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 <endian.h>
88 # include <errno.h>
89 # include <dlfcn.h>
90 # include <stdio.h>
91 # include <unistd.h>
92 # include <sys/resource.h>
93 # include <pthread.h>
94 # include <sys/stat.h>
95 # include <sys/time.h>
96 # include <sys/times.h>
97 # include <sys/utsname.h>
98 # include <sys/socket.h>
99 # include <sys/wait.h>
100 # include <pwd.h>
101 # include <poll.h>
102 # include <fcntl.h>
103 # include <string.h>
104 # include <syscall.h>
105 # include <sys/sysinfo.h>
106 # include <gnu/libc-version.h>
107 # include <sys/ipc.h>
108 # include <sys/shm.h>
109 # include <link.h>
110 # include <stdint.h>
111 # include <inttypes.h>
112 # include <sys/ioctl.h>
113
114 #ifndef _GNU_SOURCE
115 #define _GNU_SOURCE
116 #include <sched.h>
117 #undef _GNU_SOURCE
118 #else
119 #include <sched.h>
120 #endif
121
122 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
123 // getrusage() is prepared to handle the associated failure.
124 #ifndef RUSAGE_THREAD
125 #define RUSAGE_THREAD (1) /* only the calling thread */
126 #endif
127
128 #define MAX_PATH (2 * K)
129
130 #define MAX_SECS 100000000
131
132 // for timer info max values which include all bits
133 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
134
135 enum CoredumpFilterBit {
136 FILE_BACKED_PVT_BIT = 1 << 2,
137 FILE_BACKED_SHARED_BIT = 1 << 3,
138 LARGEPAGES_BIT = 1 << 6,
139 DAX_SHARED_BIT = 1 << 8
140 };
141
142 ////////////////////////////////////////////////////////////////////////////////
143 // global variables
144 julong os::Linux::_physical_memory = 0;
145
146 address os::Linux::_initial_thread_stack_bottom = NULL;
147 uintptr_t os::Linux::_initial_thread_stack_size = 0;
148
149 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
150 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
151 Mutex* os::Linux::_createThread_lock = NULL;
152 pthread_t os::Linux::_main_thread;
153 int os::Linux::_page_size = -1;
154 bool os::Linux::_supports_fast_thread_cpu_time = false;
155 uint32_t os::Linux::_os_version = 0;
156 const char * os::Linux::_glibc_version = NULL;
157 const char * os::Linux::_libpthread_version = NULL;
158
159 static jlong initial_time_count=0;
160
161 static int clock_tics_per_sec = 100;
162
163 // If the VM might have been created on the primordial thread, we need to resolve the
164 // primordial thread stack bounds and check if the current thread might be the
165 // primordial thread in places. If we know that the primordial thread is never used,
166 // such as when the VM was created by one of the standard java launchers, we can
167 // avoid this
168 static bool suppress_primordial_thread_resolution = false;
169
170 // For diagnostics to print a message once. see run_periodic_checks
171 static sigset_t check_signal_done;
172 static bool check_signals = true;
173
174 // Signal number used to suspend/resume a thread
175
176 // do not use any signal number less than SIGSEGV, see 4355769
177 static int SR_signum = SIGUSR2;
178 sigset_t SR_sigset;
179
180 // utility functions
181
182 static int SR_initialize();
183
184 julong os::available_memory() {
185 return Linux::available_memory();
186 }
187
188 julong os::Linux::available_memory() {
189 // values in struct sysinfo are "unsigned long"
190 struct sysinfo si;
191 julong avail_mem;
192
193 if (OSContainer::is_containerized()) {
194 jlong mem_limit, mem_usage;
195 if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
196 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
197 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
198 }
199 if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
200 log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
201 }
202 if (mem_limit > 0 && mem_usage > 0 ) {
203 avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
204 log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
205 return avail_mem;
206 }
207 }
208
209 sysinfo(&si);
210 avail_mem = (julong)si.freeram * si.mem_unit;
211 log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
212 return avail_mem;
213 }
214
215 julong os::physical_memory() {
216 jlong phys_mem = 0;
217 if (OSContainer::is_containerized()) {
218 jlong mem_limit;
219 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
220 log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
221 return mem_limit;
222 }
223 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
224 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
225 }
226
227 phys_mem = Linux::physical_memory();
228 log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
229 return phys_mem;
230 }
231
232 static uint64_t initial_total_ticks = 0;
233 static uint64_t initial_steal_ticks = 0;
234 static bool has_initial_tick_info = false;
235
236 static void next_line(FILE *f) {
237 int c;
238 do {
239 c = fgetc(f);
240 } while (c != '\n' && c != EOF);
241 }
242
243 bool os::Linux::get_tick_information(CPUPerfTicks* pticks, int which_logical_cpu) {
244 FILE* fh;
245 uint64_t userTicks, niceTicks, systemTicks, idleTicks;
246 // since at least kernel 2.6 : iowait: time waiting for I/O to complete
247 // irq: time servicing interrupts; softirq: time servicing softirqs
248 uint64_t iowTicks = 0, irqTicks = 0, sirqTicks= 0;
249 // steal (since kernel 2.6.11): time spent in other OS when running in a virtualized environment
250 uint64_t stealTicks = 0;
251 // guest (since kernel 2.6.24): time spent running a virtual CPU for guest OS under the
252 // control of the Linux kernel
253 uint64_t guestNiceTicks = 0;
254 int logical_cpu = -1;
255 const int required_tickinfo_count = (which_logical_cpu == -1) ? 4 : 5;
256 int n;
257
258 memset(pticks, 0, sizeof(CPUPerfTicks));
259
260 if ((fh = fopen("/proc/stat", "r")) == NULL) {
261 return false;
262 }
263
264 if (which_logical_cpu == -1) {
265 n = fscanf(fh, "cpu " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
266 UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
267 UINT64_FORMAT " " UINT64_FORMAT " ",
268 &userTicks, &niceTicks, &systemTicks, &idleTicks,
269 &iowTicks, &irqTicks, &sirqTicks,
270 &stealTicks, &guestNiceTicks);
271 } else {
272 // Move to next line
273 next_line(fh);
274
275 // find the line for requested cpu faster to just iterate linefeeds?
276 for (int i = 0; i < which_logical_cpu; i++) {
277 next_line(fh);
278 }
279
280 n = fscanf(fh, "cpu%u " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
281 UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
282 UINT64_FORMAT " " UINT64_FORMAT " ",
283 &logical_cpu, &userTicks, &niceTicks,
284 &systemTicks, &idleTicks, &iowTicks, &irqTicks, &sirqTicks,
285 &stealTicks, &guestNiceTicks);
286 }
287
288 fclose(fh);
289 if (n < required_tickinfo_count || logical_cpu != which_logical_cpu) {
290 return false;
291 }
292 pticks->used = userTicks + niceTicks;
293 pticks->usedKernel = systemTicks + irqTicks + sirqTicks;
294 pticks->total = userTicks + niceTicks + systemTicks + idleTicks +
295 iowTicks + irqTicks + sirqTicks + stealTicks + guestNiceTicks;
296
297 if (n > required_tickinfo_count + 3) {
298 pticks->steal = stealTicks;
299 pticks->has_steal_ticks = true;
300 } else {
301 pticks->steal = 0;
302 pticks->has_steal_ticks = false;
303 }
304
305 return true;
306 }
307
308 // Return true if user is running as root.
309
310 bool os::have_special_privileges() {
311 static bool init = false;
312 static bool privileges = false;
313 if (!init) {
314 privileges = (getuid() != geteuid()) || (getgid() != getegid());
315 init = true;
316 }
317 return privileges;
318 }
319
320
321 #ifndef SYS_gettid
322 // i386: 224, ia64: 1105, amd64: 186, sparc 143
323 #ifdef __ia64__
324 #define SYS_gettid 1105
325 #else
326 #ifdef __i386__
327 #define SYS_gettid 224
328 #else
329 #ifdef __amd64__
330 #define SYS_gettid 186
331 #else
332 #ifdef __sparc__
333 #define SYS_gettid 143
334 #else
335 #error define gettid for the arch
336 #endif
337 #endif
338 #endif
339 #endif
340 #endif
341
342
343 // pid_t gettid()
344 //
345 // Returns the kernel thread id of the currently running thread. Kernel
346 // thread id is used to access /proc.
347 pid_t os::Linux::gettid() {
348 int rslt = syscall(SYS_gettid);
349 assert(rslt != -1, "must be."); // old linuxthreads implementation?
350 return (pid_t)rslt;
351 }
352
353 // Most versions of linux have a bug where the number of processors are
354 // determined by looking at the /proc file system. In a chroot environment,
355 // the system call returns 1.
356 static bool unsafe_chroot_detected = false;
357 static const char *unstable_chroot_error = "/proc file system not found.\n"
358 "Java may be unstable running multithreaded in a chroot "
359 "environment on Linux when /proc filesystem is not mounted.";
360
361 void os::Linux::initialize_system_info() {
362 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
363 if (processor_count() == 1) {
364 pid_t pid = os::Linux::gettid();
365 char fname[32];
366 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
367 FILE *fp = fopen(fname, "r");
368 if (fp == NULL) {
369 unsafe_chroot_detected = true;
370 } else {
371 fclose(fp);
372 }
373 }
374 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
375 assert(processor_count() > 0, "linux error");
376 }
377
378 void os::init_system_properties_values() {
379 // The next steps are taken in the product version:
380 //
381 // Obtain the JAVA_HOME value from the location of libjvm.so.
382 // This library should be located at:
383 // <JAVA_HOME>/lib/{client|server}/libjvm.so.
384 //
385 // If "/jre/lib/" appears at the right place in the path, then we
386 // assume libjvm.so is installed in a JDK and we use this path.
387 //
388 // Otherwise exit with message: "Could not create the Java virtual machine."
389 //
390 // The following extra steps are taken in the debugging version:
391 //
392 // If "/jre/lib/" does NOT appear at the right place in the path
393 // instead of exit check for $JAVA_HOME environment variable.
394 //
395 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
396 // then we append a fake suffix "hotspot/libjvm.so" to this path so
397 // it looks like libjvm.so is installed there
398 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
399 //
400 // Otherwise exit.
401 //
402 // Important note: if the location of libjvm.so changes this
403 // code needs to be changed accordingly.
404
405 // See ld(1):
406 // The linker uses the following search paths to locate required
407 // shared libraries:
408 // 1: ...
409 // ...
410 // 7: The default directories, normally /lib and /usr/lib.
411 #ifndef OVERRIDE_LIBPATH
412 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
413 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
414 #else
415 #define DEFAULT_LIBPATH "/lib:/usr/lib"
416 #endif
417 #else
418 #define DEFAULT_LIBPATH OVERRIDE_LIBPATH
419 #endif
420
421 // Base path of extensions installed on the system.
422 #define SYS_EXT_DIR "/usr/java/packages"
423 #define EXTENSIONS_DIR "/lib/ext"
424
425 // Buffer that fits several sprintfs.
426 // Note that the space for the colon and the trailing null are provided
427 // by the nulls included by the sizeof operator.
428 const size_t bufsize =
429 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
430 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
431 char *buf = NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
432
433 // sysclasspath, java_home, dll_dir
434 {
435 char *pslash;
436 os::jvm_path(buf, bufsize);
437
438 // Found the full path to libjvm.so.
439 // Now cut the path to <java_home>/jre if we can.
440 pslash = strrchr(buf, '/');
441 if (pslash != NULL) {
442 *pslash = '\0'; // Get rid of /libjvm.so.
443 }
444 pslash = strrchr(buf, '/');
445 if (pslash != NULL) {
446 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
447 }
448 Arguments::set_dll_dir(buf);
449
450 if (pslash != NULL) {
451 pslash = strrchr(buf, '/');
452 if (pslash != NULL) {
453 *pslash = '\0'; // Get rid of /lib.
454 }
455 }
456 Arguments::set_java_home(buf);
457 if (!set_boot_path('/', ':')) {
458 vm_exit_during_initialization("Failed setting boot class path.", NULL);
459 }
460 }
461
462 // Where to look for native libraries.
463 //
464 // Note: Due to a legacy implementation, most of the library path
465 // is set in the launcher. This was to accomodate linking restrictions
466 // on legacy Linux implementations (which are no longer supported).
467 // Eventually, all the library path setting will be done here.
468 //
469 // However, to prevent the proliferation of improperly built native
470 // libraries, the new path component /usr/java/packages is added here.
471 // Eventually, all the library path setting will be done here.
472 {
473 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
474 // should always exist (until the legacy problem cited above is
475 // addressed).
476 const char *v = ::getenv("LD_LIBRARY_PATH");
477 const char *v_colon = ":";
478 if (v == NULL) { v = ""; v_colon = ""; }
479 // That's +1 for the colon and +1 for the trailing '\0'.
480 char *ld_library_path = NEW_C_HEAP_ARRAY(char,
481 strlen(v) + 1 +
482 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
483 mtInternal);
484 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
485 Arguments::set_library_path(ld_library_path);
486 FREE_C_HEAP_ARRAY(char, ld_library_path);
487 }
488
489 // Extensions directories.
490 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
491 Arguments::set_ext_dirs(buf);
492
493 FREE_C_HEAP_ARRAY(char, buf);
494
495 #undef DEFAULT_LIBPATH
496 #undef SYS_EXT_DIR
497 #undef EXTENSIONS_DIR
498 }
499
500 ////////////////////////////////////////////////////////////////////////////////
501 // breakpoint support
502
503 void os::breakpoint() {
504 BREAKPOINT;
505 }
506
507 extern "C" void breakpoint() {
508 // use debugger to set breakpoint here
509 }
510
511 ////////////////////////////////////////////////////////////////////////////////
512 // signal support
513
514 debug_only(static bool signal_sets_initialized = false);
515 static sigset_t unblocked_sigs, vm_sigs;
516
517 void os::Linux::signal_sets_init() {
518 // Should also have an assertion stating we are still single-threaded.
519 assert(!signal_sets_initialized, "Already initialized");
520 // Fill in signals that are necessarily unblocked for all threads in
521 // the VM. Currently, we unblock the following signals:
522 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
523 // by -Xrs (=ReduceSignalUsage));
524 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
525 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
526 // the dispositions or masks wrt these signals.
527 // Programs embedding the VM that want to use the above signals for their
528 // own purposes must, at this time, use the "-Xrs" option to prevent
529 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
530 // (See bug 4345157, and other related bugs).
531 // In reality, though, unblocking these signals is really a nop, since
532 // these signals are not blocked by default.
533 sigemptyset(&unblocked_sigs);
534 sigaddset(&unblocked_sigs, SIGILL);
535 sigaddset(&unblocked_sigs, SIGSEGV);
536 sigaddset(&unblocked_sigs, SIGBUS);
537 sigaddset(&unblocked_sigs, SIGFPE);
538 #if defined(PPC64)
539 sigaddset(&unblocked_sigs, SIGTRAP);
540 #endif
541 sigaddset(&unblocked_sigs, SR_signum);
542
543 if (!ReduceSignalUsage) {
544 if (!os::Posix::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
545 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
546 }
547 if (!os::Posix::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
548 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
549 }
550 if (!os::Posix::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
551 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
552 }
553 }
554 // Fill in signals that are blocked by all but the VM thread.
555 sigemptyset(&vm_sigs);
556 if (!ReduceSignalUsage) {
557 sigaddset(&vm_sigs, BREAK_SIGNAL);
558 }
559 debug_only(signal_sets_initialized = true);
560
561 }
562
563 // These are signals that are unblocked while a thread is running Java.
564 // (For some reason, they get blocked by default.)
565 sigset_t* os::Linux::unblocked_signals() {
566 assert(signal_sets_initialized, "Not initialized");
567 return &unblocked_sigs;
568 }
569
570 // These are the signals that are blocked while a (non-VM) thread is
571 // running Java. Only the VM thread handles these signals.
572 sigset_t* os::Linux::vm_signals() {
573 assert(signal_sets_initialized, "Not initialized");
574 return &vm_sigs;
575 }
576
577 void os::Linux::hotspot_sigmask(Thread* thread) {
578
579 //Save caller's signal mask before setting VM signal mask
580 sigset_t caller_sigmask;
581 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
582
583 OSThread* osthread = thread->osthread();
584 osthread->set_caller_sigmask(caller_sigmask);
585
586 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
587
588 if (!ReduceSignalUsage) {
589 if (thread->is_VM_thread()) {
590 // Only the VM thread handles BREAK_SIGNAL ...
591 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
592 } else {
593 // ... all other threads block BREAK_SIGNAL
594 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
595 }
596 }
597 }
598
599 //////////////////////////////////////////////////////////////////////////////
600 // detecting pthread library
601
602 void os::Linux::libpthread_init() {
603 // Save glibc and pthread version strings.
604 #if !defined(_CS_GNU_LIBC_VERSION) || \
605 !defined(_CS_GNU_LIBPTHREAD_VERSION)
606 #error "glibc too old (< 2.3.2)"
607 #endif
608
609 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
610 assert(n > 0, "cannot retrieve glibc version");
611 char *str = (char *)malloc(n, mtInternal);
612 confstr(_CS_GNU_LIBC_VERSION, str, n);
613 os::Linux::set_glibc_version(str);
614
615 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
616 assert(n > 0, "cannot retrieve pthread version");
617 str = (char *)malloc(n, mtInternal);
618 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
619 os::Linux::set_libpthread_version(str);
620 }
621
622 /////////////////////////////////////////////////////////////////////////////
623 // thread stack expansion
624
625 // os::Linux::manually_expand_stack() takes care of expanding the thread
626 // stack. Note that this is normally not needed: pthread stacks allocate
627 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
628 // committed. Therefore it is not necessary to expand the stack manually.
629 //
630 // Manually expanding the stack was historically needed on LinuxThreads
631 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
632 // it is kept to deal with very rare corner cases:
633 //
634 // For one, user may run the VM on an own implementation of threads
635 // whose stacks are - like the old LinuxThreads - implemented using
636 // mmap(MAP_GROWSDOWN).
637 //
638 // Also, this coding may be needed if the VM is running on the primordial
639 // thread. Normally we avoid running on the primordial thread; however,
640 // user may still invoke the VM on the primordial thread.
641 //
642 // The following historical comment describes the details about running
643 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
644
645
646 // Force Linux kernel to expand current thread stack. If "bottom" is close
647 // to the stack guard, caller should block all signals.
648 //
649 // MAP_GROWSDOWN:
650 // A special mmap() flag that is used to implement thread stacks. It tells
651 // kernel that the memory region should extend downwards when needed. This
652 // allows early versions of LinuxThreads to only mmap the first few pages
653 // when creating a new thread. Linux kernel will automatically expand thread
654 // stack as needed (on page faults).
655 //
656 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
657 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
658 // region, it's hard to tell if the fault is due to a legitimate stack
659 // access or because of reading/writing non-exist memory (e.g. buffer
660 // overrun). As a rule, if the fault happens below current stack pointer,
661 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
662 // application (see Linux kernel fault.c).
663 //
664 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
665 // stack overflow detection.
666 //
667 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
668 // not use MAP_GROWSDOWN.
669 //
670 // To get around the problem and allow stack banging on Linux, we need to
671 // manually expand thread stack after receiving the SIGSEGV.
672 //
673 // There are two ways to expand thread stack to address "bottom", we used
674 // both of them in JVM before 1.5:
675 // 1. adjust stack pointer first so that it is below "bottom", and then
676 // touch "bottom"
677 // 2. mmap() the page in question
678 //
679 // Now alternate signal stack is gone, it's harder to use 2. For instance,
680 // if current sp is already near the lower end of page 101, and we need to
681 // call mmap() to map page 100, it is possible that part of the mmap() frame
682 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
683 // That will destroy the mmap() frame and cause VM to crash.
684 //
685 // The following code works by adjusting sp first, then accessing the "bottom"
686 // page to force a page fault. Linux kernel will then automatically expand the
687 // stack mapping.
688 //
689 // _expand_stack_to() assumes its frame size is less than page size, which
690 // should always be true if the function is not inlined.
691
692 static void NOINLINE _expand_stack_to(address bottom) {
693 address sp;
694 size_t size;
695 volatile char *p;
696
697 // Adjust bottom to point to the largest address within the same page, it
698 // gives us a one-page buffer if alloca() allocates slightly more memory.
699 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
700 bottom += os::Linux::page_size() - 1;
701
702 // sp might be slightly above current stack pointer; if that's the case, we
703 // will alloca() a little more space than necessary, which is OK. Don't use
704 // os::current_stack_pointer(), as its result can be slightly below current
705 // stack pointer, causing us to not alloca enough to reach "bottom".
706 sp = (address)&sp;
707
708 if (sp > bottom) {
709 size = sp - bottom;
710 p = (volatile char *)alloca(size);
711 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
712 p[0] = '\0';
713 }
714 }
715
716 void os::Linux::expand_stack_to(address bottom) {
717 _expand_stack_to(bottom);
718 }
719
720 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
721 assert(t!=NULL, "just checking");
722 assert(t->osthread()->expanding_stack(), "expand should be set");
723 assert(t->stack_base() != NULL, "stack_base was not initialized");
724
725 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
726 sigset_t mask_all, old_sigset;
727 sigfillset(&mask_all);
728 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
729 _expand_stack_to(addr);
730 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
731 return true;
732 }
733 return false;
734 }
735
736 //////////////////////////////////////////////////////////////////////////////
737 // create new thread
738
739 // Thread start routine for all newly created threads
740 static void *thread_native_entry(Thread *thread) {
741
742 thread->record_stack_base_and_size();
743
744 // Try to randomize the cache line index of hot stack frames.
745 // This helps when threads of the same stack traces evict each other's
746 // cache lines. The threads can be either from the same JVM instance, or
747 // from different JVM instances. The benefit is especially true for
748 // processors with hyperthreading technology.
