1 /*
2 * Copyright (c) 1999, 2019, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
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/orderAccess.hpp"
57 #include "runtime/osThread.hpp"
58 #include "runtime/perfMemory.hpp"
59 #include "runtime/sharedRuntime.hpp"
60 #include "runtime/statSampler.hpp"
61 #include "runtime/stubRoutines.hpp"
62 #include "runtime/thread.inline.hpp"
63 #include "runtime/threadCritical.hpp"
64 #include "runtime/threadSMR.hpp"
65 #include "runtime/timer.hpp"
66 #include "runtime/vm_version.hpp"
67 #include "semaphore_posix.hpp"
68 #include "services/attachListener.hpp"
69 #include "services/memTracker.hpp"
70 #include "services/runtimeService.hpp"
71 #include "utilities/align.hpp"
72 #include "utilities/decoder.hpp"
73 #include "utilities/defaultStream.hpp"
74 #include "utilities/events.hpp"
75 #include "utilities/elfFile.hpp"
76 #include "utilities/growableArray.hpp"
77 #include "utilities/macros.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 <errno.h>
88 # include <dlfcn.h>
89 # include <stdio.h>
90 # include <unistd.h>
91 # include <sys/resource.h>
92 # include <pthread.h>
93 # include <sys/stat.h>
94 # include <sys/time.h>
95 # include <sys/times.h>
96 # include <sys/utsname.h>
97 # include <sys/socket.h>
98 # include <sys/wait.h>
99 # include <pwd.h>
100 # include <poll.h>
101 # include <fcntl.h>
102 # include <string.h>
103 # include <syscall.h>
104 # include <sys/sysinfo.h>
105 # include <gnu/libc-version.h>
106 # include <sys/ipc.h>
107 # include <sys/shm.h>
108 # include <link.h>
109 # include <stdint.h>
110 # include <inttypes.h>
111 # include <sys/ioctl.h>
112
113 #ifndef _GNU_SOURCE
114 #define _GNU_SOURCE
115 #include <sched.h>
116 #undef _GNU_SOURCE
117 #else
118 #include <sched.h>
119 #endif
120
121 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
122 // getrusage() is prepared to handle the associated failure.
123 #ifndef RUSAGE_THREAD
124 #define RUSAGE_THREAD (1) /* only the calling thread */
125 #endif
126
127 #define MAX_PATH (2 * K)
128
129 #define MAX_SECS 100000000
130
131 // for timer info max values which include all bits
132 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
133
134 enum CoredumpFilterBit {
135 FILE_BACKED_PVT_BIT = 1 << 2,
136 FILE_BACKED_SHARED_BIT = 1 << 3,
137 LARGEPAGES_BIT = 1 << 6,
138 DAX_SHARED_BIT = 1 << 8
139 };
140
141 ////////////////////////////////////////////////////////////////////////////////
142 // global variables
143 julong os::Linux::_physical_memory = 0;
144
145 address os::Linux::_initial_thread_stack_bottom = NULL;
146 uintptr_t os::Linux::_initial_thread_stack_size = 0;
147
148 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
149 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
150 Mutex* os::Linux::_createThread_lock = 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 uint32_t os::Linux::_os_version = 0;
155 const char * os::Linux::_glibc_version = NULL;
156 const char * os::Linux::_libpthread_version = NULL;
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 = (char *)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 = (char *)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 assert(t->stack_base() != NULL, "stack_base was not initialized");
723
724 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
725 sigset_t mask_all, old_sigset;
726 sigfillset(&mask_all);
727 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
728 _expand_stack_to(addr);
729 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
730 return true;
731 }
732 return false;
733 }
734
735 //////////////////////////////////////////////////////////////////////////////
736 // create new thread
737
738 // Thread start routine for all newly created threads
739 static void *thread_native_entry(Thread *thread) {
740
741 thread->record_stack_base_and_size();
742
743 // Try to randomize the cache line index of hot stack frames.
744 // This helps when threads of the same stack traces evict each other's
745 // cache lines. The threads can be either from the same JVM instance, or
746 // from different JVM instances. The benefit is especially true for
747 // processors with hyperthreading technology.
748 static int counter = 0;
749 int pid = os::current_process_id();
750 alloca(((pid ^ counter++) & 7) * 128);
751
752 thread->initialize_thread_current();
753
754 OSThread* osthread = thread->osthread();
755 Monitor* sync = osthread->startThread_lock();
756
757 osthread->set_thread_id(os::current_thread_id());
758
759 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
760 os::current_thread_id(), (uintx) pthread_self());
761
762 if (UseNUMA) {
763 int lgrp_id = os::numa_get_group_id();
764 if (lgrp_id != -1) {
765 thread->set_lgrp_id(lgrp_id);
766 }
767 }
768 // initialize signal mask for this thread
769 os::Linux::hotspot_sigmask(thread);
770
771 // initialize floating point control register
772 os::Linux::init_thread_fpu_state();
773
774 // handshaking with parent thread
775 {
776 MutexLocker ml(sync, Mutex::_no_safepoint_check_flag);
777
778 // notify parent thread
779 osthread->set_state(INITIALIZED);
780 sync->notify_all();
781
782 // wait until os::start_thread()
783 while (osthread->get_state() == INITIALIZED) {
784 sync->wait_without_safepoint_check();
785 }
786 }
787
788 assert(osthread->pthread_id() != 0, "pthread_id was not set as expected");
789
790 // call one more level start routine
791 thread->call_run();
792
793 // Note: at this point the thread object may already have deleted itself.
794 // Prevent dereferencing it from here on out.
795 thread = NULL;
796
797 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
798 os::current_thread_id(), (uintx) pthread_self());
799
800 return 0;
801 }
802
803 bool os::create_thread(Thread* thread, ThreadType thr_type,
804 size_t req_stack_size) {
805 assert(thread->osthread() == NULL, "caller responsible");
806
807 // Allocate the OSThread object
808 OSThread* osthread = new OSThread(NULL, NULL);
809 if (osthread == NULL) {
810 return false;
811 }
812
813 // set the correct thread state
814 osthread->set_thread_type(thr_type);
815
816 // Initial state is ALLOCATED but not INITIALIZED
817 osthread->set_state(ALLOCATED);
818
819 thread->set_osthread(osthread);
820
821 // init thread attributes
822 pthread_attr_t attr;
823 pthread_attr_init(&attr);
824 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
825
826 // Calculate stack size if it's not specified by caller.
827 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
828 // In the Linux NPTL pthread implementation the guard size mechanism
829 // is not implemented properly. The posix standard requires adding
830 // the size of the guard pages to the stack size, instead Linux
831 // takes the space out of 'stacksize'. Thus we adapt the requested
832 // stack_size by the size of the guard pages to mimick proper
833 // behaviour. However, be careful not to end up with a size
834 // of zero due to overflow. Don't add the guard page in that case.
835 size_t guard_size = os::Linux::default_guard_size(thr_type);
836 if (stack_size <= SIZE_MAX - guard_size) {
837 stack_size += guard_size;
838 }
839 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
840
841 int status = pthread_attr_setstacksize(&attr, stack_size);
842 assert_status(status == 0, status, "pthread_attr_setstacksize");
843
844 // Configure glibc guard page.
845 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
846
847 ThreadState state;
848
849 {
850 pthread_t tid;
851 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
852
853 char buf[64];
854 if (ret == 0) {
855 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
856 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
857 } else {
858 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
859 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
860 // Log some OS information which might explain why creating the thread failed.
861 log_info(os, thread)("Number of threads approx. running in the VM: %d", Threads::number_of_threads());
862 LogStream st(Log(os, thread)::info());
863 os::Posix::print_rlimit_info(&st);
864 os::print_memory_info(&st);
865 os::Linux::print_proc_sys_info(&st);
866 os::Linux::print_container_info(&st);
867 }
868
869 pthread_attr_destroy(&attr);
870
871 if (ret != 0) {
872 // Need to clean up stuff we've allocated so far
873 thread->set_osthread(NULL);
874 delete osthread;
875 return false;
876 }
877
878 // Store pthread info into the OSThread
879 osthread->set_pthread_id(tid);
880
881 // Wait until child thread is either initialized or aborted
882 {
883 Monitor* sync_with_child = osthread->startThread_lock();
884 MutexLocker ml(sync_with_child, Mutex::_no_safepoint_check_flag);
885 while ((state = osthread->get_state()) == ALLOCATED) {
886 sync_with_child->wait_without_safepoint_check();
887 }
888 }
889 }
890
891 // Aborted due to thread limit being reached
892 if (state == ZOMBIE) {
893 thread->set_osthread(NULL);
894 delete osthread;
895 return false;
896 }
897
898 // The thread is returned suspended (in state INITIALIZED),
899 // and is started higher up in the call chain
900 assert(state == INITIALIZED, "race condition");
901 return true;
902 }
903
904 /////////////////////////////////////////////////////////////////////////////
905 // attach existing thread
906
907 // bootstrap the main thread
908 bool os::create_main_thread(JavaThread* thread) {
909 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
910 return create_attached_thread(thread);
911 }
912
913 bool os::create_attached_thread(JavaThread* thread) {
914 #ifdef ASSERT
915 thread->verify_not_published();
916 #endif
917
918 // Allocate the OSThread object
919 OSThread* osthread = new OSThread(NULL, NULL);
920
921 if (osthread == NULL) {
922 return false;
923 }
924
925 // Store pthread info into the OSThread
926 osthread->set_thread_id(os::Linux::gettid());
927 osthread->set_pthread_id(::pthread_self());
928
929 // initialize floating point control register
930 os::Linux::init_thread_fpu_state();
931
932 // Initial thread state is RUNNABLE
933 osthread->set_state(RUNNABLE);
934
935 thread->set_osthread(osthread);
936
937 if (UseNUMA) {
938 int lgrp_id = os::numa_get_group_id();
939 if (lgrp_id != -1) {
940 thread->set_lgrp_id(lgrp_id);
941 }
942 }
943
944 if (os::is_primordial_thread()) {
945 // If current thread is primordial thread, its stack is mapped on demand,
946 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
947 // the entire stack region to avoid SEGV in stack banging.
948 // It is also useful to get around the heap-stack-gap problem on SuSE
949 // kernel (see 4821821 for details). We first expand stack to the top
950 // of yellow zone, then enable stack yellow zone (order is significant,
951 // enabling yellow zone first will crash JVM on SuSE Linux), so there
952 // is no gap between the last two virtual memory regions.
953
954 JavaThread *jt = (JavaThread *)thread;
955 address addr = jt->stack_reserved_zone_base();
956 assert(addr != NULL, "initialization problem?");
957 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
958
959 osthread->set_expanding_stack();
960 os::Linux::manually_expand_stack(jt, addr);
961 osthread->clear_expanding_stack();
962 }
963
964 // initialize signal mask for this thread
965 // and save the caller's signal mask
966 os::Linux::hotspot_sigmask(thread);
967
968 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
969 os::current_thread_id(), (uintx) pthread_self());
970
971 return true;
972 }
973
974 void os::pd_start_thread(Thread* thread) {
975 OSThread * osthread = thread->osthread();
976 assert(osthread->get_state() != INITIALIZED, "just checking");
977 Monitor* sync_with_child = osthread->startThread_lock();
978 MutexLocker ml(sync_with_child, Mutex::_no_safepoint_check_flag);
979 sync_with_child->notify();
980 }
981
982 // Free Linux resources related to the OSThread
983 void os::free_thread(OSThread* osthread) {
984 assert(osthread != NULL, "osthread not set");
985
986 // We are told to free resources of the argument thread,
987 // but we can only really operate on the current thread.
988 assert(Thread::current()->osthread() == osthread,
989 "os::free_thread but not current thread");
990
991 #ifdef ASSERT
992 sigset_t current;
993 sigemptyset(¤t);
994 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
995 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
996 #endif
997
998 // Restore caller's signal mask
999 sigset_t sigmask = osthread->caller_sigmask();
1000 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1001
1002 delete osthread;
1003 }
1004
1005 //////////////////////////////////////////////////////////////////////////////
1006 // primordial thread
1007
1008 // Check if current thread is the primordial thread, similar to Solaris thr_main.
1009 bool os::is_primordial_thread(void) {
1010 if (suppress_primordial_thread_resolution) {
1011 return false;
1012 }
1013 char dummy;
1014 // If called before init complete, thread stack bottom will be null.
1015 // Can be called if fatal error occurs before initialization.
1016 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
1017 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
1018 os::Linux::initial_thread_stack_size() != 0,
1019 "os::init did not locate primordial thread's stack region");
1020 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
1021 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
1022 os::Linux::initial_thread_stack_size()) {
1023 return true;
1024 } else {
1025 return false;
1026 }
1027 }
1028
1029 // Find the virtual memory area that contains addr
1030 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1031 FILE *fp = fopen("/proc/self/maps", "r");
1032 if (fp) {
1033 address low, high;
1034 while (!feof(fp)) {
1035 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1036 if (low <= addr && addr < high) {
1037 if (vma_low) *vma_low = low;
1038 if (vma_high) *vma_high = high;
1039 fclose(fp);
1040 return true;
1041 }
1042 }
1043 for (;;) {
1044 int ch = fgetc(fp);
1045 if (ch == EOF || ch == (int)'\n') break;
1046 }
1047 }
1048 fclose(fp);
1049 }
1050 return false;
1051 }
1052
1053 // Locate primordial thread stack. This special handling of primordial thread stack
1054 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1055 // bogus value for the primordial process thread. While the launcher has created
1056 // the VM in a new thread since JDK 6, we still have to allow for the use of the
1057 // JNI invocation API from a primordial thread.
1058 void os::Linux::capture_initial_stack(size_t max_size) {
1059
1060 // max_size is either 0 (which means accept OS default for thread stacks) or
1061 // a user-specified value known to be at least the minimum needed. If we
1062 // are actually on the primordial thread we can make it appear that we have a
1063 // smaller max_size stack by inserting the guard pages at that location. But we
1064 // cannot do anything to emulate a larger stack than what has been provided by
1065 // the OS or threading library. In fact if we try to use a stack greater than
1066 // what is set by rlimit then we will crash the hosting process.
1067
1068 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
1069 // If this is "unlimited" then it will be a huge value.
1070 struct rlimit rlim;
1071 getrlimit(RLIMIT_STACK, &rlim);
1072 size_t stack_size = rlim.rlim_cur;
1073
1074 // 6308388: a bug in ld.so will relocate its own .data section to the
1075 // lower end of primordial stack; reduce ulimit -s value a little bit
1076 // so we won't install guard page on ld.so's data section.
1077 // But ensure we don't underflow the stack size - allow 1 page spare
1078 if (stack_size >= (size_t)(3 * page_size())) {
1079 stack_size -= 2 * page_size();
1080 }
1081
1082 // Try to figure out where the stack base (top) is. This is harder.
1083 //
1084 // When an application is started, glibc saves the initial stack pointer in
1085 // a global variable "__libc_stack_end", which is then used by system
1086 // libraries. __libc_stack_end should be pretty close to stack top. The
1087 // variable is available since the very early days. However, because it is
1088 // a private interface, it could disappear in the future.
1089 //
1090 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1091 // to __libc_stack_end, it is very close to stack top, but isn't the real
1092 // stack top. Note that /proc may not exist if VM is running as a chroot
1093 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1094 // /proc/<pid>/stat could change in the future (though unlikely).
1095 //
1096 // We try __libc_stack_end first. If that doesn't work, look for
1097 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1098 // as a hint, which should work well in most cases.
1099
1100 uintptr_t stack_start;
1101
1102 // try __libc_stack_end first
1103 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1104 if (p && *p) {
1105 stack_start = *p;
1106 } else {
1107 // see if we can get the start_stack field from /proc/self/stat
1108 FILE *fp;
1109 int pid;
1110 char state;
1111 int ppid;
1112 int pgrp;
1113 int session;
1114 int nr;
1115 int tpgrp;
1116 unsigned long flags;
1117 unsigned long minflt;
1118 unsigned long cminflt;
1119 unsigned long majflt;
1120 unsigned long cmajflt;
1121 unsigned long utime;
1122 unsigned long stime;
1123 long cutime;
1124 long cstime;
1125 long prio;
1126 long nice;
1127 long junk;
1128 long it_real;
1129 uintptr_t start;
1130 uintptr_t vsize;
1131 intptr_t rss;
1132 uintptr_t rsslim;
1133 uintptr_t scodes;
1134 uintptr_t ecode;
1135 int i;
1136
1137 // Figure what the primordial thread stack base is. Code is inspired
1138 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1139 // followed by command name surrounded by parentheses, state, etc.
1140 char stat[2048];
1141 int statlen;
1142
1143 fp = fopen("/proc/self/stat", "r");
1144 if (fp) {
1145 statlen = fread(stat, 1, 2047, fp);
1146 stat[statlen] = '\0';
1147 fclose(fp);
1148
1149 // Skip pid and the command string. Note that we could be dealing with
1150 // weird command names, e.g. user could decide to rename java launcher
1151 // to "java 1.4.2 :)", then the stat file would look like
1152 // 1234 (java 1.4.2 :)) R ... ...
1153 // We don't really need to know the command string, just find the last
1154 // occurrence of ")" and then start parsing from there. See bug 4726580.
1155 char * s = strrchr(stat, ')');
1156
1157 i = 0;
1158 if (s) {
1159 // Skip blank chars
1160 do { s++; } while (s && isspace(*s));
1161
1162 #define _UFM UINTX_FORMAT
1163 #define _DFM INTX_FORMAT
1164
1165 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1166 // 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
1167 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,
1168 &state, // 3 %c
1169 &ppid, // 4 %d
1170 &pgrp, // 5 %d
1171 &session, // 6 %d
1172 &nr, // 7 %d
1173 &tpgrp, // 8 %d
1174 &flags, // 9 %lu
1175 &minflt, // 10 %lu
1176 &cminflt, // 11 %lu
1177 &majflt, // 12 %lu
1178 &cmajflt, // 13 %lu
1179 &utime, // 14 %lu
1180 &stime, // 15 %lu
1181 &cutime, // 16 %ld
1182 &cstime, // 17 %ld
1183 &prio, // 18 %ld
1184 &nice, // 19 %ld
1185 &junk, // 20 %ld
1186 &it_real, // 21 %ld
1187 &start, // 22 UINTX_FORMAT
1188 &vsize, // 23 UINTX_FORMAT
1189 &rss, // 24 INTX_FORMAT
1190 &rsslim, // 25 UINTX_FORMAT
1191 &scodes, // 26 UINTX_FORMAT
1192 &ecode, // 27 UINTX_FORMAT
1193 &stack_start); // 28 UINTX_FORMAT
1194 }
1195
1196 #undef _UFM
1197 #undef _DFM
1198
1199 if (i != 28 - 2) {
1200 assert(false, "Bad conversion from /proc/self/stat");
1201 // product mode - assume we are the primordial thread, good luck in the
1202 // embedded case.
1203 warning("Can't detect primordial thread stack location - bad conversion");
1204 stack_start = (uintptr_t) &rlim;
1205 }
1206 } else {
1207 // For some reason we can't open /proc/self/stat (for example, running on
1208 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1209 // most cases, so don't abort:
1210 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1211 stack_start = (uintptr_t) &rlim;
1212 }
1213 }
1214
1215 // Now we have a pointer (stack_start) very close to the stack top, the
1216 // next thing to do is to figure out the exact location of stack top. We
1217 // can find out the virtual memory area that contains stack_start by
1218 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1219 // and its upper limit is the real stack top. (again, this would fail if
1220 // running inside chroot, because /proc may not exist.)
1221
1222 uintptr_t stack_top;
1223 address low, high;
1224 if (find_vma((address)stack_start, &low, &high)) {
1225 // success, "high" is the true stack top. (ignore "low", because initial
1226 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1227 stack_top = (uintptr_t)high;
1228 } else {
1229 // failed, likely because /proc/self/maps does not exist
1230 warning("Can't detect primordial thread stack location - find_vma failed");
1231 // best effort: stack_start is normally within a few pages below the real
1232 // stack top, use it as stack top, and reduce stack size so we won't put
1233 // guard page outside stack.
1234 stack_top = stack_start;
1235 stack_size -= 16 * page_size();
1236 }
1237
1238 // stack_top could be partially down the page so align it
1239 stack_top = align_up(stack_top, page_size());
1240
1241 // Allowed stack value is minimum of max_size and what we derived from rlimit
1242 if (max_size > 0) {
1243 _initial_thread_stack_size = MIN2(max_size, stack_size);
1244 } else {
1245 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1246 // clamp it at 8MB as we do on Solaris
1247 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1248 }
1249 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1250 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1251
1252 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1253
1254 if (log_is_enabled(Info, os, thread)) {
1255 // See if we seem to be on primordial process thread
1256 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1257 uintptr_t(&rlim) < stack_top;
1258
1259 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1260 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1261 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1262 stack_top, intptr_t(_initial_thread_stack_bottom));
1263 }
1264 }
1265
1266 ////////////////////////////////////////////////////////////////////////////////
1267 // time support
1268
1269 #ifndef SUPPORTS_CLOCK_MONOTONIC
1270 #error "Build platform doesn't support clock_gettime and related functionality"
1271 #endif
1272
1273 // Time since start-up in seconds to a fine granularity.
1274 // Used by VMSelfDestructTimer and the MemProfiler.
