raw("=");
5559 err.cr();
5560 err.print_raw_cr(title);
5561 for (i = 0; i < 78; i++) err.print_raw("-");
5562 err.cr();
5563 err.print_raw_cr(message);
5564 for (i = 0; i < 78; i++) err.print_raw("=");
5565 err.cr();
5566
5567 char buf[16];
5568 // Prevent process from exiting upon "read error" without consuming all CPU
5569 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5570
5571 return buf[0] == 'y' || buf[0] == 'Y';
5572 }
5573
5574 // Is a (classpath) directory empty?
5575 bool os::dir_is_empty(const char* path) {
5576 DIR *dir = NULL;
5577 struct dirent *ptr;
5578
5579 dir = opendir(path);
5580 if (dir == NULL) return true;
5581
5582 // Scan the directory
5583 bool result = true;
5584 while (result && (ptr = readdir(dir)) != NULL) {
5585 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5586 result = false;
5587 }
5588 }
5589 closedir(dir);
5590 return result;
5591 }
5592
5593 // This code originates from JDK's sysOpen and open64_w
5594 // from src/solaris/hpi/src/system_md.c
5595
5596 int os::open(const char *path, int oflag, int mode) {
5597 if (strlen(path) > MAX_PATH - 1) {
5598 errno = ENAMETOOLONG;
5599 return -1;
5600 }
5601
5602 // All file descriptors that are opened in the Java process and not
5603 // specifically destined for a subprocess should have the close-on-exec
5604 // flag set. If we don't set it, then careless 3rd party native code
5605 // might fork and exec without closing all appropriate file descriptors
5606 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5607 // turn might:
5608 //
5609 // - cause end-of-file to fail to be detected on some file
5610 // descriptors, resulting in mysterious hangs, or
5611 //
5612 // - might cause an fopen in the subprocess to fail on a system
5613 // suffering from bug 1085341.
5614 //
5615 // (Yes, the default setting of the close-on-exec flag is a Unix
5616 // design flaw)
5617 //
5618 // See:
5619 // 1085341: 32-bit stdio routines should support file descriptors >255
5620 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5621 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5622 //
5623 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5624 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5625 // because it saves a system call and removes a small window where the flag
5626 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5627 // and we fall back to using FD_CLOEXEC (see below).
5628 #ifdef O_CLOEXEC
5629 oflag |= O_CLOEXEC;
5630 #endif
5631
5632 int fd = ::open64(path, oflag, mode);
5633 if (fd == -1) return -1;
5634
5635 //If the open succeeded, the file might still be a directory
5636 {
5637 struct stat64 buf64;
5638 int ret = ::fstat64(fd, &buf64);
5639 int st_mode = buf64.st_mode;
5640
5641 if (ret != -1) {
5642 if ((st_mode & S_IFMT) == S_IFDIR) {
5643 errno = EISDIR;
5644 ::close(fd);
5645 return -1;
5646 }
5647 } else {
5648 ::close(fd);
5649 return -1;
5650 }
5651 }
5652
5653 #ifdef FD_CLOEXEC
5654 // Validate that the use of the O_CLOEXEC flag on open above worked.
5655 // With recent kernels, we will perform this check exactly once.