749 static int counter = 0;
750 int pid = os::current_process_id();
751 alloca(((pid ^ counter++) & 7) * 128);
752
753 thread->initialize_thread_current();
754
755 OSThread* osthread = thread->osthread();
756 Monitor* sync = osthread->startThread_lock();
757
758 osthread->set_thread_id(os::current_thread_id());
759
760 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
761 os::current_thread_id(), (uintx) pthread_self());
762
763 if (UseNUMA) {
764 int lgrp_id = os::numa_get_group_id();
765 if (lgrp_id != -1) {
766 thread->set_lgrp_id(lgrp_id);
767 }
768 }
769 // initialize signal mask for this thread
770 os::Linux::hotspot_sigmask(thread);
771
772 // initialize floating point control register
773 os::Linux::init_thread_fpu_state();
774
775 // handshaking with parent thread
776 {
777 MutexLocker ml(sync, Mutex::_no_safepoint_check_flag);
778
779 // notify parent thread
780 osthread->set_state(INITIALIZED);
781 sync->notify_all();
782
783 // wait until os::start_thread()
784 while (osthread->get_state() == INITIALIZED) {
785 sync->wait_without_safepoint_check();
786 }
787 }
788
789 assert(osthread->pthread_id() != 0, "pthread_id was not set as expected");
790
791 // call one more level start routine
792 thread->call_run();
793
794 // Note: at this point the thread object may already have deleted itself.
795 // Prevent dereferencing it from here on out.
796 thread = NULL;
797
798 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
799 os::current_thread_id(), (uintx) pthread_self());
800
801 return 0;
802 }
803
804 // On Linux, glibc places static TLS blocks (for __thread variables) on
805 // the thread stack. This decreases the stack size actually available
806 // to threads.
807 //
808 // For large static TLS sizes, this may cause threads to malfunction due
809 // to insufficient stack space. This is a well-known issue in glibc:
810 // http://sourceware.org/bugzilla/show_bug.cgi?id=11787.
811 //
812 // As a workaround, we call a private but assumed-stable glibc function,
813 // __pthread_get_minstack() to obtain the minstack size and derive the
814 // static TLS size from it. We then increase the user requested stack
815 // size by this TLS size.
816 //
817 // Due to compatibility concerns, this size adjustment is opt-in and
818 // controlled via AdjustStackSizeForTLS.
819 typedef size_t (*GetMinStack)(const pthread_attr_t *attr);
820
821 GetMinStack _get_minstack_func = NULL;
822
823 static void get_minstack_init() {
824 _get_minstack_func =
825 (GetMinStack)dlsym(RTLD_DEFAULT, "__pthread_get_minstack");
826 log_info(os, thread)("Lookup of __pthread_get_minstack %s",
827 _get_minstack_func == NULL ? "failed" : "succeeded");
828 }
829
830 // Returns the size of the static TLS area glibc puts on thread stacks.
831 // The value is cached on first use, which occurs when the first thread
832 // is created during VM initialization.
833 static size_t get_static_tls_area_size(const pthread_attr_t *attr) {
834 size_t tls_size = 0;
835 if (_get_minstack_func != NULL) {
836 // Obtain the pthread minstack size by calling __pthread_get_minstack.
837 size_t minstack_size = _get_minstack_func(attr);
838
839 // Remove non-TLS area size included in minstack size returned
840 // by __pthread_get_minstack() to get the static TLS size.
841 // In glibc before 2.27, minstack size includes guard_size.
842 // In glibc 2.27 and later, guard_size is automatically added
843 // to the stack size by pthread_create and is no longer included
844 // in minstack size. In both cases, the guard_size is taken into
845 // account, so there is no need to adjust the result for that.
846 //
847 // Although __pthread_get_minstack() is a private glibc function,
848 // it is expected to have a stable behavior across future glibc
849 // versions while glibc still allocates the static TLS blocks off
850 // the stack. Following is glibc 2.28 __pthread_get_minstack():
851 //
852 // size_t
853 // __pthread_get_minstack (const pthread_attr_t *attr)
854 // {
855 // return GLRO(dl_pagesize) + __static_tls_size + PTHREAD_STACK_MIN;
856 // }
857 //
858 //
859 // The following 'minstack_size > os::vm_page_size() + PTHREAD_STACK_MIN'
860 // if check is done for precaution.
861 if (minstack_size > (size_t)os::vm_page_size() + PTHREAD_STACK_MIN) {
862 tls_size = minstack_size - os::vm_page_size() - PTHREAD_STACK_MIN;
863 }
864 }
865
866 log_info(os, thread)("Stack size adjustment for TLS is " SIZE_FORMAT,
867 tls_size);
868 return tls_size;
869 }
870
871 bool os::create_thread(Thread* thread, ThreadType thr_type,
872 size_t req_stack_size) {
873 assert(thread->osthread() == NULL, "caller responsible");
874
875 // Allocate the OSThread object
876 OSThread* osthread = new OSThread(NULL, NULL);
877 if (osthread == NULL) {
878 return false;
879 }
880
881 // set the correct thread state
882 osthread->set_thread_type(thr_type);
883
884 // Initial state is ALLOCATED but not INITIALIZED
885 osthread->set_state(ALLOCATED);
886
887 thread->set_osthread(osthread);
888
889 // init thread attributes
890 pthread_attr_t attr;
891 pthread_attr_init(&attr);
892 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
893
894 // Calculate stack size if it's not specified by caller.
895 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
896 // In glibc versions prior to 2.7 the guard size mechanism
897 // is not implemented properly. The posix standard requires adding
898 // the size of the guard pages to the stack size, instead Linux
899 // takes the space out of 'stacksize'. Thus we adapt the requested
900 // stack_size by the size of the guard pages to mimick proper
901 // behaviour. However, be careful not to end up with a size
902 // of zero due to overflow. Don't add the guard page in that case.
903 size_t guard_size = os::Linux::default_guard_size(thr_type);
904 // Configure glibc guard page. Must happen before calling
905 // get_static_tls_area_size(), which uses the guard_size.
906 pthread_attr_setguardsize(&attr, guard_size);
907
908 size_t stack_adjust_size = 0;
909 if (AdjustStackSizeForTLS) {
910 // Adjust the stack_size for on-stack TLS - see get_static_tls_area_size().
911 stack_adjust_size += get_static_tls_area_size(&attr);
912 } else {
913 stack_adjust_size += guard_size;
914 }
915
916 stack_adjust_size = align_up(stack_adjust_size, os::vm_page_size());
917 if (stack_size <= SIZE_MAX - stack_adjust_size) {
918 stack_size += stack_adjust_size;
919 }
920 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
921
922 int status = pthread_attr_setstacksize(&attr, stack_size);
923 assert_status(status == 0, status, "pthread_attr_setstacksize");
924
925 ThreadState state;
926
927 {
928 pthread_t tid;
929 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
930
931 char buf[64];
932 if (ret == 0) {
933 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
934 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
935 } else {
936 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
937 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
938 // Log some OS information which might explain why creating the thread failed.
939 log_info(os, thread)("Number of threads approx. running in the VM: %d", Threads::number_of_threads());
940 LogStream st(Log(os, thread)::info());
941 os::Posix::print_rlimit_info(&st);
942 os::print_memory_info(&st);
943 os::Linux::print_proc_sys_info(&st);
944 os::Linux::print_container_info(&st);
945 }
946
947 pthread_attr_destroy(&attr);
948
949 if (ret != 0) {
950 // Need to clean up stuff we've allocated so far
951 thread->set_osthread(NULL);
952 delete osthread;
953 return false;
954 }
955
956 // Store pthread info into the OSThread
957 osthread->set_pthread_id(tid);
958
959 // Wait until child thread is either initialized or aborted
960 {
961 Monitor* sync_with_child = osthread->startThread_lock();
962 MutexLocker ml(sync_with_child, Mutex::_no_safepoint_check_flag);
963 while ((state = osthread->get_state()) == ALLOCATED) {
964 sync_with_child->wait_without_safepoint_check();
965 }
966 }
967 }
968
969 // Aborted due to thread limit being reached
970 if (state == ZOMBIE) {
971 thread->set_osthread(NULL);
972 delete osthread;
973 return false;
974 }
975
976 // The thread is returned suspended (in state INITIALIZED),
977 // and is started higher up in the call chain
978 assert(state == INITIALIZED, "race condition");
979 return true;
980 }
981
982 /////////////////////////////////////////////////////////////////////////////
983 // attach existing thread
984
985 // bootstrap the main thread
986 bool os::create_main_thread(JavaThread* thread) {
987 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
988 return create_attached_thread(thread);
989 }
990
991 bool os::create_attached_thread(JavaThread* thread) {
992 #ifdef ASSERT
993 thread->verify_not_published();
994 #endif
995
996 // Allocate the OSThread object
997 OSThread* osthread = new OSThread(NULL, NULL);
998
999 if (osthread == NULL) {
1000 return false;
1001 }
1002
1003 // Store pthread info into the OSThread
1004 osthread->set_thread_id(os::Linux::gettid());
1005 osthread->set_pthread_id(::pthread_self());
1006
1007 // initialize floating point control register
1008 os::Linux::init_thread_fpu_state();
1009
1010 // Initial thread state is RUNNABLE
1011 osthread->set_state(RUNNABLE);
1012
1013 thread->set_osthread(osthread);
1014
1015 if (UseNUMA) {
1016 int lgrp_id = os::numa_get_group_id();
1017 if (lgrp_id != -1) {
1018 thread->set_lgrp_id(lgrp_id);
1019 }
1020 }
1021
1022 if (os::is_primordial_thread()) {
1023 // If current thread is primordial thread, its stack is mapped on demand,
1024 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1025 // the entire stack region to avoid SEGV in stack banging.
1026 // It is also useful to get around the heap-stack-gap problem on SuSE
1027 // kernel (see 4821821 for details). We first expand stack to the top
1028 // of yellow zone, then enable stack yellow zone (order is significant,
1029 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1030 // is no gap between the last two virtual memory regions.
1031
1032 JavaThread *jt = (JavaThread *)thread;
1033 address addr = jt->stack_reserved_zone_base();
1034 assert(addr != NULL, "initialization problem?");
1035 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1036
1037 osthread->set_expanding_stack();
1038 os::Linux::manually_expand_stack(jt, addr);
1039 osthread->clear_expanding_stack();
1040 }
1041
1042 // initialize signal mask for this thread
1043 // and save the caller's signal mask
1044 os::Linux::hotspot_sigmask(thread);
1045
1046 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
1047 os::current_thread_id(), (uintx) pthread_self());
1048
1049 return true;
1050 }
1051
1052 void os::pd_start_thread(Thread* thread) {
1053 OSThread * osthread = thread->osthread();
1054 assert(osthread->get_state() != INITIALIZED, "just checking");
1055 Monitor* sync_with_child = osthread->startThread_lock();
1056 MutexLocker ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1057 sync_with_child->notify();
1058 }
1059
1060 // Free Linux resources related to the OSThread
1061 void os::free_thread(OSThread* osthread) {
1062 assert(osthread != NULL, "osthread not set");
1063
1064 // We are told to free resources of the argument thread,
1065 // but we can only really operate on the current thread.
1066 assert(Thread::current()->osthread() == osthread,
1067 "os::free_thread but not current thread");
1068
1069 #ifdef ASSERT
1070 sigset_t current;
1071 sigemptyset(¤t);
1072 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
1073 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
1074 #endif
1075
1076 // Restore caller's signal mask
1077 sigset_t sigmask = osthread->caller_sigmask();
1078 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1079
1080 delete osthread;
1081 }
1082
1083 //////////////////////////////////////////////////////////////////////////////
1084 // primordial thread
1085
1086 // Check if current thread is the primordial thread, similar to Solaris thr_main.
1087 bool os::is_primordial_thread(void) {
1088 if (suppress_primordial_thread_resolution) {
1089 return false;
1090 }
1091 char dummy;
1092 // If called before init complete, thread stack bottom will be null.
1093 // Can be called if fatal error occurs before initialization.
1094 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
1095 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
1096 os::Linux::initial_thread_stack_size() != 0,
1097 "os::init did not locate primordial thread's stack region");
1098 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
1099 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
1100 os::Linux::initial_thread_stack_size()) {
1101 return true;
1102 } else {
1103 return false;
1104 }
1105 }
1106
1107 // Find the virtual memory area that contains addr
1108 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1109 FILE *fp = fopen("/proc/self/maps", "r");
1110 if (fp) {
1111 address low, high;
1112 while (!feof(fp)) {
1113 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1114 if (low <= addr && addr < high) {
1115 if (vma_low) *vma_low = low;
1116 if (vma_high) *vma_high = high;
1117 fclose(fp);
1118 return true;
1119 }
1120 }
1121 for (;;) {
1122 int ch = fgetc(fp);
1123 if (ch == EOF || ch == (int)'\n') break;
1124 }
1125 }
1126 fclose(fp);
1127 }
1128 return false;
1129 }
1130
1131 // Locate primordial thread stack. This special handling of primordial thread stack
1132 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1133 // bogus value for the primordial process thread. While the launcher has created
1134 // the VM in a new thread since JDK 6, we still have to allow for the use of the
1135 // JNI invocation API from a primordial thread.
1136 void os::Linux::capture_initial_stack(size_t max_size) {
1137
1138 // max_size is either 0 (which means accept OS default for thread stacks) or
1139 // a user-specified value known to be at least the minimum needed. If we
1140 // are actually on the primordial thread we can make it appear that we have a
1141 // smaller max_size stack by inserting the guard pages at that location. But we
1142 // cannot do anything to emulate a larger stack than what has been provided by
1143 // the OS or threading library. In fact if we try to use a stack greater than
1144 // what is set by rlimit then we will crash the hosting process.
1145
1146 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
1147 // If this is "unlimited" then it will be a huge value.
1148 struct rlimit rlim;
1149 getrlimit(RLIMIT_STACK, &rlim);
1150 size_t stack_size = rlim.rlim_cur;
1151
1152 // 6308388: a bug in ld.so will relocate its own .data section to the
1153 // lower end of primordial stack; reduce ulimit -s value a little bit
1154 // so we won't install guard page on ld.so's data section.
1155 // But ensure we don't underflow the stack size - allow 1 page spare
1156 if (stack_size >= (size_t)(3 * page_size())) {
1157 stack_size -= 2 * page_size();
1158 }
1159
1160 // Try to figure out where the stack base (top) is. This is harder.
1161 //
1162 // When an application is started, glibc saves the initial stack pointer in
1163 // a global variable "__libc_stack_end", which is then used by system
1164 // libraries. __libc_stack_end should be pretty close to stack top. The
1165 // variable is available since the very early days. However, because it is
1166 // a private interface, it could disappear in the future.
1167 //
1168 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1169 // to __libc_stack_end, it is very close to stack top, but isn't the real
1170 // stack top. Note that /proc may not exist if VM is running as a chroot
1171 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1172 // /proc/<pid>/stat could change in the future (though unlikely).
1173 //
1174 // We try __libc_stack_end first. If that doesn't work, look for
1175 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1176 // as a hint, which should work well in most cases.
1177
1178 uintptr_t stack_start;
1179
1180 // try __libc_stack_end first
1181 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1182 if (p && *p) {
1183 stack_start = *p;
1184 } else {
1185 // see if we can get the start_stack field from /proc/self/stat
1186 FILE *fp;
1187 int pid;
1188 char state;
1189 int ppid;
1190 int pgrp;
1191 int session;
1192 int nr;
1193 int tpgrp;
1194 unsigned long flags;
1195 unsigned long minflt;
1196 unsigned long cminflt;
1197 unsigned long majflt;
1198 unsigned long cmajflt;
1199 unsigned long utime;
1200 unsigned long stime;
1201 long cutime;
1202 long cstime;
1203 long prio;
1204 long nice;
1205 long junk;
1206 long it_real;
1207 uintptr_t start;
1208 uintptr_t vsize;
1209 intptr_t rss;
1210 uintptr_t rsslim;
1211 uintptr_t scodes;
1212 uintptr_t ecode;
1213 int i;
1214
1215 // Figure what the primordial thread stack base is. Code is inspired
1216 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1217 // followed by command name surrounded by parentheses, state, etc.
1218 char stat[2048];
1219 int statlen;
1220
1221 fp = fopen("/proc/self/stat", "r");
1222 if (fp) {
1223 statlen = fread(stat, 1, 2047, fp);
1224 stat[statlen] = '\0';
1225 fclose(fp);
1226
1227 // Skip pid and the command string. Note that we could be dealing with
1228 // weird command names, e.g. user could decide to rename java launcher
1229 // to "java 1.4.2 :)", then the stat file would look like
1230 // 1234 (java 1.4.2 :)) R ... ...
1231 // We don't really need to know the command string, just find the last
1232 // occurrence of ")" and then start parsing from there. See bug 4726580.
1233 char * s = strrchr(stat, ')');
1234
1235 i = 0;
1236 if (s) {
1237 // Skip blank chars
1238 do { s++; } while (s && isspace(*s));
1239
1240 #define _UFM UINTX_FORMAT
1241 #define _DFM INTX_FORMAT
1242
1243 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1244 // 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
1245 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,
1246 &state, // 3 %c
1247 &ppid, // 4 %d
1248 &pgrp, // 5 %d
1249 &session, // 6 %d
1250 &nr, // 7 %d
1251 &tpgrp, // 8 %d
1252 &flags, // 9 %lu
1253 &minflt, // 10 %lu
1254 &cminflt, // 11 %lu
1255 &majflt, // 12 %lu
1256 &cmajflt, // 13 %lu
1257 &utime, // 14 %lu
1258 &stime, // 15 %lu
1259 &cutime, // 16 %ld
1260 &cstime, // 17 %ld
1261 &prio, // 18 %ld
1262 &nice, // 19 %ld
1263 &junk, // 20 %ld
1264 &it_real, // 21 %ld
1265 &start, // 22 UINTX_FORMAT
1266 &vsize, // 23 UINTX_FORMAT
1267 &rss, // 24 INTX_FORMAT
1268 &rsslim, // 25 UINTX_FORMAT
1269 &scodes, // 26 UINTX_FORMAT
1270 &ecode, // 27 UINTX_FORMAT
1271 &stack_start); // 28 UINTX_FORMAT
1272 }
1273
1274 #undef _UFM
1275 #undef _DFM
1276
1277 if (i != 28 - 2) {
1278 assert(false, "Bad conversion from /proc/self/stat");
1279 // product mode - assume we are the primordial thread, good luck in the
1280 // embedded case.
1281 warning("Can't detect primordial thread stack location - bad conversion");
1282 stack_start = (uintptr_t) &rlim;
1283 }
1284 } else {
1285 // For some reason we can't open /proc/self/stat (for example, running on
1286 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1287 // most cases, so don't abort:
1288 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1289 stack_start = (uintptr_t) &rlim;
1290 }
1291 }
1292
1293 // Now we have a pointer (stack_start) very close to the stack top, the
1294 // next thing to do is to figure out the exact location of stack top. We
1295 // can find out the virtual memory area that contains stack_start by
1296 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1297 // and its upper limit is the real stack top. (again, this would fail if
1298 // running inside chroot, because /proc may not exist.)
1299
1300 uintptr_t stack_top;
1301 address low, high;
1302 if (find_vma((address)stack_start, &low, &high)) {
1303 // success, "high" is the true stack top. (ignore "low", because initial
1304 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1305 stack_top = (uintptr_t)high;
1306 } else {
1307 // failed, likely because /proc/self/maps does not exist
1308 warning("Can't detect primordial thread stack location - find_vma failed");
1309 // best effort: stack_start is normally within a few pages below the real
1310 // stack top, use it as stack top, and reduce stack size so we won't put
1311 // guard page outside stack.
1312 stack_top = stack_start;
1313 stack_size -= 16 * page_size();
1314 }
1315
1316 // stack_top could be partially down the page so align it
1317 stack_top = align_up(stack_top, page_size());
1318
1319 // Allowed stack value is minimum of max_size and what we derived from rlimit
1320 if (max_size > 0) {
1321 _initial_thread_stack_size = MIN2(max_size, stack_size);
1322 } else {
1323 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1324 // clamp it at 8MB as we do on Solaris
1325 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1326 }
1327 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1328 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1329
1330 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1331
1332 if (log_is_enabled(Info, os, thread)) {
1333 // See if we seem to be on primordial process thread
1334 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1335 uintptr_t(&rlim) < stack_top;
1336
1337 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1338 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1339 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1340 stack_top, intptr_t(_initial_thread_stack_bottom));
1341 }
1342 }
1343
1344 ////////////////////////////////////////////////////////////////////////////////
1345 // time support
1346
1347 #ifndef SUPPORTS_CLOCK_MONOTONIC
1348 #error "Build platform doesn't support clock_gettime and related functionality"
1349 #endif
1350
1351 // Time since start-up in seconds to a fine granularity.
1352 // Used by VMSelfDestructTimer and the MemProfiler.
1353 double os::elapsedTime() {
1354
1355 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1356 }
1357
1358 jlong os::elapsed_counter() {
1359 return javaTimeNanos() - initial_time_count;
1360 }
1361
1362 jlong os::elapsed_frequency() {
1363 return NANOSECS_PER_SEC; // nanosecond resolution
1364 }
1365
1366 bool os::supports_vtime() { return true; }
1367
1368 double os::elapsedVTime() {
1369 struct rusage usage;
1370 int retval = getrusage(RUSAGE_THREAD, &usage);
1371 if (retval == 0) {
1372 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);
1373 } else {
1374 // better than nothing, but not much
1375 return elapsedTime();
1376 }
1377 }
1378
1379 jlong os::javaTimeMillis() {
1380 timeval time;
1381 int status = gettimeofday(&time, NULL);
1382 assert(status != -1, "linux error");
1383 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1384 }
1385
1386 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1387 timeval time;
1388 int status = gettimeofday(&time, NULL);
1389 assert(status != -1, "linux error");
1390 seconds = jlong(time.tv_sec);
1391 nanos = jlong(time.tv_usec) * 1000;
1392 }
1393
1394 void os::Linux::fast_thread_clock_init() {
1395 if (!UseLinuxPosixThreadCPUClocks) {
1396 return;
1397 }
1398 clockid_t clockid;
1399 struct timespec tp;
1400 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1401 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1402
1403 // Switch to using fast clocks for thread cpu time if
1404 // the clock_getres() returns 0 error code.