1275 double os::elapsedTime() {
1276
1277 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1278 }
1279
1280 jlong os::elapsed_counter() {
1281 return javaTimeNanos() - initial_time_count;
1282 }
1283
1284 jlong os::elapsed_frequency() {
1285 return NANOSECS_PER_SEC; // nanosecond resolution
1286 }
1287
1288 bool os::supports_vtime() { return true; }
1289 bool os::enable_vtime() { return false; }
1290 bool os::vtime_enabled() { return false; }
1291
1292 double os::elapsedVTime() {
1293 struct rusage usage;
1294 int retval = getrusage(RUSAGE_THREAD, &usage);
1295 if (retval == 0) {
1296 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);
1297 } else {
1298 // better than nothing, but not much
1299 return elapsedTime();
1300 }
1301 }
1302
1303 jlong os::javaTimeMillis() {
1304 timeval time;
1305 int status = gettimeofday(&time, NULL);
1306 assert(status != -1, "linux error");
1307 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1308 }
1309
1310 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1311 timeval time;
1312 int status = gettimeofday(&time, NULL);
1313 assert(status != -1, "linux error");
1314 seconds = jlong(time.tv_sec);
1315 nanos = jlong(time.tv_usec) * 1000;
1316 }
1317
1318 void os::Linux::fast_thread_clock_init() {
1319 if (!UseLinuxPosixThreadCPUClocks) {
1320 return;
1321 }
1322 clockid_t clockid;
1323 struct timespec tp;
1324 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1325 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1326
1327 // Switch to using fast clocks for thread cpu time if
1328 // the clock_getres() returns 0 error code.
1329 // Note, that some kernels may support the current thread
1330 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1331 // returned by the pthread_getcpuclockid().
1332 // If the fast Posix clocks are supported then the clock_getres()
1333 // must return at least tp.tv_sec == 0 which means a resolution
1334 // better than 1 sec. This is extra check for reliability.
1335
1336 if (pthread_getcpuclockid_func &&
1337 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1338 os::Posix::clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1339 _supports_fast_thread_cpu_time = true;
1340 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1341 }
1342 }
1343
1344 jlong os::javaTimeNanos() {
1345 if (os::supports_monotonic_clock()) {
1346 struct timespec tp;
1347 int status = os::Posix::clock_gettime(CLOCK_MONOTONIC, &tp);
1348 assert(status == 0, "gettime error");
1349 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1350 return result;
1351 } else {
1352 timeval time;
1353 int status = gettimeofday(&time, NULL);
1354 assert(status != -1, "linux error");
1355 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1356 return 1000 * usecs;
1357 }
1358 }
1359
1360 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1361 if (os::supports_monotonic_clock()) {
1362 info_ptr->max_value = ALL_64_BITS;
1363
1364 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1365 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1366 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1367 } else {
1368 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1369 info_ptr->max_value = ALL_64_BITS;
1370
1371 // gettimeofday is a real time clock so it skips
1372 info_ptr->may_skip_backward = true;
1373 info_ptr->may_skip_forward = true;
1374 }
1375
1376 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1377 }
1378
1379 // Return the real, user, and system times in seconds from an
1380 // arbitrary fixed point in the past.
1381 bool os::getTimesSecs(double* process_real_time,
1382 double* process_user_time,
1383 double* process_system_time) {
1384 struct tms ticks;
1385 clock_t real_ticks = times(&ticks);
1386
1387 if (real_ticks == (clock_t) (-1)) {
1388 return false;
1389 } else {
1390 double ticks_per_second = (double) clock_tics_per_sec;
1391 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1392 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1393 *process_real_time = ((double) real_ticks) / ticks_per_second;
1394
1395 return true;
1396 }
1397 }
1398
1399
1400 char * os::local_time_string(char *buf, size_t buflen) {
1401 struct tm t;
1402 time_t long_time;
1403 time(&long_time);
1404 localtime_r(&long_time, &t);
1405 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1406 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1407 t.tm_hour, t.tm_min, t.tm_sec);
1408 return buf;
1409 }
1410
1411 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1412 return localtime_r(clock, res);
1413 }
1414
1415 ////////////////////////////////////////////////////////////////////////////////
1416 // runtime exit support
1417
1418 // Note: os::shutdown() might be called very early during initialization, or
1419 // called from signal handler. Before adding something to os::shutdown(), make
1420 // sure it is async-safe and can handle partially initialized VM.
1421 void os::shutdown() {
1422
1423 // allow PerfMemory to attempt cleanup of any persistent resources
1424 perfMemory_exit();
1425
1426 // needs to remove object in file system
1427 AttachListener::abort();
1428
1429 // flush buffered output, finish log files
1430 ostream_abort();
1431
1432 // Check for abort hook
1433 abort_hook_t abort_hook = Arguments::abort_hook();
1434 if (abort_hook != NULL) {
1435 abort_hook();
1436 }
1437
1438 }
1439
1440 // Note: os::abort() might be called very early during initialization, or
1441 // called from signal handler. Before adding something to os::abort(), make
1442 // sure it is async-safe and can handle partially initialized VM.
1443 void os::abort(bool dump_core, void* siginfo, const void* context) {
1444 os::shutdown();
1445 if (dump_core) {
1446 if (DumpPrivateMappingsInCore) {
1447 ClassLoader::close_jrt_image();
1448 }
1449 #ifndef PRODUCT
1450 fdStream out(defaultStream::output_fd());
1451 out.print_raw("Current thread is ");
1452 char buf[16];
1453 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1454 out.print_raw_cr(buf);
1455 out.print_raw_cr("Dumping core ...");
1456 #endif
1457 ::abort(); // dump core
1458 }
1459
1460 ::exit(1);
1461 }
1462
1463 // Die immediately, no exit hook, no abort hook, no cleanup.
1464 // Dump a core file, if possible, for debugging.
1465 void os::die() {
1466 if (TestUnresponsiveErrorHandler && !CreateCoredumpOnCrash) {
1467 // For TimeoutInErrorHandlingTest.java, we just kill the VM
1468 // and don't take the time to generate a core file.
1469 os::signal_raise(SIGKILL);
1470 } else {
1471 ::abort();
1472 }
1473 }
1474
1475 // thread_id is kernel thread id (similar to Solaris LWP id)
1476 intx os::current_thread_id() { return os::Linux::gettid(); }
1477 int os::current_process_id() {
1478 return ::getpid();
1479 }
1480
1481 // DLL functions
1482
1483 const char* os::dll_file_extension() { return ".so"; }
1484
1485 // This must be hard coded because it's the system's temporary
1486 // directory not the java application's temp directory, ala java.io.tmpdir.
1487 const char* os::get_temp_directory() { return "/tmp"; }
1488
1489 static bool file_exists(const char* filename) {
1490 struct stat statbuf;
1491 if (filename == NULL || strlen(filename) == 0) {
1492 return false;
1493 }
1494 return os::stat(filename, &statbuf) == 0;
1495 }
1496
1497 // check if addr is inside libjvm.so
1498 bool os::address_is_in_vm(address addr) {
1499 static address libjvm_base_addr;
1500 Dl_info dlinfo;
1501
1502 if (libjvm_base_addr == NULL) {
1503 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1504 libjvm_base_addr = (address)dlinfo.dli_fbase;
1505 }
1506 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1507 }
1508
1509 if (dladdr((void *)addr, &dlinfo) != 0) {
1510 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1511 }
1512
1513 return false;
1514 }
1515
1516 bool os::dll_address_to_function_name(address addr, char *buf,
1517 int buflen, int *offset,
1518 bool demangle) {
1519 // buf is not optional, but offset is optional
1520 assert(buf != NULL, "sanity check");
1521
1522 Dl_info dlinfo;
1523
1524 if (dladdr((void*)addr, &dlinfo) != 0) {
1525 // see if we have a matching symbol
1526 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1527 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1528 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1529 }
1530 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1531 return true;
1532 }
1533 // no matching symbol so try for just file info
1534 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1535 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1536 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1537 return true;
1538 }
1539 }
1540 }
1541
1542 buf[0] = '\0';
1543 if (offset != NULL) *offset = -1;
1544 return false;
1545 }
1546
1547 struct _address_to_library_name {
1548 address addr; // input : memory address
1549 size_t buflen; // size of fname
1550 char* fname; // output: library name
1551 address base; // library base addr
1552 };
1553
1554 static int address_to_library_name_callback(struct dl_phdr_info *info,
1555 size_t size, void *data) {
1556 int i;
1557 bool found = false;
1558 address libbase = NULL;
1559 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1560
1561 // iterate through all loadable segments
1562 for (i = 0; i < info->dlpi_phnum; i++) {
1563 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1564 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1565 // base address of a library is the lowest address of its loaded
1566 // segments.
1567 if (libbase == NULL || libbase > segbase) {
1568 libbase = segbase;
1569 }
1570 // see if 'addr' is within current segment
1571 if (segbase <= d->addr &&
1572 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1573 found = true;
1574 }
1575 }
1576 }
1577
1578 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1579 // so dll_address_to_library_name() can fall through to use dladdr() which
1580 // can figure out executable name from argv[0].
1581 if (found && info->dlpi_name && info->dlpi_name[0]) {
1582 d->base = libbase;
1583 if (d->fname) {
1584 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1585 }
1586 return 1;
1587 }
1588 return 0;
1589 }
1590
1591 bool os::dll_address_to_library_name(address addr, char* buf,
1592 int buflen, int* offset) {
1593 // buf is not optional, but offset is optional
1594 assert(buf != NULL, "sanity check");
1595
1596 Dl_info dlinfo;
1597 struct _address_to_library_name data;
1598
1599 // There is a bug in old glibc dladdr() implementation that it could resolve
1600 // to wrong library name if the .so file has a base address != NULL. Here
1601 // we iterate through the program headers of all loaded libraries to find
1602 // out which library 'addr' really belongs to. This workaround can be
1603 // removed once the minimum requirement for glibc is moved to 2.3.x.
1604 data.addr = addr;
1605 data.fname = buf;
1606 data.buflen = buflen;
1607 data.base = NULL;
1608 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1609
1610 if (rslt) {
1611 // buf already contains library name
1612 if (offset) *offset = addr - data.base;
1613 return true;
1614 }
1615 if (dladdr((void*)addr, &dlinfo) != 0) {
1616 if (dlinfo.dli_fname != NULL) {
1617 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1618 }
1619 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1620 *offset = addr - (address)dlinfo.dli_fbase;
1621 }
1622 return true;
1623 }
1624
1625 buf[0] = '\0';
1626 if (offset) *offset = -1;
1627 return false;
1628 }
1629
1630 // Loads .dll/.so and
1631 // in case of error it checks if .dll/.so was built for the
1632 // same architecture as Hotspot is running on
1633
1634
1635 // Remember the stack's state. The Linux dynamic linker will change
1636 // the stack to 'executable' at most once, so we must safepoint only once.
1637 bool os::Linux::_stack_is_executable = false;
1638
1639 // VM operation that loads a library. This is necessary if stack protection
1640 // of the Java stacks can be lost during loading the library. If we
1641 // do not stop the Java threads, they can stack overflow before the stacks
1642 // are protected again.
1643 class VM_LinuxDllLoad: public VM_Operation {
1644 private:
1645 const char *_filename;
1646 char *_ebuf;
1647 int _ebuflen;
1648 void *_lib;
1649 public:
1650 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1651 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1652 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1653 void doit() {
1654 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1655 os::Linux::_stack_is_executable = true;
1656 }
1657 void* loaded_library() { return _lib; }
1658 };
1659
1660 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1661 void * result = NULL;
1662 bool load_attempted = false;
1663
1664 // Check whether the library to load might change execution rights
1665 // of the stack. If they are changed, the protection of the stack
1666 // guard pages will be lost. We need a safepoint to fix this.
1667 //
1668 // See Linux man page execstack(8) for more info.
1669 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1670 if (!ElfFile::specifies_noexecstack(filename)) {
1671 if (!is_init_completed()) {
1672 os::Linux::_stack_is_executable = true;
1673 // This is OK - No Java threads have been created yet, and hence no
1674 // stack guard pages to fix.
1675 //
1676 // Dynamic loader will make all stacks executable after
1677 // this function returns, and will not do that again.
1678 assert(Threads::number_of_threads() == 0, "no Java threads should exist yet.");
1679 } else {
1680 warning("You have loaded library %s which might have disabled stack guard. "
1681 "The VM will try to fix the stack guard now.\n"
1682 "It's highly recommended that you fix the library with "
1683 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1684 filename);
1685
1686 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1687 JavaThread *jt = JavaThread::current();
1688 if (jt->thread_state() != _thread_in_native) {
1689 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1690 // that requires ExecStack. Cannot enter safe point. Let's give up.
1691 warning("Unable to fix stack guard. Giving up.");
1692 } else {
1693 if (!LoadExecStackDllInVMThread) {
1694 // This is for the case where the DLL has an static
1695 // constructor function that executes JNI code. We cannot
1696 // load such DLLs in the VMThread.
1697 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1698 }
1699
1700 ThreadInVMfromNative tiv(jt);
1701 debug_only(VMNativeEntryWrapper vew;)
1702
1703 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1704 VMThread::execute(&op);
1705 if (LoadExecStackDllInVMThread) {
1706 result = op.loaded_library();
1707 }
1708 load_attempted = true;
1709 }
1710 }
1711 }
1712 }
1713
1714 if (!load_attempted) {
1715 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1716 }
1717
1718 if (result != NULL) {
1719 // Successful loading
1720 return result;
1721 }
1722
1723 Elf32_Ehdr elf_head;
1724 int diag_msg_max_length=ebuflen-strlen(ebuf);
1725 char* diag_msg_buf=ebuf+strlen(ebuf);
1726
1727 if (diag_msg_max_length==0) {
1728 // No more space in ebuf for additional diagnostics message
1729 return NULL;
1730 }
1731
1732
1733 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1734
1735 if (file_descriptor < 0) {
1736 // Can't open library, report dlerror() message
1737 return NULL;
1738 }
1739
1740 bool failed_to_read_elf_head=
1741 (sizeof(elf_head)!=
1742 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1743
1744 ::close(file_descriptor);
1745 if (failed_to_read_elf_head) {
1746 // file i/o error - report dlerror() msg
1747 return NULL;
1748 }
1749
1750 typedef struct {
1751 Elf32_Half code; // Actual value as defined in elf.h
1752 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1753 unsigned char elf_class; // 32 or 64 bit
1754 unsigned char endianess; // MSB or LSB
1755 char* name; // String representation
1756 } arch_t;
1757
1758 #ifndef EM_486
1759 #define EM_486 6 /* Intel 80486 */
1760 #endif
1761 #ifndef EM_AARCH64
1762 #define EM_AARCH64 183 /* ARM AARCH64 */
1763 #endif
1764
1765 static const arch_t arch_array[]={
1766 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1767 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1768 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1769 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1770 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1771 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1772 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1773 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1774 #if defined(VM_LITTLE_ENDIAN)
1775 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1776 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1777 #else
1778 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1779 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1780 #endif
1781 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1782 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1783 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1784 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1785 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1786 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1787 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1788 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1789 };
1790
1791 #if (defined IA32)
1792 static Elf32_Half running_arch_code=EM_386;
1793 #elif (defined AMD64) || (defined X32)
1794 static Elf32_Half running_arch_code=EM_X86_64;
1795 #elif (defined IA64)
1796 static Elf32_Half running_arch_code=EM_IA_64;
1797 #elif (defined __sparc) && (defined _LP64)
1798 static Elf32_Half running_arch_code=EM_SPARCV9;
1799 #elif (defined __sparc) && (!defined _LP64)
1800 static Elf32_Half running_arch_code=EM_SPARC;
1801 #elif (defined __powerpc64__)
1802 static Elf32_Half running_arch_code=EM_PPC64;
1803 #elif (defined __powerpc__)
1804 static Elf32_Half running_arch_code=EM_PPC;
1805 #elif (defined AARCH64)
1806 static Elf32_Half running_arch_code=EM_AARCH64;
1807 #elif (defined ARM)
1808 static Elf32_Half running_arch_code=EM_ARM;
1809 #elif (defined S390)
1810 static Elf32_Half running_arch_code=EM_S390;
1811 #elif (defined ALPHA)
1812 static Elf32_Half running_arch_code=EM_ALPHA;
1813 #elif (defined MIPSEL)
1814 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1815 #elif (defined PARISC)
1816 static Elf32_Half running_arch_code=EM_PARISC;
1817 #elif (defined MIPS)
1818 static Elf32_Half running_arch_code=EM_MIPS;
1819 #elif (defined M68K)
1820 static Elf32_Half running_arch_code=EM_68K;
1821 #elif (defined SH)
1822 static Elf32_Half running_arch_code=EM_SH;
1823 #else
1824 #error Method os::dll_load requires that one of following is defined:\
1825 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1826 #endif
1827
1828 // Identify compatability class for VM's architecture and library's architecture
1829 // Obtain string descriptions for architectures
1830
1831 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1832 int running_arch_index=-1;
1833
1834 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1835 if (running_arch_code == arch_array[i].code) {
1836 running_arch_index = i;
1837 }
1838 if (lib_arch.code == arch_array[i].code) {
1839 lib_arch.compat_class = arch_array[i].compat_class;
1840 lib_arch.name = arch_array[i].name;
1841 }
1842 }
1843
1844 assert(running_arch_index != -1,
1845 "Didn't find running architecture code (running_arch_code) in arch_array");
1846 if (running_arch_index == -1) {
1847 // Even though running architecture detection failed
1848 // we may still continue with reporting dlerror() message
1849 return NULL;
1850 }
1851
1852 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1853 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1854 return NULL;
1855 }
1856
1857 #ifndef S390
1858 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1859 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1860 return NULL;
1861 }
1862 #endif // !S390
1863
1864 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1865 if (lib_arch.name!=NULL) {
1866 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1867 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1868 lib_arch.name, arch_array[running_arch_index].name);
1869 } else {
1870 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1871 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1872 lib_arch.code,
1873 arch_array[running_arch_index].name);
1874 }
1875 }
1876
1877 return NULL;
1878 }
1879
1880 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1881 int ebuflen) {
1882 void * result = ::dlopen(filename, RTLD_LAZY);
1883 if (result == NULL) {
1884 const char* error_report = ::dlerror();
1885 if (error_report == NULL) {
1886 error_report = "dlerror returned no error description";
1887 }
1888 if (ebuf != NULL && ebuflen > 0) {
1889 ::strncpy(ebuf, error_report, ebuflen-1);
1890 ebuf[ebuflen-1]='\0';
1891 }
1892 Events::log(NULL, "Loading shared library %s failed, %s", filename, error_report);
1893 } else {
1894 Events::log(NULL, "Loaded shared library %s", filename);
1895 }
1896 return result;
1897 }
1898
1899 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1900 int ebuflen) {
1901 void * result = NULL;
1902 if (LoadExecStackDllInVMThread) {
1903 result = dlopen_helper(filename, ebuf, ebuflen);
1904 }
1905
1906 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1907 // library that requires an executable stack, or which does not have this
1908 // stack attribute set, dlopen changes the stack attribute to executable. The
1909 // read protection of the guard pages gets lost.
1910 //
1911 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1912 // may have been queued at the same time.
1913
1914 if (!_stack_is_executable) {
1915 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
1916 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1917 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1918 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1919 warning("Attempt to reguard stack yellow zone failed.");
1920 }
1921 }
1922 }
1923 }
1924
1925 return result;
1926 }
1927
1928 void* os::dll_lookup(void* handle, const char* name) {
1929 void* res = dlsym(handle, name);
1930 return res;
1931 }
1932
1933 void* os::get_default_process_handle() {
1934 return (void*)::dlopen(NULL, RTLD_LAZY);
1935 }
1936
1937 static bool _print_ascii_file(const char* filename, outputStream* st, const char* hdr = NULL) {
1938 int fd = ::open(filename, O_RDONLY);
1939 if (fd == -1) {
1940 return false;
1941 }
1942
1943 if (hdr != NULL) {
1944 st->print_cr("%s", hdr);
1945 }
1946
1947 char buf[33];
1948 int bytes;
1949 buf[32] = '\0';
1950 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1951 st->print_raw(buf, bytes);
1952 }
1953
1954 ::close(fd);
1955
1956 return true;
1957 }
1958
1959 void os::print_dll_info(outputStream *st) {
1960 st->print_cr("Dynamic libraries:");
1961
1962 char fname[32];
1963 pid_t pid = os::Linux::gettid();
1964
1965 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1966
1967 if (!_print_ascii_file(fname, st)) {
1968 st->print("Can not get library information for pid = %d\n", pid);
1969 }
1970 }
1971
1972 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1973 FILE *procmapsFile = NULL;
1974
1975 // Open the procfs maps file for the current process
1976 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1977 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1978 char line[PATH_MAX + 100];
1979
1980 // Read line by line from 'file'
1981 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1982 u8 base, top, offset, inode;
1983 char permissions[5];
1984 char device[6];
1985 char name[PATH_MAX + 1];
1986
1987 // Parse fields from line
1988 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %7s " INT64_FORMAT " %s",
1989 &base, &top, permissions, &offset, device, &inode, name);
1990
1991 // Filter by device id '00:00' so that we only get file system mapped files.
1992 if (strcmp(device, "00:00") != 0) {
1993
1994 // Call callback with the fields of interest
1995 if(callback(name, (address)base, (address)top, param)) {
1996 // Oops abort, callback aborted
1997 fclose(procmapsFile);
1998 return 1;
1999 }
2000 }
2001 }
2002 fclose(procmapsFile);
2003 }
2004 return 0;
2005 }
2006
2007 void os::print_os_info_brief(outputStream* st) {
2008 os::Linux::print_distro_info(st);
2009
2010 os::Posix::print_uname_info(st);
2011
2012 os::Linux::print_libversion_info(st);
2013
2014 }
2015
2016 void os::print_os_info(outputStream* st) {
2017 st->print("OS:");
2018
2019 os::Linux::print_distro_info(st);
2020
2021 os::Posix::print_uname_info(st);
2022
2023 // Print warning if unsafe chroot environment detected
2024 if (unsafe_chroot_detected) {
2025 st->print("WARNING!! ");
2026 st->print_cr("%s", unstable_chroot_error);
2027 }
2028
2029 os::Linux::print_libversion_info(st);
2030
2031 os::Posix::print_rlimit_info(st);
2032
2033 os::Posix::print_load_average(st);
2034
2035 os::Linux::print_full_memory_info(st);
2036
2037 os::Linux::print_proc_sys_info(st);
2038
2039 os::Linux::print_ld_preload_file(st);
2040
2041 os::Linux::print_container_info(st);
2042
2043 VM_Version::print_platform_virtualization_info(st);
2044
2045 os::Linux::print_steal_info(st);
2046 }
2047
2048 // Try to identify popular distros.