5656 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5657 if (!O_CLOEXEC_is_known_to_work) {
5658 int flags = ::fcntl(fd, F_GETFD);
5659 if (flags != -1) {
5660 if ((flags & FD_CLOEXEC) != 0)
5661 O_CLOEXEC_is_known_to_work = 1;
5662 else
5663 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5664 }
5665 }
5666 #endif
5667
5668 return fd;
5669 }
5670
5671
5672 // create binary file, rewriting existing file if required
5673 int os::create_binary_file(const char* path, bool rewrite_existing) {
5674 int oflags = O_WRONLY | O_CREAT;
5675 if (!rewrite_existing) {
5676 oflags |= O_EXCL;
5677 }
5678 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5679 }
5680
5681 // return current position of file pointer
5682 jlong os::current_file_offset(int fd) {
5683 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5684 }
5685
5686 // move file pointer to the specified offset
5687 jlong os::seek_to_file_offset(int fd, jlong offset) {
5688 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5689 }
5690
5691 // This code originates from JDK's sysAvailable
5692 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5693
5694 int os::available(int fd, jlong *bytes) {
5695 jlong cur, end;
5696 int mode;
5697 struct stat64 buf64;
5698
5699 if (::fstat64(fd, &buf64) >= 0) {
5700 mode = buf64.st_mode;
5701 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5702 int n;
5703 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5704 *bytes = n;
5705 return 1;
5706 }
5707 }
5708 }
5709 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5710 return 0;
5711 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5712 return 0;
5713 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5714 return 0;
5715 }
5716 *bytes = end - cur;
5717 return 1;
5718 }
5719
5720 // Map a block of memory.
5721 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5722 char *addr, size_t bytes, bool read_only,
5723 bool allow_exec) {
5724 int prot;
5725 int flags = MAP_PRIVATE;
5726
5727 if (read_only) {
5728 prot = PROT_READ;
5729 } else {
5730 prot = PROT_READ | PROT_WRITE;
5731 }
5732
5733 if (allow_exec) {
5734 prot |= PROT_EXEC;
5735 }
5736
5737 if (addr != NULL) {
5738 flags |= MAP_FIXED;
5739 }
5740
5741 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5742 fd, file_offset);
5743 if (mapped_address == MAP_FAILED) {
5744 return NULL;
5745 }
5746 return mapped_address;
5747 }
5748
5749
5750 // Remap a block of memory.
5751 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5752 char *addr, size_t bytes, bool read_only,
5753 bool allow_exec) {
5754 // same as map_memory() on this OS
5755 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5756 allow_exec);
5757 }
5758
5759
5760 // Unmap a block of memory.
5761 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5762 return munmap(addr, bytes) == 0;
5763 }
5764
5765 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5766
5767 static jlong fast_cpu_time(Thread *thread) {
5768 clockid_t clockid;
5769 int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(),
5770 &clockid);
5771 if (rc == 0) {
5772 return os::Linux::fast_thread_cpu_time(clockid);
5773 } else {
5774 // It's possible to encounter a terminated native thread that failed
5775 // to detach itself from the VM - which should result in ESRCH.
5776 assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed");
5777 return -1;
5778 }
5779 }
5780
5781 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5782 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5783 // of a thread.
5784 //
5785 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5786 // the fast estimate available on the platform.
5787
5788 jlong os::current_thread_cpu_time() {
5789 if (os::Linux::supports_fast_thread_cpu_time()) {
5790 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5791 } else {
5792 // return user + sys since the cost is the same
5793 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5794 }
5795 }
5796
5797 jlong os::thread_cpu_time(Thread* thread) {
5798 // consistent with what current_thread_cpu_time() returns
5799 if (os::Linux::supports_fast_thread_cpu_time()) {
5800 return fast_cpu_time(thread);
5801 } else {
5802 return slow_thread_cpu_time(thread, true /* user + sys */);
5803 }
5804 }
5805
5806 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5807 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5808 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5809 } else {
5810 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5811 }
5812 }
5813
5814 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5815 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5816 return fast_cpu_time(thread);
5817 } else {
5818 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5819 }
5820 }
5821
5822 // -1 on error.
5823 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5824 pid_t tid = thread->osthread()->thread_id();
5825 char *s;
5826 char stat[2048];
5827 int statlen;
5828 char proc_name[64];
5829 int count;
5830 long sys_time, user_time;
5831 char cdummy;
5832 int idummy;
5833 long ldummy;
5834 FILE *fp;
5835
5836 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5837 fp = fopen(proc_name, "r");
5838 if (fp == NULL) return -1;
5839 statlen = fread(stat, 1, 2047, fp);
5840 stat[statlen] = '\0';
5841 fclose(fp);
5842
5843 // Skip pid and the command string. Note that we could be dealing with
5844 // weird command names, e.g. user could decide to rename java launcher
5845 // to "java 1.4.2 :)", then the stat file would look like
5846 // 1234 (java 1.4.2 :)) R ... ...