1405 // Note, that some kernels may support the current thread
1406 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1407 // returned by the pthread_getcpuclockid().
1408 // If the fast Posix clocks are supported then the clock_getres()
1409 // must return at least tp.tv_sec == 0 which means a resolution
1410 // better than 1 sec. This is extra check for reliability.
1411
1412 if (pthread_getcpuclockid_func &&
1413 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1414 os::Posix::clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1415 _supports_fast_thread_cpu_time = true;
1416 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1417 }
1418 }
1419
1420 jlong os::javaTimeNanos() {
1421 if (os::supports_monotonic_clock()) {
1422 struct timespec tp;
1423 int status = os::Posix::clock_gettime(CLOCK_MONOTONIC, &tp);
1424 assert(status == 0, "gettime error");
1425 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1426 return result;
1427 } else {
1428 timeval time;
1429 int status = gettimeofday(&time, NULL);
1430 assert(status != -1, "linux error");
1431 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1432 return 1000 * usecs;
1433 }
1434 }
1435
1436 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1437 if (os::supports_monotonic_clock()) {
1438 info_ptr->max_value = ALL_64_BITS;
1439
1440 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1441 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1442 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1443 } else {
1444 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1445 info_ptr->max_value = ALL_64_BITS;
1446
1447 // gettimeofday is a real time clock so it skips
1448 info_ptr->may_skip_backward = true;
1449 info_ptr->may_skip_forward = true;
1450 }
1451
1452 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1453 }
1454
1455 // Return the real, user, and system times in seconds from an
1456 // arbitrary fixed point in the past.
1457 bool os::getTimesSecs(double* process_real_time,
1458 double* process_user_time,
1459 double* process_system_time) {
1460 struct tms ticks;
1461 clock_t real_ticks = times(&ticks);
1462
1463 if (real_ticks == (clock_t) (-1)) {
1464 return false;
1465 } else {
1466 double ticks_per_second = (double) clock_tics_per_sec;
1467 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1468 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1469 *process_real_time = ((double) real_ticks) / ticks_per_second;
1470
1471 return true;
1472 }
1473 }
1474
1475
1476 char * os::local_time_string(char *buf, size_t buflen) {
1477 struct tm t;
1478 time_t long_time;
1479 time(&long_time);
1480 localtime_r(&long_time, &t);
1481 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1482 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1483 t.tm_hour, t.tm_min, t.tm_sec);
1484 return buf;
1485 }
1486
1487 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1488 return localtime_r(clock, res);
1489 }
1490
1491 ////////////////////////////////////////////////////////////////////////////////
1492 // runtime exit support
1493
1494 // Note: os::shutdown() might be called very early during initialization, or
1495 // called from signal handler. Before adding something to os::shutdown(), make
1496 // sure it is async-safe and can handle partially initialized VM.
1497 void os::shutdown() {
1498
1499 // allow PerfMemory to attempt cleanup of any persistent resources
1500 perfMemory_exit();
1501
1502 // needs to remove object in file system
1503 AttachListener::abort();
1504
1505 // flush buffered output, finish log files
1506 ostream_abort();
1507
1508 // Check for abort hook
1509 abort_hook_t abort_hook = Arguments::abort_hook();
1510 if (abort_hook != NULL) {
1511 abort_hook();
1512 }
1513
1514 }
1515
1516 // Note: os::abort() might be called very early during initialization, or
1517 // called from signal handler. Before adding something to os::abort(), make
1518 // sure it is async-safe and can handle partially initialized VM.
1519 void os::abort(bool dump_core, void* siginfo, const void* context) {
1520 os::shutdown();
1521 if (dump_core) {
1522 if (DumpPrivateMappingsInCore) {
1523 ClassLoader::close_jrt_image();
1524 }
1525 #ifndef PRODUCT
1526 fdStream out(defaultStream::output_fd());
1527 out.print_raw("Current thread is ");
1528 char buf[16];
1529 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1530 out.print_raw_cr(buf);
1531 out.print_raw_cr("Dumping core ...");
1532 #endif
1533 ::abort(); // dump core
1534 }
1535
1536 ::exit(1);
1537 }
1538
1539 // Die immediately, no exit hook, no abort hook, no cleanup.
1540 // Dump a core file, if possible, for debugging.
1541 void os::die() {
1542 if (TestUnresponsiveErrorHandler && !CreateCoredumpOnCrash) {
1543 // For TimeoutInErrorHandlingTest.java, we just kill the VM
1544 // and don't take the time to generate a core file.
1545 os::signal_raise(SIGKILL);
1546 } else {
1547 ::abort();
1548 }
1549 }
1550
1551 // thread_id is kernel thread id (similar to Solaris LWP id)
1552 intx os::current_thread_id() { return os::Linux::gettid(); }
1553 int os::current_process_id() {
1554 return ::getpid();
1555 }
1556
1557 // DLL functions
1558
1559 const char* os::dll_file_extension() { return ".so"; }
1560
1561 // This must be hard coded because it's the system's temporary
1562 // directory not the java application's temp directory, ala java.io.tmpdir.
1563 const char* os::get_temp_directory() { return "/tmp"; }
1564
1565 static bool file_exists(const char* filename) {
1566 struct stat statbuf;
1567 if (filename == NULL || strlen(filename) == 0) {
1568 return false;
1569 }
1570 return os::stat(filename, &statbuf) == 0;
1571 }
1572
1573 // check if addr is inside libjvm.so
1574 bool os::address_is_in_vm(address addr) {
1575 static address libjvm_base_addr;
1576 Dl_info dlinfo;
1577
1578 if (libjvm_base_addr == NULL) {
1579 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1580 libjvm_base_addr = (address)dlinfo.dli_fbase;
1581 }
1582 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1583 }
1584
1585 if (dladdr((void *)addr, &dlinfo) != 0) {
1586 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1587 }
1588
1589 return false;
1590 }
1591
1592 bool os::dll_address_to_function_name(address addr, char *buf,
1593 int buflen, int *offset,
1594 bool demangle) {
1595 // buf is not optional, but offset is optional
1596 assert(buf != NULL, "sanity check");
1597
1598 Dl_info dlinfo;
1599
1600 if (dladdr((void*)addr, &dlinfo) != 0) {
1601 // see if we have a matching symbol
1602 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1603 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1604 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1605 }
1606 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1607 return true;
1608 }
1609 // no matching symbol so try for just file info
1610 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1611 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1612 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1613 return true;
1614 }
1615 }
1616 }
1617
1618 buf[0] = '\0';
1619 if (offset != NULL) *offset = -1;
1620 return false;
1621 }
1622
1623 struct _address_to_library_name {
1624 address addr; // input : memory address
1625 size_t buflen; // size of fname
1626 char* fname; // output: library name
1627 address base; // library base addr
1628 };
1629
1630 static int address_to_library_name_callback(struct dl_phdr_info *info,
1631 size_t size, void *data) {
1632 int i;
1633 bool found = false;
1634 address libbase = NULL;
1635 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1636
1637 // iterate through all loadable segments
1638 for (i = 0; i < info->dlpi_phnum; i++) {
1639 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1640 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1641 // base address of a library is the lowest address of its loaded
1642 // segments.
1643 if (libbase == NULL || libbase > segbase) {
1644 libbase = segbase;
1645 }
1646 // see if 'addr' is within current segment
1647 if (segbase <= d->addr &&
1648 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1649 found = true;
1650 }
1651 }
1652 }
1653
1654 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1655 // so dll_address_to_library_name() can fall through to use dladdr() which
1656 // can figure out executable name from argv[0].
1657 if (found && info->dlpi_name && info->dlpi_name[0]) {
1658 d->base = libbase;
1659 if (d->fname) {
1660 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1661 }
1662 return 1;
1663 }
1664 return 0;
1665 }
1666
1667 bool os::dll_address_to_library_name(address addr, char* buf,
1668 int buflen, int* offset) {
1669 // buf is not optional, but offset is optional
1670 assert(buf != NULL, "sanity check");
1671
1672 Dl_info dlinfo;
1673 struct _address_to_library_name data;
1674
1675 // There is a bug in old glibc dladdr() implementation that it could resolve
1676 // to wrong library name if the .so file has a base address != NULL. Here
1677 // we iterate through the program headers of all loaded libraries to find
1678 // out which library 'addr' really belongs to. This workaround can be
1679 // removed once the minimum requirement for glibc is moved to 2.3.x.
1680 data.addr = addr;
1681 data.fname = buf;
1682 data.buflen = buflen;
1683 data.base = NULL;
1684 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1685
1686 if (rslt) {
1687 // buf already contains library name
1688 if (offset) *offset = addr - data.base;
1689 return true;
1690 }
1691 if (dladdr((void*)addr, &dlinfo) != 0) {
1692 if (dlinfo.dli_fname != NULL) {
1693 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1694 }
1695 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1696 *offset = addr - (address)dlinfo.dli_fbase;
1697 }
1698 return true;
1699 }
1700
1701 buf[0] = '\0';
1702 if (offset) *offset = -1;
1703 return false;
1704 }
1705
1706 // Loads .dll/.so and
1707 // in case of error it checks if .dll/.so was built for the
1708 // same architecture as Hotspot is running on
1709
1710
1711 // Remember the stack's state. The Linux dynamic linker will change
1712 // the stack to 'executable' at most once, so we must safepoint only once.
1713 bool os::Linux::_stack_is_executable = false;
1714
1715 // VM operation that loads a library. This is necessary if stack protection
1716 // of the Java stacks can be lost during loading the library. If we
1717 // do not stop the Java threads, they can stack overflow before the stacks
1718 // are protected again.
1719 class VM_LinuxDllLoad: public VM_Operation {
1720 private:
1721 const char *_filename;
1722 char *_ebuf;
1723 int _ebuflen;
1724 void *_lib;
1725 public:
1726 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1727 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1728 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1729 void doit() {
1730 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1731 os::Linux::_stack_is_executable = true;
1732 }
1733 void* loaded_library() { return _lib; }
1734 };
1735
1736 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1737 void * result = NULL;
1738 bool load_attempted = false;
1739
1740 log_info(os)("attempting shared library load of %s", filename);
1741
1742 // Check whether the library to load might change execution rights
1743 // of the stack. If they are changed, the protection of the stack
1744 // guard pages will be lost. We need a safepoint to fix this.
1745 //
1746 // See Linux man page execstack(8) for more info.
1747 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1748 if (!ElfFile::specifies_noexecstack(filename)) {
1749 if (!is_init_completed()) {
1750 os::Linux::_stack_is_executable = true;
1751 // This is OK - No Java threads have been created yet, and hence no
1752 // stack guard pages to fix.
1753 //
1754 // Dynamic loader will make all stacks executable after
1755 // this function returns, and will not do that again.
1756 assert(Threads::number_of_threads() == 0, "no Java threads should exist yet.");
1757 } else {
1758 warning("You have loaded library %s which might have disabled stack guard. "
1759 "The VM will try to fix the stack guard now.\n"
1760 "It's highly recommended that you fix the library with "
1761 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1762 filename);
1763
1764 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1765 JavaThread *jt = JavaThread::current();
1766 if (jt->thread_state() != _thread_in_native) {
1767 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1768 // that requires ExecStack. Cannot enter safe point. Let's give up.
1769 warning("Unable to fix stack guard. Giving up.");
1770 } else {
1771 if (!LoadExecStackDllInVMThread) {
1772 // This is for the case where the DLL has an static
1773 // constructor function that executes JNI code. We cannot
1774 // load such DLLs in the VMThread.
1775 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1776 }
1777
1778 ThreadInVMfromNative tiv(jt);
1779 debug_only(VMNativeEntryWrapper vew;)
1780
1781 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1782 VMThread::execute(&op);
1783 if (LoadExecStackDllInVMThread) {
1784 result = op.loaded_library();
1785 }
1786 load_attempted = true;
1787 }
1788 }
1789 }
1790 }
1791
1792 if (!load_attempted) {
1793 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1794 }
1795
1796 if (result != NULL) {
1797 // Successful loading
1798 return result;
1799 }
1800
1801 Elf32_Ehdr elf_head;
1802 int diag_msg_max_length=ebuflen-strlen(ebuf);
1803 char* diag_msg_buf=ebuf+strlen(ebuf);
1804
1805 if (diag_msg_max_length==0) {
1806 // No more space in ebuf for additional diagnostics message
1807 return NULL;
1808 }
1809
1810
1811 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1812
1813 if (file_descriptor < 0) {
1814 // Can't open library, report dlerror() message
1815 return NULL;
1816 }
1817
1818 bool failed_to_read_elf_head=
1819 (sizeof(elf_head)!=
1820 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1821
1822 ::close(file_descriptor);
1823 if (failed_to_read_elf_head) {
1824 // file i/o error - report dlerror() msg
1825 return NULL;
1826 }
1827
1828 if (elf_head.e_ident[EI_DATA] != LITTLE_ENDIAN_ONLY(ELFDATA2LSB) BIG_ENDIAN_ONLY(ELFDATA2MSB)) {
1829 // handle invalid/out of range endianness values
1830 if (elf_head.e_ident[EI_DATA] == 0 || elf_head.e_ident[EI_DATA] > 2) {
1831 return NULL;
1832 }
1833
1834 #if defined(VM_LITTLE_ENDIAN)
1835 // VM is LE, shared object BE
1836 elf_head.e_machine = be16toh(elf_head.e_machine);
1837 #else
1838 // VM is BE, shared object LE
1839 elf_head.e_machine = le16toh(elf_head.e_machine);
1840 #endif
1841 }
1842
1843 typedef struct {
1844 Elf32_Half code; // Actual value as defined in elf.h
1845 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1846 unsigned char elf_class; // 32 or 64 bit
1847 unsigned char endianness; // MSB or LSB
1848 char* name; // String representation
1849 } arch_t;
1850
1851 #ifndef EM_486
1852 #define EM_486 6 /* Intel 80486 */
1853 #endif
1854 #ifndef EM_AARCH64
1855 #define EM_AARCH64 183 /* ARM AARCH64 */
1856 #endif
1857
1858 static const arch_t arch_array[]={
1859 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1860 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1861 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1862 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1863 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1864 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1865 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1866 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1867 #if defined(VM_LITTLE_ENDIAN)
1868 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1869 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1870 #else
1871 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1872 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1873 #endif
1874 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1875 // we only support 64 bit z architecture
1876 {EM_S390, EM_S390, ELFCLASS64, ELFDATA2MSB, (char*)"IBM System/390"},
1877 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1878 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1879 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1880 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1881 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1882 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1883 };
1884
1885 #if (defined IA32)
1886 static Elf32_Half running_arch_code=EM_386;
1887 #elif (defined AMD64) || (defined X32)
1888 static Elf32_Half running_arch_code=EM_X86_64;
1889 #elif (defined IA64)
1890 static Elf32_Half running_arch_code=EM_IA_64;
1891 #elif (defined __sparc) && (defined _LP64)
1892 static Elf32_Half running_arch_code=EM_SPARCV9;
1893 #elif (defined __sparc) && (!defined _LP64)
1894 static Elf32_Half running_arch_code=EM_SPARC;
1895 #elif (defined __powerpc64__)
1896 static Elf32_Half running_arch_code=EM_PPC64;
1897 #elif (defined __powerpc__)
1898 static Elf32_Half running_arch_code=EM_PPC;
1899 #elif (defined AARCH64)
1900 static Elf32_Half running_arch_code=EM_AARCH64;
1901 #elif (defined ARM)
1902 static Elf32_Half running_arch_code=EM_ARM;
1903 #elif (defined S390)
1904 static Elf32_Half running_arch_code=EM_S390;
1905 #elif (defined ALPHA)
1906 static Elf32_Half running_arch_code=EM_ALPHA;
1907 #elif (defined MIPSEL)
1908 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1909 #elif (defined PARISC)
1910 static Elf32_Half running_arch_code=EM_PARISC;
1911 #elif (defined MIPS)
1912 static Elf32_Half running_arch_code=EM_MIPS;
1913 #elif (defined M68K)
1914 static Elf32_Half running_arch_code=EM_68K;
1915 #elif (defined SH)
1916 static Elf32_Half running_arch_code=EM_SH;
1917 #else
1918 #error Method os::dll_load requires that one of following is defined:\
1919 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1920 #endif
1921
1922 // Identify compatibility class for VM's architecture and library's architecture
1923 // Obtain string descriptions for architectures
1924
1925 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1926 int running_arch_index=-1;
1927
1928 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1929 if (running_arch_code == arch_array[i].code) {
1930 running_arch_index = i;
1931 }
1932 if (lib_arch.code == arch_array[i].code) {
1933 lib_arch.compat_class = arch_array[i].compat_class;
1934 lib_arch.name = arch_array[i].name;
1935 }
1936 }
1937
1938 assert(running_arch_index != -1,
1939 "Didn't find running architecture code (running_arch_code) in arch_array");
1940 if (running_arch_index == -1) {
1941 // Even though running architecture detection failed
1942 // we may still continue with reporting dlerror() message
1943 return NULL;
1944 }
1945
1946 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1947 if (lib_arch.name != NULL) {
1948 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1949 " (Possible cause: can't load %s .so on a %s platform)",
1950 lib_arch.name, arch_array[running_arch_index].name);
1951 } else {
1952 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1953 " (Possible cause: can't load this .so (machine code=0x%x) on a %s platform)",
1954 lib_arch.code, arch_array[running_arch_index].name);
1955 }
1956 return NULL;
1957 }
1958
1959 if (lib_arch.endianness != arch_array[running_arch_index].endianness) {
1960 ::snprintf(diag_msg_buf, diag_msg_max_length-1, " (Possible cause: endianness mismatch)");
1961 return NULL;
1962 }
1963
1964 // ELF file class/capacity : 0 - invalid, 1 - 32bit, 2 - 64bit
1965 if (lib_arch.elf_class > 2 || lib_arch.elf_class < 1) {
1966 ::snprintf(diag_msg_buf, diag_msg_max_length-1, " (Possible cause: invalid ELF file class)");
1967 return NULL;
1968 }
1969
1970 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1971 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1972 " (Possible cause: architecture word width mismatch, can't load %d-bit .so on a %d-bit platform)",
1973 (int) lib_arch.elf_class * 32, arch_array[running_arch_index].elf_class * 32);
1974 return NULL;
1975 }
1976
1977 return NULL;
1978 }
1979
1980 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1981 int ebuflen) {
1982 void * result = ::dlopen(filename, RTLD_LAZY);
1983 if (result == NULL) {
1984 const char* error_report = ::dlerror();
1985 if (error_report == NULL) {
1986 error_report = "dlerror returned no error description";
1987 }
1988 if (ebuf != NULL && ebuflen > 0) {
1989 ::strncpy(ebuf, error_report, ebuflen-1);
1990 ebuf[ebuflen-1]='\0';
1991 }
1992 Events::log(NULL, "Loading shared library %s failed, %s", filename, error_report);
1993 log_info(os)("shared library load of %s failed, %s", filename, error_report);
1994 } else {
1995 Events::log(NULL, "Loaded shared library %s", filename);
1996 log_info(os)("shared library load of %s was successful", filename);
1997 }
1998 return result;
1999 }
2000
2001 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
2002 int ebuflen) {
2003 void * result = NULL;
2004 if (LoadExecStackDllInVMThread) {
2005 result = dlopen_helper(filename, ebuf, ebuflen);
2006 }
2007
2008 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2009 // library that requires an executable stack, or which does not have this
2010 // stack attribute set, dlopen changes the stack attribute to executable. The
2011 // read protection of the guard pages gets lost.
2012 //
2013 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2014 // may have been queued at the same time.
2015
2016 if (!_stack_is_executable) {
2017 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
2018 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2019 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
2020 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
2021 warning("Attempt to reguard stack yellow zone failed.");
2022 }
2023 }
2024 }
2025 }
2026
2027 return result;
2028 }
2029
2030 void* os::dll_lookup(void* handle, const char* name) {
2031 void* res = dlsym(handle, name);
2032 return res;
2033 }
2034
2035 void* os::get_default_process_handle() {
2036 return (void*)::dlopen(NULL, RTLD_LAZY);
2037 }
2038
2039 static bool _print_ascii_file(const char* filename, outputStream* st, const char* hdr = NULL) {
2040 int fd = ::open(filename, O_RDONLY);
2041 if (fd == -1) {
2042 return false;
2043 }
2044
2045 if (hdr != NULL) {
2046 st->print_cr("%s", hdr);
2047 }
2048
2049 char buf[33];
2050 int bytes;
2051 buf[32] = '\0';
2052 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
2053 st->print_raw(buf, bytes);
2054 }
2055
2056 ::close(fd);
2057
2058 return true;
2059 }
2060
2061 void os::print_dll_info(outputStream *st) {
2062 st->print_cr("Dynamic libraries:");
2063
2064 char fname[32];
2065 pid_t pid = os::Linux::gettid();
2066
2067 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2068
2069 if (!_print_ascii_file(fname, st)) {
2070 st->print("Can not get library information for pid = %d\n", pid);
2071 }
2072 }
2073
2074 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2075 FILE *procmapsFile = NULL;
2076
2077 // Open the procfs maps file for the current process
2078 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2079 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2080 char line[PATH_MAX + 100];
2081
2082 // Read line by line from 'file'
2083 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2084 u8 base, top, offset, inode;
2085 char permissions[5];
2086 char device[6];
2087 char name[PATH_MAX + 1];
2088
2089 // Parse fields from line
2090 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %7s " INT64_FORMAT " %s",
2091 &base, &top, permissions, &offset, device, &inode, name);
2092
2093 // Filter by device id '00:00' so that we only get file system mapped files.