2049 // Most Linux distributions have a /etc/XXX-release file, which contains
2050 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2051 // file that also contains the OS version string. Some have more than one
2052 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2053 // /etc/redhat-release.), so the order is important.
2054 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2055 // their own specific XXX-release file as well as a redhat-release file.
2056 // Because of this the XXX-release file needs to be searched for before the
2057 // redhat-release file.
2058 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2059 // search for redhat-release / SuSE-release needs to be before lsb-release.
2060 // Since the lsb-release file is the new standard it needs to be searched
2061 // before the older style release files.
2062 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2063 // next to last resort. The os-release file is a new standard that contains
2064 // distribution information and the system-release file seems to be an old
2065 // standard that has been replaced by the lsb-release and os-release files.
2066 // Searching for the debian_version file is the last resort. It contains
2067 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2068 // "Debian " is printed before the contents of the debian_version file.
2069
2070 const char* distro_files[] = {
2071 "/etc/oracle-release",
2072 "/etc/mandriva-release",
2073 "/etc/mandrake-release",
2074 "/etc/sun-release",
2075 "/etc/redhat-release",
2076 "/etc/SuSE-release",
2077 "/etc/lsb-release",
2078 "/etc/turbolinux-release",
2079 "/etc/gentoo-release",
2080 "/etc/ltib-release",
2081 "/etc/angstrom-version",
2082 "/etc/system-release",
2083 "/etc/os-release",
2084 NULL };
2085
2086 void os::Linux::print_distro_info(outputStream* st) {
2087 for (int i = 0;; i++) {
2088 const char* file = distro_files[i];
2089 if (file == NULL) {
2090 break; // done
2091 }
2092 // If file prints, we found it.
2093 if (_print_ascii_file(file, st)) {
2094 return;
2095 }
2096 }
2097
2098 if (file_exists("/etc/debian_version")) {
2099 st->print("Debian ");
2100 _print_ascii_file("/etc/debian_version", st);
2101 } else {
2102 st->print("Linux");
2103 }
2104 st->cr();
2105 }
2106
2107 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2108 char buf[256];
2109 while (fgets(buf, sizeof(buf), fp)) {
2110 // Edit out extra stuff in expected format
2111 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2112 char* ptr = strstr(buf, "\""); // the name is in quotes
2113 if (ptr != NULL) {
2114 ptr++; // go beyond first quote
2115 char* nl = strchr(ptr, '\"');
2116 if (nl != NULL) *nl = '\0';
2117 strncpy(distro, ptr, length);
2118 } else {
2119 ptr = strstr(buf, "=");
2120 ptr++; // go beyond equals then
2121 char* nl = strchr(ptr, '\n');
2122 if (nl != NULL) *nl = '\0';
2123 strncpy(distro, ptr, length);
2124 }
2125 return;
2126 } else if (get_first_line) {
2127 char* nl = strchr(buf, '\n');
2128 if (nl != NULL) *nl = '\0';
2129 strncpy(distro, buf, length);
2130 return;
2131 }
2132 }
2133 // print last line and close
2134 char* nl = strchr(buf, '\n');
2135 if (nl != NULL) *nl = '\0';
2136 strncpy(distro, buf, length);
2137 }
2138
2139 static void parse_os_info(char* distro, size_t length, const char* file) {
2140 FILE* fp = fopen(file, "r");
2141 if (fp != NULL) {
2142 // if suse format, print out first line
2143 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2144 parse_os_info_helper(fp, distro, length, get_first_line);
2145 fclose(fp);
2146 }
2147 }
2148
2149 void os::get_summary_os_info(char* buf, size_t buflen) {
2150 for (int i = 0;; i++) {
2151 const char* file = distro_files[i];
2152 if (file == NULL) {
2153 break; // ran out of distro_files
2154 }
2155 if (file_exists(file)) {
2156 parse_os_info(buf, buflen, file);
2157 return;
2158 }
2159 }
2160 // special case for debian
2161 if (file_exists("/etc/debian_version")) {
2162 strncpy(buf, "Debian ", buflen);
2163 if (buflen > 7) {
2164 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2165 }
2166 } else {
2167 strncpy(buf, "Linux", buflen);
2168 }
2169 }
2170
2171 void os::Linux::print_libversion_info(outputStream* st) {
2172 // libc, pthread
2173 st->print("libc:");
2174 st->print("%s ", os::Linux::glibc_version());
2175 st->print("%s ", os::Linux::libpthread_version());
2176 st->cr();
2177 }
2178
2179 void os::Linux::print_proc_sys_info(outputStream* st) {
2180 st->cr();
2181 st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):");
2182 _print_ascii_file("/proc/sys/kernel/threads-max", st);
2183 st->cr();
2184 st->cr();
2185
2186 st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):");
2187 _print_ascii_file("/proc/sys/vm/max_map_count", st);
2188 st->cr();
2189 st->cr();
2190
2191 st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):");
2192 _print_ascii_file("/proc/sys/kernel/pid_max", st);
2193 st->cr();
2194 st->cr();
2195 }
2196
2197 void os::Linux::print_full_memory_info(outputStream* st) {
2198 st->print("\n/proc/meminfo:\n");
2199 _print_ascii_file("/proc/meminfo", st);
2200 st->cr();
2201 }
2202
2203 void os::Linux::print_ld_preload_file(outputStream* st) {
2204 _print_ascii_file("/etc/ld.so.preload", st, "\n/etc/ld.so.preload:");
2205 st->cr();
2206 }
2207
2208 void os::Linux::print_container_info(outputStream* st) {
2209 if (!OSContainer::is_containerized()) {
2210 return;
2211 }
2212
2213 st->print("container (cgroup) information:\n");
2214
2215 const char *p_ct = OSContainer::container_type();
2216 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "not supported");
2217
2218 char *p = OSContainer::cpu_cpuset_cpus();
2219 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "not supported");
2220 free(p);
2221
2222 p = OSContainer::cpu_cpuset_memory_nodes();
2223 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "not supported");
2224 free(p);
2225
2226 int i = OSContainer::active_processor_count();
2227 st->print("active_processor_count: ");
2228 if (i > 0) {
2229 st->print("%d\n", i);
2230 } else {
2231 st->print("not supported\n");
2232 }
2233
2234 i = OSContainer::cpu_quota();
2235 st->print("cpu_quota: ");
2236 if (i > 0) {
2237 st->print("%d\n", i);
2238 } else {
2239 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no quota");
2240 }
2241
2242 i = OSContainer::cpu_period();
2243 st->print("cpu_period: ");
2244 if (i > 0) {
2245 st->print("%d\n", i);
2246 } else {
2247 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no period");
2248 }
2249
2250 i = OSContainer::cpu_shares();
2251 st->print("cpu_shares: ");
2252 if (i > 0) {
2253 st->print("%d\n", i);
2254 } else {
2255 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no shares");
2256 }
2257
2258 jlong j = OSContainer::memory_limit_in_bytes();
2259 st->print("memory_limit_in_bytes: ");
2260 if (j > 0) {
2261 st->print(JLONG_FORMAT "\n", j);
2262 } else {
2263 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2264 }
2265
2266 j = OSContainer::memory_and_swap_limit_in_bytes();
2267 st->print("memory_and_swap_limit_in_bytes: ");
2268 if (j > 0) {
2269 st->print(JLONG_FORMAT "\n", j);
2270 } else {
2271 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2272 }
2273
2274 j = OSContainer::memory_soft_limit_in_bytes();
2275 st->print("memory_soft_limit_in_bytes: ");
2276 if (j > 0) {
2277 st->print(JLONG_FORMAT "\n", j);
2278 } else {
2279 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2280 }
2281
2282 j = OSContainer::OSContainer::memory_usage_in_bytes();
2283 st->print("memory_usage_in_bytes: ");
2284 if (j > 0) {
2285 st->print(JLONG_FORMAT "\n", j);
2286 } else {
2287 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2288 }
2289
2290 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2291 st->print("memory_max_usage_in_bytes: ");
2292 if (j > 0) {
2293 st->print(JLONG_FORMAT "\n", j);
2294 } else {
2295 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2296 }
2297 st->cr();
2298 }
2299
2300 void os::Linux::print_steal_info(outputStream* st) {
2301 if (has_initial_tick_info) {
2302 CPUPerfTicks pticks;
2303 bool res = os::Linux::get_tick_information(&pticks, -1);
2304
2305 if (res && pticks.has_steal_ticks) {
2306 uint64_t steal_ticks_difference = pticks.steal - initial_steal_ticks;
2307 uint64_t total_ticks_difference = pticks.total - initial_total_ticks;
2308 double steal_ticks_perc = 0.0;
2309 if (total_ticks_difference != 0) {
2310 steal_ticks_perc = (double) steal_ticks_difference / total_ticks_difference;
2311 }
2312 st->print_cr("Steal ticks since vm start: " UINT64_FORMAT, steal_ticks_difference);
2313 st->print_cr("Steal ticks percentage since vm start:%7.3f", steal_ticks_perc);
2314 }
2315 }
2316 }
2317
2318 void os::print_memory_info(outputStream* st) {
2319
2320 st->print("Memory:");
2321 st->print(" %dk page", os::vm_page_size()>>10);
2322
2323 // values in struct sysinfo are "unsigned long"
2324 struct sysinfo si;
2325 sysinfo(&si);
2326
2327 st->print(", physical " UINT64_FORMAT "k",
2328 os::physical_memory() >> 10);
2329 st->print("(" UINT64_FORMAT "k free)",
2330 os::available_memory() >> 10);
2331 st->print(", swap " UINT64_FORMAT "k",
2332 ((jlong)si.totalswap * si.mem_unit) >> 10);
2333 st->print("(" UINT64_FORMAT "k free)",
2334 ((jlong)si.freeswap * si.mem_unit) >> 10);
2335 st->cr();
2336 }
2337
2338 // Print the first "model name" line and the first "flags" line
2339 // that we find and nothing more. We assume "model name" comes
2340 // before "flags" so if we find a second "model name", then the
2341 // "flags" field is considered missing.
2342 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2343 #if defined(IA32) || defined(AMD64)
2344 // Other platforms have less repetitive cpuinfo files
2345 FILE *fp = fopen("/proc/cpuinfo", "r");
2346 if (fp) {
2347 while (!feof(fp)) {
2348 if (fgets(buf, buflen, fp)) {
2349 // Assume model name comes before flags
2350 bool model_name_printed = false;
2351 if (strstr(buf, "model name") != NULL) {
2352 if (!model_name_printed) {
2353 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2354 st->print_raw(buf);
2355 model_name_printed = true;
2356 } else {
2357 // model name printed but not flags? Odd, just return
2358 fclose(fp);
2359 return true;
2360 }
2361 }
2362 // print the flags line too
2363 if (strstr(buf, "flags") != NULL) {
2364 st->print_raw(buf);
2365 fclose(fp);
2366 return true;
2367 }
2368 }
2369 }
2370 fclose(fp);
2371 }
2372 #endif // x86 platforms
2373 return false;
2374 }
2375
2376 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2377 // Only print the model name if the platform provides this as a summary
2378 if (!print_model_name_and_flags(st, buf, buflen)) {
2379 st->print("\n/proc/cpuinfo:\n");
2380 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2381 st->print_cr(" <Not Available>");
2382 }
2383 }
2384 }
2385
2386 #if defined(AMD64) || defined(IA32) || defined(X32)
2387 const char* search_string = "model name";
2388 #elif defined(M68K)
2389 const char* search_string = "CPU";
2390 #elif defined(PPC64)
2391 const char* search_string = "cpu";
2392 #elif defined(S390)
2393 const char* search_string = "machine =";
2394 #elif defined(SPARC)
2395 const char* search_string = "cpu";
2396 #else
2397 const char* search_string = "Processor";
2398 #endif
2399
2400 // Parses the cpuinfo file for string representing the model name.
2401 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2402 FILE* fp = fopen("/proc/cpuinfo", "r");
2403 if (fp != NULL) {
2404 while (!feof(fp)) {
2405 char buf[256];
2406 if (fgets(buf, sizeof(buf), fp)) {
2407 char* start = strstr(buf, search_string);
2408 if (start != NULL) {
2409 char *ptr = start + strlen(search_string);
2410 char *end = buf + strlen(buf);
2411 while (ptr != end) {
2412 // skip whitespace and colon for the rest of the name.
2413 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2414 break;
2415 }
2416 ptr++;
2417 }
2418 if (ptr != end) {
2419 // reasonable string, get rid of newline and keep the rest
2420 char* nl = strchr(buf, '\n');
2421 if (nl != NULL) *nl = '\0';
2422 strncpy(cpuinfo, ptr, length);
2423 fclose(fp);
2424 return;
2425 }
2426 }
2427 }
2428 }
2429 fclose(fp);
2430 }
2431 // cpuinfo not found or parsing failed, just print generic string. The entire
2432 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2433 #if defined(AARCH64)
2434 strncpy(cpuinfo, "AArch64", length);
2435 #elif defined(AMD64)
2436 strncpy(cpuinfo, "x86_64", length);
2437 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2438 strncpy(cpuinfo, "ARM", length);
2439 #elif defined(IA32)
2440 strncpy(cpuinfo, "x86_32", length);
2441 #elif defined(IA64)
2442 strncpy(cpuinfo, "IA64", length);
2443 #elif defined(PPC)
2444 strncpy(cpuinfo, "PPC64", length);
2445 #elif defined(S390)
2446 strncpy(cpuinfo, "S390", length);
2447 #elif defined(SPARC)
2448 strncpy(cpuinfo, "sparcv9", length);
2449 #elif defined(ZERO_LIBARCH)
2450 strncpy(cpuinfo, ZERO_LIBARCH, length);
2451 #else
2452 strncpy(cpuinfo, "unknown", length);
2453 #endif
2454 }
2455
2456 static void print_signal_handler(outputStream* st, int sig,
2457 char* buf, size_t buflen);
2458
2459 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2460 st->print_cr("Signal Handlers:");
2461 print_signal_handler(st, SIGSEGV, buf, buflen);
2462 print_signal_handler(st, SIGBUS , buf, buflen);
2463 print_signal_handler(st, SIGFPE , buf, buflen);
2464 print_signal_handler(st, SIGPIPE, buf, buflen);
2465 print_signal_handler(st, SIGXFSZ, buf, buflen);
2466 print_signal_handler(st, SIGILL , buf, buflen);
2467 print_signal_handler(st, SR_signum, buf, buflen);
2468 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2469 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2470 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2471 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2472 #if defined(PPC64)
2473 print_signal_handler(st, SIGTRAP, buf, buflen);
2474 #endif
2475 }
2476
2477 static char saved_jvm_path[MAXPATHLEN] = {0};
2478
2479 // Find the full path to the current module, libjvm.so
2480 void os::jvm_path(char *buf, jint buflen) {
2481 // Error checking.
2482 if (buflen < MAXPATHLEN) {
2483 assert(false, "must use a large-enough buffer");
2484 buf[0] = '\0';
2485 return;
2486 }
2487 // Lazy resolve the path to current module.
2488 if (saved_jvm_path[0] != 0) {
2489 strcpy(buf, saved_jvm_path);
2490 return;
2491 }
2492
2493 char dli_fname[MAXPATHLEN];
2494 bool ret = dll_address_to_library_name(
2495 CAST_FROM_FN_PTR(address, os::jvm_path),
2496 dli_fname, sizeof(dli_fname), NULL);
2497 assert(ret, "cannot locate libjvm");
2498 char *rp = NULL;
2499 if (ret && dli_fname[0] != '\0') {
2500 rp = os::Posix::realpath(dli_fname, buf, buflen);
2501 }
2502 if (rp == NULL) {
2503 return;
2504 }
2505
2506 if (Arguments::sun_java_launcher_is_altjvm()) {
2507 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2508 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2509 // If "/jre/lib/" appears at the right place in the string, then
2510 // assume we are installed in a JDK and we're done. Otherwise, check
2511 // for a JAVA_HOME environment variable and fix up the path so it
2512 // looks like libjvm.so is installed there (append a fake suffix
2513 // hotspot/libjvm.so).
2514 const char *p = buf + strlen(buf) - 1;
2515 for (int count = 0; p > buf && count < 5; ++count) {
2516 for (--p; p > buf && *p != '/'; --p)
2517 /* empty */ ;
2518 }
2519
2520 if (strncmp(p, "/jre/lib/", 9) != 0) {
2521 // Look for JAVA_HOME in the environment.
2522 char* java_home_var = ::getenv("JAVA_HOME");
2523 if (java_home_var != NULL && java_home_var[0] != 0) {
2524 char* jrelib_p;
2525 int len;
2526
2527 // Check the current module name "libjvm.so".
2528 p = strrchr(buf, '/');
2529 if (p == NULL) {
2530 return;
2531 }
2532 assert(strstr(p, "/libjvm") == p, "invalid library name");
2533
2534 rp = os::Posix::realpath(java_home_var, buf, buflen);
2535 if (rp == NULL) {
2536 return;
2537 }
2538
2539 // determine if this is a legacy image or modules image
2540 // modules image doesn't have "jre" subdirectory
2541 len = strlen(buf);
2542 assert(len < buflen, "Ran out of buffer room");
2543 jrelib_p = buf + len;
2544 snprintf(jrelib_p, buflen-len, "/jre/lib");
2545 if (0 != access(buf, F_OK)) {
2546 snprintf(jrelib_p, buflen-len, "/lib");
2547 }
2548
2549 if (0 == access(buf, F_OK)) {
2550 // Use current module name "libjvm.so"
2551 len = strlen(buf);
2552 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2553 } else {
2554 // Go back to path of .so
2555 rp = os::Posix::realpath(dli_fname, buf, buflen);
2556 if (rp == NULL) {
2557 return;
2558 }
2559 }
2560 }
2561 }
2562 }
2563
2564 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2565 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2566 }
2567
2568 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2569 // no prefix required, not even "_"
2570 }
2571
2572 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2573 // no suffix required
2574 }
2575
2576 ////////////////////////////////////////////////////////////////////////////////
2577 // sun.misc.Signal support
2578
2579 static volatile jint sigint_count = 0;
2580
2581 static void UserHandler(int sig, void *siginfo, void *context) {
2582 // 4511530 - sem_post is serialized and handled by the manager thread. When
2583 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2584 // don't want to flood the manager thread with sem_post requests.
2585 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2586 return;
2587 }
2588
2589 // Ctrl-C is pressed during error reporting, likely because the error
2590 // handler fails to abort. Let VM die immediately.
2591 if (sig == SIGINT && VMError::is_error_reported()) {
2592 os::die();
2593 }
2594
2595 os::signal_notify(sig);
2596 }
2597
2598 void* os::user_handler() {
2599 return CAST_FROM_FN_PTR(void*, UserHandler);
2600 }
2601
2602 extern "C" {
2603 typedef void (*sa_handler_t)(int);
2604 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2605 }
2606
2607 void* os::signal(int signal_number, void* handler) {
2608 struct sigaction sigAct, oldSigAct;
2609
2610 sigfillset(&(sigAct.sa_mask));
2611 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2612 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2613
2614 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2615 // -1 means registration failed
2616 return (void *)-1;
2617 }
2618
2619 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2620 }
2621
2622 void os::signal_raise(int signal_number) {
2623 ::raise(signal_number);
2624 }
2625
2626 // The following code is moved from os.cpp for making this
2627 // code platform specific, which it is by its very nature.
2628
2629 // Will be modified when max signal is changed to be dynamic
2630 int os::sigexitnum_pd() {
2631 return NSIG;
2632 }
2633
2634 // a counter for each possible signal value
2635 static volatile jint pending_signals[NSIG+1] = { 0 };
2636
2637 // Linux(POSIX) specific hand shaking semaphore.
2638 static Semaphore* sig_sem = NULL;
2639 static PosixSemaphore sr_semaphore;
2640
2641 static void jdk_misc_signal_init() {
2642 // Initialize signal structures
2643 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2644
2645 // Initialize signal semaphore
2646 sig_sem = new Semaphore();
2647 }
2648
2649 void os::signal_notify(int sig) {
2650 if (sig_sem != NULL) {
2651 Atomic::inc(&pending_signals[sig]);
2652 sig_sem->signal();
2653 } else {
2654 // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2655 // initialization isn't called.
2656 assert(ReduceSignalUsage, "signal semaphore should be created");
2657 }
2658 }
2659
2660 static int check_pending_signals() {
2661 Atomic::store(0, &sigint_count);
2662 for (;;) {
2663 for (int i = 0; i < NSIG + 1; i++) {
2664 jint n = pending_signals[i];
2665 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2666 return i;
2667 }
2668 }
2669 JavaThread *thread = JavaThread::current();
2670 ThreadBlockInVM tbivm(thread);
2671
2672 bool threadIsSuspended;
2673 do {
2674 thread->set_suspend_equivalent();
2675 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2676 sig_sem->wait();
2677
2678 // were we externally suspended while we were waiting?
2679 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2680 if (threadIsSuspended) {
2681 // The semaphore has been incremented, but while we were waiting
2682 // another thread suspended us. We don't want to continue running
2683 // while suspended because that would surprise the thread that
2684 // suspended us.