5847 // We don't really need to know the command string, just find the last
5848 // occurrence of ")" and then start parsing from there. See bug 4726580.
5849 s = strrchr(stat, ')');
5850 if (s == NULL) return -1;
5851
5852 // Skip blank chars
5853 do { s++; } while (s && isspace(*s));
5854
5855 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5856 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5857 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5858 &user_time, &sys_time);
5859 if (count != 13) return -1;
5860 if (user_sys_cpu_time) {
5861 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5862 } else {
5863 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5864 }
5865 }
5866
5867 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5868 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5869 info_ptr->may_skip_backward = false; // elapsed time not wall time
5870 info_ptr->may_skip_forward = false; // elapsed time not wall time
5871 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5872 }
5873
5874 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5875 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5876 info_ptr->may_skip_backward = false; // elapsed time not wall time
5877 info_ptr->may_skip_forward = false; // elapsed time not wall time
5878 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5879 }
5880
5881 bool os::is_thread_cpu_time_supported() {
5882 return true;
5883 }
5884
5885 // System loadavg support. Returns -1 if load average cannot be obtained.
5886 // Linux doesn't yet have a (official) notion of processor sets,
5887 // so just return the system wide load average.
5888 int os::loadavg(double loadavg[], int nelem) {
5889 return ::getloadavg(loadavg, nelem);
5890 }
5891
5892 void os::pause() {
5893 char filename[MAX_PATH];
5894 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5895 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5896 } else {
5897 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5898 }
5899
5900 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5901 if (fd != -1) {
5902 struct stat buf;
5903 ::close(fd);
5904 while (::stat(filename, &buf) == 0) {
5905 (void)::poll(NULL, 0, 100);
5906 }
5907 } else {
5908 jio_fprintf(stderr,
5909 "Could not open pause file '%s', continuing immediately.\n", filename);
5910 }
5911 }
5912
5913 extern char** environ;
5914
5915 // Run the specified command in a separate process. Return its exit value,
5916 // or -1 on failure (e.g. can't fork a new process).
5917 // Unlike system(), this function can be called from signal handler. It
5918 // doesn't block SIGINT et al.
5919 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) {
5920 const char * argv[4] = {"sh", "-c", cmd, NULL};
5921
5922 pid_t pid ;
5923
5924 if (use_vfork_if_available) {
5925 pid = vfork();
5926 } else {
5927 pid = fork();
5928 }
5929
5930 if (pid < 0) {
5931 // fork failed
5932 return -1;
5933
5934 } else if (pid == 0) {
5935 // child process
5936
5937 execve("/bin/sh", (char* const*)argv, environ);
5938
5939 // execve failed
5940 _exit(-1);
5941
5942 } else {
5943 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5944 // care about the actual exit code, for now.
5945
5946 int status;
5947
5948 // Wait for the child process to exit. This returns immediately if
5949 // the child has already exited. */
5950 while (waitpid(pid, &status, 0) < 0) {
5951 switch (errno) {
5952 case ECHILD: return 0;
5953 case EINTR: break;
5954 default: return -1;
5955 }
5956 }
5957
5958 if (WIFEXITED(status)) {
5959 // The child exited normally; get its exit code.
5960 return WEXITSTATUS(status);
5961 } else if (WIFSIGNALED(status)) {
5962 // The child exited because of a signal
5963 // The best value to return is 0x80 + signal number,
5964 // because that is what all Unix shells do, and because
5965 // it allows callers to distinguish between process exit and
5966 // process death by signal.