2094 if (strcmp(device, "00:00") != 0) {
2095
2096 // Call callback with the fields of interest
2097 if(callback(name, (address)base, (address)top, param)) {
2098 // Oops abort, callback aborted
2099 fclose(procmapsFile);
2100 return 1;
2101 }
2102 }
2103 }
2104 fclose(procmapsFile);
2105 }
2106 return 0;
2107 }
2108
2109 void os::print_os_info_brief(outputStream* st) {
2110 os::Linux::print_distro_info(st);
2111
2112 os::Posix::print_uname_info(st);
2113
2114 os::Linux::print_libversion_info(st);
2115
2116 }
2117
2118 void os::print_os_info(outputStream* st) {
2119 st->print("OS:");
2120
2121 os::Linux::print_distro_info(st);
2122
2123 os::Posix::print_uname_info(st);
2124
2125 // Print warning if unsafe chroot environment detected
2126 if (unsafe_chroot_detected) {
2127 st->print("WARNING!! ");
2128 st->print_cr("%s", unstable_chroot_error);
2129 }
2130
2131 os::Linux::print_libversion_info(st);
2132
2133 os::Posix::print_rlimit_info(st);
2134
2135 os::Posix::print_load_average(st);
2136
2137 os::Linux::print_full_memory_info(st);
2138
2139 os::Linux::print_proc_sys_info(st);
2140
2141 os::Linux::print_ld_preload_file(st);
2142
2143 os::Linux::print_container_info(st);
2144
2145 VM_Version::print_platform_virtualization_info(st);
2146
2147 os::Linux::print_steal_info(st);
2148 }
2149
2150 // Try to identify popular distros.
2151 // Most Linux distributions have a /etc/XXX-release file, which contains
2152 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2153 // file that also contains the OS version string. Some have more than one
2154 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2155 // /etc/redhat-release.), so the order is important.
2156 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2157 // their own specific XXX-release file as well as a redhat-release file.
2158 // Because of this the XXX-release file needs to be searched for before the
2159 // redhat-release file.
2160 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2161 // search for redhat-release / SuSE-release needs to be before lsb-release.
2162 // Since the lsb-release file is the new standard it needs to be searched
2163 // before the older style release files.
2164 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2165 // next to last resort. The os-release file is a new standard that contains
2166 // distribution information and the system-release file seems to be an old
2167 // standard that has been replaced by the lsb-release and os-release files.
2168 // Searching for the debian_version file is the last resort. It contains
2169 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2170 // "Debian " is printed before the contents of the debian_version file.
2171
2172 const char* distro_files[] = {
2173 "/etc/oracle-release",
2174 "/etc/mandriva-release",
2175 "/etc/mandrake-release",
2176 "/etc/sun-release",
2177 "/etc/redhat-release",
2178 "/etc/SuSE-release",
2179 "/etc/lsb-release",
2180 "/etc/turbolinux-release",
2181 "/etc/gentoo-release",
2182 "/etc/ltib-release",
2183 "/etc/angstrom-version",
2184 "/etc/system-release",
2185 "/etc/os-release",
2186 NULL };
2187
2188 void os::Linux::print_distro_info(outputStream* st) {
2189 for (int i = 0;; i++) {
2190 const char* file = distro_files[i];
2191 if (file == NULL) {
2192 break; // done
2193 }
2194 // If file prints, we found it.
2195 if (_print_ascii_file(file, st)) {
2196 return;
2197 }
2198 }
2199
2200 if (file_exists("/etc/debian_version")) {
2201 st->print("Debian ");
2202 _print_ascii_file("/etc/debian_version", st);
2203 } else {
2204 st->print("Linux");
2205 }
2206 st->cr();
2207 }
2208
2209 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2210 char buf[256];
2211 while (fgets(buf, sizeof(buf), fp)) {
2212 // Edit out extra stuff in expected format
2213 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2214 char* ptr = strstr(buf, "\""); // the name is in quotes
2215 if (ptr != NULL) {
2216 ptr++; // go beyond first quote
2217 char* nl = strchr(ptr, '\"');
2218 if (nl != NULL) *nl = '\0';
2219 strncpy(distro, ptr, length);
2220 } else {
2221 ptr = strstr(buf, "=");
2222 ptr++; // go beyond equals then
2223 char* nl = strchr(ptr, '\n');
2224 if (nl != NULL) *nl = '\0';
2225 strncpy(distro, ptr, length);
2226 }
2227 return;
2228 } else if (get_first_line) {
2229 char* nl = strchr(buf, '\n');
2230 if (nl != NULL) *nl = '\0';
2231 strncpy(distro, buf, length);
2232 return;
2233 }
2234 }
2235 // print last line and close
2236 char* nl = strchr(buf, '\n');
2237 if (nl != NULL) *nl = '\0';
2238 strncpy(distro, buf, length);
2239 }
2240
2241 static void parse_os_info(char* distro, size_t length, const char* file) {
2242 FILE* fp = fopen(file, "r");
2243 if (fp != NULL) {
2244 // if suse format, print out first line
2245 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2246 parse_os_info_helper(fp, distro, length, get_first_line);
2247 fclose(fp);
2248 }
2249 }
2250
2251 void os::get_summary_os_info(char* buf, size_t buflen) {
2252 for (int i = 0;; i++) {
2253 const char* file = distro_files[i];
2254 if (file == NULL) {
2255 break; // ran out of distro_files
2256 }
2257 if (file_exists(file)) {
2258 parse_os_info(buf, buflen, file);
2259 return;
2260 }
2261 }
2262 // special case for debian
2263 if (file_exists("/etc/debian_version")) {
2264 strncpy(buf, "Debian ", buflen);
2265 if (buflen > 7) {
2266 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2267 }
2268 } else {
2269 strncpy(buf, "Linux", buflen);
2270 }
2271 }
2272
2273 void os::Linux::print_libversion_info(outputStream* st) {
2274 // libc, pthread
2275 st->print("libc:");
2276 st->print("%s ", os::Linux::glibc_version());
2277 st->print("%s ", os::Linux::libpthread_version());
2278 st->cr();
2279 }
2280
2281 void os::Linux::print_proc_sys_info(outputStream* st) {
2282 st->cr();
2283 st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):");
2284 _print_ascii_file("/proc/sys/kernel/threads-max", st);
2285 st->cr();
2286 st->cr();
2287
2288 st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):");
2289 _print_ascii_file("/proc/sys/vm/max_map_count", st);
2290 st->cr();
2291 st->cr();
2292
2293 st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):");
2294 _print_ascii_file("/proc/sys/kernel/pid_max", st);
2295 st->cr();
2296 st->cr();
2297 }
2298
2299 void os::Linux::print_full_memory_info(outputStream* st) {
2300 st->print("\n/proc/meminfo:\n");
2301 _print_ascii_file("/proc/meminfo", st);
2302 st->cr();
2303 }
2304
2305 void os::Linux::print_ld_preload_file(outputStream* st) {
2306 _print_ascii_file("/etc/ld.so.preload", st, "\n/etc/ld.so.preload:");
2307 st->cr();
2308 }
2309
2310 void os::Linux::print_container_info(outputStream* st) {
2311 if (!OSContainer::is_containerized()) {
2312 return;
2313 }
2314
2315 st->print("container (cgroup) information:\n");
2316
2317 const char *p_ct = OSContainer::container_type();
2318 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "not supported");
2319
2320 char *p = OSContainer::cpu_cpuset_cpus();
2321 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "not supported");
2322 free(p);
2323
2324 p = OSContainer::cpu_cpuset_memory_nodes();
2325 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "not supported");
2326 free(p);
2327
2328 int i = OSContainer::active_processor_count();
2329 st->print("active_processor_count: ");
2330 if (i > 0) {
2331 st->print("%d\n", i);
2332 } else {
2333 st->print("not supported\n");
2334 }
2335
2336 i = OSContainer::cpu_quota();
2337 st->print("cpu_quota: ");
2338 if (i > 0) {
2339 st->print("%d\n", i);
2340 } else {
2341 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no quota");
2342 }
2343
2344 i = OSContainer::cpu_period();
2345 st->print("cpu_period: ");
2346 if (i > 0) {
2347 st->print("%d\n", i);
2348 } else {
2349 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no period");
2350 }
2351
2352 i = OSContainer::cpu_shares();
2353 st->print("cpu_shares: ");
2354 if (i > 0) {
2355 st->print("%d\n", i);
2356 } else {
2357 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no shares");
2358 }
2359
2360 jlong j = OSContainer::memory_limit_in_bytes();
2361 st->print("memory_limit_in_bytes: ");
2362 if (j > 0) {
2363 st->print(JLONG_FORMAT "\n", j);
2364 } else {
2365 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2366 }
2367
2368 j = OSContainer::memory_and_swap_limit_in_bytes();
2369 st->print("memory_and_swap_limit_in_bytes: ");
2370 if (j > 0) {
2371 st->print(JLONG_FORMAT "\n", j);
2372 } else {
2373 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2374 }
2375
2376 j = OSContainer::memory_soft_limit_in_bytes();
2377 st->print("memory_soft_limit_in_bytes: ");
2378 if (j > 0) {
2379 st->print(JLONG_FORMAT "\n", j);
2380 } else {
2381 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2382 }
2383
2384 j = OSContainer::OSContainer::memory_usage_in_bytes();
2385 st->print("memory_usage_in_bytes: ");
2386 if (j > 0) {
2387 st->print(JLONG_FORMAT "\n", j);
2388 } else {
2389 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2390 }
2391
2392 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2393 st->print("memory_max_usage_in_bytes: ");
2394 if (j > 0) {
2395 st->print(JLONG_FORMAT "\n", j);
2396 } else {
2397 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2398 }
2399 st->cr();
2400 }
2401
2402 void os::Linux::print_steal_info(outputStream* st) {
2403 if (has_initial_tick_info) {
2404 CPUPerfTicks pticks;
2405 bool res = os::Linux::get_tick_information(&pticks, -1);
2406
2407 if (res && pticks.has_steal_ticks) {
2408 uint64_t steal_ticks_difference = pticks.steal - initial_steal_ticks;
2409 uint64_t total_ticks_difference = pticks.total - initial_total_ticks;
2410 double steal_ticks_perc = 0.0;
2411 if (total_ticks_difference != 0) {
2412 steal_ticks_perc = (double) steal_ticks_difference / total_ticks_difference;
2413 }
2414 st->print_cr("Steal ticks since vm start: " UINT64_FORMAT, steal_ticks_difference);
2415 st->print_cr("Steal ticks percentage since vm start:%7.3f", steal_ticks_perc);
2416 }
2417 }
2418 }
2419
2420 void os::print_memory_info(outputStream* st) {
2421
2422 st->print("Memory:");
2423 st->print(" %dk page", os::vm_page_size()>>10);
2424
2425 // values in struct sysinfo are "unsigned long"
2426 struct sysinfo si;
2427 sysinfo(&si);
2428
2429 st->print(", physical " UINT64_FORMAT "k",
2430 os::physical_memory() >> 10);
2431 st->print("(" UINT64_FORMAT "k free)",
2432 os::available_memory() >> 10);
2433 st->print(", swap " UINT64_FORMAT "k",
2434 ((jlong)si.totalswap * si.mem_unit) >> 10);
2435 st->print("(" UINT64_FORMAT "k free)",
2436 ((jlong)si.freeswap * si.mem_unit) >> 10);
2437 st->cr();
2438 }
2439
2440 // Print the first "model name" line and the first "flags" line
2441 // that we find and nothing more. We assume "model name" comes
2442 // before "flags" so if we find a second "model name", then the
2443 // "flags" field is considered missing.
2444 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2445 #if defined(IA32) || defined(AMD64)
2446 // Other platforms have less repetitive cpuinfo files
2447 FILE *fp = fopen("/proc/cpuinfo", "r");
2448 if (fp) {
2449 while (!feof(fp)) {
2450 if (fgets(buf, buflen, fp)) {
2451 // Assume model name comes before flags
2452 bool model_name_printed = false;
2453 if (strstr(buf, "model name") != NULL) {
2454 if (!model_name_printed) {
2455 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2456 st->print_raw(buf);
2457 model_name_printed = true;
2458 } else {
2459 // model name printed but not flags? Odd, just return
2460 fclose(fp);
2461 return true;
2462 }
2463 }
2464 // print the flags line too
2465 if (strstr(buf, "flags") != NULL) {
2466 st->print_raw(buf);
2467 fclose(fp);
2468 return true;
2469 }
2470 }
2471 }
2472 fclose(fp);
2473 }
2474 #endif // x86 platforms
2475 return false;
2476 }
2477
2478 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2479 // Only print the model name if the platform provides this as a summary
2480 if (!print_model_name_and_flags(st, buf, buflen)) {
2481 st->print("\n/proc/cpuinfo:\n");
2482 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2483 st->print_cr(" <Not Available>");
2484 }
2485 }
2486 }
2487
2488 #if defined(AMD64) || defined(IA32) || defined(X32)
2489 const char* search_string = "model name";
2490 #elif defined(M68K)
2491 const char* search_string = "CPU";
2492 #elif defined(PPC64)
2493 const char* search_string = "cpu";
2494 #elif defined(S390)
2495 const char* search_string = "machine =";
2496 #elif defined(SPARC)
2497 const char* search_string = "cpu";
2498 #else
2499 const char* search_string = "Processor";
2500 #endif
2501
2502 // Parses the cpuinfo file for string representing the model name.
2503 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2504 FILE* fp = fopen("/proc/cpuinfo", "r");
2505 if (fp != NULL) {
2506 while (!feof(fp)) {
2507 char buf[256];
2508 if (fgets(buf, sizeof(buf), fp)) {
2509 char* start = strstr(buf, search_string);
2510 if (start != NULL) {
2511 char *ptr = start + strlen(search_string);
2512 char *end = buf + strlen(buf);
2513 while (ptr != end) {
2514 // skip whitespace and colon for the rest of the name.
2515 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2516 break;
2517 }
2518 ptr++;
2519 }
2520 if (ptr != end) {
2521 // reasonable string, get rid of newline and keep the rest
2522 char* nl = strchr(buf, '\n');
2523 if (nl != NULL) *nl = '\0';
2524 strncpy(cpuinfo, ptr, length);
2525 fclose(fp);
2526 return;
2527 }
2528 }
2529 }
2530 }
2531 fclose(fp);
2532 }
2533 // cpuinfo not found or parsing failed, just print generic string. The entire
2534 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2535 #if defined(AARCH64)
2536 strncpy(cpuinfo, "AArch64", length);
2537 #elif defined(AMD64)
2538 strncpy(cpuinfo, "x86_64", length);
2539 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2540 strncpy(cpuinfo, "ARM", length);
2541 #elif defined(IA32)
2542 strncpy(cpuinfo, "x86_32", length);
2543 #elif defined(IA64)
2544 strncpy(cpuinfo, "IA64", length);
2545 #elif defined(PPC)
2546 strncpy(cpuinfo, "PPC64", length);
2547 #elif defined(S390)
2548 strncpy(cpuinfo, "S390", length);
2549 #elif defined(SPARC)
2550 strncpy(cpuinfo, "sparcv9", length);
2551 #elif defined(ZERO_LIBARCH)
2552 strncpy(cpuinfo, ZERO_LIBARCH, length);
2553 #else
2554 strncpy(cpuinfo, "unknown", length);
2555 #endif
2556 }
2557
2558 static void print_signal_handler(outputStream* st, int sig,
2559 char* buf, size_t buflen);
2560
2561 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2562 st->print_cr("Signal Handlers:");
2563 print_signal_handler(st, SIGSEGV, buf, buflen);
2564 print_signal_handler(st, SIGBUS , buf, buflen);
2565 print_signal_handler(st, SIGFPE , buf, buflen);
2566 print_signal_handler(st, SIGPIPE, buf, buflen);
2567 print_signal_handler(st, SIGXFSZ, buf, buflen);
2568 print_signal_handler(st, SIGILL , buf, buflen);
2569 print_signal_handler(st, SR_signum, buf, buflen);
2570 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2571 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2572 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2573 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2574 #if defined(PPC64)
2575 print_signal_handler(st, SIGTRAP, buf, buflen);
2576 #endif
2577 }
2578
2579 static char saved_jvm_path[MAXPATHLEN] = {0};
2580
2581 // Find the full path to the current module, libjvm.so
2582 void os::jvm_path(char *buf, jint buflen) {
2583 // Error checking.
2584 if (buflen < MAXPATHLEN) {
2585 assert(false, "must use a large-enough buffer");
2586 buf[0] = '\0';
2587 return;
2588 }
2589 // Lazy resolve the path to current module.
2590 if (saved_jvm_path[0] != 0) {
2591 strcpy(buf, saved_jvm_path);
2592 return;
2593 }
2594
2595 char dli_fname[MAXPATHLEN];
2596 bool ret = dll_address_to_library_name(
2597 CAST_FROM_FN_PTR(address, os::jvm_path),
2598 dli_fname, sizeof(dli_fname), NULL);
2599 assert(ret, "cannot locate libjvm");
2600 char *rp = NULL;
2601 if (ret && dli_fname[0] != '\0') {
2602 rp = os::Posix::realpath(dli_fname, buf, buflen);
2603 }
2604 if (rp == NULL) {
2605 return;
2606 }
2607
2608 if (Arguments::sun_java_launcher_is_altjvm()) {
2609 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2610 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2611 // If "/jre/lib/" appears at the right place in the string, then
2612 // assume we are installed in a JDK and we're done. Otherwise, check
2613 // for a JAVA_HOME environment variable and fix up the path so it
2614 // looks like libjvm.so is installed there (append a fake suffix
2615 // hotspot/libjvm.so).
2616 const char *p = buf + strlen(buf) - 1;
2617 for (int count = 0; p > buf && count < 5; ++count) {
2618 for (--p; p > buf && *p != '/'; --p)
2619 /* empty */ ;
2620 }
2621
2622 if (strncmp(p, "/jre/lib/", 9) != 0) {
2623 // Look for JAVA_HOME in the environment.
2624 char* java_home_var = ::getenv("JAVA_HOME");
2625 if (java_home_var != NULL && java_home_var[0] != 0) {
2626 char* jrelib_p;
2627 int len;
2628
2629 // Check the current module name "libjvm.so".
2630 p = strrchr(buf, '/');
2631 if (p == NULL) {
2632 return;
2633 }
2634 assert(strstr(p, "/libjvm") == p, "invalid library name");
2635
2636 rp = os::Posix::realpath(java_home_var, buf, buflen);
2637 if (rp == NULL) {
2638 return;
2639 }
2640
2641 // determine if this is a legacy image or modules image
2642 // modules image doesn't have "jre" subdirectory
2643 len = strlen(buf);
2644 assert(len < buflen, "Ran out of buffer room");
2645 jrelib_p = buf + len;
2646 snprintf(jrelib_p, buflen-len, "/jre/lib");
2647 if (0 != access(buf, F_OK)) {
2648 snprintf(jrelib_p, buflen-len, "/lib");
2649 }
2650
2651 if (0 == access(buf, F_OK)) {
2652 // Use current module name "libjvm.so"
2653 len = strlen(buf);
2654 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2655 } else {
2656 // Go back to path of .so
2657 rp = os::Posix::realpath(dli_fname, buf, buflen);
2658 if (rp == NULL) {
2659 return;
2660 }
2661 }
2662 }
2663 }
2664 }
2665
2666 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2667 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2668 }
2669
2670 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2671 // no prefix required, not even "_"
2672 }
2673
2674 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2675 // no suffix required
2676 }
2677
2678 ////////////////////////////////////////////////////////////////////////////////
2679 // sun.misc.Signal support
2680
2681 static void UserHandler(int sig, void *siginfo, void *context) {
2682 // Ctrl-C is pressed during error reporting, likely because the error
2683 // handler fails to abort. Let VM die immediately.
2684 if (sig == SIGINT && VMError::is_error_reported()) {
2685 os::die();
2686 }
2687
2688 os::signal_notify(sig);
2689 }
2690
2691 void* os::user_handler() {
2692 return CAST_FROM_FN_PTR(void*, UserHandler);
2693 }
2694
2695 extern "C" {
2696 typedef void (*sa_handler_t)(int);
2697 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2698 }
2699
2700 void* os::signal(int signal_number, void* handler) {
2701 struct sigaction sigAct, oldSigAct;
2702
2703 sigfillset(&(sigAct.sa_mask));
2704 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2705 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2706
2707 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2708 // -1 means registration failed
2709 return (void *)-1;
2710 }
2711
2712 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2713 }
2714
2715 void os::signal_raise(int signal_number) {
2716 ::raise(signal_number);
2717 }
2718
2719 // The following code is moved from os.cpp for making this
2720 // code platform specific, which it is by its very nature.
2721
2722 // Will be modified when max signal is changed to be dynamic
2723 int os::sigexitnum_pd() {
2724 return NSIG;
2725 }
2726
2727 // a counter for each possible signal value
2728 static volatile jint pending_signals[NSIG+1] = { 0 };
2729
2730 // Linux(POSIX) specific hand shaking semaphore.
2731 static Semaphore* sig_sem = NULL;
2732 static PosixSemaphore sr_semaphore;
2733
2734 static void jdk_misc_signal_init() {
2735 // Initialize signal structures
2736 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2737
2738 // Initialize signal semaphore
2739 sig_sem = new Semaphore();
2740 }
2741
2742 void os::signal_notify(int sig) {
2743 if (sig_sem != NULL) {
2744 Atomic::inc(&pending_signals[sig]);
2745 sig_sem->signal();
2746 } else {
2747 // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2748 // initialization isn't called.