2685 sig_sem->signal();
2686
2687 thread->java_suspend_self();
2688 }
2689 } while (threadIsSuspended);
2690 }
2691 }
2692
2693 int os::signal_wait() {
2694 return check_pending_signals();
2695 }
2696
2697 ////////////////////////////////////////////////////////////////////////////////
2698 // Virtual Memory
2699
2700 int os::vm_page_size() {
2701 // Seems redundant as all get out
2702 assert(os::Linux::page_size() != -1, "must call os::init");
2703 return os::Linux::page_size();
2704 }
2705
2706 // Solaris allocates memory by pages.
2707 int os::vm_allocation_granularity() {
2708 assert(os::Linux::page_size() != -1, "must call os::init");
2709 return os::Linux::page_size();
2710 }
2711
2712 // Rationale behind this function:
2713 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2714 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2715 // samples for JITted code. Here we create private executable mapping over the code cache
2716 // and then we can use standard (well, almost, as mapping can change) way to provide
2717 // info for the reporting script by storing timestamp and location of symbol
2718 void linux_wrap_code(char* base, size_t size) {
2719 static volatile jint cnt = 0;
2720
2721 if (!UseOprofile) {
2722 return;
2723 }
2724
2725 char buf[PATH_MAX+1];
2726 int num = Atomic::add(1, &cnt);
2727
2728 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2729 os::get_temp_directory(), os::current_process_id(), num);
2730 unlink(buf);
2731
2732 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2733
2734 if (fd != -1) {
2735 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2736 if (rv != (off_t)-1) {
2737 if (::write(fd, "", 1) == 1) {
2738 mmap(base, size,
2739 PROT_READ|PROT_WRITE|PROT_EXEC,
2740 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2741 }
2742 }
2743 ::close(fd);
2744 unlink(buf);
2745 }
2746 }
2747
2748 static bool recoverable_mmap_error(int err) {
2749 // See if the error is one we can let the caller handle. This
2750 // list of errno values comes from JBS-6843484. I can't find a
2751 // Linux man page that documents this specific set of errno
2752 // values so while this list currently matches Solaris, it may
2753 // change as we gain experience with this failure mode.
2754 switch (err) {
2755 case EBADF:
2756 case EINVAL:
2757 case ENOTSUP:
2758 // let the caller deal with these errors
2759 return true;
2760
2761 default:
2762 // Any remaining errors on this OS can cause our reserved mapping
2763 // to be lost. That can cause confusion where different data
2764 // structures think they have the same memory mapped. The worst
2765 // scenario is if both the VM and a library think they have the
2766 // same memory mapped.
2767 return false;
2768 }
2769 }
2770
2771 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2772 int err) {
2773 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2774 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2775 os::strerror(err), err);
2776 }
2777
2778 static void warn_fail_commit_memory(char* addr, size_t size,
2779 size_t alignment_hint, bool exec,
2780 int err) {
2781 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2782 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2783 alignment_hint, exec, os::strerror(err), err);
2784 }
2785
2786 // NOTE: Linux kernel does not really reserve the pages for us.
2787 // All it does is to check if there are enough free pages
2788 // left at the time of mmap(). This could be a potential
2789 // problem.
2790 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2791 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2792 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2793 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2794 if (res != (uintptr_t) MAP_FAILED) {
2795 if (UseNUMAInterleaving) {
2796 numa_make_global(addr, size);
2797 }
2798 return 0;
2799 }
2800
2801 int err = errno; // save errno from mmap() call above
2802
2803 if (!recoverable_mmap_error(err)) {
2804 warn_fail_commit_memory(addr, size, exec, err);
2805 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2806 }
2807
2808 return err;
2809 }
2810
2811 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2812 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2813 }
2814
2815 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2816 const char* mesg) {
2817 assert(mesg != NULL, "mesg must be specified");
2818 int err = os::Linux::commit_memory_impl(addr, size, exec);
2819 if (err != 0) {
2820 // the caller wants all commit errors to exit with the specified mesg:
2821 warn_fail_commit_memory(addr, size, exec, err);
2822 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2823 }
2824 }
2825
2826 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2827 #ifndef MAP_HUGETLB
2828 #define MAP_HUGETLB 0x40000
2829 #endif
2830
2831 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2832 #ifndef MADV_HUGEPAGE
2833 #define MADV_HUGEPAGE 14
2834 #endif
2835
2836 int os::Linux::commit_memory_impl(char* addr, size_t size,
2837 size_t alignment_hint, bool exec) {
2838 int err = os::Linux::commit_memory_impl(addr, size, exec);
2839 if (err == 0) {
2840 realign_memory(addr, size, alignment_hint);
2841 }
2842 return err;
2843 }
2844
2845 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2846 bool exec) {
2847 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2848 }
2849
2850 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2851 size_t alignment_hint, bool exec,
2852 const char* mesg) {
2853 assert(mesg != NULL, "mesg must be specified");
2854 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2855 if (err != 0) {
2856 // the caller wants all commit errors to exit with the specified mesg:
2857 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2858 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2859 }
2860 }
2861
2862 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2863 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2864 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2865 // be supported or the memory may already be backed by huge pages.
2866 ::madvise(addr, bytes, MADV_HUGEPAGE);
2867 }
2868 }
2869
2870 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2871 // This method works by doing an mmap over an existing mmaping and effectively discarding
2872 // the existing pages. However it won't work for SHM-based large pages that cannot be
2873 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2874 // small pages on top of the SHM segment. This method always works for small pages, so we
2875 // allow that in any case.
2876 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2877 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2878 }
2879 }
2880
2881 void os::numa_make_global(char *addr, size_t bytes) {
2882 Linux::numa_interleave_memory(addr, bytes);
2883 }
2884
2885 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2886 // bind policy to MPOL_PREFERRED for the current thread.
2887 #define USE_MPOL_PREFERRED 0
2888
2889 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2890 // To make NUMA and large pages more robust when both enabled, we need to ease
2891 // the requirements on where the memory should be allocated. MPOL_BIND is the
2892 // default policy and it will force memory to be allocated on the specified
2893 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2894 // the specified node, but will not force it. Using this policy will prevent
2895 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2896 // free large pages.
2897 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2898 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2899 }
2900
2901 bool os::numa_topology_changed() { return false; }
2902
2903 size_t os::numa_get_groups_num() {
2904 // Return just the number of nodes in which it's possible to allocate memory
2905 // (in numa terminology, configured nodes).
2906 return Linux::numa_num_configured_nodes();
2907 }
2908
2909 int os::numa_get_group_id() {
2910 int cpu_id = Linux::sched_getcpu();
2911 if (cpu_id != -1) {
2912 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2913 if (lgrp_id != -1) {
2914 return lgrp_id;
2915 }
2916 }
2917 return 0;
2918 }
2919
2920 int os::Linux::get_existing_num_nodes() {
2921 int node;
2922 int highest_node_number = Linux::numa_max_node();
2923 int num_nodes = 0;
2924
2925 // Get the total number of nodes in the system including nodes without memory.
2926 for (node = 0; node <= highest_node_number; node++) {
2927 if (is_node_in_existing_nodes(node)) {
2928 num_nodes++;
2929 }
2930 }
2931 return num_nodes;
2932 }
2933
2934 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2935 int highest_node_number = Linux::numa_max_node();
2936 size_t i = 0;
2937
2938 // Map all node ids in which it is possible to allocate memory. Also nodes are
2939 // not always consecutively available, i.e. available from 0 to the highest
2940 // node number. If the nodes have been bound explicitly using numactl membind,
2941 // then allocate memory from those nodes only.
2942 for (int node = 0; node <= highest_node_number; node++) {
2943 if (Linux::is_node_in_bound_nodes((unsigned int)node)) {
2944 ids[i++] = node;
2945 }
2946 }
2947 return i;
2948 }
2949
2950 bool os::get_page_info(char *start, page_info* info) {
2951 return false;
2952 }
2953
2954 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2955 page_info* page_found) {
2956 return end;
2957 }
2958
2959
2960 int os::Linux::sched_getcpu_syscall(void) {
2961 unsigned int cpu = 0;
2962 int retval = -1;
2963
2964 #if defined(IA32)
2965 #ifndef SYS_getcpu
2966 #define SYS_getcpu 318
2967 #endif
2968 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2969 #elif defined(AMD64)
2970 // Unfortunately we have to bring all these macros here from vsyscall.h
2971 // to be able to compile on old linuxes.
2972 #define __NR_vgetcpu 2
2973 #define VSYSCALL_START (-10UL << 20)
2974 #define VSYSCALL_SIZE 1024
2975 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2976 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2977 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2978 retval = vgetcpu(&cpu, NULL, NULL);
2979 #endif
2980
2981 return (retval == -1) ? retval : cpu;
2982 }
2983
2984 void os::Linux::sched_getcpu_init() {
2985 // sched_getcpu() should be in libc.
2986 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2987 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2988
2989 // If it's not, try a direct syscall.
2990 if (sched_getcpu() == -1) {
2991 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2992 (void*)&sched_getcpu_syscall));
2993 }
2994
2995 if (sched_getcpu() == -1) {
2996 vm_exit_during_initialization("getcpu(2) system call not supported by kernel");
2997 }
2998 }
2999
3000 // Something to do with the numa-aware allocator needs these symbols
3001 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
3002 extern "C" JNIEXPORT void numa_error(char *where) { }
3003
3004 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
3005 // load symbol from base version instead.
3006 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
3007 void *f = dlvsym(handle, name, "libnuma_1.1");
3008 if (f == NULL) {
3009 f = dlsym(handle, name);
3010 }
3011 return f;
3012 }
3013
3014 // Handle request to load libnuma symbol version 1.2 (API v2) only.
3015 // Return NULL if the symbol is not defined in this particular version.
3016 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
3017 return dlvsym(handle, name, "libnuma_1.2");
3018 }
3019
3020 bool os::Linux::libnuma_init() {
3021 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
3022 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
3023 if (handle != NULL) {
3024 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
3025 libnuma_dlsym(handle, "numa_node_to_cpus")));
3026 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
3027 libnuma_dlsym(handle, "numa_max_node")));
3028 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
3029 libnuma_dlsym(handle, "numa_num_configured_nodes")));
3030 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
3031 libnuma_dlsym(handle, "numa_available")));
3032 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
3033 libnuma_dlsym(handle, "numa_tonode_memory")));
3034 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
3035 libnuma_dlsym(handle, "numa_interleave_memory")));
3036 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
3037 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
3038 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
3039 libnuma_dlsym(handle, "numa_set_bind_policy")));
3040 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
3041 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
3042 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
3043 libnuma_dlsym(handle, "numa_distance")));
3044 set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t,
3045 libnuma_v2_dlsym(handle, "numa_get_membind")));
3046 set_numa_get_interleave_mask(CAST_TO_FN_PTR(numa_get_interleave_mask_func_t,
3047 libnuma_v2_dlsym(handle, "numa_get_interleave_mask")));
3048
3049 if (numa_available() != -1) {
3050 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
3051 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
3052 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
3053 set_numa_interleave_bitmask(_numa_get_interleave_mask());
3054 set_numa_membind_bitmask(_numa_get_membind());
3055 // Create an index -> node mapping, since nodes are not always consecutive
3056 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3057 rebuild_nindex_to_node_map();
3058 // Create a cpu -> node mapping
3059 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3060 rebuild_cpu_to_node_map();
3061 return true;
3062 }
3063 }
3064 }
3065 return false;
3066 }
3067
3068 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
3069 // Creating guard page is very expensive. Java thread has HotSpot
3070 // guard pages, only enable glibc guard page for non-Java threads.
3071 // (Remember: compiler thread is a Java thread, too!)
3072 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
3073 }
3074
3075 void os::Linux::rebuild_nindex_to_node_map() {
3076 int highest_node_number = Linux::numa_max_node();
3077
3078 nindex_to_node()->clear();
3079 for (int node = 0; node <= highest_node_number; node++) {
3080 if (Linux::is_node_in_existing_nodes(node)) {
3081 nindex_to_node()->append(node);
3082 }
3083 }
3084 }
3085
3086 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
3087 // The table is later used in get_node_by_cpu().
3088 void os::Linux::rebuild_cpu_to_node_map() {
3089 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
3090 // in libnuma (possible values are starting from 16,
3091 // and continuing up with every other power of 2, but less
3092 // than the maximum number of CPUs supported by kernel), and
3093 // is a subject to change (in libnuma version 2 the requirements
3094 // are more reasonable) we'll just hardcode the number they use
3095 // in the library.
3096 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
3097
3098 size_t cpu_num = processor_count();
3099 size_t cpu_map_size = NCPUS / BitsPerCLong;
3100 size_t cpu_map_valid_size =
3101 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
3102
3103 cpu_to_node()->clear();
3104 cpu_to_node()->at_grow(cpu_num - 1);
3105
3106 size_t node_num = get_existing_num_nodes();
3107
3108 int distance = 0;
3109 int closest_distance = INT_MAX;
3110 int closest_node = 0;
3111 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
3112 for (size_t i = 0; i < node_num; i++) {
3113 // Check if node is configured (not a memory-less node). If it is not, find
3114 // the closest configured node. Check also if node is bound, i.e. it's allowed
3115 // to allocate memory from the node. If it's not allowed, map cpus in that node
3116 // to the closest node from which memory allocation is allowed.
3117 if (!is_node_in_configured_nodes(nindex_to_node()->at(i)) ||
3118 !is_node_in_bound_nodes(nindex_to_node()->at(i))) {
3119 closest_distance = INT_MAX;
3120 // Check distance from all remaining nodes in the system. Ignore distance
3121 // from itself, from another non-configured node, and from another non-bound
3122 // node.
3123 for (size_t m = 0; m < node_num; m++) {
3124 if (m != i &&
3125 is_node_in_configured_nodes(nindex_to_node()->at(m)) &&
3126 is_node_in_bound_nodes(nindex_to_node()->at(m))) {
3127 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3128 // If a closest node is found, update. There is always at least one
3129 // configured and bound node in the system so there is always at least
3130 // one node close.
3131 if (distance != 0 && distance < closest_distance) {
3132 closest_distance = distance;
3133 closest_node = nindex_to_node()->at(m);
3134 }
3135 }
3136 }
3137 } else {
3138 // Current node is already a configured node.
3139 closest_node = nindex_to_node()->at(i);
3140 }
3141
3142 // Get cpus from the original node and map them to the closest node. If node
3143 // is a configured node (not a memory-less node), then original node and
3144 // closest node are the same.
3145 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3146 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3147 if (cpu_map[j] != 0) {
3148 for (size_t k = 0; k < BitsPerCLong; k++) {
3149 if (cpu_map[j] & (1UL << k)) {
3150 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3151 }
3152 }
3153 }
3154 }
3155 }
3156 }
3157 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3158 }
3159
3160 int os::Linux::get_node_by_cpu(int cpu_id) {
3161 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3162 return cpu_to_node()->at(cpu_id);
3163 }
3164 return -1;
3165 }
3166
3167 GrowableArray<int>* os::Linux::_cpu_to_node;
3168 GrowableArray<int>* os::Linux::_nindex_to_node;
3169 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3170 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3171 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3172 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3173 os::Linux::numa_available_func_t os::Linux::_numa_available;
3174 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3175 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3176 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3177 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3178 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3179 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3180 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind;
3181 os::Linux::numa_get_interleave_mask_func_t os::Linux::_numa_get_interleave_mask;
3182 os::Linux::NumaAllocationPolicy os::Linux::_current_numa_policy;
3183 unsigned long* os::Linux::_numa_all_nodes;
3184 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3185 struct bitmask* os::Linux::_numa_nodes_ptr;
3186 struct bitmask* os::Linux::_numa_interleave_bitmask;
3187 struct bitmask* os::Linux::_numa_membind_bitmask;
3188
3189 bool os::pd_uncommit_memory(char* addr, size_t size) {
3190 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3191 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3192 return res != (uintptr_t) MAP_FAILED;
3193 }
3194
3195 static address get_stack_commited_bottom(address bottom, size_t size) {
3196 address nbot = bottom;
3197 address ntop = bottom + size;
3198
3199 size_t page_sz = os::vm_page_size();
3200 unsigned pages = size / page_sz;
3201
3202 unsigned char vec[1];
3203 unsigned imin = 1, imax = pages + 1, imid;
3204 int mincore_return_value = 0;
3205
3206 assert(imin <= imax, "Unexpected page size");
3207
3208 while (imin < imax) {
3209 imid = (imax + imin) / 2;
3210 nbot = ntop - (imid * page_sz);
3211
3212 // Use a trick with mincore to check whether the page is mapped or not.
3213 // mincore sets vec to 1 if page resides in memory and to 0 if page
3214 // is swapped output but if page we are asking for is unmapped
3215 // it returns -1,ENOMEM
3216 mincore_return_value = mincore(nbot, page_sz, vec);
3217
3218 if (mincore_return_value == -1) {
3219 // Page is not mapped go up
3220 // to find first mapped page
3221 if (errno != EAGAIN) {
3222 assert(errno == ENOMEM, "Unexpected mincore errno");
3223 imax = imid;
3224 }
3225 } else {
3226 // Page is mapped go down
3227 // to find first not mapped page
3228 imin = imid + 1;
3229 }
3230 }
3231
3232 nbot = nbot + page_sz;
3233
3234 // Adjust stack bottom one page up if last checked page is not mapped
3235 if (mincore_return_value == -1) {
3236 nbot = nbot + page_sz;
3237 }
3238
3239 return nbot;
3240 }
3241
3242 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3243 int mincore_return_value;
3244 const size_t stripe = 1024; // query this many pages each time
3245 unsigned char vec[stripe + 1];
3246 // set a guard
3247 vec[stripe] = 'X';
3248
3249 const size_t page_sz = os::vm_page_size();
3250 size_t pages = size / page_sz;
3251
3252 assert(is_aligned(start, page_sz), "Start address must be page aligned");
3253 assert(is_aligned(size, page_sz), "Size must be page aligned");
3254
3255 committed_start = NULL;
3256
3257 int loops = (pages + stripe - 1) / stripe;
3258 int committed_pages = 0;
3259 address loop_base = start;
3260 bool found_range = false;
3261
3262 for (int index = 0; index < loops && !found_range; index ++) {
3263 assert(pages > 0, "Nothing to do");
3264 int pages_to_query = (pages >= stripe) ? stripe : pages;
3265 pages -= pages_to_query;
3266
3267 // Get stable read
3268 while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3269
3270 // During shutdown, some memory goes away without properly notifying NMT,
3271 // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3272 // Bailout and return as not committed for now.
3273 if (mincore_return_value == -1 && errno == ENOMEM) {
3274 return false;
3275 }
3276
3277 assert(vec[stripe] == 'X', "overflow guard");
3278 assert(mincore_return_value == 0, "Range must be valid");
3279 // Process this stripe
3280 for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3281 if ((vec[vecIdx] & 0x01) == 0) { // not committed
3282 // End of current contiguous region
3283 if (committed_start != NULL) {
3284 found_range = true;
3285 break;
3286 }
3287 } else { // committed
3288 // Start of region
3289 if (committed_start == NULL) {
3290 committed_start = loop_base + page_sz * vecIdx;
3291 }
3292 committed_pages ++;
3293 }
3294 }
3295
3296 loop_base += pages_to_query * page_sz;
3297 }
3298
3299 if (committed_start != NULL) {
3300 assert(committed_pages > 0, "Must have committed region");
3301 assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3302 assert(committed_start >= start && committed_start < start + size, "Out of range");
3303 committed_size = page_sz * committed_pages;
3304 return true;
3305 } else {
3306 assert(committed_pages == 0, "Should not have committed region");
3307 return false;
3308 }
3309 }
3310
3311
3312 // Linux uses a growable mapping for the stack, and if the mapping for
3313 // the stack guard pages is not removed when we detach a thread the
3314 // stack cannot grow beyond the pages where the stack guard was
3315 // mapped. If at some point later in the process the stack expands to
3316 // that point, the Linux kernel cannot expand the stack any further
3317 // because the guard pages are in the way, and a segfault occurs.
3318 //
3319 // However, it's essential not to split the stack region by unmapping
3320 // a region (leaving a hole) that's already part of the stack mapping,
3321 // so if the stack mapping has already grown beyond the guard pages at
3322 // the time we create them, we have to truncate the stack mapping.
3323 // So, we need to know the extent of the stack mapping when
3324 // create_stack_guard_pages() is called.
3325
3326 // We only need this for stacks that are growable: at the time of
3327 // writing thread stacks don't use growable mappings (i.e. those
3328 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3329 // only applies to the main thread.
3330
3331 // If the (growable) stack mapping already extends beyond the point
3332 // where we're going to put our guard pages, truncate the mapping at
3333 // that point by munmap()ping it. This ensures that when we later
3334 // munmap() the guard pages we don't leave a hole in the stack
3335 // mapping. This only affects the main/primordial thread
3336
3337 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3338 if (os::is_primordial_thread()) {
3339 // As we manually grow stack up to bottom inside create_attached_thread(),
3340 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3341 // we don't need to do anything special.
3342 // Check it first, before calling heavy function.
3343 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3344 unsigned char vec[1];
3345
3346 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3347 // Fallback to slow path on all errors, including EAGAIN
3348 stack_extent = (uintptr_t) get_stack_commited_bottom(
3349 os::Linux::initial_thread_stack_bottom(),
3350 (size_t)addr - stack_extent);
3351 }
3352
3353 if (stack_extent < (uintptr_t)addr) {
3354 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3355 }
3356 }
3357
3358 return os::commit_memory(addr, size, !ExecMem);
3359 }
3360
3361 // If this is a growable mapping, remove the guard pages entirely by
3362 // munmap()ping them. If not, just call uncommit_memory(). This only
3363 // affects the main/primordial thread, but guard against future OS changes.