5967 return 0x80 + WTERMSIG(status);
5968 } else {
5969 // Unknown exit code; pass it through
5970 return status;
5971 }
5972 }
5973 }
5974
5975 // Get the default path to the core file
5976 // Returns the length of the string
5977 int os::get_core_path(char* buffer, size_t bufferSize) {
5978 /*
5979 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5980 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5981 */
5982 const int core_pattern_len = 129;
5983 char core_pattern[core_pattern_len] = {0};
5984
5985 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5986 if (core_pattern_file == -1) {
5987 return -1;
5988 }
5989
5990 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5991 ::close(core_pattern_file);
5992 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5993 return -1;
5994 }
5995 if (core_pattern[ret-1] == '\n') {
5996 core_pattern[ret-1] = '\0';
5997 } else {
5998 core_pattern[ret] = '\0';
5999 }
6000
6001 // Replace the %p in the core pattern with the process id. NOTE: we do this
6002 // only if the pattern doesn't start with "|", and we support only one %p in
6003 // the pattern.
6004 char *pid_pos = strstr(core_pattern, "%p");
6005 const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : ""; // skip over the "%p"
6006 int written;
6007
6008 if (core_pattern[0] == '/') {
6009 if (pid_pos != NULL) {
6010 *pid_pos = '\0';
6011 written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern,
6012 current_process_id(), tail);
6013 } else {
6014 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
6015 }
6016 } else {
6017 char cwd[PATH_MAX];
6018
6019 const char* p = get_current_directory(cwd, PATH_MAX);
6020 if (p == NULL) {
6021 return -1;
6022 }
6023
6024 if (core_pattern[0] == '|') {
6025 written = jio_snprintf(buffer, bufferSize,
6026 "\"%s\" (or dumping to %s/core.%d)",
6027 &core_pattern[1], p, current_process_id());
6028 } else if (pid_pos != NULL) {
6029 *pid_pos = '\0';
6030 written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern,
6031 current_process_id(), tail);
6032 } else {
6033 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6034 }
6035 }
6036
6037 if (written < 0) {
6038 return -1;
6039 }
6040
6041 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6042 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6043
6044 if (core_uses_pid_file != -1) {
6045 char core_uses_pid = 0;
6046 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6047 ::close(core_uses_pid_file);
6048
6049 if (core_uses_pid == '1') {
6050 jio_snprintf(buffer + written, bufferSize - written,
6051 ".%d", current_process_id());
6052 }
6053 }
6054 }
6055
6056 return strlen(buffer);
6057 }
6058
6059 bool os::start_debugging(char *buf, int buflen) {
6060 int len = (int)strlen(buf);
6061 char *p = &buf[len];
6062
6063 jio_snprintf(p, buflen-len,
6064 "\n\n"
6065 "Do you want to debug the problem?\n\n"
6066 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
6067 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
6068 "Otherwise, press RETURN to abort...",
6069 os::current_process_id(), os::current_process_id(),
6070 os::current_thread_id(), os::current_thread_id());
6071
6072 bool yes = os::message_box("Unexpected Error", buf);
6073
6074 if (yes) {
6075 // yes, user asked VM to launch debugger
6076 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
6077 os::current_process_id(), os::current_process_id());
6078
6079 os::fork_and_exec(buf);
6080 yes = false;
6081 }
6082 return yes;
6083 }
6084
6085
6086 // Java/Compiler thread:
6087 //
6088 // Low memory addresses
6089 // P0 +------------------------+
6090 // | |\ Java thread created by VM does not have glibc
6091 // | glibc guard page | - guard page, attached Java thread usually has
6092 // | |/ 1 glibc guard page.
6093 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6094 // | |\
6095 // | HotSpot Guard Pages | - red, yellow and reserved pages
6096 // | |/
6097 // +------------------------+ JavaThread::stack_reserved_zone_base()
6098 // | |\
6099 // | Normal Stack | -
6100 // | |/
6101 // P2 +------------------------+ Thread::stack_base()
6102 //
6103 // Non-Java thread:
6104 //
6105 // Low memory addresses
6106 // P0 +------------------------+
6107 // | |\
6108 // | glibc guard page | - usually 1 page
6109 // | |/
6110 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6111 // | |\
6112 // | Normal Stack | -
6113 // | |/
6114 // P2 +------------------------+ Thread::stack_base()
6115 //
6116 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
6117 // returned from pthread_attr_getstack().