2749 assert(ReduceSignalUsage, "signal semaphore should be created");
2750 }
2751 }
2752
2753 static int check_pending_signals() {
2754 for (;;) {
2755 for (int i = 0; i < NSIG + 1; i++) {
2756 jint n = pending_signals[i];
2757 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2758 return i;
2759 }
2760 }
2761 JavaThread *thread = JavaThread::current();
2762 ThreadBlockInVM tbivm(thread);
2763
2764 bool threadIsSuspended;
2765 do {
2766 thread->set_suspend_equivalent();
2767 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2768 sig_sem->wait();
2769
2770 // were we externally suspended while we were waiting?
2771 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2772 if (threadIsSuspended) {
2773 // The semaphore has been incremented, but while we were waiting
2774 // another thread suspended us. We don't want to continue running
2775 // while suspended because that would surprise the thread that
2776 // suspended us.
2777 sig_sem->signal();
2778
2779 thread->java_suspend_self();
2780 }
2781 } while (threadIsSuspended);
2782 }
2783 }
2784
2785 int os::signal_wait() {
2786 return check_pending_signals();
2787 }
2788
2789 ////////////////////////////////////////////////////////////////////////////////
2790 // Virtual Memory
2791
2792 int os::vm_page_size() {
2793 // Seems redundant as all get out
2794 assert(os::Linux::page_size() != -1, "must call os::init");
2795 return os::Linux::page_size();
2796 }
2797
2798 // Solaris allocates memory by pages.
2799 int os::vm_allocation_granularity() {
2800 assert(os::Linux::page_size() != -1, "must call os::init");
2801 return os::Linux::page_size();
2802 }
2803
2804 // Rationale behind this function:
2805 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2806 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2807 // samples for JITted code. Here we create private executable mapping over the code cache
2808 // and then we can use standard (well, almost, as mapping can change) way to provide
2809 // info for the reporting script by storing timestamp and location of symbol
2810 void linux_wrap_code(char* base, size_t size) {
2811 static volatile jint cnt = 0;
2812
2813 if (!UseOprofile) {
2814 return;
2815 }
2816
2817 char buf[PATH_MAX+1];
2818 int num = Atomic::add(1, &cnt);
2819
2820 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2821 os::get_temp_directory(), os::current_process_id(), num);
2822 unlink(buf);
2823
2824 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2825
2826 if (fd != -1) {
2827 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2828 if (rv != (off_t)-1) {
2829 if (::write(fd, "", 1) == 1) {
2830 mmap(base, size,
2831 PROT_READ|PROT_WRITE|PROT_EXEC,
2832 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2833 }
2834 }
2835 ::close(fd);
2836 unlink(buf);
2837 }
2838 }
2839
2840 static bool recoverable_mmap_error(int err) {
2841 // See if the error is one we can let the caller handle. This
2842 // list of errno values comes from JBS-6843484. I can't find a
2843 // Linux man page that documents this specific set of errno
2844 // values so while this list currently matches Solaris, it may
2845 // change as we gain experience with this failure mode.
2846 switch (err) {
2847 case EBADF:
2848 case EINVAL:
2849 case ENOTSUP:
2850 // let the caller deal with these errors
2851 return true;
2852
2853 default:
2854 // Any remaining errors on this OS can cause our reserved mapping
2855 // to be lost. That can cause confusion where different data
2856 // structures think they have the same memory mapped. The worst
2857 // scenario is if both the VM and a library think they have the
2858 // same memory mapped.
2859 return false;
2860 }
2861 }
2862
2863 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2864 int err) {
2865 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2866 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2867 os::strerror(err), err);
2868 }
2869
2870 static void warn_fail_commit_memory(char* addr, size_t size,
2871 size_t alignment_hint, bool exec,
2872 int err) {
2873 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2874 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2875 alignment_hint, exec, os::strerror(err), err);
2876 }
2877
2878 // NOTE: Linux kernel does not really reserve the pages for us.
2879 // All it does is to check if there are enough free pages
2880 // left at the time of mmap(). This could be a potential
2881 // problem.
2882 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2883 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2884 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2885 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2886 if (res != (uintptr_t) MAP_FAILED) {
2887 if (UseNUMAInterleaving) {
2888 numa_make_global(addr, size);
2889 }
2890 return 0;
2891 }
2892
2893 int err = errno; // save errno from mmap() call above
2894
2895 if (!recoverable_mmap_error(err)) {
2896 warn_fail_commit_memory(addr, size, exec, err);
2897 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2898 }
2899
2900 return err;
2901 }
2902
2903 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2904 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2905 }
2906
2907 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2908 const char* mesg) {
2909 assert(mesg != NULL, "mesg must be specified");
2910 int err = os::Linux::commit_memory_impl(addr, size, exec);
2911 if (err != 0) {
2912 // the caller wants all commit errors to exit with the specified mesg:
2913 warn_fail_commit_memory(addr, size, exec, err);
2914 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2915 }
2916 }
2917
2918 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2919 #ifndef MAP_HUGETLB
2920 #define MAP_HUGETLB 0x40000
2921 #endif
2922
2923 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2924 #ifndef MADV_HUGEPAGE
2925 #define MADV_HUGEPAGE 14
2926 #endif
2927
2928 int os::Linux::commit_memory_impl(char* addr, size_t size,
2929 size_t alignment_hint, bool exec) {
2930 int err = os::Linux::commit_memory_impl(addr, size, exec);
2931 if (err == 0) {
2932 realign_memory(addr, size, alignment_hint);
2933 }
2934 return err;
2935 }
2936
2937 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2938 bool exec) {
2939 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2940 }
2941
2942 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2943 size_t alignment_hint, bool exec,
2944 const char* mesg) {
2945 assert(mesg != NULL, "mesg must be specified");
2946 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2947 if (err != 0) {
2948 // the caller wants all commit errors to exit with the specified mesg:
2949 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2950 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2951 }
2952 }
2953
2954 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2955 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2956 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2957 // be supported or the memory may already be backed by huge pages.
2958 ::madvise(addr, bytes, MADV_HUGEPAGE);
2959 }
2960 }
2961
2962 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2963 // This method works by doing an mmap over an existing mmaping and effectively discarding
2964 // the existing pages. However it won't work for SHM-based large pages that cannot be
2965 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2966 // small pages on top of the SHM segment. This method always works for small pages, so we
2967 // allow that in any case.
2968 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2969 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2970 }
2971 }
2972
2973 void os::numa_make_global(char *addr, size_t bytes) {
2974 Linux::numa_interleave_memory(addr, bytes);
2975 }
2976
2977 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2978 // bind policy to MPOL_PREFERRED for the current thread.
2979 #define USE_MPOL_PREFERRED 0
2980
2981 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2982 // To make NUMA and large pages more robust when both enabled, we need to ease
2983 // the requirements on where the memory should be allocated. MPOL_BIND is the
2984 // default policy and it will force memory to be allocated on the specified
2985 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2986 // the specified node, but will not force it. Using this policy will prevent
2987 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2988 // free large pages.
2989 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2990 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2991 }
2992
2993 bool os::numa_topology_changed() { return false; }
2994
2995 size_t os::numa_get_groups_num() {
2996 // Return just the number of nodes in which it's possible to allocate memory
2997 // (in numa terminology, configured nodes).
2998 return Linux::numa_num_configured_nodes();
2999 }
3000
3001 int os::numa_get_group_id() {
3002 int cpu_id = Linux::sched_getcpu();
3003 if (cpu_id != -1) {
3004 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
3005 if (lgrp_id != -1) {
3006 return lgrp_id;
3007 }
3008 }
3009 return 0;
3010 }
3011
3012 int os::Linux::get_existing_num_nodes() {
3013 int node;
3014 int highest_node_number = Linux::numa_max_node();
3015 int num_nodes = 0;
3016
3017 // Get the total number of nodes in the system including nodes without memory.
3018 for (node = 0; node <= highest_node_number; node++) {
3019 if (is_node_in_existing_nodes(node)) {
3020 num_nodes++;
3021 }
3022 }
3023 return num_nodes;
3024 }
3025
3026 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
3027 int highest_node_number = Linux::numa_max_node();
3028 size_t i = 0;
3029
3030 // Map all node ids in which it is possible to allocate memory. Also nodes are
3031 // not always consecutively available, i.e. available from 0 to the highest
3032 // node number. If the nodes have been bound explicitly using numactl membind,
3033 // then allocate memory from those nodes only.
3034 for (int node = 0; node <= highest_node_number; node++) {
3035 if (Linux::is_node_in_bound_nodes((unsigned int)node)) {
3036 ids[i++] = node;
3037 }
3038 }
3039 return i;
3040 }
3041
3042 bool os::get_page_info(char *start, page_info* info) {
3043 return false;
3044 }
3045
3046 char *os::scan_pages(char *start, char* end, page_info* page_expected,
3047 page_info* page_found) {
3048 return end;
3049 }
3050
3051
3052 int os::Linux::sched_getcpu_syscall(void) {
3053 unsigned int cpu = 0;
3054 int retval = -1;
3055
3056 #if defined(IA32)
3057 #ifndef SYS_getcpu
3058 #define SYS_getcpu 318
3059 #endif
3060 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
3061 #elif defined(AMD64)
3062 // Unfortunately we have to bring all these macros here from vsyscall.h
3063 // to be able to compile on old linuxes.
3064 #define __NR_vgetcpu 2
3065 #define VSYSCALL_START (-10UL << 20)
3066 #define VSYSCALL_SIZE 1024
3067 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
3068 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
3069 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
3070 retval = vgetcpu(&cpu, NULL, NULL);
3071 #endif
3072
3073 return (retval == -1) ? retval : cpu;
3074 }
3075
3076 void os::Linux::sched_getcpu_init() {
3077 // sched_getcpu() should be in libc.
3078 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
3079 dlsym(RTLD_DEFAULT, "sched_getcpu")));
3080
3081 // If it's not, try a direct syscall.
3082 if (sched_getcpu() == -1) {
3083 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
3084 (void*)&sched_getcpu_syscall));
3085 }
3086
3087 if (sched_getcpu() == -1) {
3088 vm_exit_during_initialization("getcpu(2) system call not supported by kernel");
3089 }
3090 }
3091
3092 // Something to do with the numa-aware allocator needs these symbols
3093 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
3094 extern "C" JNIEXPORT void numa_error(char *where) { }
3095
3096 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
3097 // load symbol from base version instead.
3098 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
3099 void *f = dlvsym(handle, name, "libnuma_1.1");
3100 if (f == NULL) {
3101 f = dlsym(handle, name);
3102 }
3103 return f;
3104 }
3105
3106 // Handle request to load libnuma symbol version 1.2 (API v2) only.
3107 // Return NULL if the symbol is not defined in this particular version.
3108 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
3109 return dlvsym(handle, name, "libnuma_1.2");
3110 }
3111
3112 bool os::Linux::libnuma_init() {
3113 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
3114 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
3115 if (handle != NULL) {
3116 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
3117 libnuma_dlsym(handle, "numa_node_to_cpus")));
3118 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
3119 libnuma_dlsym(handle, "numa_max_node")));
3120 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
3121 libnuma_dlsym(handle, "numa_num_configured_nodes")));
3122 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
3123 libnuma_dlsym(handle, "numa_available")));
3124 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
3125 libnuma_dlsym(handle, "numa_tonode_memory")));
3126 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
3127 libnuma_dlsym(handle, "numa_interleave_memory")));
3128 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
3129 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
3130 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
3131 libnuma_dlsym(handle, "numa_set_bind_policy")));
3132 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
3133 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
3134 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
3135 libnuma_dlsym(handle, "numa_distance")));
3136 set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t,
3137 libnuma_v2_dlsym(handle, "numa_get_membind")));
3138 set_numa_get_interleave_mask(CAST_TO_FN_PTR(numa_get_interleave_mask_func_t,
3139 libnuma_v2_dlsym(handle, "numa_get_interleave_mask")));
3140
3141 if (numa_available() != -1) {
3142 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
3143 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
3144 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
3145 set_numa_interleave_bitmask(_numa_get_interleave_mask());
3146 set_numa_membind_bitmask(_numa_get_membind());
3147 // Create an index -> node mapping, since nodes are not always consecutive
3148 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3149 rebuild_nindex_to_node_map();
3150 // Create a cpu -> node mapping
3151 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3152 rebuild_cpu_to_node_map();
3153 return true;
3154 }
3155 }
3156 }
3157 return false;
3158 }
3159
3160 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
3161 // Creating guard page is very expensive. Java thread has HotSpot
3162 // guard pages, only enable glibc guard page for non-Java threads.
3163 // (Remember: compiler thread is a Java thread, too!)
3164 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
3165 }
3166
3167 void os::Linux::rebuild_nindex_to_node_map() {
3168 int highest_node_number = Linux::numa_max_node();
3169
3170 nindex_to_node()->clear();
3171 for (int node = 0; node <= highest_node_number; node++) {
3172 if (Linux::is_node_in_existing_nodes(node)) {
3173 nindex_to_node()->append(node);
3174 }
3175 }
3176 }
3177
3178 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
3179 // The table is later used in get_node_by_cpu().
3180 void os::Linux::rebuild_cpu_to_node_map() {
3181 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
3182 // in libnuma (possible values are starting from 16,
3183 // and continuing up with every other power of 2, but less
3184 // than the maximum number of CPUs supported by kernel), and
3185 // is a subject to change (in libnuma version 2 the requirements
3186 // are more reasonable) we'll just hardcode the number they use
3187 // in the library.
3188 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
3189
3190 size_t cpu_num = processor_count();
3191 size_t cpu_map_size = NCPUS / BitsPerCLong;
3192 size_t cpu_map_valid_size =
3193 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
3194
3195 cpu_to_node()->clear();
3196 cpu_to_node()->at_grow(cpu_num - 1);
3197
3198 size_t node_num = get_existing_num_nodes();
3199
3200 int distance = 0;
3201 int closest_distance = INT_MAX;
3202 int closest_node = 0;
3203 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
3204 for (size_t i = 0; i < node_num; i++) {
3205 // Check if node is configured (not a memory-less node). If it is not, find
3206 // the closest configured node. Check also if node is bound, i.e. it's allowed
3207 // to allocate memory from the node. If it's not allowed, map cpus in that node
3208 // to the closest node from which memory allocation is allowed.
3209 if (!is_node_in_configured_nodes(nindex_to_node()->at(i)) ||
3210 !is_node_in_bound_nodes(nindex_to_node()->at(i))) {
3211 closest_distance = INT_MAX;
3212 // Check distance from all remaining nodes in the system. Ignore distance
3213 // from itself, from another non-configured node, and from another non-bound
3214 // node.
3215 for (size_t m = 0; m < node_num; m++) {
3216 if (m != i &&
3217 is_node_in_configured_nodes(nindex_to_node()->at(m)) &&
3218 is_node_in_bound_nodes(nindex_to_node()->at(m))) {
3219 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3220 // If a closest node is found, update. There is always at least one
3221 // configured and bound node in the system so there is always at least
3222 // one node close.
3223 if (distance != 0 && distance < closest_distance) {
3224 closest_distance = distance;
3225 closest_node = nindex_to_node()->at(m);
3226 }
3227 }
3228 }
3229 } else {
3230 // Current node is already a configured node.
3231 closest_node = nindex_to_node()->at(i);
3232 }
3233
3234 // Get cpus from the original node and map them to the closest node. If node
3235 // is a configured node (not a memory-less node), then original node and
3236 // closest node are the same.
3237 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3238 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3239 if (cpu_map[j] != 0) {
3240 for (size_t k = 0; k < BitsPerCLong; k++) {
3241 if (cpu_map[j] & (1UL << k)) {
3242 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3243 }
3244 }
3245 }
3246 }
3247 }
3248 }
3249 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3250 }
3251
3252 int os::Linux::get_node_by_cpu(int cpu_id) {
3253 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3254 return cpu_to_node()->at(cpu_id);
3255 }
3256 return -1;
3257 }
3258
3259 GrowableArray<int>* os::Linux::_cpu_to_node;
3260 GrowableArray<int>* os::Linux::_nindex_to_node;
3261 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3262 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3263 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3264 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3265 os::Linux::numa_available_func_t os::Linux::_numa_available;
3266 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3267 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3268 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3269 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3270 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3271 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3272 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind;
3273 os::Linux::numa_get_interleave_mask_func_t os::Linux::_numa_get_interleave_mask;
3274 os::Linux::NumaAllocationPolicy os::Linux::_current_numa_policy;
3275 unsigned long* os::Linux::_numa_all_nodes;
3276 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3277 struct bitmask* os::Linux::_numa_nodes_ptr;
3278 struct bitmask* os::Linux::_numa_interleave_bitmask;
3279 struct bitmask* os::Linux::_numa_membind_bitmask;
3280
3281 bool os::pd_uncommit_memory(char* addr, size_t size) {
3282 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3283 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3284 return res != (uintptr_t) MAP_FAILED;
3285 }
3286
3287 static address get_stack_commited_bottom(address bottom, size_t size) {
3288 address nbot = bottom;
3289 address ntop = bottom + size;
3290
3291 size_t page_sz = os::vm_page_size();
3292 unsigned pages = size / page_sz;
3293
3294 unsigned char vec[1];
3295 unsigned imin = 1, imax = pages + 1, imid;
3296 int mincore_return_value = 0;
3297
3298 assert(imin <= imax, "Unexpected page size");
3299
3300 while (imin < imax) {
3301 imid = (imax + imin) / 2;
3302 nbot = ntop - (imid * page_sz);
3303
3304 // Use a trick with mincore to check whether the page is mapped or not.
3305 // mincore sets vec to 1 if page resides in memory and to 0 if page
3306 // is swapped output but if page we are asking for is unmapped
3307 // it returns -1,ENOMEM
3308 mincore_return_value = mincore(nbot, page_sz, vec);
3309
3310 if (mincore_return_value == -1) {
3311 // Page is not mapped go up
3312 // to find first mapped page
3313 if (errno != EAGAIN) {
3314 assert(errno == ENOMEM, "Unexpected mincore errno");
3315 imax = imid;
3316 }
3317 } else {
3318 // Page is mapped go down
3319 // to find first not mapped page
3320 imin = imid + 1;
3321 }
3322 }
3323
3324 nbot = nbot + page_sz;
3325
3326 // Adjust stack bottom one page up if last checked page is not mapped
3327 if (mincore_return_value == -1) {
3328 nbot = nbot + page_sz;
3329 }
3330
3331 return nbot;
3332 }
3333
3334 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3335 int mincore_return_value;
3336 const size_t stripe = 1024; // query this many pages each time
3337 unsigned char vec[stripe + 1];
3338 // set a guard
3339 vec[stripe] = 'X';
3340
3341 const size_t page_sz = os::vm_page_size();
3342 size_t pages = size / page_sz;
3343
3344 assert(is_aligned(start, page_sz), "Start address must be page aligned");
3345 assert(is_aligned(size, page_sz), "Size must be page aligned");
3346
3347 committed_start = NULL;
3348
3349 int loops = (pages + stripe - 1) / stripe;
3350 int committed_pages = 0;
3351 address loop_base = start;
3352 bool found_range = false;
3353
3354 for (int index = 0; index < loops && !found_range; index ++) {
3355 assert(pages > 0, "Nothing to do");
3356 int pages_to_query = (pages >= stripe) ? stripe : pages;
3357 pages -= pages_to_query;
3358
3359 // Get stable read
3360 while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3361
3362 // During shutdown, some memory goes away without properly notifying NMT,
3363 // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3364 // Bailout and return as not committed for now.
3365 if (mincore_return_value == -1 && errno == ENOMEM) {
3366 return false;
3367 }
3368
3369 assert(vec[stripe] == 'X', "overflow guard");
3370 assert(mincore_return_value == 0, "Range must be valid");
3371 // Process this stripe
3372 for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3373 if ((vec[vecIdx] & 0x01) == 0) { // not committed
3374 // End of current contiguous region
3375 if (committed_start != NULL) {
3376 found_range = true;
3377 break;
3378 }
3379 } else { // committed
3380 // Start of region
3381 if (committed_start == NULL) {
3382 committed_start = loop_base + page_sz * vecIdx;
3383 }
3384 committed_pages ++;
3385 }
3386 }
3387
3388 loop_base += pages_to_query * page_sz;
3389 }
3390
3391 if (committed_start != NULL) {
3392 assert(committed_pages > 0, "Must have committed region");
3393 assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3394 assert(committed_start >= start && committed_start < start + size, "Out of range");
3395 committed_size = page_sz * committed_pages;
3396 return true;
3397 } else {
3398 assert(committed_pages == 0, "Should not have committed region");
3399 return false;
3400 }
3401 }
3402
3403
3404 // Linux uses a growable mapping for the stack, and if the mapping for
3405 // the stack guard pages is not removed when we detach a thread the
3406 // stack cannot grow beyond the pages where the stack guard was
3407 // mapped. If at some point later in the process the stack expands to
3408 // that point, the Linux kernel cannot expand the stack any further
3409 // because the guard pages are in the way, and a segfault occurs.
3410 //
3411 // However, it's essential not to split the stack region by unmapping
3412 // a region (leaving a hole) that's already part of the stack mapping,
3413 // so if the stack mapping has already grown beyond the guard pages at
3414 // the time we create them, we have to truncate the stack mapping.
3415 // So, we need to know the extent of the stack mapping when
3416 // create_stack_guard_pages() is called.
3417
3418 // We only need this for stacks that are growable: at the time of
3419 // writing thread stacks don't use growable mappings (i.e. those
3420 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3421 // only applies to the main thread.
3422
3423 // If the (growable) stack mapping already extends beyond the point
3424 // where we're going to put our guard pages, truncate the mapping at
3425 // that point by munmap()ping it. This ensures that when we later
3426 // munmap() the guard pages we don't leave a hole in the stack
3427 // mapping. This only affects the main/primordial thread
3428
3429 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3430 if (os::is_primordial_thread()) {
3431 // As we manually grow stack up to bottom inside create_attached_thread(),
3432 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3433 // we don't need to do anything special.
3434 // Check it first, before calling heavy function.