3364 // It's safe to always unmap guard pages for primordial thread because we
3365 // always place it right after end of the mapped region.
3366
3367 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3368 uintptr_t stack_extent, stack_base;
3369
3370 if (os::is_primordial_thread()) {
3371 return ::munmap(addr, size) == 0;
3372 }
3373
3374 return os::uncommit_memory(addr, size);
3375 }
3376
3377 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3378 // at 'requested_addr'. If there are existing memory mappings at the same
3379 // location, however, they will be overwritten. If 'fixed' is false,
3380 // 'requested_addr' is only treated as a hint, the return value may or
3381 // may not start from the requested address. Unlike Linux mmap(), this
3382 // function returns NULL to indicate failure.
3383 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3384 char * addr;
3385 int flags;
3386
3387 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3388 if (fixed) {
3389 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3390 flags |= MAP_FIXED;
3391 }
3392
3393 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3394 // touch an uncommitted page. Otherwise, the read/write might
3395 // succeed if we have enough swap space to back the physical page.
3396 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3397 flags, -1, 0);
3398
3399 return addr == MAP_FAILED ? NULL : addr;
3400 }
3401
3402 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3403 // (req_addr != NULL) or with a given alignment.
3404 // - bytes shall be a multiple of alignment.
3405 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3406 // - alignment sets the alignment at which memory shall be allocated.
3407 // It must be a multiple of allocation granularity.
3408 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3409 // req_addr or NULL.
3410 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3411
3412 size_t extra_size = bytes;
3413 if (req_addr == NULL && alignment > 0) {
3414 extra_size += alignment;
3415 }
3416
3417 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3418 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3419 -1, 0);
3420 if (start == MAP_FAILED) {
3421 start = NULL;
3422 } else {
3423 if (req_addr != NULL) {
3424 if (start != req_addr) {
3425 ::munmap(start, extra_size);
3426 start = NULL;
3427 }
3428 } else {
3429 char* const start_aligned = align_up(start, alignment);
3430 char* const end_aligned = start_aligned + bytes;
3431 char* const end = start + extra_size;
3432 if (start_aligned > start) {
3433 ::munmap(start, start_aligned - start);
3434 }
3435 if (end_aligned < end) {
3436 ::munmap(end_aligned, end - end_aligned);
3437 }
3438 start = start_aligned;
3439 }
3440 }
3441 return start;
3442 }
3443
3444 static int anon_munmap(char * addr, size_t size) {
3445 return ::munmap(addr, size) == 0;
3446 }
3447
3448 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3449 size_t alignment_hint) {
3450 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3451 }
3452
3453 bool os::pd_release_memory(char* addr, size_t size) {
3454 return anon_munmap(addr, size);
3455 }
3456
3457 static bool linux_mprotect(char* addr, size_t size, int prot) {
3458 // Linux wants the mprotect address argument to be page aligned.
3459 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3460
3461 // According to SUSv3, mprotect() should only be used with mappings
3462 // established by mmap(), and mmap() always maps whole pages. Unaligned
3463 // 'addr' likely indicates problem in the VM (e.g. trying to change
3464 // protection of malloc'ed or statically allocated memory). Check the
3465 // caller if you hit this assert.
3466 assert(addr == bottom, "sanity check");
3467
3468 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3469 Events::log(NULL, "Protecting memory [" INTPTR_FORMAT "," INTPTR_FORMAT "] with protection modes %x", p2i(bottom), p2i(bottom+size), prot);
3470 return ::mprotect(bottom, size, prot) == 0;
3471 }
3472
3473 // Set protections specified
3474 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3475 bool is_committed) {
3476 unsigned int p = 0;
3477 switch (prot) {
3478 case MEM_PROT_NONE: p = PROT_NONE; break;
3479 case MEM_PROT_READ: p = PROT_READ; break;
3480 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3481 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3482 default:
3483 ShouldNotReachHere();
3484 }
3485 // is_committed is unused.
3486 return linux_mprotect(addr, bytes, p);
3487 }
3488
3489 bool os::guard_memory(char* addr, size_t size) {
3490 return linux_mprotect(addr, size, PROT_NONE);
3491 }
3492
3493 bool os::unguard_memory(char* addr, size_t size) {
3494 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3495 }
3496
3497 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3498 size_t page_size) {
3499 bool result = false;
3500 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3501 MAP_ANONYMOUS|MAP_PRIVATE,
3502 -1, 0);
3503 if (p != MAP_FAILED) {
3504 void *aligned_p = align_up(p, page_size);
3505
3506 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3507
3508 munmap(p, page_size * 2);
3509 }
3510
3511 if (warn && !result) {
3512 warning("TransparentHugePages is not supported by the operating system.");
3513 }
3514
3515 return result;
3516 }
3517
3518 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3519 bool result = false;
3520 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3521 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3522 -1, 0);
3523
3524 if (p != MAP_FAILED) {
3525 // We don't know if this really is a huge page or not.
3526 FILE *fp = fopen("/proc/self/maps", "r");
3527 if (fp) {
3528 while (!feof(fp)) {
3529 char chars[257];
3530 long x = 0;
3531 if (fgets(chars, sizeof(chars), fp)) {
3532 if (sscanf(chars, "%lx-%*x", &x) == 1
3533 && x == (long)p) {
3534 if (strstr (chars, "hugepage")) {
3535 result = true;
3536 break;
3537 }
3538 }
3539 }
3540 }
3541 fclose(fp);
3542 }
3543 munmap(p, page_size);
3544 }
3545
3546 if (warn && !result) {
3547 warning("HugeTLBFS is not supported by the operating system.");
3548 }
3549
3550 return result;
3551 }
3552
3553 // From the coredump_filter documentation:
3554 //
3555 // - (bit 0) anonymous private memory
3556 // - (bit 1) anonymous shared memory
3557 // - (bit 2) file-backed private memory
3558 // - (bit 3) file-backed shared memory
3559 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3560 // effective only if the bit 2 is cleared)
3561 // - (bit 5) hugetlb private memory
3562 // - (bit 6) hugetlb shared memory
3563 // - (bit 7) dax private memory
3564 // - (bit 8) dax shared memory
3565 //
3566 static void set_coredump_filter(CoredumpFilterBit bit) {
3567 FILE *f;
3568 long cdm;
3569
3570 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3571 return;
3572 }
3573
3574 if (fscanf(f, "%lx", &cdm) != 1) {
3575 fclose(f);
3576 return;
3577 }
3578
3579 long saved_cdm = cdm;
3580 rewind(f);
3581 cdm |= bit;
3582
3583 if (cdm != saved_cdm) {
3584 fprintf(f, "%#lx", cdm);
3585 }
3586
3587 fclose(f);
3588 }
3589
3590 // Large page support
3591
3592 static size_t _large_page_size = 0;
3593
3594 size_t os::Linux::find_large_page_size() {
3595 size_t large_page_size = 0;
3596
3597 // large_page_size on Linux is used to round up heap size. x86 uses either
3598 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3599 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3600 // page as large as 256M.
3601 //
3602 // Here we try to figure out page size by parsing /proc/meminfo and looking
3603 // for a line with the following format:
3604 // Hugepagesize: 2048 kB
3605 //
3606 // If we can't determine the value (e.g. /proc is not mounted, or the text
3607 // format has been changed), we'll use the largest page size supported by
3608 // the processor.
3609
3610 #ifndef ZERO
3611 large_page_size =
3612 AARCH64_ONLY(2 * M)
3613 AMD64_ONLY(2 * M)
3614 ARM32_ONLY(2 * M)
3615 IA32_ONLY(4 * M)
3616 IA64_ONLY(256 * M)
3617 PPC_ONLY(4 * M)
3618 S390_ONLY(1 * M)
3619 SPARC_ONLY(4 * M);
3620 #endif // ZERO
3621
3622 FILE *fp = fopen("/proc/meminfo", "r");
3623 if (fp) {
3624 while (!feof(fp)) {
3625 int x = 0;
3626 char buf[16];
3627 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3628 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3629 large_page_size = x * K;
3630 break;
3631 }
3632 } else {
3633 // skip to next line
3634 for (;;) {
3635 int ch = fgetc(fp);
3636 if (ch == EOF || ch == (int)'\n') break;
3637 }
3638 }
3639 }
3640 fclose(fp);
3641 }
3642
3643 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3644 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3645 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3646 proper_unit_for_byte_size(large_page_size));
3647 }
3648
3649 return large_page_size;
3650 }
3651
3652 size_t os::Linux::setup_large_page_size() {
3653 _large_page_size = Linux::find_large_page_size();
3654 const size_t default_page_size = (size_t)Linux::page_size();
3655 if (_large_page_size > default_page_size) {
3656 _page_sizes[0] = _large_page_size;
3657 _page_sizes[1] = default_page_size;
3658 _page_sizes[2] = 0;
3659 }
3660
3661 return _large_page_size;
3662 }
3663
3664 bool os::Linux::setup_large_page_type(size_t page_size) {
3665 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3666 FLAG_IS_DEFAULT(UseSHM) &&
3667 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3668
3669 // The type of large pages has not been specified by the user.
3670
3671 // Try UseHugeTLBFS and then UseSHM.
3672 UseHugeTLBFS = UseSHM = true;
3673
3674 // Don't try UseTransparentHugePages since there are known
3675 // performance issues with it turned on. This might change in the future.
3676 UseTransparentHugePages = false;
3677 }
3678
3679 if (UseTransparentHugePages) {
3680 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3681 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3682 UseHugeTLBFS = false;
3683 UseSHM = false;
3684 return true;
3685 }
3686 UseTransparentHugePages = false;
3687 }
3688
3689 if (UseHugeTLBFS) {
3690 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3691 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3692 UseSHM = false;
3693 return true;
3694 }
3695 UseHugeTLBFS = false;
3696 }
3697
3698 return UseSHM;
3699 }
3700
3701 void os::large_page_init() {
3702 if (!UseLargePages &&
3703 !UseTransparentHugePages &&
3704 !UseHugeTLBFS &&
3705 !UseSHM) {
3706 // Not using large pages.
3707 return;
3708 }
3709
3710 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3711 // The user explicitly turned off large pages.
3712 // Ignore the rest of the large pages flags.
3713 UseTransparentHugePages = false;
3714 UseHugeTLBFS = false;
3715 UseSHM = false;
3716 return;
3717 }
3718
3719 size_t large_page_size = Linux::setup_large_page_size();
3720 UseLargePages = Linux::setup_large_page_type(large_page_size);
3721
3722 set_coredump_filter(LARGEPAGES_BIT);
3723 }
3724
3725 #ifndef SHM_HUGETLB
3726 #define SHM_HUGETLB 04000
3727 #endif
3728
3729 #define shm_warning_format(format, ...) \
3730 do { \
3731 if (UseLargePages && \
3732 (!FLAG_IS_DEFAULT(UseLargePages) || \
3733 !FLAG_IS_DEFAULT(UseSHM) || \
3734 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3735 warning(format, __VA_ARGS__); \
3736 } \
3737 } while (0)
3738
3739 #define shm_warning(str) shm_warning_format("%s", str)
3740
3741 #define shm_warning_with_errno(str) \
3742 do { \
3743 int err = errno; \
3744 shm_warning_format(str " (error = %d)", err); \
3745 } while (0)
3746
3747 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3748 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3749
3750 if (!is_aligned(alignment, SHMLBA)) {
3751 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3752 return NULL;
3753 }
3754
3755 // To ensure that we get 'alignment' aligned memory from shmat,
3756 // we pre-reserve aligned virtual memory and then attach to that.
3757
3758 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3759 if (pre_reserved_addr == NULL) {
3760 // Couldn't pre-reserve aligned memory.
3761 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3762 return NULL;
3763 }
3764
3765 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3766 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3767
3768 if ((intptr_t)addr == -1) {
3769 int err = errno;
3770 shm_warning_with_errno("Failed to attach shared memory.");
3771
3772 assert(err != EACCES, "Unexpected error");
3773 assert(err != EIDRM, "Unexpected error");
3774 assert(err != EINVAL, "Unexpected error");
3775
3776 // Since we don't know if the kernel unmapped the pre-reserved memory area
3777 // we can't unmap it, since that would potentially unmap memory that was
3778 // mapped from other threads.
3779 return NULL;
3780 }
3781
3782 return addr;
3783 }
3784
3785 static char* shmat_at_address(int shmid, char* req_addr) {
3786 if (!is_aligned(req_addr, SHMLBA)) {
3787 assert(false, "Requested address needs to be SHMLBA aligned");
3788 return NULL;
3789 }
3790
3791 char* addr = (char*)shmat(shmid, req_addr, 0);
3792
3793 if ((intptr_t)addr == -1) {
3794 shm_warning_with_errno("Failed to attach shared memory.");
3795 return NULL;
3796 }
3797
3798 return addr;
3799 }
3800
3801 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3802 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3803 if (req_addr != NULL) {
3804 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3805 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3806 return shmat_at_address(shmid, req_addr);
3807 }
3808
3809 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3810 // return large page size aligned memory addresses when req_addr == NULL.
3811 // However, if the alignment is larger than the large page size, we have
3812 // to manually ensure that the memory returned is 'alignment' aligned.
3813 if (alignment > os::large_page_size()) {
3814 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3815 return shmat_with_alignment(shmid, bytes, alignment);
3816 } else {
3817 return shmat_at_address(shmid, NULL);
3818 }
3819 }
3820
3821 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3822 char* req_addr, bool exec) {
3823 // "exec" is passed in but not used. Creating the shared image for
3824 // the code cache doesn't have an SHM_X executable permission to check.
3825 assert(UseLargePages && UseSHM, "only for SHM large pages");
3826 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3827 assert(is_aligned(req_addr, alignment), "Unaligned address");
3828
3829 if (!is_aligned(bytes, os::large_page_size())) {
3830 return NULL; // Fallback to small pages.
3831 }
3832
3833 // Create a large shared memory region to attach to based on size.
3834 // Currently, size is the total size of the heap.
3835 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3836 if (shmid == -1) {
3837 // Possible reasons for shmget failure:
3838 // 1. shmmax is too small for Java heap.
3839 // > check shmmax value: cat /proc/sys/kernel/shmmax
3840 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3841 // 2. not enough large page memory.
3842 // > check available large pages: cat /proc/meminfo
3843 // > increase amount of large pages:
3844 // echo new_value > /proc/sys/vm/nr_hugepages
3845 // Note 1: different Linux may use different name for this property,
3846 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3847 // Note 2: it's possible there's enough physical memory available but
3848 // they are so fragmented after a long run that they can't
3849 // coalesce into large pages. Try to reserve large pages when
3850 // the system is still "fresh".
3851 shm_warning_with_errno("Failed to reserve shared memory.");
3852 return NULL;
3853 }
3854
3855 // Attach to the region.
3856 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3857
3858 // Remove shmid. If shmat() is successful, the actual shared memory segment
3859 // will be deleted when it's detached by shmdt() or when the process
3860 // terminates. If shmat() is not successful this will remove the shared
3861 // segment immediately.
3862 shmctl(shmid, IPC_RMID, NULL);
3863
3864 return addr;
3865 }
3866
3867 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3868 int error) {
3869 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3870
3871 bool warn_on_failure = UseLargePages &&
3872 (!FLAG_IS_DEFAULT(UseLargePages) ||
3873 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3874 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3875
3876 if (warn_on_failure) {
3877 char msg[128];
3878 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3879 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3880 warning("%s", msg);
3881 }
3882 }
3883
3884 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3885 char* req_addr,
3886 bool exec) {
3887 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3888 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3889 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3890
3891 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3892 char* addr = (char*)::mmap(req_addr, bytes, prot,
3893 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3894 -1, 0);
3895
3896 if (addr == MAP_FAILED) {
3897 warn_on_large_pages_failure(req_addr, bytes, errno);
3898 return NULL;
3899 }
3900
3901 assert(is_aligned(addr, os::large_page_size()), "Must be");
3902
3903 return addr;
3904 }
3905
3906 // Reserve memory using mmap(MAP_HUGETLB).
3907 // - bytes shall be a multiple of alignment.
3908 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3909 // - alignment sets the alignment at which memory shall be allocated.
3910 // It must be a multiple of allocation granularity.
3911 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3912 // req_addr or NULL.
3913 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3914 size_t alignment,
3915 char* req_addr,
3916 bool exec) {
3917 size_t large_page_size = os::large_page_size();
3918 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3919
3920 assert(is_aligned(req_addr, alignment), "Must be");
3921 assert(is_aligned(bytes, alignment), "Must be");
3922
3923 // First reserve - but not commit - the address range in small pages.
3924 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3925
3926 if (start == NULL) {
3927 return NULL;
3928 }
3929
3930 assert(is_aligned(start, alignment), "Must be");
3931
3932 char* end = start + bytes;
3933
3934 // Find the regions of the allocated chunk that can be promoted to large pages.
3935 char* lp_start = align_up(start, large_page_size);
3936 char* lp_end = align_down(end, large_page_size);
3937
3938 size_t lp_bytes = lp_end - lp_start;
3939
3940 assert(is_aligned(lp_bytes, large_page_size), "Must be");
3941
3942 if (lp_bytes == 0) {
3943 // The mapped region doesn't even span the start and the end of a large page.
3944 // Fall back to allocate a non-special area.
3945 ::munmap(start, end - start);
3946 return NULL;
3947 }
3948
3949 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3950
3951 void* result;
3952
3953 // Commit small-paged leading area.
3954 if (start != lp_start) {
3955 result = ::mmap(start, lp_start - start, prot,
3956 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3957 -1, 0);
3958 if (result == MAP_FAILED) {
3959 ::munmap(lp_start, end - lp_start);
3960 return NULL;
3961 }
3962 }
3963
3964 // Commit large-paged area.
3965 result = ::mmap(lp_start, lp_bytes, prot,
3966 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3967 -1, 0);
3968 if (result == MAP_FAILED) {
3969 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3970 // If the mmap above fails, the large pages region will be unmapped and we
3971 // have regions before and after with small pages. Release these regions.
3972 //
3973 // | mapped | unmapped | mapped |
3974 // ^ ^ ^ ^
3975 // start lp_start lp_end end
3976 //
3977 ::munmap(start, lp_start - start);
3978 ::munmap(lp_end, end - lp_end);
3979 return NULL;
3980 }
3981
3982 // Commit small-paged trailing area.
3983 if (lp_end != end) {
3984 result = ::mmap(lp_end, end - lp_end, prot,
3985 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3986 -1, 0);
3987 if (result == MAP_FAILED) {
3988 ::munmap(start, lp_end - start);
3989 return NULL;
3990 }
3991 }
3992
3993 return start;
3994 }
3995
3996 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3997 size_t alignment,
3998 char* req_addr,
3999 bool exec) {
4000 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
4001 assert(is_aligned(req_addr, alignment), "Must be");
4002 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
4003 assert(is_power_of_2(os::large_page_size()), "Must be");
4004 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
4005
4006 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
4007 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
4008 } else {
4009 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
4010 }
4011 }
4012
4013 char* os::reserve_memory_special(size_t bytes, size_t alignment,
4014 char* req_addr, bool exec) {
4015 assert(UseLargePages, "only for large pages");
4016
4017 char* addr;
4018 if (UseSHM) {
4019 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
4020 } else {
4021 assert(UseHugeTLBFS, "must be");
4022 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
4023 }
4024
4025 if (addr != NULL) {
4026 if (UseNUMAInterleaving) {
4027 numa_make_global(addr, bytes);
4028 }
4029
4030 // The memory is committed
4031 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
4032 }
4033
4034 return addr;
4035 }
4036
4037 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
4038 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
4039 return shmdt(base) == 0;
4040 }
4041
4042 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
4043 return pd_release_memory(base, bytes);
4044 }
4045
4046 bool os::release_memory_special(char* base, size_t bytes) {
4047 bool res;
4048 if (MemTracker::tracking_level() > NMT_minimal) {
4049 Tracker tkr(Tracker::release);
4050 res = os::Linux::release_memory_special_impl(base, bytes);
4051 if (res) {
4052 tkr.record((address)base, bytes);
4053 }
4054
4055 } else {
4056 res = os::Linux::release_memory_special_impl(base, bytes);
4057 }
4058 return res;
4059 }
4060
4061 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
4062 assert(UseLargePages, "only for large pages");
4063 bool res;
4064
4065 if (UseSHM) {
4066 res = os::Linux::release_memory_special_shm(base, bytes);
4067 } else {
4068 assert(UseHugeTLBFS, "must be");
4069 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
4070 }
4071 return res;
4072 }
4073
4074 size_t os::large_page_size() {
4075 return _large_page_size;
4076 }
4077
4078 // With SysV SHM the entire memory region must be allocated as shared
4079 // memory.
4080 // HugeTLBFS allows application to commit large page memory on demand.
4081 // However, when committing memory with HugeTLBFS fails, the region
4082 // that was supposed to be committed will lose the old reservation
4083 // and allow other threads to steal that memory region. Because of this
4084 // behavior we can't commit HugeTLBFS memory.
4085 bool os::can_commit_large_page_memory() {
4086 return UseTransparentHugePages;
4087 }
4088
4089 bool os::can_execute_large_page_memory() {
4090 return UseTransparentHugePages || UseHugeTLBFS;
4091 }
4092
4093 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
4094 assert(file_desc >= 0, "file_desc is not valid");
4095 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
4096 if (result != NULL) {
4097 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
4098 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
4099 }
4100 }
4101 return result;
4102 }
4103
4104 // Reserve memory at an arbitrary address, only if that area is
4105 // available (and not reserved for something else).