6118 // ** Due to NPTL implementation error, linux takes the glibc guard page out
6119 // of the stack size given in pthread_attr. We work around this for
6120 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
6121 //
6122 #ifndef ZERO
6123 static void current_stack_region(address * bottom, size_t * size) {
6124 if (os::is_primordial_thread()) {
6125 // primordial thread needs special handling because pthread_getattr_np()
6126 // may return bogus value.
6127 *bottom = os::Linux::initial_thread_stack_bottom();
6128 *size = os::Linux::initial_thread_stack_size();
6129 } else {
6130 pthread_attr_t attr;
6131
6132 int rslt = pthread_getattr_np(pthread_self(), &attr);
6133
6134 // JVM needs to know exact stack location, abort if it fails
6135 if (rslt != 0) {
6136 if (rslt == ENOMEM) {
6137 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
6138 } else {
6139 fatal("pthread_getattr_np failed with error = %d", rslt);
6140 }
6141 }
6142
6143 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
6144 fatal("Cannot locate current stack attributes!");
6145 }
6146
6147 // Work around NPTL stack guard error.
6148 size_t guard_size = 0;
6149 rslt = pthread_attr_getguardsize(&attr, &guard_size);
6150 if (rslt != 0) {
6151 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
6152 }
6153 *bottom += guard_size;
6154 *size -= guard_size;
6155
6156 pthread_attr_destroy(&attr);
6157
6158 }
6159 assert(os::current_stack_pointer() >= *bottom &&
6160 os::current_stack_pointer() < *bottom + *size, "just checking");
6161 }
6162
6163 address os::current_stack_base() {
6164 address bottom;
6165 size_t size;
6166 current_stack_region(&bottom, &size);
6167 return (bottom + size);
6168 }
6169
6170 size_t os::current_stack_size() {
6171 // This stack size includes the usable stack and HotSpot guard pages
6172 // (for the threads that have Hotspot guard pages).
6173 address bottom;
6174 size_t size;
6175 current_stack_region(&bottom, &size);
6176 return size;
6177 }
6178 #endif
6179
6180 static inline struct timespec get_mtime(const char* filename) {
6181 struct stat st;
6182 int ret = os::stat(filename, &st);
6183 assert(ret == 0, "failed to stat() file '%s': %s", filename, os::strerror(errno));
6184 return st.st_mtim;
6185 }
6186
6187 int os::compare_file_modified_times(const char* file1, const char* file2) {
6188 struct timespec filetime1 = get_mtime(file1);
6189 struct timespec filetime2 = get_mtime(file2);
6190 int diff = filetime1.tv_sec - filetime2.tv_sec;
6191 if (diff == 0) {
6192 return filetime1.tv_nsec - filetime2.tv_nsec;
6193 }
6194 return diff;
6195 }
6196
6197 bool os::supports_map_sync() {
6198 return true;
6199 }
6200
6201 /////////////// Unit tests ///////////////
6202
6203 #ifndef PRODUCT
6204
6205 class TestReserveMemorySpecial : AllStatic {
6206 public:
6207 static void small_page_write(void* addr, size_t size) {
6208 size_t page_size = os::vm_page_size();
6209
6210 char* end = (char*)addr + size;
6211 for (char* p = (char*)addr; p < end; p += page_size) {
6212 *p = 1;
6213 }
6214 }
6215
6216 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6217 if (!UseHugeTLBFS) {
6218 return;
6219 }
6220
6221 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6222
6223 if (addr != NULL) {
6224 small_page_write(addr, size);
6225
6226 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6227 }
6228 }
6229
6230 static void test_reserve_memory_special_huge_tlbfs_only() {
6231 if (!UseHugeTLBFS) {
6232 return;
6233 }
6234
6235 size_t lp = os::large_page_size();
6236
6237 for (size_t size = lp; size <= lp * 10; size += lp) {
6238 test_reserve_memory_special_huge_tlbfs_only(size);
6239 }
6240 }
6241
6242 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6243 size_t lp = os::large_page_size();
6244 size_t ag = os::vm_allocation_granularity();
6245
6246 // sizes to test
6247 const size_t sizes[] = {
6248 lp, lp + ag, lp + lp / 2, lp * 2,
6249 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6250 lp * 10, lp * 10 + lp / 2
6251 };
6252 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6253
6254 // For each size/alignment combination, we test three scenarios:
6255 // 1) with req_addr == NULL
6256 // 2) with a non-null req_addr at which we expect to successfully allocate
6257 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6258 // expect the allocation to either fail or to ignore req_addr
6259
6260 // Pre-allocate two areas; they shall be as large as the largest allocation
6261 // and aligned to the largest alignment we will be testing.