3435 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3436 unsigned char vec[1];
3437
3438 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3439 // Fallback to slow path on all errors, including EAGAIN
3440 stack_extent = (uintptr_t) get_stack_commited_bottom(
3441 os::Linux::initial_thread_stack_bottom(),
3442 (size_t)addr - stack_extent);
3443 }
3444
3445 if (stack_extent < (uintptr_t)addr) {
3446 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3447 }
3448 }
3449
3450 return os::commit_memory(addr, size, !ExecMem);
3451 }
3452
3453 // If this is a growable mapping, remove the guard pages entirely by
3454 // munmap()ping them. If not, just call uncommit_memory(). This only
3455 // affects the main/primordial thread, but guard against future OS changes.
3456 // It's safe to always unmap guard pages for primordial thread because we
3457 // always place it right after end of the mapped region.
3458
3459 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3460 uintptr_t stack_extent, stack_base;
3461
3462 if (os::is_primordial_thread()) {
3463 return ::munmap(addr, size) == 0;
3464 }
3465
3466 return os::uncommit_memory(addr, size);
3467 }
3468
3469 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3470 // at 'requested_addr'. If there are existing memory mappings at the same
3471 // location, however, they will be overwritten. If 'fixed' is false,
3472 // 'requested_addr' is only treated as a hint, the return value may or
3473 // may not start from the requested address. Unlike Linux mmap(), this
3474 // function returns NULL to indicate failure.
3475 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3476 char * addr;
3477 int flags;
3478
3479 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3480 if (fixed) {
3481 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3482 flags |= MAP_FIXED;
3483 }
3484
3485 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3486 // touch an uncommitted page. Otherwise, the read/write might
3487 // succeed if we have enough swap space to back the physical page.
3488 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3489 flags, -1, 0);
3490
3491 return addr == MAP_FAILED ? NULL : addr;
3492 }
3493
3494 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3495 // (req_addr != NULL) or with a given alignment.
3496 // - bytes shall be a multiple of alignment.
3497 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3498 // - alignment sets the alignment at which memory shall be allocated.
3499 // It must be a multiple of allocation granularity.
3500 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3501 // req_addr or NULL.
3502 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3503
3504 size_t extra_size = bytes;
3505 if (req_addr == NULL && alignment > 0) {
3506 extra_size += alignment;
3507 }
3508
3509 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3510 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3511 -1, 0);
3512 if (start == MAP_FAILED) {
3513 start = NULL;
3514 } else {
3515 if (req_addr != NULL) {
3516 if (start != req_addr) {
3517 ::munmap(start, extra_size);
3518 start = NULL;
3519 }
3520 } else {
3521 char* const start_aligned = align_up(start, alignment);
3522 char* const end_aligned = start_aligned + bytes;
3523 char* const end = start + extra_size;
3524 if (start_aligned > start) {
3525 ::munmap(start, start_aligned - start);
3526 }
3527 if (end_aligned < end) {
3528 ::munmap(end_aligned, end - end_aligned);
3529 }
3530 start = start_aligned;
3531 }
3532 }
3533 return start;
3534 }
3535
3536 static int anon_munmap(char * addr, size_t size) {
3537 return ::munmap(addr, size) == 0;
3538 }
3539
3540 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3541 size_t alignment_hint) {
3542 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3543 }
3544
3545 bool os::pd_release_memory(char* addr, size_t size) {
3546 return anon_munmap(addr, size);
3547 }
3548
3549 static bool linux_mprotect(char* addr, size_t size, int prot) {
3550 // Linux wants the mprotect address argument to be page aligned.
3551 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3552
3553 // According to SUSv3, mprotect() should only be used with mappings
3554 // established by mmap(), and mmap() always maps whole pages. Unaligned
3555 // 'addr' likely indicates problem in the VM (e.g. trying to change
3556 // protection of malloc'ed or statically allocated memory). Check the
3557 // caller if you hit this assert.
3558 assert(addr == bottom, "sanity check");
3559
3560 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3561 Events::log(NULL, "Protecting memory [" INTPTR_FORMAT "," INTPTR_FORMAT "] with protection modes %x", p2i(bottom), p2i(bottom+size), prot);
3562 return ::mprotect(bottom, size, prot) == 0;
3563 }
3564
3565 // Set protections specified
3566 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3567 bool is_committed) {
3568 unsigned int p = 0;
3569 switch (prot) {
3570 case MEM_PROT_NONE: p = PROT_NONE; break;
3571 case MEM_PROT_READ: p = PROT_READ; break;
3572 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3573 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3574 default:
3575 ShouldNotReachHere();
3576 }
3577 // is_committed is unused.
3578 return linux_mprotect(addr, bytes, p);
3579 }
3580
3581 bool os::guard_memory(char* addr, size_t size) {
3582 return linux_mprotect(addr, size, PROT_NONE);
3583 }
3584
3585 bool os::unguard_memory(char* addr, size_t size) {
3586 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3587 }
3588
3589 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3590 size_t page_size) {
3591 bool result = false;
3592 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3593 MAP_ANONYMOUS|MAP_PRIVATE,
3594 -1, 0);
3595 if (p != MAP_FAILED) {
3596 void *aligned_p = align_up(p, page_size);
3597
3598 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3599
3600 munmap(p, page_size * 2);
3601 }
3602
3603 if (warn && !result) {
3604 warning("TransparentHugePages is not supported by the operating system.");
3605 }
3606
3607 return result;
3608 }
3609
3610 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3611 bool result = false;
3612 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3613 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3614 -1, 0);
3615
3616 if (p != MAP_FAILED) {
3617 // We don't know if this really is a huge page or not.
3618 FILE *fp = fopen("/proc/self/maps", "r");
3619 if (fp) {
3620 while (!feof(fp)) {
3621 char chars[257];
3622 long x = 0;
3623 if (fgets(chars, sizeof(chars), fp)) {
3624 if (sscanf(chars, "%lx-%*x", &x) == 1
3625 && x == (long)p) {
3626 if (strstr (chars, "hugepage")) {
3627 result = true;
3628 break;
3629 }
3630 }
3631 }
3632 }
3633 fclose(fp);
3634 }
3635 munmap(p, page_size);
3636 }
3637
3638 if (warn && !result) {
3639 warning("HugeTLBFS is not supported by the operating system.");
3640 }
3641
3642 return result;
3643 }
3644
3645 // From the coredump_filter documentation:
3646 //
3647 // - (bit 0) anonymous private memory
3648 // - (bit 1) anonymous shared memory
3649 // - (bit 2) file-backed private memory
3650 // - (bit 3) file-backed shared memory
3651 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3652 // effective only if the bit 2 is cleared)
3653 // - (bit 5) hugetlb private memory
3654 // - (bit 6) hugetlb shared memory
3655 // - (bit 7) dax private memory
3656 // - (bit 8) dax shared memory
3657 //
3658 static void set_coredump_filter(CoredumpFilterBit bit) {
3659 FILE *f;
3660 long cdm;
3661
3662 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3663 return;
3664 }
3665
3666 if (fscanf(f, "%lx", &cdm) != 1) {
3667 fclose(f);
3668 return;
3669 }
3670
3671 long saved_cdm = cdm;
3672 rewind(f);
3673 cdm |= bit;
3674
3675 if (cdm != saved_cdm) {
3676 fprintf(f, "%#lx", cdm);
3677 }
3678
3679 fclose(f);
3680 }
3681
3682 // Large page support
3683
3684 static size_t _large_page_size = 0;
3685
3686 size_t os::Linux::find_large_page_size() {
3687 size_t large_page_size = 0;
3688
3689 // large_page_size on Linux is used to round up heap size. x86 uses either
3690 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3691 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3692 // page as large as 256M.
3693 //
3694 // Here we try to figure out page size by parsing /proc/meminfo and looking
3695 // for a line with the following format:
3696 // Hugepagesize: 2048 kB
3697 //
3698 // If we can't determine the value (e.g. /proc is not mounted, or the text
3699 // format has been changed), we'll use the largest page size supported by
3700 // the processor.
3701
3702 #ifndef ZERO
3703 large_page_size =
3704 AARCH64_ONLY(2 * M)
3705 AMD64_ONLY(2 * M)
3706 ARM32_ONLY(2 * M)
3707 IA32_ONLY(4 * M)
3708 IA64_ONLY(256 * M)
3709 PPC_ONLY(4 * M)
3710 S390_ONLY(1 * M)
3711 SPARC_ONLY(4 * M);
3712 #endif // ZERO
3713
3714 FILE *fp = fopen("/proc/meminfo", "r");
3715 if (fp) {
3716 while (!feof(fp)) {
3717 int x = 0;
3718 char buf[16];
3719 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3720 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3721 large_page_size = x * K;
3722 break;
3723 }
3724 } else {
3725 // skip to next line
3726 for (;;) {
3727 int ch = fgetc(fp);
3728 if (ch == EOF || ch == (int)'\n') break;
3729 }
3730 }
3731 }
3732 fclose(fp);
3733 }
3734
3735 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3736 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3737 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3738 proper_unit_for_byte_size(large_page_size));
3739 }
3740
3741 return large_page_size;
3742 }
3743
3744 size_t os::Linux::setup_large_page_size() {
3745 _large_page_size = Linux::find_large_page_size();
3746 const size_t default_page_size = (size_t)Linux::page_size();
3747 if (_large_page_size > default_page_size) {
3748 _page_sizes[0] = _large_page_size;
3749 _page_sizes[1] = default_page_size;
3750 _page_sizes[2] = 0;
3751 }
3752
3753 return _large_page_size;
3754 }
3755
3756 bool os::Linux::setup_large_page_type(size_t page_size) {
3757 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3758 FLAG_IS_DEFAULT(UseSHM) &&
3759 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3760
3761 // The type of large pages has not been specified by the user.
3762
3763 // Try UseHugeTLBFS and then UseSHM.
3764 UseHugeTLBFS = UseSHM = true;
3765
3766 // Don't try UseTransparentHugePages since there are known
3767 // performance issues with it turned on. This might change in the future.
3768 UseTransparentHugePages = false;
3769 }
3770
3771 if (UseTransparentHugePages) {
3772 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3773 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3774 UseHugeTLBFS = false;
3775 UseSHM = false;
3776 return true;
3777 }
3778 UseTransparentHugePages = false;
3779 }
3780
3781 if (UseHugeTLBFS) {
3782 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3783 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3784 UseSHM = false;
3785 return true;
3786 }
3787 UseHugeTLBFS = false;
3788 }
3789
3790 return UseSHM;
3791 }
3792
3793 void os::large_page_init() {
3794 if (!UseLargePages &&
3795 !UseTransparentHugePages &&
3796 !UseHugeTLBFS &&
3797 !UseSHM) {
3798 // Not using large pages.
3799 return;
3800 }
3801
3802 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3803 // The user explicitly turned off large pages.
3804 // Ignore the rest of the large pages flags.
3805 UseTransparentHugePages = false;
3806 UseHugeTLBFS = false;
3807 UseSHM = false;
3808 return;
3809 }
3810
3811 size_t large_page_size = Linux::setup_large_page_size();
3812 UseLargePages = Linux::setup_large_page_type(large_page_size);
3813
3814 set_coredump_filter(LARGEPAGES_BIT);
3815 }
3816
3817 #ifndef SHM_HUGETLB
3818 #define SHM_HUGETLB 04000
3819 #endif
3820
3821 #define shm_warning_format(format, ...) \
3822 do { \
3823 if (UseLargePages && \
3824 (!FLAG_IS_DEFAULT(UseLargePages) || \
3825 !FLAG_IS_DEFAULT(UseSHM) || \
3826 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3827 warning(format, __VA_ARGS__); \
3828 } \
3829 } while (0)
3830
3831 #define shm_warning(str) shm_warning_format("%s", str)
3832
3833 #define shm_warning_with_errno(str) \
3834 do { \
3835 int err = errno; \
3836 shm_warning_format(str " (error = %d)", err); \
3837 } while (0)
3838
3839 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3840 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3841
3842 if (!is_aligned(alignment, SHMLBA)) {
3843 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3844 return NULL;
3845 }
3846
3847 // To ensure that we get 'alignment' aligned memory from shmat,
3848 // we pre-reserve aligned virtual memory and then attach to that.
3849
3850 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3851 if (pre_reserved_addr == NULL) {
3852 // Couldn't pre-reserve aligned memory.
3853 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3854 return NULL;
3855 }
3856
3857 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3858 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3859
3860 if ((intptr_t)addr == -1) {
3861 int err = errno;
3862 shm_warning_with_errno("Failed to attach shared memory.");
3863
3864 assert(err != EACCES, "Unexpected error");
3865 assert(err != EIDRM, "Unexpected error");
3866 assert(err != EINVAL, "Unexpected error");
3867
3868 // Since we don't know if the kernel unmapped the pre-reserved memory area
3869 // we can't unmap it, since that would potentially unmap memory that was
3870 // mapped from other threads.
3871 return NULL;
3872 }
3873
3874 return addr;
3875 }
3876
3877 static char* shmat_at_address(int shmid, char* req_addr) {
3878 if (!is_aligned(req_addr, SHMLBA)) {
3879 assert(false, "Requested address needs to be SHMLBA aligned");
3880 return NULL;
3881 }
3882
3883 char* addr = (char*)shmat(shmid, req_addr, 0);
3884
3885 if ((intptr_t)addr == -1) {
3886 shm_warning_with_errno("Failed to attach shared memory.");
3887 return NULL;
3888 }
3889
3890 return addr;
3891 }
3892
3893 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3894 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3895 if (req_addr != NULL) {
3896 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3897 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3898 return shmat_at_address(shmid, req_addr);
3899 }
3900
3901 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3902 // return large page size aligned memory addresses when req_addr == NULL.
3903 // However, if the alignment is larger than the large page size, we have
3904 // to manually ensure that the memory returned is 'alignment' aligned.
3905 if (alignment > os::large_page_size()) {
3906 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3907 return shmat_with_alignment(shmid, bytes, alignment);
3908 } else {
3909 return shmat_at_address(shmid, NULL);
3910 }
3911 }
3912
3913 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3914 char* req_addr, bool exec) {
3915 // "exec" is passed in but not used. Creating the shared image for
3916 // the code cache doesn't have an SHM_X executable permission to check.
3917 assert(UseLargePages && UseSHM, "only for SHM large pages");
3918 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3919 assert(is_aligned(req_addr, alignment), "Unaligned address");
3920
3921 if (!is_aligned(bytes, os::large_page_size())) {
3922 return NULL; // Fallback to small pages.
3923 }
3924
3925 // Create a large shared memory region to attach to based on size.
3926 // Currently, size is the total size of the heap.
3927 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3928 if (shmid == -1) {
3929 // Possible reasons for shmget failure:
3930 // 1. shmmax is too small for Java heap.
3931 // > check shmmax value: cat /proc/sys/kernel/shmmax
3932 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3933 // 2. not enough large page memory.
3934 // > check available large pages: cat /proc/meminfo
3935 // > increase amount of large pages:
3936 // echo new_value > /proc/sys/vm/nr_hugepages
3937 // Note 1: different Linux may use different name for this property,
3938 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3939 // Note 2: it's possible there's enough physical memory available but
3940 // they are so fragmented after a long run that they can't
3941 // coalesce into large pages. Try to reserve large pages when
3942 // the system is still "fresh".
3943 shm_warning_with_errno("Failed to reserve shared memory.");
3944 return NULL;
3945 }
3946
3947 // Attach to the region.
3948 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3949
3950 // Remove shmid. If shmat() is successful, the actual shared memory segment
3951 // will be deleted when it's detached by shmdt() or when the process
3952 // terminates. If shmat() is not successful this will remove the shared
3953 // segment immediately.
3954 shmctl(shmid, IPC_RMID, NULL);
3955
3956 return addr;
3957 }
3958
3959 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3960 int error) {
3961 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3962
3963 bool warn_on_failure = UseLargePages &&
3964 (!FLAG_IS_DEFAULT(UseLargePages) ||
3965 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3966 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3967
3968 if (warn_on_failure) {
3969 char msg[128];
3970 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3971 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3972 warning("%s", msg);
3973 }
3974 }
3975
3976 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3977 char* req_addr,
3978 bool exec) {
3979 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3980 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3981 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3982
3983 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3984 char* addr = (char*)::mmap(req_addr, bytes, prot,
3985 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3986 -1, 0);
3987
3988 if (addr == MAP_FAILED) {
3989 warn_on_large_pages_failure(req_addr, bytes, errno);
3990 return NULL;
3991 }
3992
3993 assert(is_aligned(addr, os::large_page_size()), "Must be");
3994
3995 return addr;
3996 }
3997
3998 // Reserve memory using mmap(MAP_HUGETLB).
3999 // - bytes shall be a multiple of alignment.
4000 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
4001 // - alignment sets the alignment at which memory shall be allocated.
4002 // It must be a multiple of allocation granularity.
4003 // Returns address of memory or NULL. If req_addr was not NULL, will only return
4004 // req_addr or NULL.
4005 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
4006 size_t alignment,
4007 char* req_addr,
4008 bool exec) {
4009 size_t large_page_size = os::large_page_size();
4010 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
4011
4012 assert(is_aligned(req_addr, alignment), "Must be");
4013 assert(is_aligned(bytes, alignment), "Must be");
4014
4015 // First reserve - but not commit - the address range in small pages.
4016 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
4017
4018 if (start == NULL) {
4019 return NULL;
4020 }
4021
4022 assert(is_aligned(start, alignment), "Must be");
4023
4024 char* end = start + bytes;
4025
4026 // Find the regions of the allocated chunk that can be promoted to large pages.
4027 char* lp_start = align_up(start, large_page_size);
4028 char* lp_end = align_down(end, large_page_size);
4029
4030 size_t lp_bytes = lp_end - lp_start;
4031
4032 assert(is_aligned(lp_bytes, large_page_size), "Must be");
4033
4034 if (lp_bytes == 0) {
4035 // The mapped region doesn't even span the start and the end of a large page.
4036 // Fall back to allocate a non-special area.
4037 ::munmap(start, end - start);
4038 return NULL;
4039 }
4040
4041 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
4042
4043 void* result;
4044
4045 // Commit small-paged leading area.
4046 if (start != lp_start) {
4047 result = ::mmap(start, lp_start - start, prot,
4048 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
4049 -1, 0);
4050 if (result == MAP_FAILED) {
4051 ::munmap(lp_start, end - lp_start);
4052 return NULL;
4053 }
4054 }
4055
4056 // Commit large-paged area.
4057 result = ::mmap(lp_start, lp_bytes, prot,
4058 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
4059 -1, 0);
4060 if (result == MAP_FAILED) {
4061 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
4062 // If the mmap above fails, the large pages region will be unmapped and we
4063 // have regions before and after with small pages. Release these regions.
4064 //
4065 // | mapped | unmapped | mapped |
4066 // ^ ^ ^ ^
4067 // start lp_start lp_end end
4068 //
4069 ::munmap(start, lp_start - start);
4070 ::munmap(lp_end, end - lp_end);
4071 return NULL;
4072 }
4073
4074 // Commit small-paged trailing area.
4075 if (lp_end != end) {
4076 result = ::mmap(lp_end, end - lp_end, prot,
4077 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
4078 -1, 0);
4079 if (result == MAP_FAILED) {
4080 ::munmap(start, lp_end - start);
4081 return NULL;
4082 }
4083 }
4084
4085 return start;
4086 }
4087
4088 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
4089 size_t alignment,
4090 char* req_addr,
4091 bool exec) {
4092 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
4093 assert(is_aligned(req_addr, alignment), "Must be");
4094 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
4095 assert(is_power_of_2(os::large_page_size()), "Must be");
4096 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
4097
4098 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
4099 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
4100 } else {
4101 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
4102 }
4103 }
4104
4105 char* os::reserve_memory_special(size_t bytes, size_t alignment,
4106 char* req_addr, bool exec) {
4107 assert(UseLargePages, "only for large pages");
4108
4109 char* addr;
4110 if (UseSHM) {
4111 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
4112 } else {
4113 assert(UseHugeTLBFS, "must be");
4114 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
4115 }
4116
4117 if (addr != NULL) {
4118 if (UseNUMAInterleaving) {
4119 numa_make_global(addr, bytes);
4120 }
4121
4122 // The memory is committed
4123 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
4124 }
4125
4126 return addr;
4127 }
4128
4129 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
4130 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
4131 return shmdt(base) == 0;
4132 }
4133
4134 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
4135 return pd_release_memory(base, bytes);
4136 }
4137
4138 bool os::release_memory_special(char* base, size_t bytes) {
4139 bool res;
4140 if (MemTracker::tracking_level() > NMT_minimal) {
4141 Tracker tkr(Tracker::release);
4142 res = os::Linux::release_memory_special_impl(base, bytes);
4143 if (res) {
4144 tkr.record((address)base, bytes);
4145 }
4146
4147 } else {
4148 res = os::Linux::release_memory_special_impl(base, bytes);
4149 }
4150 return res;
4151 }
4152
4153 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
4154 assert(UseLargePages, "only for large pages");
4155 bool res;
4156
4157 if (UseSHM) {
4158 res = os::Linux::release_memory_special_shm(base, bytes);
4159 } else {
4160 assert(UseHugeTLBFS, "must be");
4161 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
4162 }
4163 return res;
4164 }
4165
4166 size_t os::large_page_size() {
4167 return _large_page_size;
4168 }
4169
4170 // With SysV SHM the entire memory region must be allocated as shared
4171 // memory.
4172 // HugeTLBFS allows application to commit large page memory on demand.
4173 // However, when committing memory with HugeTLBFS fails, the region
4174 // that was supposed to be committed will lose the old reservation
4175 // and allow other threads to steal that memory region. Because of this
4176 // behavior we can't commit HugeTLBFS memory.