4106
4107 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
4108 // Assert only that the size is a multiple of the page size, since
4109 // that's all that mmap requires, and since that's all we really know
4110 // about at this low abstraction level. If we need higher alignment,
4111 // we can either pass an alignment to this method or verify alignment
4112 // in one of the methods further up the call chain. See bug 5044738.
4113 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
4114
4115 // Repeatedly allocate blocks until the block is allocated at the
4116 // right spot.
4117
4118 // Linux mmap allows caller to pass an address as hint; give it a try first,
4119 // if kernel honors the hint then we can return immediately.
4120 char * addr = anon_mmap(requested_addr, bytes, false);
4121 if (addr == requested_addr) {
4122 return requested_addr;
4123 }
4124
4125 if (addr != NULL) {
4126 // mmap() is successful but it fails to reserve at the requested address
4127 anon_munmap(addr, bytes);
4128 }
4129
4130 return NULL;
4131 }
4132
4133 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4134 void os::infinite_sleep() {
4135 while (true) { // sleep forever ...
4136 ::sleep(100); // ... 100 seconds at a time
4137 }
4138 }
4139
4140 // Used to convert frequent JVM_Yield() to nops
4141 bool os::dont_yield() {
4142 return DontYieldALot;
4143 }
4144
4145 // Linux CFS scheduler (since 2.6.23) does not guarantee sched_yield(2) will
4146 // actually give up the CPU. Since skip buddy (v2.6.28):
4147 //
4148 // * Sets the yielding task as skip buddy for current CPU's run queue.
4149 // * Picks next from run queue, if empty, picks a skip buddy (can be the yielding task).
4150 // * Clears skip buddies for this run queue (yielding task no longer a skip buddy).
4151 //
4152 // An alternative is calling os::naked_short_nanosleep with a small number to avoid
4153 // getting re-scheduled immediately.
4154 //
4155 void os::naked_yield() {
4156 sched_yield();
4157 }
4158
4159 ////////////////////////////////////////////////////////////////////////////////
4160 // thread priority support
4161
4162 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4163 // only supports dynamic priority, static priority must be zero. For real-time
4164 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4165 // However, for large multi-threaded applications, SCHED_RR is not only slower
4166 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4167 // of 5 runs - Sep 2005).
4168 //
4169 // The following code actually changes the niceness of kernel-thread/LWP. It
4170 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4171 // not the entire user process, and user level threads are 1:1 mapped to kernel
4172 // threads. It has always been the case, but could change in the future. For
4173 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4174 // It is only used when ThreadPriorityPolicy=1 and may require system level permission
4175 // (e.g., root privilege or CAP_SYS_NICE capability).
4176
4177 int os::java_to_os_priority[CriticalPriority + 1] = {
4178 19, // 0 Entry should never be used
4179
4180 4, // 1 MinPriority
4181 3, // 2
4182 2, // 3
4183
4184 1, // 4
4185 0, // 5 NormPriority
4186 -1, // 6
4187
4188 -2, // 7
4189 -3, // 8
4190 -4, // 9 NearMaxPriority
4191
4192 -5, // 10 MaxPriority
4193
4194 -5 // 11 CriticalPriority
4195 };
4196
4197 static int prio_init() {
4198 if (ThreadPriorityPolicy == 1) {
4199 if (geteuid() != 0) {
4200 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4201 warning("-XX:ThreadPriorityPolicy=1 may require system level permission, " \
4202 "e.g., being the root user. If the necessary permission is not " \
4203 "possessed, changes to priority will be silently ignored.");
4204 }
4205 }
4206 }
4207 if (UseCriticalJavaThreadPriority) {
4208 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4209 }
4210 return 0;
4211 }
4212
4213 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4214 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4215
4216 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4217 return (ret == 0) ? OS_OK : OS_ERR;
4218 }
4219
4220 OSReturn os::get_native_priority(const Thread* const thread,
4221 int *priority_ptr) {
4222 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4223 *priority_ptr = java_to_os_priority[NormPriority];
4224 return OS_OK;
4225 }
4226
4227 errno = 0;
4228 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4229 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4230 }
4231
4232 ////////////////////////////////////////////////////////////////////////////////
4233 // suspend/resume support
4234
4235 // The low-level signal-based suspend/resume support is a remnant from the
4236 // old VM-suspension that used to be for java-suspension, safepoints etc,
4237 // within hotspot. Currently used by JFR's OSThreadSampler
4238 //
4239 // The remaining code is greatly simplified from the more general suspension
4240 // code that used to be used.
4241 //
4242 // The protocol is quite simple:
4243 // - suspend:
4244 // - sends a signal to the target thread
4245 // - polls the suspend state of the osthread using a yield loop
4246 // - target thread signal handler (SR_handler) sets suspend state
4247 // and blocks in sigsuspend until continued
4248 // - resume:
4249 // - sets target osthread state to continue
4250 // - sends signal to end the sigsuspend loop in the SR_handler
4251 //
4252 // Note that the SR_lock plays no role in this suspend/resume protocol,
4253 // but is checked for NULL in SR_handler as a thread termination indicator.
4254 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4255 //
4256 // Note that resume_clear_context() and suspend_save_context() are needed
4257 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4258 // which in part is used by:
4259 // - Forte Analyzer: AsyncGetCallTrace()
4260 // - StackBanging: get_frame_at_stack_banging_point()
4261
4262 static void resume_clear_context(OSThread *osthread) {
4263 osthread->set_ucontext(NULL);
4264 osthread->set_siginfo(NULL);
4265 }
4266
4267 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4268 ucontext_t* context) {
4269 osthread->set_ucontext(context);
4270 osthread->set_siginfo(siginfo);
4271 }
4272
4273 // Handler function invoked when a thread's execution is suspended or
4274 // resumed. We have to be careful that only async-safe functions are
4275 // called here (Note: most pthread functions are not async safe and
4276 // should be avoided.)
4277 //
4278 // Note: sigwait() is a more natural fit than sigsuspend() from an
4279 // interface point of view, but sigwait() prevents the signal hander
4280 // from being run. libpthread would get very confused by not having
4281 // its signal handlers run and prevents sigwait()'s use with the
4282 // mutex granting granting signal.
4283 //
4284 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4285 //
4286 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4287 // Save and restore errno to avoid confusing native code with EINTR
4288 // after sigsuspend.
4289 int old_errno = errno;
4290
4291 Thread* thread = Thread::current_or_null_safe();
4292 assert(thread != NULL, "Missing current thread in SR_handler");
4293
4294 // On some systems we have seen signal delivery get "stuck" until the signal
4295 // mask is changed as part of thread termination. Check that the current thread
4296 // has not already terminated (via SR_lock()) - else the following assertion
4297 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4298 // destructor has completed.
4299
4300 if (thread->SR_lock() == NULL) {
4301 return;
4302 }
4303
4304 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4305
4306 OSThread* osthread = thread->osthread();
4307
4308 os::SuspendResume::State current = osthread->sr.state();
4309 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4310 suspend_save_context(osthread, siginfo, context);
4311
4312 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4313 os::SuspendResume::State state = osthread->sr.suspended();
4314 if (state == os::SuspendResume::SR_SUSPENDED) {
4315 sigset_t suspend_set; // signals for sigsuspend()
4316 sigemptyset(&suspend_set);
4317 // get current set of blocked signals and unblock resume signal
4318 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4319 sigdelset(&suspend_set, SR_signum);
4320
4321 sr_semaphore.signal();
4322 // wait here until we are resumed
4323 while (1) {
4324 sigsuspend(&suspend_set);
4325
4326 os::SuspendResume::State result = osthread->sr.running();
4327 if (result == os::SuspendResume::SR_RUNNING) {
4328 sr_semaphore.signal();
4329 break;
4330 }
4331 }
4332
4333 } else if (state == os::SuspendResume::SR_RUNNING) {
4334 // request was cancelled, continue
4335 } else {
4336 ShouldNotReachHere();
4337 }
4338
4339 resume_clear_context(osthread);
4340 } else if (current == os::SuspendResume::SR_RUNNING) {
4341 // request was cancelled, continue
4342 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4343 // ignore
4344 } else {
4345 // ignore
4346 }
4347
4348 errno = old_errno;
4349 }
4350
4351 static int SR_initialize() {
4352 struct sigaction act;
4353 char *s;
4354
4355 // Get signal number to use for suspend/resume
4356 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4357 int sig = ::strtol(s, 0, 10);
4358 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4359 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4360 SR_signum = sig;
4361 } else {
4362 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4363 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4364 }
4365 }
4366
4367 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4368 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4369
4370 sigemptyset(&SR_sigset);
4371 sigaddset(&SR_sigset, SR_signum);
4372
4373 // Set up signal handler for suspend/resume
4374 act.sa_flags = SA_RESTART|SA_SIGINFO;
4375 act.sa_handler = (void (*)(int)) SR_handler;
4376
4377 // SR_signum is blocked by default.
4378 // 4528190 - We also need to block pthread restart signal (32 on all
4379 // supported Linux platforms). Note that LinuxThreads need to block
4380 // this signal for all threads to work properly. So we don't have
4381 // to use hard-coded signal number when setting up the mask.
4382 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4383
4384 if (sigaction(SR_signum, &act, 0) == -1) {
4385 return -1;
4386 }
4387
4388 // Save signal flag
4389 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4390 return 0;
4391 }
4392
4393 static int sr_notify(OSThread* osthread) {
4394 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4395 assert_status(status == 0, status, "pthread_kill");
4396 return status;
4397 }
4398
4399 // "Randomly" selected value for how long we want to spin
4400 // before bailing out on suspending a thread, also how often
4401 // we send a signal to a thread we want to resume
4402 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4403 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4404
4405 // returns true on success and false on error - really an error is fatal
4406 // but this seems the normal response to library errors
4407 static bool do_suspend(OSThread* osthread) {
4408 assert(osthread->sr.is_running(), "thread should be running");
4409 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4410
4411 // mark as suspended and send signal
4412 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4413 // failed to switch, state wasn't running?
4414 ShouldNotReachHere();
4415 return false;
4416 }
4417
4418 if (sr_notify(osthread) != 0) {
4419 ShouldNotReachHere();
4420 }
4421
4422 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4423 while (true) {
4424 if (sr_semaphore.timedwait(2)) {
4425 break;
4426 } else {
4427 // timeout
4428 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4429 if (cancelled == os::SuspendResume::SR_RUNNING) {
4430 return false;
4431 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4432 // make sure that we consume the signal on the semaphore as well
4433 sr_semaphore.wait();
4434 break;
4435 } else {
4436 ShouldNotReachHere();
4437 return false;
4438 }
4439 }
4440 }
4441
4442 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4443 return true;
4444 }
4445
4446 static void do_resume(OSThread* osthread) {
4447 assert(osthread->sr.is_suspended(), "thread should be suspended");
4448 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4449
4450 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4451 // failed to switch to WAKEUP_REQUEST
4452 ShouldNotReachHere();
4453 return;
4454 }
4455
4456 while (true) {
4457 if (sr_notify(osthread) == 0) {
4458 if (sr_semaphore.timedwait(2)) {
4459 if (osthread->sr.is_running()) {
4460 return;
4461 }
4462 }
4463 } else {
4464 ShouldNotReachHere();
4465 }
4466 }
4467
4468 guarantee(osthread->sr.is_running(), "Must be running!");
4469 }
4470
4471 ///////////////////////////////////////////////////////////////////////////////////
4472 // signal handling (except suspend/resume)
4473
4474 // This routine may be used by user applications as a "hook" to catch signals.
4475 // The user-defined signal handler must pass unrecognized signals to this
4476 // routine, and if it returns true (non-zero), then the signal handler must
4477 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4478 // routine will never retun false (zero), but instead will execute a VM panic
4479 // routine kill the process.
4480 //
4481 // If this routine returns false, it is OK to call it again. This allows
4482 // the user-defined signal handler to perform checks either before or after
4483 // the VM performs its own checks. Naturally, the user code would be making
4484 // a serious error if it tried to handle an exception (such as a null check
4485 // or breakpoint) that the VM was generating for its own correct operation.
4486 //
4487 // This routine may recognize any of the following kinds of signals:
4488 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4489 // It should be consulted by handlers for any of those signals.
4490 //
4491 // The caller of this routine must pass in the three arguments supplied
4492 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4493 // field of the structure passed to sigaction(). This routine assumes that
4494 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4495 //
4496 // Note that the VM will print warnings if it detects conflicting signal
4497 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4498 //
4499 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4500 siginfo_t* siginfo,
4501 void* ucontext,
4502 int abort_if_unrecognized);
4503
4504 static void signalHandler(int sig, siginfo_t* info, void* uc) {
4505 assert(info != NULL && uc != NULL, "it must be old kernel");
4506 int orig_errno = errno; // Preserve errno value over signal handler.
4507 JVM_handle_linux_signal(sig, info, uc, true);
4508 errno = orig_errno;
4509 }
4510
4511
4512 // This boolean allows users to forward their own non-matching signals
4513 // to JVM_handle_linux_signal, harmlessly.
4514 bool os::Linux::signal_handlers_are_installed = false;
4515
4516 // For signal-chaining
4517 bool os::Linux::libjsig_is_loaded = false;
4518 typedef struct sigaction *(*get_signal_t)(int);
4519 get_signal_t os::Linux::get_signal_action = NULL;
4520
4521 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4522 struct sigaction *actp = NULL;
4523
4524 if (libjsig_is_loaded) {
4525 // Retrieve the old signal handler from libjsig
4526 actp = (*get_signal_action)(sig);
4527 }
4528 if (actp == NULL) {
4529 // Retrieve the preinstalled signal handler from jvm
4530 actp = os::Posix::get_preinstalled_handler(sig);
4531 }
4532
4533 return actp;
4534 }
4535
4536 static bool call_chained_handler(struct sigaction *actp, int sig,
4537 siginfo_t *siginfo, void *context) {
4538 // Call the old signal handler
4539 if (actp->sa_handler == SIG_DFL) {
4540 // It's more reasonable to let jvm treat it as an unexpected exception
4541 // instead of taking the default action.
4542 return false;
4543 } else if (actp->sa_handler != SIG_IGN) {
4544 if ((actp->sa_flags & SA_NODEFER) == 0) {
4545 // automaticlly block the signal
4546 sigaddset(&(actp->sa_mask), sig);
4547 }
4548
4549 sa_handler_t hand = NULL;
4550 sa_sigaction_t sa = NULL;
4551 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4552 // retrieve the chained handler
4553 if (siginfo_flag_set) {
4554 sa = actp->sa_sigaction;
4555 } else {
4556 hand = actp->sa_handler;
4557 }
4558
4559 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4560 actp->sa_handler = SIG_DFL;
4561 }
4562
4563 // try to honor the signal mask
4564 sigset_t oset;
4565 sigemptyset(&oset);
4566 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4567
4568 // call into the chained handler
4569 if (siginfo_flag_set) {
4570 (*sa)(sig, siginfo, context);
4571 } else {
4572 (*hand)(sig);
4573 }
4574
4575 // restore the signal mask
4576 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4577 }
4578 // Tell jvm's signal handler the signal is taken care of.
4579 return true;
4580 }
4581
4582 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4583 bool chained = false;
4584 // signal-chaining
4585 if (UseSignalChaining) {
4586 struct sigaction *actp = get_chained_signal_action(sig);
4587 if (actp != NULL) {
4588 chained = call_chained_handler(actp, sig, siginfo, context);
4589 }
4590 }
4591 return chained;
4592 }
4593
4594 // for diagnostic
4595 int sigflags[NSIG];
4596
4597 int os::Linux::get_our_sigflags(int sig) {
4598 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4599 return sigflags[sig];
4600 }
4601
4602 void os::Linux::set_our_sigflags(int sig, int flags) {
4603 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4604 if (sig > 0 && sig < NSIG) {
4605 sigflags[sig] = flags;
4606 }
4607 }
4608
4609 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4610 // Check for overwrite.
4611 struct sigaction oldAct;
4612 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4613
4614 void* oldhand = oldAct.sa_sigaction
4615 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4616 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4617 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4618 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4619 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4620 if (AllowUserSignalHandlers || !set_installed) {
4621 // Do not overwrite; user takes responsibility to forward to us.
4622 return;
4623 } else if (UseSignalChaining) {
4624 // save the old handler in jvm
4625 os::Posix::save_preinstalled_handler(sig, oldAct);
4626 // libjsig also interposes the sigaction() call below and saves the
4627 // old sigaction on it own.
4628 } else {
4629 fatal("Encountered unexpected pre-existing sigaction handler "
4630 "%#lx for signal %d.", (long)oldhand, sig);
4631 }
4632 }
4633
4634 struct sigaction sigAct;
4635 sigfillset(&(sigAct.sa_mask));
4636 sigAct.sa_handler = SIG_DFL;
4637 if (!set_installed) {
4638 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4639 } else {
4640 sigAct.sa_sigaction = signalHandler;
4641 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4642 }
4643 // Save flags, which are set by ours
4644 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4645 sigflags[sig] = sigAct.sa_flags;
4646
4647 int ret = sigaction(sig, &sigAct, &oldAct);
4648 assert(ret == 0, "check");
4649
4650 void* oldhand2 = oldAct.sa_sigaction
4651 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4652 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4653 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4654 }
4655
4656 // install signal handlers for signals that HotSpot needs to
4657 // handle in order to support Java-level exception handling.
4658
4659 void os::Linux::install_signal_handlers() {
4660 if (!signal_handlers_are_installed) {
4661 signal_handlers_are_installed = true;
4662
4663 // signal-chaining
4664 typedef void (*signal_setting_t)();
4665 signal_setting_t begin_signal_setting = NULL;
4666 signal_setting_t end_signal_setting = NULL;
4667 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4668 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4669 if (begin_signal_setting != NULL) {
4670 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4671 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4672 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4673 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4674 libjsig_is_loaded = true;
4675 assert(UseSignalChaining, "should enable signal-chaining");
4676 }
4677 if (libjsig_is_loaded) {
4678 // Tell libjsig jvm is setting signal handlers
4679 (*begin_signal_setting)();
4680 }
4681
4682 set_signal_handler(SIGSEGV, true);
4683 set_signal_handler(SIGPIPE, true);
4684 set_signal_handler(SIGBUS, true);
4685 set_signal_handler(SIGILL, true);
4686 set_signal_handler(SIGFPE, true);
4687 #if defined(PPC64)
4688 set_signal_handler(SIGTRAP, true);
4689 #endif
4690 set_signal_handler(SIGXFSZ, true);
4691
4692 if (libjsig_is_loaded) {
4693 // Tell libjsig jvm finishes setting signal handlers
4694 (*end_signal_setting)();
4695 }
4696
4697 // We don't activate signal checker if libjsig is in place, we trust ourselves
4698 // and if UserSignalHandler is installed all bets are off.
4699 // Log that signal checking is off only if -verbose:jni is specified.
4700 if (CheckJNICalls) {
4701 if (libjsig_is_loaded) {
4702 if (PrintJNIResolving) {
4703 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4704 }
4705 check_signals = false;
4706 }
4707 if (AllowUserSignalHandlers) {
4708 if (PrintJNIResolving) {
4709 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4710 }
4711 check_signals = false;
4712 }
4713 }
4714 }
4715 }
4716
4717 // This is the fastest way to get thread cpu time on Linux.
4718 // Returns cpu time (user+sys) for any thread, not only for current.
4719 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4720 // It might work on 2.6.10+ with a special kernel/glibc patch.
4721 // For reference, please, see IEEE Std 1003.1-2004:
4722 // http://www.unix.org/single_unix_specification
4723
4724 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4725 struct timespec tp;
4726 int rc = os::Posix::clock_gettime(clockid, &tp);
4727 assert(rc == 0, "clock_gettime is expected to return 0 code");
4728
4729 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4730 }
4731
4732 void os::Linux::initialize_os_info() {
4733 assert(_os_version == 0, "OS info already initialized");
4734
4735 struct utsname _uname;
4736
4737 uint32_t major;
4738 uint32_t minor;
4739 uint32_t fix;
4740
4741 int rc;
4742
4743 // Kernel version is unknown if
4744 // verification below fails.
4745 _os_version = 0x01000000;
4746
4747 rc = uname(&_uname);
4748 if (rc != -1) {
4749
4750 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4751 if (rc == 3) {
4752
4753 if (major < 256 && minor < 256 && fix < 256) {
4754 // Kernel version format is as expected,
4755 // set it overriding unknown state.
4756 _os_version = (major << 16) |
4757 (minor << 8 ) |
4758 (fix << 0 ) ;
4759 }
4760 }
4761 }
4762 }
4763
4764 uint32_t os::Linux::os_version() {
4765 assert(_os_version != 0, "not initialized");
4766 return _os_version & 0x00FFFFFF;
4767 }
4768
4769 bool os::Linux::os_version_is_known() {
4770 assert(_os_version != 0, "not initialized");
4771 return _os_version & 0x01000000 ? false : true;
4772 }
4773
4774 /////
4775 // glibc on Linux platform uses non-documented flag
4776 // to indicate, that some special sort of signal
4777 // trampoline is used.