6262 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6263 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6264 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6265 -1, 0);
6266 assert(mapping1 != MAP_FAILED, "should work");
6267
6268 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6269 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6270 -1, 0);
6271 assert(mapping2 != MAP_FAILED, "should work");
6272
6273 // Unmap the first mapping, but leave the second mapping intact: the first
6274 // mapping will serve as a value for a "good" req_addr (case 2). The second
6275 // mapping, still intact, as "bad" req_addr (case 3).
6276 ::munmap(mapping1, mapping_size);
6277
6278 // Case 1
6279 for (int i = 0; i < num_sizes; i++) {
6280 const size_t size = sizes[i];
6281 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6282 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6283 if (p != NULL) {
6284 assert(is_aligned(p, alignment), "must be");
6285 small_page_write(p, size);
6286 os::Linux::release_memory_special_huge_tlbfs(p, size);
6287 }
6288 }
6289 }
6290
6291 // Case 2
6292 for (int i = 0; i < num_sizes; i++) {
6293 const size_t size = sizes[i];
6294 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6295 char* const req_addr = align_up(mapping1, alignment);
6296 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6297 if (p != NULL) {
6298 assert(p == req_addr, "must be");
6299 small_page_write(p, size);
6300 os::Linux::release_memory_special_huge_tlbfs(p, size);
6301 }
6302 }
6303 }
6304
6305 // Case 3
6306 for (int i = 0; i < num_sizes; i++) {
6307 const size_t size = sizes[i];
6308 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6309 char* const req_addr = align_up(mapping2, alignment);
6310 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6311 // as the area around req_addr contains already existing mappings, the API should always
6312 // return NULL (as per contract, it cannot return another address)
6313 assert(p == NULL, "must be");
6314 }
6315 }
6316
6317 ::munmap(mapping2, mapping_size);
6318
6319 }
6320
6321 static void test_reserve_memory_special_huge_tlbfs() {
6322 if (!UseHugeTLBFS) {
6323 return;
6324 }
6325
6326 test_reserve_memory_special_huge_tlbfs_only();
6327 test_reserve_memory_special_huge_tlbfs_mixed();
6328 }
6329
6330 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6331 if (!UseSHM) {
6332 return;
6333 }
6334
6335 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6336
6337 if (addr != NULL) {
6338 assert(is_aligned(addr, alignment), "Check");
6339 assert(is_aligned(addr, os::large_page_size()), "Check");
6340
6341 small_page_write(addr, size);
6342
6343 os::Linux::release_memory_special_shm(addr, size);
6344 }
6345 }
6346
6347 static void test_reserve_memory_special_shm() {
6348 size_t lp = os::large_page_size();
6349 size_t ag = os::vm_allocation_granularity();
6350
6351 for (size_t size = ag; size < lp * 3; size += ag) {
6352 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6353 test_reserve_memory_special_shm(size, alignment);
6354 }
6355 }
6356 }
6357
6358 static void test() {
6359 test_reserve_memory_special_huge_tlbfs();
6360 test_reserve_memory_special_shm();
6361 }
6362 };
6363
6364 void TestReserveMemorySpecial_test() {
6365 TestReserveMemorySpecial::test();
6366 }
6367
6368 #endif