4177 bool os::can_commit_large_page_memory() {
4178 return UseTransparentHugePages;
4179 }
4180
4181 bool os::can_execute_large_page_memory() {
4182 return UseTransparentHugePages || UseHugeTLBFS;
4183 }
4184
4185 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
4186 assert(file_desc >= 0, "file_desc is not valid");
4187 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
4188 if (result != NULL) {
4189 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
4190 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
4191 }
4192 }
4193 return result;
4194 }
4195
4196 // Reserve memory at an arbitrary address, only if that area is
4197 // available (and not reserved for something else).
4198
4199 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
4200 // Assert only that the size is a multiple of the page size, since
4201 // that's all that mmap requires, and since that's all we really know
4202 // about at this low abstraction level. If we need higher alignment,
4203 // we can either pass an alignment to this method or verify alignment
4204 // in one of the methods further up the call chain. See bug 5044738.
4205 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
4206
4207 // Repeatedly allocate blocks until the block is allocated at the
4208 // right spot.
4209
4210 // Linux mmap allows caller to pass an address as hint; give it a try first,
4211 // if kernel honors the hint then we can return immediately.
4212 char * addr = anon_mmap(requested_addr, bytes, false);
4213 if (addr == requested_addr) {
4214 return requested_addr;
4215 }
4216
4217 if (addr != NULL) {
4218 // mmap() is successful but it fails to reserve at the requested address
4219 anon_munmap(addr, bytes);
4220 }
4221
4222 return NULL;
4223 }
4224
4225 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4226 void os::infinite_sleep() {
4227 while (true) { // sleep forever ...
4228 ::sleep(100); // ... 100 seconds at a time
4229 }
4230 }
4231
4232 // Used to convert frequent JVM_Yield() to nops
4233 bool os::dont_yield() {
4234 return DontYieldALot;
4235 }
4236
4237 // Linux CFS scheduler (since 2.6.23) does not guarantee sched_yield(2) will
4238 // actually give up the CPU. Since skip buddy (v2.6.28):
4239 //
4240 // * Sets the yielding task as skip buddy for current CPU's run queue.
4241 // * Picks next from run queue, if empty, picks a skip buddy (can be the yielding task).
4242 // * Clears skip buddies for this run queue (yielding task no longer a skip buddy).
4243 //
4244 // An alternative is calling os::naked_short_nanosleep with a small number to avoid
4245 // getting re-scheduled immediately.
4246 //
4247 void os::naked_yield() {
4248 sched_yield();
4249 }
4250
4251 ////////////////////////////////////////////////////////////////////////////////
4252 // thread priority support
4253
4254 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4255 // only supports dynamic priority, static priority must be zero. For real-time
4256 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4257 // However, for large multi-threaded applications, SCHED_RR is not only slower
4258 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4259 // of 5 runs - Sep 2005).
4260 //
4261 // The following code actually changes the niceness of kernel-thread/LWP. It
4262 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4263 // not the entire user process, and user level threads are 1:1 mapped to kernel
4264 // threads. It has always been the case, but could change in the future. For
4265 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4266 // It is only used when ThreadPriorityPolicy=1 and may require system level permission
4267 // (e.g., root privilege or CAP_SYS_NICE capability).
4268
4269 int os::java_to_os_priority[CriticalPriority + 1] = {
4270 19, // 0 Entry should never be used
4271
4272 4, // 1 MinPriority
4273 3, // 2
4274 2, // 3
4275
4276 1, // 4
4277 0, // 5 NormPriority
4278 -1, // 6
4279
4280 -2, // 7
4281 -3, // 8
4282 -4, // 9 NearMaxPriority
4283
4284 -5, // 10 MaxPriority
4285
4286 -5 // 11 CriticalPriority
4287 };
4288
4289 static int prio_init() {
4290 if (ThreadPriorityPolicy == 1) {
4291 if (geteuid() != 0) {
4292 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4293 warning("-XX:ThreadPriorityPolicy=1 may require system level permission, " \
4294 "e.g., being the root user. If the necessary permission is not " \
4295 "possessed, changes to priority will be silently ignored.");
4296 }
4297 }
4298 }
4299 if (UseCriticalJavaThreadPriority) {
4300 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4301 }
4302 return 0;
4303 }
4304
4305 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4306 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4307
4308 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4309 return (ret == 0) ? OS_OK : OS_ERR;
4310 }
4311
4312 OSReturn os::get_native_priority(const Thread* const thread,
4313 int *priority_ptr) {
4314 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4315 *priority_ptr = java_to_os_priority[NormPriority];
4316 return OS_OK;
4317 }
4318
4319 errno = 0;
4320 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4321 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4322 }
4323
4324 ////////////////////////////////////////////////////////////////////////////////
4325 // suspend/resume support
4326
4327 // The low-level signal-based suspend/resume support is a remnant from the
4328 // old VM-suspension that used to be for java-suspension, safepoints etc,
4329 // within hotspot. Currently used by JFR's OSThreadSampler
4330 //
4331 // The remaining code is greatly simplified from the more general suspension
4332 // code that used to be used.
4333 //
4334 // The protocol is quite simple:
4335 // - suspend:
4336 // - sends a signal to the target thread
4337 // - polls the suspend state of the osthread using a yield loop
4338 // - target thread signal handler (SR_handler) sets suspend state
4339 // and blocks in sigsuspend until continued
4340 // - resume:
4341 // - sets target osthread state to continue
4342 // - sends signal to end the sigsuspend loop in the SR_handler
4343 //
4344 // Note that the SR_lock plays no role in this suspend/resume protocol,
4345 // but is checked for NULL in SR_handler as a thread termination indicator.
4346 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4347 //
4348 // Note that resume_clear_context() and suspend_save_context() are needed
4349 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4350 // which in part is used by:
4351 // - Forte Analyzer: AsyncGetCallTrace()
4352 // - StackBanging: get_frame_at_stack_banging_point()
4353
4354 static void resume_clear_context(OSThread *osthread) {
4355 osthread->set_ucontext(NULL);
4356 osthread->set_siginfo(NULL);
4357 }
4358
4359 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4360 ucontext_t* context) {
4361 osthread->set_ucontext(context);
4362 osthread->set_siginfo(siginfo);
4363 }
4364
4365 // Handler function invoked when a thread's execution is suspended or
4366 // resumed. We have to be careful that only async-safe functions are
4367 // called here (Note: most pthread functions are not async safe and
4368 // should be avoided.)
4369 //
4370 // Note: sigwait() is a more natural fit than sigsuspend() from an
4371 // interface point of view, but sigwait() prevents the signal hander
4372 // from being run. libpthread would get very confused by not having
4373 // its signal handlers run and prevents sigwait()'s use with the
4374 // mutex granting granting signal.
4375 //
4376 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4377 //
4378 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4379 // Save and restore errno to avoid confusing native code with EINTR
4380 // after sigsuspend.
4381 int old_errno = errno;
4382
4383 Thread* thread = Thread::current_or_null_safe();
4384 assert(thread != NULL, "Missing current thread in SR_handler");
4385
4386 // On some systems we have seen signal delivery get "stuck" until the signal
4387 // mask is changed as part of thread termination. Check that the current thread
4388 // has not already terminated (via SR_lock()) - else the following assertion
4389 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4390 // destructor has completed.
4391
4392 if (thread->SR_lock() == NULL) {
4393 return;
4394 }
4395
4396 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4397
4398 OSThread* osthread = thread->osthread();
4399
4400 os::SuspendResume::State current = osthread->sr.state();
4401 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4402 suspend_save_context(osthread, siginfo, context);
4403
4404 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4405 os::SuspendResume::State state = osthread->sr.suspended();
4406 if (state == os::SuspendResume::SR_SUSPENDED) {
4407 sigset_t suspend_set; // signals for sigsuspend()
4408 sigemptyset(&suspend_set);
4409 // get current set of blocked signals and unblock resume signal
4410 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4411 sigdelset(&suspend_set, SR_signum);
4412
4413 sr_semaphore.signal();
4414 // wait here until we are resumed
4415 while (1) {
4416 sigsuspend(&suspend_set);
4417
4418 os::SuspendResume::State result = osthread->sr.running();
4419 if (result == os::SuspendResume::SR_RUNNING) {
4420 sr_semaphore.signal();
4421 break;
4422 }
4423 }
4424
4425 } else if (state == os::SuspendResume::SR_RUNNING) {
4426 // request was cancelled, continue
4427 } else {
4428 ShouldNotReachHere();
4429 }
4430
4431 resume_clear_context(osthread);
4432 } else if (current == os::SuspendResume::SR_RUNNING) {
4433 // request was cancelled, continue
4434 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4435 // ignore
4436 } else {
4437 // ignore
4438 }
4439
4440 errno = old_errno;
4441 }
4442
4443 static int SR_initialize() {
4444 struct sigaction act;
4445 char *s;
4446
4447 // Get signal number to use for suspend/resume
4448 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4449 int sig = ::strtol(s, 0, 10);
4450 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4451 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4452 SR_signum = sig;
4453 } else {
4454 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4455 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4456 }
4457 }
4458
4459 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4460 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4461
4462 sigemptyset(&SR_sigset);
4463 sigaddset(&SR_sigset, SR_signum);
4464
4465 // Set up signal handler for suspend/resume
4466 act.sa_flags = SA_RESTART|SA_SIGINFO;
4467 act.sa_handler = (void (*)(int)) SR_handler;
4468
4469 // SR_signum is blocked by default.
4470 // 4528190 - We also need to block pthread restart signal (32 on all
4471 // supported Linux platforms). Note that LinuxThreads need to block
4472 // this signal for all threads to work properly. So we don't have
4473 // to use hard-coded signal number when setting up the mask.
4474 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4475
4476 if (sigaction(SR_signum, &act, 0) == -1) {
4477 return -1;
4478 }
4479
4480 // Save signal flag
4481 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4482 return 0;
4483 }
4484
4485 static int sr_notify(OSThread* osthread) {
4486 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4487 assert_status(status == 0, status, "pthread_kill");
4488 return status;
4489 }
4490
4491 // "Randomly" selected value for how long we want to spin
4492 // before bailing out on suspending a thread, also how often
4493 // we send a signal to a thread we want to resume
4494 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4495 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4496
4497 // returns true on success and false on error - really an error is fatal
4498 // but this seems the normal response to library errors
4499 static bool do_suspend(OSThread* osthread) {
4500 assert(osthread->sr.is_running(), "thread should be running");
4501 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4502
4503 // mark as suspended and send signal
4504 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4505 // failed to switch, state wasn't running?
4506 ShouldNotReachHere();
4507 return false;
4508 }
4509
4510 if (sr_notify(osthread) != 0) {
4511 ShouldNotReachHere();
4512 }
4513
4514 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4515 while (true) {
4516 if (sr_semaphore.timedwait(2)) {
4517 break;
4518 } else {
4519 // timeout
4520 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4521 if (cancelled == os::SuspendResume::SR_RUNNING) {
4522 return false;
4523 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4524 // make sure that we consume the signal on the semaphore as well
4525 sr_semaphore.wait();
4526 break;
4527 } else {
4528 ShouldNotReachHere();
4529 return false;
4530 }
4531 }
4532 }
4533
4534 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4535 return true;
4536 }
4537
4538 static void do_resume(OSThread* osthread) {
4539 assert(osthread->sr.is_suspended(), "thread should be suspended");
4540 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4541
4542 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4543 // failed to switch to WAKEUP_REQUEST
4544 ShouldNotReachHere();
4545 return;
4546 }
4547
4548 while (true) {
4549 if (sr_notify(osthread) == 0) {
4550 if (sr_semaphore.timedwait(2)) {
4551 if (osthread->sr.is_running()) {
4552 return;
4553 }
4554 }
4555 } else {
4556 ShouldNotReachHere();
4557 }
4558 }
4559
4560 guarantee(osthread->sr.is_running(), "Must be running!");
4561 }
4562
4563 ///////////////////////////////////////////////////////////////////////////////////
4564 // signal handling (except suspend/resume)
4565
4566 // This routine may be used by user applications as a "hook" to catch signals.
4567 // The user-defined signal handler must pass unrecognized signals to this
4568 // routine, and if it returns true (non-zero), then the signal handler must
4569 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4570 // routine will never retun false (zero), but instead will execute a VM panic
4571 // routine kill the process.
4572 //
4573 // If this routine returns false, it is OK to call it again. This allows
4574 // the user-defined signal handler to perform checks either before or after
4575 // the VM performs its own checks. Naturally, the user code would be making
4576 // a serious error if it tried to handle an exception (such as a null check
4577 // or breakpoint) that the VM was generating for its own correct operation.
4578 //
4579 // This routine may recognize any of the following kinds of signals:
4580 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4581 // It should be consulted by handlers for any of those signals.
4582 //
4583 // The caller of this routine must pass in the three arguments supplied
4584 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4585 // field of the structure passed to sigaction(). This routine assumes that
4586 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4587 //
4588 // Note that the VM will print warnings if it detects conflicting signal
4589 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4590 //
4591 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4592 siginfo_t* siginfo,
4593 void* ucontext,
4594 int abort_if_unrecognized);
4595
4596 static void signalHandler(int sig, siginfo_t* info, void* uc) {
4597 assert(info != NULL && uc != NULL, "it must be old kernel");
4598 int orig_errno = errno; // Preserve errno value over signal handler.
4599 JVM_handle_linux_signal(sig, info, uc, true);
4600 errno = orig_errno;
4601 }
4602
4603
4604 // This boolean allows users to forward their own non-matching signals
4605 // to JVM_handle_linux_signal, harmlessly.
4606 bool os::Linux::signal_handlers_are_installed = false;
4607
4608 // For signal-chaining
4609 bool os::Linux::libjsig_is_loaded = false;
4610 typedef struct sigaction *(*get_signal_t)(int);
4611 get_signal_t os::Linux::get_signal_action = NULL;
4612
4613 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4614 struct sigaction *actp = NULL;
4615
4616 if (libjsig_is_loaded) {
4617 // Retrieve the old signal handler from libjsig
4618 actp = (*get_signal_action)(sig);
4619 }
4620 if (actp == NULL) {
4621 // Retrieve the preinstalled signal handler from jvm
4622 actp = os::Posix::get_preinstalled_handler(sig);
4623 }
4624
4625 return actp;
4626 }
4627
4628 static bool call_chained_handler(struct sigaction *actp, int sig,
4629 siginfo_t *siginfo, void *context) {
4630 // Call the old signal handler
4631 if (actp->sa_handler == SIG_DFL) {
4632 // It's more reasonable to let jvm treat it as an unexpected exception
4633 // instead of taking the default action.
4634 return false;
4635 } else if (actp->sa_handler != SIG_IGN) {
4636 if ((actp->sa_flags & SA_NODEFER) == 0) {
4637 // automaticlly block the signal
4638 sigaddset(&(actp->sa_mask), sig);
4639 }
4640
4641 sa_handler_t hand = NULL;
4642 sa_sigaction_t sa = NULL;
4643 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4644 // retrieve the chained handler
4645 if (siginfo_flag_set) {
4646 sa = actp->sa_sigaction;
4647 } else {
4648 hand = actp->sa_handler;
4649 }
4650
4651 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4652 actp->sa_handler = SIG_DFL;
4653 }
4654
4655 // try to honor the signal mask
4656 sigset_t oset;
4657 sigemptyset(&oset);
4658 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4659
4660 // call into the chained handler
4661 if (siginfo_flag_set) {
4662 (*sa)(sig, siginfo, context);
4663 } else {
4664 (*hand)(sig);
4665 }
4666
4667 // restore the signal mask
4668 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4669 }
4670 // Tell jvm's signal handler the signal is taken care of.
4671 return true;
4672 }
4673
4674 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4675 bool chained = false;
4676 // signal-chaining
4677 if (UseSignalChaining) {
4678 struct sigaction *actp = get_chained_signal_action(sig);
4679 if (actp != NULL) {
4680 chained = call_chained_handler(actp, sig, siginfo, context);
4681 }
4682 }
4683 return chained;
4684 }
4685
4686 // for diagnostic
4687 int sigflags[NSIG];
4688
4689 int os::Linux::get_our_sigflags(int sig) {
4690 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4691 return sigflags[sig];
4692 }
4693
4694 void os::Linux::set_our_sigflags(int sig, int flags) {
4695 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4696 if (sig > 0 && sig < NSIG) {
4697 sigflags[sig] = flags;
4698 }
4699 }
4700
4701 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4702 // Check for overwrite.
4703 struct sigaction oldAct;
4704 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4705
4706 void* oldhand = oldAct.sa_sigaction
4707 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4708 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4709 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4710 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4711 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4712 if (AllowUserSignalHandlers || !set_installed) {
4713 // Do not overwrite; user takes responsibility to forward to us.
4714 return;
4715 } else if (UseSignalChaining) {
4716 // save the old handler in jvm
4717 os::Posix::save_preinstalled_handler(sig, oldAct);
4718 // libjsig also interposes the sigaction() call below and saves the
4719 // old sigaction on it own.
4720 } else {
4721 fatal("Encountered unexpected pre-existing sigaction handler "
4722 "%#lx for signal %d.", (long)oldhand, sig);
4723 }
4724 }
4725
4726 struct sigaction sigAct;
4727 sigfillset(&(sigAct.sa_mask));
4728 sigAct.sa_handler = SIG_DFL;
4729 if (!set_installed) {
4730 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4731 } else {
4732 sigAct.sa_sigaction = signalHandler;
4733 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4734 }
4735 // Save flags, which are set by ours
4736 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4737 sigflags[sig] = sigAct.sa_flags;
4738
4739 int ret = sigaction(sig, &sigAct, &oldAct);
4740 assert(ret == 0, "check");
4741
4742 void* oldhand2 = oldAct.sa_sigaction
4743 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4744 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4745 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4746 }
4747
4748 // install signal handlers for signals that HotSpot needs to
4749 // handle in order to support Java-level exception handling.
4750
4751 void os::Linux::install_signal_handlers() {
4752 if (!signal_handlers_are_installed) {
4753 signal_handlers_are_installed = true;
4754
4755 // signal-chaining
4756 typedef void (*signal_setting_t)();
4757 signal_setting_t begin_signal_setting = NULL;
4758 signal_setting_t end_signal_setting = NULL;
4759 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4760 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4761 if (begin_signal_setting != NULL) {
4762 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4763 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4764 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4765 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4766 libjsig_is_loaded = true;
4767 assert(UseSignalChaining, "should enable signal-chaining");
4768 }
4769 if (libjsig_is_loaded) {
4770 // Tell libjsig jvm is setting signal handlers
4771 (*begin_signal_setting)();
4772 }
4773
4774 set_signal_handler(SIGSEGV, true);
4775 set_signal_handler(SIGPIPE, true);
4776 set_signal_handler(SIGBUS, true);
4777 set_signal_handler(SIGILL, true);
4778 set_signal_handler(SIGFPE, true);
4779 #if defined(PPC64)
4780 set_signal_handler(SIGTRAP, true);
4781 #endif
4782 set_signal_handler(SIGXFSZ, true);
4783
4784 if (libjsig_is_loaded) {
4785 // Tell libjsig jvm finishes setting signal handlers
4786 (*end_signal_setting)();
4787 }
4788
4789 // We don't activate signal checker if libjsig is in place, we trust ourselves
4790 // and if UserSignalHandler is installed all bets are off.
4791 // Log that signal checking is off only if -verbose:jni is specified.
4792 if (CheckJNICalls) {
4793 if (libjsig_is_loaded) {
4794 if (PrintJNIResolving) {
4795 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4796 }
4797 check_signals = false;
4798 }
4799 if (AllowUserSignalHandlers) {
4800 if (PrintJNIResolving) {
4801 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4802 }
4803 check_signals = false;
4804 }
4805 }
4806 }
4807 }
4808
4809 // This is the fastest way to get thread cpu time on Linux.
4810 // Returns cpu time (user+sys) for any thread, not only for current.
4811 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4812 // It might work on 2.6.10+ with a special kernel/glibc patch.
4813 // For reference, please, see IEEE Std 1003.1-2004:
4814 // http://www.unix.org/single_unix_specification
4815
4816 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4817 struct timespec tp;
4818 int rc = os::Posix::clock_gettime(clockid, &tp);
4819 assert(rc == 0, "clock_gettime is expected to return 0 code");
4820
4821 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4822 }
4823
4824 void os::Linux::initialize_os_info() {
4825 assert(_os_version == 0, "OS info already initialized");
4826
4827 struct utsname _uname;
4828
4829 uint32_t major;
4830 uint32_t minor;
4831 uint32_t fix;
4832
4833 int rc;
4834
4835 // Kernel version is unknown if
4836 // verification below fails.
4837 _os_version = 0x01000000;
4838
4839 rc = uname(&_uname);
4840 if (rc != -1) {
4841
4842 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4843 if (rc == 3) {
4844
4845 if (major < 256 && minor < 256 && fix < 256) {
4846 // Kernel version format is as expected,
4847 // set it overriding unknown state.
4848 _os_version = (major << 16) |
4849 (minor << 8 ) |
4850 (fix << 0 ) ;
4851 }
4852 }
4853 }
4854 }
4855
4856 uint32_t os::Linux::os_version() {
4857 assert(_os_version != 0, "not initialized");
4858 return _os_version & 0x00FFFFFF;
4859 }
4860
4861 /////
4862 // glibc on Linux platform uses non-documented flag
4863 // to indicate, that some special sort of signal
4864 // trampoline is used.