4778 // We will never set this flag, and we should
4779 // ignore this flag in our diagnostic
4780 #ifdef SIGNIFICANT_SIGNAL_MASK
4781 #undef SIGNIFICANT_SIGNAL_MASK
4782 #endif
4783 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4784
4785 static const char* get_signal_handler_name(address handler,
4786 char* buf, int buflen) {
4787 int offset = 0;
4788 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4789 if (found) {
4790 // skip directory names
4791 const char *p1, *p2;
4792 p1 = buf;
4793 size_t len = strlen(os::file_separator());
4794 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4795 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4796 } else {
4797 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4798 }
4799 return buf;
4800 }
4801
4802 static void print_signal_handler(outputStream* st, int sig,
4803 char* buf, size_t buflen) {
4804 struct sigaction sa;
4805
4806 sigaction(sig, NULL, &sa);
4807
4808 // See comment for SIGNIFICANT_SIGNAL_MASK define
4809 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4810
4811 st->print("%s: ", os::exception_name(sig, buf, buflen));
4812
4813 address handler = (sa.sa_flags & SA_SIGINFO)
4814 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4815 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4816
4817 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4818 st->print("SIG_DFL");
4819 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4820 st->print("SIG_IGN");
4821 } else {
4822 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4823 }
4824
4825 st->print(", sa_mask[0]=");
4826 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4827
4828 address rh = VMError::get_resetted_sighandler(sig);
4829 // May be, handler was resetted by VMError?
4830 if (rh != NULL) {
4831 handler = rh;
4832 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4833 }
4834
4835 st->print(", sa_flags=");
4836 os::Posix::print_sa_flags(st, sa.sa_flags);
4837
4838 // Check: is it our handler?
4839 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4840 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4841 // It is our signal handler
4842 // check for flags, reset system-used one!
4843 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4844 st->print(
4845 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4846 os::Linux::get_our_sigflags(sig));
4847 }
4848 }
4849 st->cr();
4850 }
4851
4852
4853 #define DO_SIGNAL_CHECK(sig) \
4854 do { \
4855 if (!sigismember(&check_signal_done, sig)) { \
4856 os::Linux::check_signal_handler(sig); \
4857 } \
4858 } while (0)
4859
4860 // This method is a periodic task to check for misbehaving JNI applications
4861 // under CheckJNI, we can add any periodic checks here
4862
4863 void os::run_periodic_checks() {
4864 if (check_signals == false) return;
4865
4866 // SEGV and BUS if overridden could potentially prevent
4867 // generation of hs*.log in the event of a crash, debugging
4868 // such a case can be very challenging, so we absolutely
4869 // check the following for a good measure:
4870 DO_SIGNAL_CHECK(SIGSEGV);
4871 DO_SIGNAL_CHECK(SIGILL);
4872 DO_SIGNAL_CHECK(SIGFPE);
4873 DO_SIGNAL_CHECK(SIGBUS);
4874 DO_SIGNAL_CHECK(SIGPIPE);
4875 DO_SIGNAL_CHECK(SIGXFSZ);
4876 #if defined(PPC64)
4877 DO_SIGNAL_CHECK(SIGTRAP);
4878 #endif
4879
4880 // ReduceSignalUsage allows the user to override these handlers
4881 // see comments at the very top and jvm_md.h
4882 if (!ReduceSignalUsage) {
4883 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4884 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4885 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4886 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4887 }
4888
4889 DO_SIGNAL_CHECK(SR_signum);
4890 }
4891
4892 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4893
4894 static os_sigaction_t os_sigaction = NULL;
4895
4896 void os::Linux::check_signal_handler(int sig) {
4897 char buf[O_BUFLEN];
4898 address jvmHandler = NULL;
4899
4900
4901 struct sigaction act;
4902 if (os_sigaction == NULL) {
4903 // only trust the default sigaction, in case it has been interposed
4904 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4905 if (os_sigaction == NULL) return;
4906 }
4907
4908 os_sigaction(sig, (struct sigaction*)NULL, &act);
4909
4910
4911 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4912
4913 address thisHandler = (act.sa_flags & SA_SIGINFO)
4914 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4915 : CAST_FROM_FN_PTR(address, act.sa_handler);
4916
4917
4918 switch (sig) {
4919 case SIGSEGV:
4920 case SIGBUS:
4921 case SIGFPE:
4922 case SIGPIPE:
4923 case SIGILL:
4924 case SIGXFSZ:
4925 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4926 break;
4927
4928 case SHUTDOWN1_SIGNAL:
4929 case SHUTDOWN2_SIGNAL:
4930 case SHUTDOWN3_SIGNAL:
4931 case BREAK_SIGNAL:
4932 jvmHandler = (address)user_handler();
4933 break;
4934
4935 default:
4936 if (sig == SR_signum) {
4937 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4938 } else {
4939 return;
4940 }
4941 break;
4942 }
4943
4944 if (thisHandler != jvmHandler) {
4945 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4946 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4947 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4948 // No need to check this sig any longer
4949 sigaddset(&check_signal_done, sig);
4950 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4951 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4952 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4953 exception_name(sig, buf, O_BUFLEN));
4954 }
4955 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4956 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4957 tty->print("expected:");
4958 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4959 tty->cr();
4960 tty->print(" found:");
4961 os::Posix::print_sa_flags(tty, act.sa_flags);
4962 tty->cr();
4963 // No need to check this sig any longer
4964 sigaddset(&check_signal_done, sig);
4965 }
4966
4967 // Dump all the signal
4968 if (sigismember(&check_signal_done, sig)) {
4969 print_signal_handlers(tty, buf, O_BUFLEN);
4970 }
4971 }
4972
4973 extern void report_error(char* file_name, int line_no, char* title,
4974 char* format, ...);
4975
4976 // this is called _before_ most of the global arguments have been parsed
4977 void os::init(void) {
4978 char dummy; // used to get a guess on initial stack address
4979
4980 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4981
4982 init_random(1234567);
4983
4984 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4985 if (Linux::page_size() == -1) {
4986 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4987 os::strerror(errno));
4988 }
4989 init_page_sizes((size_t) Linux::page_size());
4990
4991 Linux::initialize_system_info();
4992
4993 Linux::initialize_os_info();
4994
4995 os::Linux::CPUPerfTicks pticks;
4996 bool res = os::Linux::get_tick_information(&pticks, -1);
4997
4998 if (res && pticks.has_steal_ticks) {
4999 has_initial_tick_info = true;
5000 initial_total_ticks = pticks.total;
5001 initial_steal_ticks = pticks.steal;
5002 }
5003
5004 // _main_thread points to the thread that created/loaded the JVM.
5005 Linux::_main_thread = pthread_self();
5006
5007 // retrieve entry point for pthread_setname_np
5008 Linux::_pthread_setname_np =
5009 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
5010
5011 os::Posix::init();
5012
5013 initial_time_count = javaTimeNanos();
5014
5015 // Always warn if no monotonic clock available
5016 if (!os::Posix::supports_monotonic_clock()) {
5017 warning("No monotonic clock was available - timed services may " \
5018 "be adversely affected if the time-of-day clock changes");
5019 }
5020 }
5021
5022 // To install functions for atexit system call
5023 extern "C" {
5024 static void perfMemory_exit_helper() {
5025 perfMemory_exit();
5026 }
5027 }
5028
5029 void os::pd_init_container_support() {
5030 OSContainer::init();
5031 }
5032
5033 void os::Linux::numa_init() {
5034
5035 // Java can be invoked as
5036 // 1. Without numactl and heap will be allocated/configured on all nodes as
5037 // per the system policy.
5038 // 2. With numactl --interleave:
5039 // Use numa_get_interleave_mask(v2) API to get nodes bitmask. The same
5040 // API for membind case bitmask is reset.
5041 // Interleave is only hint and Kernel can fallback to other nodes if
5042 // no memory is available on the target nodes.
5043 // 3. With numactl --membind:
5044 // Use numa_get_membind(v2) API to get nodes bitmask. The same API for
5045 // interleave case returns bitmask of all nodes.
5046 // numa_all_nodes_ptr holds bitmask of all nodes.
5047 // numa_get_interleave_mask(v2) and numa_get_membind(v2) APIs returns correct
5048 // bitmask when externally configured to run on all or fewer nodes.
5049
5050 if (!Linux::libnuma_init()) {
5051 UseNUMA = false;
5052 } else {
5053 if ((Linux::numa_max_node() < 1) || Linux::is_bound_to_single_node()) {
5054 // If there's only one node (they start from 0) or if the process
5055 // is bound explicitly to a single node using membind, disable NUMA.
5056 UseNUMA = false;
5057 } else {
5058
5059 LogTarget(Info,os) log;
5060 LogStream ls(log);
5061
5062 Linux::set_configured_numa_policy(Linux::identify_numa_policy());
5063
5064 struct bitmask* bmp = Linux::_numa_membind_bitmask;
5065 const char* numa_mode = "membind";
5066
5067 if (Linux::is_running_in_interleave_mode()) {
5068 bmp = Linux::_numa_interleave_bitmask;
5069 numa_mode = "interleave";
5070 }
5071
5072 ls.print("UseNUMA is enabled and invoked in '%s' mode."
5073 " Heap will be configured using NUMA memory nodes:", numa_mode);
5074
5075 for (int node = 0; node <= Linux::numa_max_node(); node++) {
5076 if (Linux::_numa_bitmask_isbitset(bmp, node)) {
5077 ls.print(" %d", node);
5078 }
5079 }
5080 }
5081 }
5082
5083 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5084 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5085 // we can make the adaptive lgrp chunk resizing work. If the user specified both
5086 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5087 // and disable adaptive resizing.
5088 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5089 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5090 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5091 UseAdaptiveSizePolicy = false;
5092 UseAdaptiveNUMAChunkSizing = false;
5093 }
5094 }
5095
5096 if (!UseNUMA && ForceNUMA) {
5097 UseNUMA = true;
5098 }
5099 }
5100
5101 // this is called _after_ the global arguments have been parsed
5102 jint os::init_2(void) {
5103
5104 // This could be set after os::Posix::init() but all platforms
5105 // have to set it the same so we have to mirror Solaris.
5106 DEBUG_ONLY(os::set_mutex_init_done();)
5107
5108 os::Posix::init_2();
5109
5110 Linux::fast_thread_clock_init();
5111
5112 // initialize suspend/resume support - must do this before signal_sets_init()
5113 if (SR_initialize() != 0) {
5114 perror("SR_initialize failed");
5115 return JNI_ERR;
5116 }
5117
5118 Linux::signal_sets_init();
5119 Linux::install_signal_handlers();
5120 // Initialize data for jdk.internal.misc.Signal
5121 if (!ReduceSignalUsage) {
5122 jdk_misc_signal_init();
5123 }
5124
5125 // Check and sets minimum stack sizes against command line options
5126 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
5127 return JNI_ERR;
5128 }
5129
5130 #if defined(IA32)
5131 // Need to ensure we've determined the process's initial stack to
5132 // perform the workaround
5133 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5134 workaround_expand_exec_shield_cs_limit();
5135 #else
5136 suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
5137 if (!suppress_primordial_thread_resolution) {
5138 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5139 }
5140 #endif
5141
5142 Linux::libpthread_init();
5143 Linux::sched_getcpu_init();
5144 log_info(os)("HotSpot is running with %s, %s",
5145 Linux::glibc_version(), Linux::libpthread_version());
5146
5147 if (UseNUMA) {
5148 Linux::numa_init();
5149 }
5150
5151 if (MaxFDLimit) {
5152 // set the number of file descriptors to max. print out error
5153 // if getrlimit/setrlimit fails but continue regardless.
5154 struct rlimit nbr_files;
5155 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5156 if (status != 0) {
5157 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5158 } else {
5159 nbr_files.rlim_cur = nbr_files.rlim_max;
5160 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5161 if (status != 0) {
5162 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5163 }
5164 }
5165 }
5166
5167 // Initialize lock used to serialize thread creation (see os::create_thread)
5168 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5169
5170 // at-exit methods are called in the reverse order of their registration.
5171 // atexit functions are called on return from main or as a result of a
5172 // call to exit(3C). There can be only 32 of these functions registered
5173 // and atexit() does not set errno.
5174
5175 if (PerfAllowAtExitRegistration) {
5176 // only register atexit functions if PerfAllowAtExitRegistration is set.
5177 // atexit functions can be delayed until process exit time, which
5178 // can be problematic for embedded VM situations. Embedded VMs should
5179 // call DestroyJavaVM() to assure that VM resources are released.
5180
5181 // note: perfMemory_exit_helper atexit function may be removed in
5182 // the future if the appropriate cleanup code can be added to the
5183 // VM_Exit VMOperation's doit method.
5184 if (atexit(perfMemory_exit_helper) != 0) {
5185 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5186 }
5187 }
5188
5189 // initialize thread priority policy
5190 prio_init();
5191
5192 if (!FLAG_IS_DEFAULT(AllocateHeapAt) || !FLAG_IS_DEFAULT(AllocateOldGenAt)) {
5193 set_coredump_filter(DAX_SHARED_BIT);
5194 }
5195
5196 if (DumpPrivateMappingsInCore) {
5197 set_coredump_filter(FILE_BACKED_PVT_BIT);
5198 }
5199
5200 if (DumpSharedMappingsInCore) {
5201 set_coredump_filter(FILE_BACKED_SHARED_BIT);
5202 }
5203
5204 return JNI_OK;
5205 }
5206
5207 // Mark the polling page as unreadable
5208 void os::make_polling_page_unreadable(void) {
5209 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5210 fatal("Could not disable polling page");
5211 }
5212 }
5213
5214 // Mark the polling page as readable
5215 void os::make_polling_page_readable(void) {
5216 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5217 fatal("Could not enable polling page");
5218 }
5219 }
5220
5221 // older glibc versions don't have this macro (which expands to
5222 // an optimized bit-counting function) so we have to roll our own
5223 #ifndef CPU_COUNT
5224
5225 static int _cpu_count(const cpu_set_t* cpus) {
5226 int count = 0;
5227 // only look up to the number of configured processors
5228 for (int i = 0; i < os::processor_count(); i++) {
5229 if (CPU_ISSET(i, cpus)) {
5230 count++;
5231 }
5232 }
5233 return count;
5234 }
5235
5236 #define CPU_COUNT(cpus) _cpu_count(cpus)
5237
5238 #endif // CPU_COUNT
5239
5240 // Get the current number of available processors for this process.
5241 // This value can change at any time during a process's lifetime.
5242 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5243 // If it appears there may be more than 1024 processors then we do a
5244 // dynamic check - see 6515172 for details.
5245 // If anything goes wrong we fallback to returning the number of online
5246 // processors - which can be greater than the number available to the process.
5247 int os::Linux::active_processor_count() {
5248 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5249 cpu_set_t* cpus_p = &cpus;
5250 int cpus_size = sizeof(cpu_set_t);
5251
5252 int configured_cpus = os::processor_count(); // upper bound on available cpus
5253 int cpu_count = 0;
5254
5255 // old build platforms may not support dynamic cpu sets
5256 #ifdef CPU_ALLOC
5257
5258 // To enable easy testing of the dynamic path on different platforms we
5259 // introduce a diagnostic flag: UseCpuAllocPath
5260 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5261 // kernel may use a mask bigger than cpu_set_t
5262 log_trace(os)("active_processor_count: using dynamic path %s"
5263 "- configured processors: %d",
5264 UseCpuAllocPath ? "(forced) " : "",
5265 configured_cpus);
5266 cpus_p = CPU_ALLOC(configured_cpus);
5267 if (cpus_p != NULL) {
5268 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5269 // zero it just to be safe
5270 CPU_ZERO_S(cpus_size, cpus_p);
5271 }
5272 else {
5273 // failed to allocate so fallback to online cpus
5274 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5275 log_trace(os)("active_processor_count: "
5276 "CPU_ALLOC failed (%s) - using "
5277 "online processor count: %d",
5278 os::strerror(errno), online_cpus);
5279 return online_cpus;
5280 }
5281 }
5282 else {
5283 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5284 configured_cpus);
5285 }
5286 #else // CPU_ALLOC
5287 // these stubs won't be executed
5288 #define CPU_COUNT_S(size, cpus) -1
5289 #define CPU_FREE(cpus)
5290
5291 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5292 configured_cpus);
5293 #endif // CPU_ALLOC
5294
5295 // pid 0 means the current thread - which we have to assume represents the process
5296 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5297 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5298 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5299 }
5300 else {
5301 cpu_count = CPU_COUNT(cpus_p);
5302 }
5303 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5304 }
5305 else {
5306 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5307 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5308 "which may exceed available processors", os::strerror(errno), cpu_count);
5309 }
5310
5311 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5312 CPU_FREE(cpus_p);
5313 }
5314
5315 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5316 return cpu_count;
5317 }
5318
5319 // Determine the active processor count from one of
5320 // three different sources:
5321 //
5322 // 1. User option -XX:ActiveProcessorCount
5323 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5324 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5325 //
5326 // Option 1, if specified, will always override.
5327 // If the cgroup subsystem is active and configured, we
5328 // will return the min of the cgroup and option 2 results.
5329 // This is required since tools, such as numactl, that
5330 // alter cpu affinity do not update cgroup subsystem
5331 // cpuset configuration files.
5332 int os::active_processor_count() {
5333 // User has overridden the number of active processors
5334 if (ActiveProcessorCount > 0) {
5335 log_trace(os)("active_processor_count: "
5336 "active processor count set by user : %d",
5337 ActiveProcessorCount);
5338 return ActiveProcessorCount;
5339 }
5340
5341 int active_cpus;
5342 if (OSContainer::is_containerized()) {
5343 active_cpus = OSContainer::active_processor_count();
5344 log_trace(os)("active_processor_count: determined by OSContainer: %d",
5345 active_cpus);
5346 } else {
5347 active_cpus = os::Linux::active_processor_count();
5348 }
5349
5350 return active_cpus;
5351 }
5352
5353 uint os::processor_id() {
5354 const int id = Linux::sched_getcpu();
5355 assert(id >= 0 && id < _processor_count, "Invalid processor id");
5356 return (uint)id;
5357 }
5358
5359 void os::set_native_thread_name(const char *name) {
5360 if (Linux::_pthread_setname_np) {
5361 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5362 snprintf(buf, sizeof(buf), "%s", name);
5363 buf[sizeof(buf) - 1] = '\0';
5364 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5365 // ERANGE should not happen; all other errors should just be ignored.
5366 assert(rc != ERANGE, "pthread_setname_np failed");
5367 }
5368 }
5369
5370 bool os::distribute_processes(uint length, uint* distribution) {
5371 // Not yet implemented.
5372 return false;
5373 }
5374
5375 bool os::bind_to_processor(uint processor_id) {
5376 // Not yet implemented.
5377 return false;
5378 }
5379
5380 ///
5381
5382 void os::SuspendedThreadTask::internal_do_task() {
5383 if (do_suspend(_thread->osthread())) {
5384 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5385 do_task(context);
5386 do_resume(_thread->osthread());
5387 }
5388 }
5389
5390 ////////////////////////////////////////////////////////////////////////////////
5391 // debug support
5392
5393 bool os::find(address addr, outputStream* st) {
5394 Dl_info dlinfo;
5395 memset(&dlinfo, 0, sizeof(dlinfo));
5396 if (dladdr(addr, &dlinfo) != 0) {
5397 st->print(PTR_FORMAT ": ", p2i(addr));
5398 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5399 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5400 p2i(addr) - p2i(dlinfo.dli_saddr));
5401 } else if (dlinfo.dli_fbase != NULL) {
5402 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5403 } else {
5404 st->print("<absolute address>");
5405 }
5406 if (dlinfo.dli_fname != NULL) {
5407 st->print(" in %s", dlinfo.dli_fname);
5408 }
5409 if (dlinfo.dli_fbase != NULL) {
5410 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5411 }
5412 st->cr();
5413
5414 if (Verbose) {
5415 // decode some bytes around the PC
5416 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5417 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5418 address lowest = (address) dlinfo.dli_sname;
5419 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5420 if (begin < lowest) begin = lowest;
5421 Dl_info dlinfo2;
5422 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5423 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5424 end = (address) dlinfo2.dli_saddr;
5425 }
5426 Disassembler::decode(begin, end, st);
5427 }
5428 return true;
5429 }
5430 return false;
5431 }
5432
5433 ////////////////////////////////////////////////////////////////////////////////
5434 // misc
5435
5436 // This does not do anything on Linux. This is basically a hook for being
5437 // able to use structured exception handling (thread-local exception filters)
5438 // on, e.g., Win32.
5439 void
5440 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5441 JavaCallArguments* args, Thread* thread) {
5442 f(value, method, args, thread);
5443 }
5444
5445 void os::print_statistics() {
5446 }
5447
5448 bool os::message_box(const char* title, const char* message) {
5449 int i;
5450 fdStream err(defaultStream::error_fd());
5451 for (i = 0; i < 78; i++) err.print_raw("=");
5452 err.cr();
5453 err.print_raw_cr(title);
5454 for (i = 0; i < 78; i++) err.print_raw("-");
5455 err.cr();
5456 err.print_raw_cr(message);
5457 for (i = 0; i < 78; i++) err.print_raw("=");
5458 err.cr();
5459
5460 char buf[16];
5461 // Prevent process from exiting upon "read error" without consuming all CPU
5462 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5463
5464 return buf[0] == 'y' || buf[0] == 'Y';
5465 }
5466
5467 // Is a (classpath) directory empty?
5468 bool os::dir_is_empty(const char* path) {
5469 DIR *dir = NULL;
5470 struct dirent *ptr;
5471
5472 dir = opendir(path);
5473 if (dir == NULL) return true;
5474
5475 // Scan the directory
5476 bool result = true;
5477 while (result && (ptr = readdir(dir)) != NULL) {
5478 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5479 result = false;
5480 }
5481 }
5482 closedir(dir);
5483 return result;
5484 }
5485
5486 // This code originates from JDK's sysOpen and open64_w
5487 // from src/solaris/hpi/src/system_md.c
5488
5489 int os::open(const char *path, int oflag, int mode) {
5490 if (strlen(path) > MAX_PATH - 1) {
5491 errno = ENAMETOOLONG;
5492 return -1;
5493 }
5494
5495 // All file descriptors that are opened in the Java process and not
5496 // specifically destined for a subprocess should have the close-on-exec
5497 // flag set. If we don't set it, then careless 3rd party native code
5498 // might fork and exec without closing all appropriate file descriptors
5499 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5500 // turn might:
5501 //
5502 // - cause end-of-file to fail to be detected on some file
5503 // descriptors, resulting in mysterious hangs, or
5504 //
5505 // - might cause an fopen in the subprocess to fail on a system
5506 // suffering from bug 1085341.
5507 //
5508 // (Yes, the default setting of the close-on-exec flag is a Unix
5509 // design flaw)
5510 //
5511 // See:
5512 // 1085341: 32-bit stdio routines should support file descriptors >255
5513 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5514 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5515 //
5516 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5517 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5518 // because it saves a system call and removes a small window where the flag
5519 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5520 // and we fall back to using FD_CLOEXEC (see below).
5521 #ifdef O_CLOEXEC
5522 oflag |= O_CLOEXEC;
5523 #endif
5524
5525 int fd = ::open64(path, oflag, mode);
5526 if (fd == -1) return -1;
5527
5528 //If the open succeeded, the file might still be a directory
5529 {
5530 struct stat64 buf64;
5531 int ret = ::fstat64(fd, &buf64);
5532 int st_mode = buf64.st_mode;
5533
5534 if (ret != -1) {
5535 if ((st_mode & S_IFMT) == S_IFDIR) {
5536 errno = EISDIR;
5537 ::close(fd);
5538 return -1;
5539 }
5540 } else {
5541 ::close(fd);
5542 return -1;
5543 }
5544 }
5545
5546 #ifdef FD_CLOEXEC
5547 // Validate that the use of the O_CLOEXEC flag on open above worked.
5548 // With recent kernels, we will perform this check exactly once.
5549 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5550 if (!O_CLOEXEC_is_known_to_work) {
5551 int flags = ::fcntl(fd, F_GETFD);
5552 if (flags != -1) {
5553 if ((flags & FD_CLOEXEC) != 0)
5554 O_CLOEXEC_is_known_to_work = 1;
5555 else
5556 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5557 }
5558 }
5559 #endif
5560
5561 return fd;
5562 }
5563
5564
5565 // create binary file, rewriting existing file if required
5566 int os::create_binary_file(const char* path, bool rewrite_existing) {
5567 int oflags = O_WRONLY | O_CREAT;
5568 if (!rewrite_existing) {
5569 oflags |= O_EXCL;
5570 }
5571 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5572 }
5573
5574 // return current position of file pointer
5575 jlong os::current_file_offset(int fd) {
5576 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5577 }
5578
5579 // move file pointer to the specified offset
5580 jlong os::seek_to_file_offset(int fd, jlong offset) {
5581 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5582 }
5583
5584 // This code originates from JDK's sysAvailable
5585 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5586
5587 int os::available(int fd, jlong *bytes) {
5588 jlong cur, end;
5589 int mode;
5590 struct stat64 buf64;
5591
5592 if (::fstat64(fd, &buf64) >= 0) {
5593 mode = buf64.st_mode;
5594 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5595 int n;
5596 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5597 *bytes = n;
5598 return 1;
5599 }
5600 }
5601 }
5602 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5603 return 0;
5604 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5605 return 0;
5606 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5607 return 0;
5608 }
5609 *bytes = end - cur;
5610 return 1;
5611 }
5612
5613 // Map a block of memory.
5614 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5615 char *addr, size_t bytes, bool read_only,
5616 bool allow_exec) {
5617 int prot;
5618 int flags = MAP_PRIVATE;
5619
5620 if (read_only) {
5621 prot = PROT_READ;
5622 } else {
5623 prot = PROT_READ | PROT_WRITE;
5624 }
5625
5626 if (allow_exec) {
5627 prot |= PROT_EXEC;
5628 }
5629
5630 if (addr != NULL) {
5631 flags |= MAP_FIXED;
5632 }
5633
5634 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5635 fd, file_offset);
5636 if (mapped_address == MAP_FAILED) {
5637 return NULL;
5638 }
5639 return mapped_address;
5640 }
5641
5642
5643 // Remap a block of memory.
5644 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5645 char *addr, size_t bytes, bool read_only,
5646 bool allow_exec) {
5647 // same as map_memory() on this OS
5648 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5649 allow_exec);
5650 }
5651
5652
5653 // Unmap a block of memory.
5654 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5655 return munmap(addr, bytes) == 0;
5656 }
5657
5658 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5659
5660 static jlong fast_cpu_time(Thread *thread) {
5661 clockid_t clockid;
5662 int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(),
5663 &clockid);
5664 if (rc == 0) {
5665 return os::Linux::fast_thread_cpu_time(clockid);
5666 } else {
5667 // It's possible to encounter a terminated native thread that failed
5668 // to detach itself from the VM - which should result in ESRCH.
5669 assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed");
5670 return -1;
5671 }
5672 }
5673
5674 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5675 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5676 // of a thread.
5677 //
5678 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5679 // the fast estimate available on the platform.
5680
5681 jlong os::current_thread_cpu_time() {
5682 if (os::Linux::supports_fast_thread_cpu_time()) {
5683 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5684 } else {
5685 // return user + sys since the cost is the same
5686 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5687 }
5688 }
5689
5690 jlong os::thread_cpu_time(Thread* thread) {
5691 // consistent with what current_thread_cpu_time() returns
5692 if (os::Linux::supports_fast_thread_cpu_time()) {
5693 return fast_cpu_time(thread);
5694 } else {
5695 return slow_thread_cpu_time(thread, true /* user + sys */);
5696 }
5697 }
5698
5699 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5700 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5701 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5702 } else {
5703 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5704 }
5705 }
5706
5707 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5708 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5709 return fast_cpu_time(thread);
5710 } else {
5711 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5712 }
5713 }
5714
5715 // -1 on error.
5716 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5717 pid_t tid = thread->osthread()->thread_id();
5718 char *s;
5719 char stat[2048];
5720 int statlen;
5721 char proc_name[64];
5722 int count;
5723 long sys_time, user_time;
5724 char cdummy;
5725 int idummy;
5726 long ldummy;
5727 FILE *fp;
5728
5729 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5730 fp = fopen(proc_name, "r");
5731 if (fp == NULL) return -1;
5732 statlen = fread(stat, 1, 2047, fp);
5733 stat[statlen] = '\0';
5734 fclose(fp);
5735
5736 // Skip pid and the command string. Note that we could be dealing with
5737 // weird command names, e.g. user could decide to rename java launcher
5738 // to "java 1.4.2 :)", then the stat file would look like
5739 // 1234 (java 1.4.2 :)) R ... ...
5740 // We don't really need to know the command string, just find the last
5741 // occurrence of ")" and then start parsing from there. See bug 4726580.
5742 s = strrchr(stat, ')');
5743 if (s == NULL) return -1;
5744
5745 // Skip blank chars
5746 do { s++; } while (s && isspace(*s));
5747
5748 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5749 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5750 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5751 &user_time, &sys_time);
5752 if (count != 13) return -1;
5753 if (user_sys_cpu_time) {
5754 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5755 } else {
5756 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5757 }
5758 }
5759
5760 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5761 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5762 info_ptr->may_skip_backward = false; // elapsed time not wall time
5763 info_ptr->may_skip_forward = false; // elapsed time not wall time
5764 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5765 }
5766
5767 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5768 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5769 info_ptr->may_skip_backward = false; // elapsed time not wall time
5770 info_ptr->may_skip_forward = false; // elapsed time not wall time
5771 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5772 }
5773
5774 bool os::is_thread_cpu_time_supported() {
5775 return true;
5776 }
5777
5778 // System loadavg support. Returns -1 if load average cannot be obtained.
5779 // Linux doesn't yet have a (official) notion of processor sets,
5780 // so just return the system wide load average.
5781 int os::loadavg(double loadavg[], int nelem) {
5782 return ::getloadavg(loadavg, nelem);
5783 }
5784
5785 void os::pause() {
5786 char filename[MAX_PATH];
5787 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5788 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5789 } else {
5790 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5791 }
5792
5793 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5794 if (fd != -1) {
5795 struct stat buf;
5796 ::close(fd);
5797 while (::stat(filename, &buf) == 0) {
5798 (void)::poll(NULL, 0, 100);
5799 }
5800 } else {
5801 jio_fprintf(stderr,
5802 "Could not open pause file '%s', continuing immediately.\n", filename);
5803 }
5804 }
5805
5806 extern char** environ;
5807
5808 // Run the specified command in a separate process. Return its exit value,
5809 // or -1 on failure (e.g. can't fork a new process).
5810 // Unlike system(), this function can be called from signal handler. It
5811 // doesn't block SIGINT et al.
5812 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) {
5813 const char * argv[4] = {"sh", "-c", cmd, NULL};
5814
5815 pid_t pid ;
5816
5817 if (use_vfork_if_available) {
5818 pid = vfork();
5819 } else {
5820 pid = fork();
5821 }
5822
5823 if (pid < 0) {
5824 // fork failed
5825 return -1;
5826
5827 } else if (pid == 0) {
5828 // child process
5829
5830 execve("/bin/sh", (char* const*)argv, environ);
5831
5832 // execve failed
5833 _exit(-1);
5834
5835 } else {
5836 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5837 // care about the actual exit code, for now.
5838
5839 int status;
5840
5841 // Wait for the child process to exit. This returns immediately if
5842 // the child has already exited. */
5843 while (waitpid(pid, &status, 0) < 0) {
5844 switch (errno) {
5845 case ECHILD: return 0;
5846 case EINTR: break;
5847 default: return -1;
5848 }
5849 }
5850
5851 if (WIFEXITED(status)) {
5852 // The child exited normally; get its exit code.
5853 return WEXITSTATUS(status);
5854 } else if (WIFSIGNALED(status)) {
5855 // The child exited because of a signal
5856 // The best value to return is 0x80 + signal number,
5857 // because that is what all Unix shells do, and because
5858 // it allows callers to distinguish between process exit and
5859 // process death by signal.
5860 return 0x80 + WTERMSIG(status);
5861 } else {
5862 // Unknown exit code; pass it through
5863 return status;
5864 }
5865 }
5866 }
5867
5868 // Get the default path to the core file
5869 // Returns the length of the string
5870 int os::get_core_path(char* buffer, size_t bufferSize) {
5871 /*
5872 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5873 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5874 */
5875 const int core_pattern_len = 129;
5876 char core_pattern[core_pattern_len] = {0};
5877
5878 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5879 if (core_pattern_file == -1) {
5880 return -1;
5881 }
5882
5883 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5884 ::close(core_pattern_file);
5885 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5886 return -1;
5887 }
5888 if (core_pattern[ret-1] == '\n') {
5889 core_pattern[ret-1] = '\0';
5890 } else {
5891 core_pattern[ret] = '\0';
5892 }
5893
5894 // Replace the %p in the core pattern with the process id. NOTE: we do this
5895 // only if the pattern doesn't start with "|", and we support only one %p in
5896 // the pattern.
5897 char *pid_pos = strstr(core_pattern, "%p");
5898 const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : ""; // skip over the "%p"
5899 int written;
5900
5901 if (core_pattern[0] == '/') {
5902 if (pid_pos != NULL) {
5903 *pid_pos = '\0';
5904 written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern,
5905 current_process_id(), tail);
5906 } else {
5907 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5908 }
5909 } else {
5910 char cwd[PATH_MAX];
5911
5912 const char* p = get_current_directory(cwd, PATH_MAX);
5913 if (p == NULL) {
5914 return -1;
5915 }
5916
5917 if (core_pattern[0] == '|') {
5918 written = jio_snprintf(buffer, bufferSize,
5919 "\"%s\" (or dumping to %s/core.%d)",
5920 &core_pattern[1], p, current_process_id());
5921 } else if (pid_pos != NULL) {
5922 *pid_pos = '\0';
5923 written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern,
5924 current_process_id(), tail);
5925 } else {
5926 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5927 }
5928 }
5929
5930 if (written < 0) {
5931 return -1;
5932 }
5933
5934 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5935 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5936
5937 if (core_uses_pid_file != -1) {
5938 char core_uses_pid = 0;
5939 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5940 ::close(core_uses_pid_file);
5941
5942 if (core_uses_pid == '1') {
5943 jio_snprintf(buffer + written, bufferSize - written,
5944 ".%d", current_process_id());
5945 }
5946 }
5947 }
5948
5949 return strlen(buffer);
5950 }
5951
5952 bool os::start_debugging(char *buf, int buflen) {
5953 int len = (int)strlen(buf);
5954 char *p = &buf[len];
5955
5956 jio_snprintf(p, buflen-len,
5957 "\n\n"
5958 "Do you want to debug the problem?\n\n"
5959 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5960 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5961 "Otherwise, press RETURN to abort...",
5962 os::current_process_id(), os::current_process_id(),
5963 os::current_thread_id(), os::current_thread_id());
5964
5965 bool yes = os::message_box("Unexpected Error", buf);
5966
5967 if (yes) {
5968 // yes, user asked VM to launch debugger
5969 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5970 os::current_process_id(), os::current_process_id());
5971
5972 os::fork_and_exec(buf);
5973 yes = false;
5974 }
5975 return yes;
5976 }
5977
5978
5979 // Java/Compiler thread:
5980 //
5981 // Low memory addresses
5982 // P0 +------------------------+
5983 // | |\ Java thread created by VM does not have glibc
5984 // | glibc guard page | - guard page, attached Java thread usually has
5985 // | |/ 1 glibc guard page.
5986 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5987 // | |\
5988 // | HotSpot Guard Pages | - red, yellow and reserved pages
5989 // | |/
5990 // +------------------------+ JavaThread::stack_reserved_zone_base()
5991 // | |\
5992 // | Normal Stack | -
5993 // | |/
5994 // P2 +------------------------+ Thread::stack_base()
5995 //
5996 // Non-Java thread:
5997 //
5998 // Low memory addresses
5999 // P0 +------------------------+
6000 // | |\
6001 // | glibc guard page | - usually 1 page
6002 // | |/
6003 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6004 // | |\
6005 // | Normal Stack | -
6006 // | |/
6007 // P2 +------------------------+ Thread::stack_base()
6008 //
6009 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
6010 // returned from pthread_attr_getstack().
6011 // ** Due to NPTL implementation error, linux takes the glibc guard page out
6012 // of the stack size given in pthread_attr. We work around this for
6013 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
6014 //
6015 #ifndef ZERO
6016 static void current_stack_region(address * bottom, size_t * size) {
6017 if (os::is_primordial_thread()) {
6018 // primordial thread needs special handling because pthread_getattr_np()
6019 // may return bogus value.
6020 *bottom = os::Linux::initial_thread_stack_bottom();
6021 *size = os::Linux::initial_thread_stack_size();
6022 } else {
6023 pthread_attr_t attr;
6024
6025 int rslt = pthread_getattr_np(pthread_self(), &attr);
6026
6027 // JVM needs to know exact stack location, abort if it fails
6028 if (rslt != 0) {
6029 if (rslt == ENOMEM) {
6030 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
6031 } else {
6032 fatal("pthread_getattr_np failed with error = %d", rslt);
6033 }
6034 }
6035
6036 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
6037 fatal("Cannot locate current stack attributes!");
6038 }
6039
6040 // Work around NPTL stack guard error.
6041 size_t guard_size = 0;
6042 rslt = pthread_attr_getguardsize(&attr, &guard_size);
6043 if (rslt != 0) {
6044 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
6045 }
6046 *bottom += guard_size;
6047 *size -= guard_size;
6048
6049 pthread_attr_destroy(&attr);
6050
6051 }
6052 assert(os::current_stack_pointer() >= *bottom &&
6053 os::current_stack_pointer() < *bottom + *size, "just checking");
6054 }
6055
6056 address os::current_stack_base() {
6057 address bottom;
6058 size_t size;
6059 current_stack_region(&bottom, &size);
6060 return (bottom + size);
6061 }
6062
6063 size_t os::current_stack_size() {
6064 // This stack size includes the usable stack and HotSpot guard pages
6065 // (for the threads that have Hotspot guard pages).
6066 address bottom;
6067 size_t size;
6068 current_stack_region(&bottom, &size);
6069 return size;
6070 }
6071 #endif
6072
6073 static inline struct timespec get_mtime(const char* filename) {
6074 struct stat st;
6075 int ret = os::stat(filename, &st);
6076 assert(ret == 0, "failed to stat() file '%s': %s", filename, os::strerror(errno));
6077 return st.st_mtim;
6078 }
6079
6080 int os::compare_file_modified_times(const char* file1, const char* file2) {
6081 struct timespec filetime1 = get_mtime(file1);
6082 struct timespec filetime2 = get_mtime(file2);
6083 int diff = filetime1.tv_sec - filetime2.tv_sec;
6084 if (diff == 0) {
6085 return filetime1.tv_nsec - filetime2.tv_nsec;
6086 }
6087 return diff;
6088 }
6089
6090 /////////////// Unit tests ///////////////
6091
6092 #ifndef PRODUCT
6093
6094 class TestReserveMemorySpecial : AllStatic {
6095 public:
6096 static void small_page_write(void* addr, size_t size) {
6097 size_t page_size = os::vm_page_size();
6098
6099 char* end = (char*)addr + size;
6100 for (char* p = (char*)addr; p < end; p += page_size) {
6101 *p = 1;
6102 }
6103 }
6104
6105 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6106 if (!UseHugeTLBFS) {
6107 return;
6108 }
6109
6110 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6111
6112 if (addr != NULL) {
6113 small_page_write(addr, size);
6114
6115 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6116 }
6117 }
6118
6119 static void test_reserve_memory_special_huge_tlbfs_only() {
6120 if (!UseHugeTLBFS) {
6121 return;
6122 }
6123
6124 size_t lp = os::large_page_size();
6125
6126 for (size_t size = lp; size <= lp * 10; size += lp) {
6127 test_reserve_memory_special_huge_tlbfs_only(size);
6128 }
6129 }
6130
6131 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6132 size_t lp = os::large_page_size();
6133 size_t ag = os::vm_allocation_granularity();
6134
6135 // sizes to test
6136 const size_t sizes[] = {
6137 lp, lp + ag, lp + lp / 2, lp * 2,
6138 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6139 lp * 10, lp * 10 + lp / 2
6140 };
6141 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6142
6143 // For each size/alignment combination, we test three scenarios:
6144 // 1) with req_addr == NULL
6145 // 2) with a non-null req_addr at which we expect to successfully allocate
6146 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6147 // expect the allocation to either fail or to ignore req_addr
6148
6149 // Pre-allocate two areas; they shall be as large as the largest allocation
6150 // and aligned to the largest alignment we will be testing.
6151 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6152 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6153 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6154 -1, 0);
6155 assert(mapping1 != MAP_FAILED, "should work");
6156
6157 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6158 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6159 -1, 0);
6160 assert(mapping2 != MAP_FAILED, "should work");
6161
6162 // Unmap the first mapping, but leave the second mapping intact: the first
6163 // mapping will serve as a value for a "good" req_addr (case 2). The second
6164 // mapping, still intact, as "bad" req_addr (case 3).
6165 ::munmap(mapping1, mapping_size);
6166
6167 // Case 1
6168 for (int i = 0; i < num_sizes; i++) {
6169 const size_t size = sizes[i];
6170 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6171 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6172 if (p != NULL) {
6173 assert(is_aligned(p, alignment), "must be");
6174 small_page_write(p, size);
6175 os::Linux::release_memory_special_huge_tlbfs(p, size);
6176 }
6177 }
6178 }
6179
6180 // Case 2
6181 for (int i = 0; i < num_sizes; i++) {
6182 const size_t size = sizes[i];
6183 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6184 char* const req_addr = align_up(mapping1, alignment);
6185 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6186 if (p != NULL) {
6187 assert(p == req_addr, "must be");
6188 small_page_write(p, size);
6189 os::Linux::release_memory_special_huge_tlbfs(p, size);
6190 }
6191 }
6192 }
6193
6194 // Case 3
6195 for (int i = 0; i < num_sizes; i++) {
6196 const size_t size = sizes[i];
6197 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6198 char* const req_addr = align_up(mapping2, alignment);
6199 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6200 // as the area around req_addr contains already existing mappings, the API should always
6201 // return NULL (as per contract, it cannot return another address)
6202 assert(p == NULL, "must be");
6203 }
6204 }
6205
6206 ::munmap(mapping2, mapping_size);
6207
6208 }
6209
6210 static void test_reserve_memory_special_huge_tlbfs() {
6211 if (!UseHugeTLBFS) {
6212 return;
6213 }
6214
6215 test_reserve_memory_special_huge_tlbfs_only();
6216 test_reserve_memory_special_huge_tlbfs_mixed();
6217 }
6218
6219 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6220 if (!UseSHM) {
6221 return;
6222 }
6223
6224 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6225
6226 if (addr != NULL) {
6227 assert(is_aligned(addr, alignment), "Check");
6228 assert(is_aligned(addr, os::large_page_size()), "Check");
6229
6230 small_page_write(addr, size);
6231
6232 os::Linux::release_memory_special_shm(addr, size);
6233 }
6234 }
6235
6236 static void test_reserve_memory_special_shm() {
6237 size_t lp = os::large_page_size();
6238 size_t ag = os::vm_allocation_granularity();
6239
6240 for (size_t size = ag; size < lp * 3; size += ag) {
6241 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6242 test_reserve_memory_special_shm(size, alignment);
6243 }
6244 }
6245 }
6246
6247 static void test() {
6248 test_reserve_memory_special_huge_tlbfs();
6249 test_reserve_memory_special_shm();
6250 }
6251 };
6252
6253 void TestReserveMemorySpecial_test() {
6254 TestReserveMemorySpecial::test();
6255 }
6256
6257 #endif