4865 // We will never set this flag, and we should
4866 // ignore this flag in our diagnostic
4867 #ifdef SIGNIFICANT_SIGNAL_MASK
4868 #undef SIGNIFICANT_SIGNAL_MASK
4869 #endif
4870 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4871
4872 static const char* get_signal_handler_name(address handler,
4873 char* buf, int buflen) {
4874 int offset = 0;
4875 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4876 if (found) {
4877 // skip directory names
4878 const char *p1, *p2;
4879 p1 = buf;
4880 size_t len = strlen(os::file_separator());
4881 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4882 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4883 } else {
4884 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4885 }
4886 return buf;
4887 }
4888
4889 static void print_signal_handler(outputStream* st, int sig,
4890 char* buf, size_t buflen) {
4891 struct sigaction sa;
4892
4893 sigaction(sig, NULL, &sa);
4894
4895 // See comment for SIGNIFICANT_SIGNAL_MASK define
4896 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4897
4898 st->print("%s: ", os::exception_name(sig, buf, buflen));
4899
4900 address handler = (sa.sa_flags & SA_SIGINFO)
4901 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4902 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4903
4904 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4905 st->print("SIG_DFL");
4906 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4907 st->print("SIG_IGN");
4908 } else {
4909 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4910 }
4911
4912 st->print(", sa_mask[0]=");
4913 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4914
4915 address rh = VMError::get_resetted_sighandler(sig);
4916 // May be, handler was resetted by VMError?
4917 if (rh != NULL) {
4918 handler = rh;
4919 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4920 }
4921
4922 st->print(", sa_flags=");
4923 os::Posix::print_sa_flags(st, sa.sa_flags);
4924
4925 // Check: is it our handler?
4926 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4927 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4928 // It is our signal handler
4929 // check for flags, reset system-used one!
4930 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4931 st->print(
4932 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4933 os::Linux::get_our_sigflags(sig));
4934 }
4935 }
4936 st->cr();
4937 }
4938
4939
4940 #define DO_SIGNAL_CHECK(sig) \
4941 do { \
4942 if (!sigismember(&check_signal_done, sig)) { \
4943 os::Linux::check_signal_handler(sig); \
4944 } \
4945 } while (0)
4946
4947 // This method is a periodic task to check for misbehaving JNI applications
4948 // under CheckJNI, we can add any periodic checks here
4949
4950 void os::run_periodic_checks() {
4951 if (check_signals == false) return;
4952
4953 // SEGV and BUS if overridden could potentially prevent
4954 // generation of hs*.log in the event of a crash, debugging
4955 // such a case can be very challenging, so we absolutely
4956 // check the following for a good measure:
4957 DO_SIGNAL_CHECK(SIGSEGV);
4958 DO_SIGNAL_CHECK(SIGILL);
4959 DO_SIGNAL_CHECK(SIGFPE);
4960 DO_SIGNAL_CHECK(SIGBUS);
4961 DO_SIGNAL_CHECK(SIGPIPE);
4962 DO_SIGNAL_CHECK(SIGXFSZ);
4963 #if defined(PPC64)
4964 DO_SIGNAL_CHECK(SIGTRAP);
4965 #endif
4966
4967 // ReduceSignalUsage allows the user to override these handlers
4968 // see comments at the very top and jvm_md.h
4969 if (!ReduceSignalUsage) {
4970 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4971 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4972 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4973 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4974 }
4975
4976 DO_SIGNAL_CHECK(SR_signum);
4977 }
4978
4979 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4980
4981 static os_sigaction_t os_sigaction = NULL;
4982
4983 void os::Linux::check_signal_handler(int sig) {
4984 char buf[O_BUFLEN];
4985 address jvmHandler = NULL;
4986
4987
4988 struct sigaction act;
4989 if (os_sigaction == NULL) {
4990 // only trust the default sigaction, in case it has been interposed
4991 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4992 if (os_sigaction == NULL) return;
4993 }
4994
4995 os_sigaction(sig, (struct sigaction*)NULL, &act);
4996
4997
4998 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4999
5000 address thisHandler = (act.sa_flags & SA_SIGINFO)
5001 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
5002 : CAST_FROM_FN_PTR(address, act.sa_handler);
5003
5004
5005 switch (sig) {
5006 case SIGSEGV:
5007 case SIGBUS:
5008 case SIGFPE:
5009 case SIGPIPE:
5010 case SIGILL:
5011 case SIGXFSZ:
5012 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
5013 break;
5014
5015 case SHUTDOWN1_SIGNAL:
5016 case SHUTDOWN2_SIGNAL:
5017 case SHUTDOWN3_SIGNAL:
5018 case BREAK_SIGNAL:
5019 jvmHandler = (address)user_handler();
5020 break;
5021
5022 default:
5023 if (sig == SR_signum) {
5024 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
5025 } else {
5026 return;
5027 }
5028 break;
5029 }
5030
5031 if (thisHandler != jvmHandler) {
5032 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
5033 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
5034 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
5035 // No need to check this sig any longer
5036 sigaddset(&check_signal_done, sig);
5037 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
5038 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
5039 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
5040 exception_name(sig, buf, O_BUFLEN));
5041 }
5042 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
5043 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
5044 tty->print("expected:");
5045 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
5046 tty->cr();
5047 tty->print(" found:");
5048 os::Posix::print_sa_flags(tty, act.sa_flags);
5049 tty->cr();
5050 // No need to check this sig any longer
5051 sigaddset(&check_signal_done, sig);
5052 }
5053
5054 // Dump all the signal
5055 if (sigismember(&check_signal_done, sig)) {
5056 print_signal_handlers(tty, buf, O_BUFLEN);
5057 }
5058 }
5059
5060 extern void report_error(char* file_name, int line_no, char* title,
5061 char* format, ...);
5062
5063 // this is called _before_ most of the global arguments have been parsed
5064 void os::init(void) {
5065 char dummy; // used to get a guess on initial stack address
5066
5067 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
5068
5069 init_random(1234567);
5070
5071 Linux::set_page_size(sysconf(_SC_PAGESIZE));
5072 if (Linux::page_size() == -1) {
5073 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
5074 os::strerror(errno));
5075 }
5076 init_page_sizes((size_t) Linux::page_size());
5077
5078 Linux::initialize_system_info();
5079
5080 Linux::initialize_os_info();
5081
5082 os::Linux::CPUPerfTicks pticks;
5083 bool res = os::Linux::get_tick_information(&pticks, -1);
5084
5085 if (res && pticks.has_steal_ticks) {
5086 has_initial_tick_info = true;
5087 initial_total_ticks = pticks.total;
5088 initial_steal_ticks = pticks.steal;
5089 }
5090
5091 // _main_thread points to the thread that created/loaded the JVM.
5092 Linux::_main_thread = pthread_self();
5093
5094 // retrieve entry point for pthread_setname_np
5095 Linux::_pthread_setname_np =
5096 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
5097
5098 os::Posix::init();
5099
5100 initial_time_count = javaTimeNanos();
5101
5102 // Always warn if no monotonic clock available
5103 if (!os::Posix::supports_monotonic_clock()) {
5104 warning("No monotonic clock was available - timed services may " \
5105 "be adversely affected if the time-of-day clock changes");
5106 }
5107 }
5108
5109 // To install functions for atexit system call
5110 extern "C" {
5111 static void perfMemory_exit_helper() {
5112 perfMemory_exit();
5113 }
5114 }
5115
5116 void os::pd_init_container_support() {
5117 OSContainer::init();
5118 }
5119
5120 void os::Linux::numa_init() {
5121
5122 // Java can be invoked as
5123 // 1. Without numactl and heap will be allocated/configured on all nodes as
5124 // per the system policy.
5125 // 2. With numactl --interleave:
5126 // Use numa_get_interleave_mask(v2) API to get nodes bitmask. The same
5127 // API for membind case bitmask is reset.
5128 // Interleave is only hint and Kernel can fallback to other nodes if
5129 // no memory is available on the target nodes.
5130 // 3. With numactl --membind:
5131 // Use numa_get_membind(v2) API to get nodes bitmask. The same API for
5132 // interleave case returns bitmask of all nodes.
5133 // numa_all_nodes_ptr holds bitmask of all nodes.
5134 // numa_get_interleave_mask(v2) and numa_get_membind(v2) APIs returns correct
5135 // bitmask when externally configured to run on all or fewer nodes.
5136
5137 if (!Linux::libnuma_init()) {
5138 UseNUMA = false;
5139 } else {
5140 if ((Linux::numa_max_node() < 1) || Linux::is_bound_to_single_node()) {
5141 // If there's only one node (they start from 0) or if the process
5142 // is bound explicitly to a single node using membind, disable NUMA.
5143 UseNUMA = false;
5144 } else {
5145
5146 LogTarget(Info,os) log;
5147 LogStream ls(log);
5148
5149 Linux::set_configured_numa_policy(Linux::identify_numa_policy());
5150
5151 struct bitmask* bmp = Linux::_numa_membind_bitmask;
5152 const char* numa_mode = "membind";
5153
5154 if (Linux::is_running_in_interleave_mode()) {
5155 bmp = Linux::_numa_interleave_bitmask;
5156 numa_mode = "interleave";
5157 }
5158
5159 ls.print("UseNUMA is enabled and invoked in '%s' mode."
5160 " Heap will be configured using NUMA memory nodes:", numa_mode);
5161
5162 for (int node = 0; node <= Linux::numa_max_node(); node++) {
5163 if (Linux::_numa_bitmask_isbitset(bmp, node)) {
5164 ls.print(" %d", node);
5165 }
5166 }
5167 }
5168 }
5169
5170 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5171 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5172 // we can make the adaptive lgrp chunk resizing work. If the user specified both
5173 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5174 // and disable adaptive resizing.
5175 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5176 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5177 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5178 UseAdaptiveSizePolicy = false;
5179 UseAdaptiveNUMAChunkSizing = false;
5180 }
5181 }
5182
5183 if (!UseNUMA && ForceNUMA) {
5184 UseNUMA = true;
5185 }
5186 }
5187
5188 // this is called _after_ the global arguments have been parsed
5189 jint os::init_2(void) {
5190
5191 // This could be set after os::Posix::init() but all platforms
5192 // have to set it the same so we have to mirror Solaris.
5193 DEBUG_ONLY(os::set_mutex_init_done();)
5194
5195 os::Posix::init_2();
5196
5197 Linux::fast_thread_clock_init();
5198
5199 // initialize suspend/resume support - must do this before signal_sets_init()
5200 if (SR_initialize() != 0) {
5201 perror("SR_initialize failed");
5202 return JNI_ERR;
5203 }
5204
5205 Linux::signal_sets_init();
5206 Linux::install_signal_handlers();
5207 // Initialize data for jdk.internal.misc.Signal
5208 if (!ReduceSignalUsage) {
5209 jdk_misc_signal_init();
5210 }
5211
5212 if (AdjustStackSizeForTLS) {
5213 get_minstack_init();
5214 }
5215
5216 // Check and sets minimum stack sizes against command line options
5217 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
5218 return JNI_ERR;
5219 }
5220
5221 #if defined(IA32)
5222 // Need to ensure we've determined the process's initial stack to
5223 // perform the workaround
5224 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5225 workaround_expand_exec_shield_cs_limit();
5226 #else
5227 suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
5228 if (!suppress_primordial_thread_resolution) {
5229 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5230 }
5231 #endif
5232
5233 Linux::libpthread_init();
5234 Linux::sched_getcpu_init();
5235 log_info(os)("HotSpot is running with %s, %s",
5236 Linux::glibc_version(), Linux::libpthread_version());
5237
5238 if (UseNUMA) {
5239 Linux::numa_init();
5240 }
5241
5242 if (MaxFDLimit) {
5243 // set the number of file descriptors to max. print out error
5244 // if getrlimit/setrlimit fails but continue regardless.
5245 struct rlimit nbr_files;
5246 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5247 if (status != 0) {
5248 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5249 } else {
5250 nbr_files.rlim_cur = nbr_files.rlim_max;
5251 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5252 if (status != 0) {
5253 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5254 }
5255 }
5256 }
5257
5258 // Initialize lock used to serialize thread creation (see os::create_thread)
5259 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5260
5261 // at-exit methods are called in the reverse order of their registration.
5262 // atexit functions are called on return from main or as a result of a
5263 // call to exit(3C). There can be only 32 of these functions registered
5264 // and atexit() does not set errno.
5265
5266 if (PerfAllowAtExitRegistration) {
5267 // only register atexit functions if PerfAllowAtExitRegistration is set.
5268 // atexit functions can be delayed until process exit time, which
5269 // can be problematic for embedded VM situations. Embedded VMs should
5270 // call DestroyJavaVM() to assure that VM resources are released.
5271
5272 // note: perfMemory_exit_helper atexit function may be removed in
5273 // the future if the appropriate cleanup code can be added to the
5274 // VM_Exit VMOperation's doit method.
5275 if (atexit(perfMemory_exit_helper) != 0) {
5276 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5277 }
5278 }
5279
5280 // initialize thread priority policy
5281 prio_init();
5282
5283 if (!FLAG_IS_DEFAULT(AllocateHeapAt) || !FLAG_IS_DEFAULT(AllocateOldGenAt)) {
5284 set_coredump_filter(DAX_SHARED_BIT);
5285 }
5286
5287 if (DumpPrivateMappingsInCore) {
5288 set_coredump_filter(FILE_BACKED_PVT_BIT);
5289 }
5290
5291 if (DumpSharedMappingsInCore) {
5292 set_coredump_filter(FILE_BACKED_SHARED_BIT);
5293 }
5294
5295 return JNI_OK;
5296 }
5297
5298 // Mark the polling page as unreadable
5299 void os::make_polling_page_unreadable(void) {
5300 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5301 fatal("Could not disable polling page");
5302 }
5303 }
5304
5305 // Mark the polling page as readable
5306 void os::make_polling_page_readable(void) {
5307 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5308 fatal("Could not enable polling page");
5309 }
5310 }
5311
5312 // older glibc versions don't have this macro (which expands to
5313 // an optimized bit-counting function) so we have to roll our own
5314 #ifndef CPU_COUNT
5315
5316 static int _cpu_count(const cpu_set_t* cpus) {
5317 int count = 0;
5318 // only look up to the number of configured processors
5319 for (int i = 0; i < os::processor_count(); i++) {
5320 if (CPU_ISSET(i, cpus)) {
5321 count++;
5322 }
5323 }
5324 return count;
5325 }
5326
5327 #define CPU_COUNT(cpus) _cpu_count(cpus)
5328
5329 #endif // CPU_COUNT
5330
5331 // Get the current number of available processors for this process.
5332 // This value can change at any time during a process's lifetime.
5333 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5334 // If it appears there may be more than 1024 processors then we do a
5335 // dynamic check - see 6515172 for details.
5336 // If anything goes wrong we fallback to returning the number of online
5337 // processors - which can be greater than the number available to the process.
5338 int os::Linux::active_processor_count() {
5339 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5340 cpu_set_t* cpus_p = &cpus;
5341 int cpus_size = sizeof(cpu_set_t);
5342
5343 int configured_cpus = os::processor_count(); // upper bound on available cpus
5344 int cpu_count = 0;
5345
5346 // old build platforms may not support dynamic cpu sets
5347 #ifdef CPU_ALLOC
5348
5349 // To enable easy testing of the dynamic path on different platforms we
5350 // introduce a diagnostic flag: UseCpuAllocPath
5351 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5352 // kernel may use a mask bigger than cpu_set_t
5353 log_trace(os)("active_processor_count: using dynamic path %s"
5354 "- configured processors: %d",
5355 UseCpuAllocPath ? "(forced) " : "",
5356 configured_cpus);
5357 cpus_p = CPU_ALLOC(configured_cpus);
5358 if (cpus_p != NULL) {
5359 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5360 // zero it just to be safe
5361 CPU_ZERO_S(cpus_size, cpus_p);
5362 }
5363 else {
5364 // failed to allocate so fallback to online cpus
5365 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5366 log_trace(os)("active_processor_count: "
5367 "CPU_ALLOC failed (%s) - using "
5368 "online processor count: %d",
5369 os::strerror(errno), online_cpus);
5370 return online_cpus;
5371 }
5372 }
5373 else {
5374 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5375 configured_cpus);
5376 }
5377 #else // CPU_ALLOC
5378 // these stubs won't be executed
5379 #define CPU_COUNT_S(size, cpus) -1
5380 #define CPU_FREE(cpus)
5381
5382 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5383 configured_cpus);
5384 #endif // CPU_ALLOC
5385
5386 // pid 0 means the current thread - which we have to assume represents the process
5387 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5388 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5389 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5390 }
5391 else {
5392 cpu_count = CPU_COUNT(cpus_p);
5393 }
5394 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5395 }
5396 else {
5397 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5398 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5399 "which may exceed available processors", os::strerror(errno), cpu_count);
5400 }
5401
5402 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5403 CPU_FREE(cpus_p);
5404 }
5405
5406 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5407 return cpu_count;
5408 }
5409
5410 // Determine the active processor count from one of
5411 // three different sources:
5412 //
5413 // 1. User option -XX:ActiveProcessorCount
5414 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5415 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5416 //
5417 // Option 1, if specified, will always override.
5418 // If the cgroup subsystem is active and configured, we
5419 // will return the min of the cgroup and option 2 results.
5420 // This is required since tools, such as numactl, that
5421 // alter cpu affinity do not update cgroup subsystem
5422 // cpuset configuration files.
5423 int os::active_processor_count() {
5424 // User has overridden the number of active processors
5425 if (ActiveProcessorCount > 0) {
5426 log_trace(os)("active_processor_count: "
5427 "active processor count set by user : %d",
5428 ActiveProcessorCount);
5429 return ActiveProcessorCount;
5430 }
5431
5432 int active_cpus;
5433 if (OSContainer::is_containerized()) {
5434 active_cpus = OSContainer::active_processor_count();
5435 log_trace(os)("active_processor_count: determined by OSContainer: %d",
5436 active_cpus);
5437 } else {
5438 active_cpus = os::Linux::active_processor_count();
5439 }
5440
5441 return active_cpus;
5442 }
5443
5444 uint os::processor_id() {
5445 const int id = Linux::sched_getcpu();
5446 assert(id >= 0 && id < _processor_count, "Invalid processor id");
5447 return (uint)id;
5448 }
5449
5450 void os::set_native_thread_name(const char *name) {
5451 if (Linux::_pthread_setname_np) {
5452 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5453 snprintf(buf, sizeof(buf), "%s", name);
5454 buf[sizeof(buf) - 1] = '\0';
5455 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5456 // ERANGE should not happen; all other errors should just be ignored.
5457 assert(rc != ERANGE, "pthread_setname_np failed");
5458 }
5459 }
5460
5461 bool os::bind_to_processor(uint processor_id) {
5462 // Not yet implemented.
5463 return false;
5464 }
5465
5466 ///
5467
5468 void os::SuspendedThreadTask::internal_do_task() {
5469 if (do_suspend(_thread->osthread())) {
5470 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5471 do_task(context);
5472 do_resume(_thread->osthread());
5473 }
5474 }
5475
5476 ////////////////////////////////////////////////////////////////////////////////
5477 // debug support
5478
5479 bool os::find(address addr, outputStream* st) {
5480 Dl_info dlinfo;
5481 memset(&dlinfo, 0, sizeof(dlinfo));
5482 if (dladdr(addr, &dlinfo) != 0) {
5483 st->print(PTR_FORMAT ": ", p2i(addr));
5484 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5485 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5486 p2i(addr) - p2i(dlinfo.dli_saddr));
5487 } else if (dlinfo.dli_fbase != NULL) {
5488 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5489 } else {
5490 st->print("<absolute address>");
5491 }
5492 if (dlinfo.dli_fname != NULL) {
5493 st->print(" in %s", dlinfo.dli_fname);
5494 }
5495 if (dlinfo.dli_fbase != NULL) {
5496 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5497 }
5498 st->cr();
5499
5500 if (Verbose) {
5501 // decode some bytes around the PC
5502 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5503 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5504 address lowest = (address) dlinfo.dli_sname;
5505 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5506 if (begin < lowest) begin = lowest;
5507 Dl_info dlinfo2;
5508 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5509 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5510 end = (address) dlinfo2.dli_saddr;
5511 }
5512 Disassembler::decode(begin, end, st);
5513 }
5514 return true;
5515 }
5516 return false;
5517 }
5518
5519 ////////////////////////////////////////////////////////////////////////////////
5520 // misc
5521
5522 // This does not do anything on Linux. This is basically a hook for being
5523 // able to use structured exception handling (thread-local exception filters)
5524 // on, e.g., Win32.
5525 void
5526 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5527 JavaCallArguments* args, Thread* thread) {
5528 f(value, method, args, thread);
5529 }
5530
5531 void os::print_statistics() {
5532 }
5533
5534 bool os::message_box(const char* title, const char* message) {
5535 int i;
5536 fdStream err(defaultStream::error_fd());
5537 for (i = 0; i < 78; i++) err.print_raw("=");
5538 err.cr();
5539 err.print_raw_cr(title);
5540 for (i = 0; i < 78; i++) err.print_raw("-");
5541 err.cr();
5542 err.print_raw_cr(message);
5543 for (i = 0; i < 78; i++) err.print_raw("=");
5544 err.cr();
5545
5546 char buf[16];
5547 // Prevent process from exiting upon "read error" without consuming all CPU
5548 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5549
5550 return buf[0] == 'y' || buf[0] == 'Y';
5551 }
5552
5553 // Is a (classpath) directory empty?
5554 bool os::dir_is_empty(const char* path) {
5555 DIR *dir = NULL;
5556 struct dirent *ptr;
5557
5558 dir = opendir(path);
5559 if (dir == NULL) return true;
5560
5561 // Scan the directory
5562 bool result = true;
5563 while (result && (ptr = readdir(dir)) != NULL) {
5564 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5565 result = false;
5566 }
5567 }
5568 closedir(dir);
5569 return result;
5570 }
5571
5572 // This code originates from JDK's sysOpen and open64_w
5573 // from src/solaris/hpi/src/system_md.c
5574
5575 int os::open(const char *path, int oflag, int mode) {
5576 if (strlen(path) > MAX_PATH - 1) {
5577 errno = ENAMETOOLONG;
5578 return -1;
5579 }
5580
5581 // All file descriptors that are opened in the Java process and not
5582 // specifically destined for a subprocess should have the close-on-exec
5583 // flag set. If we don't set it, then careless 3rd party native code
5584 // might fork and exec without closing all appropriate file descriptors
5585 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5586 // turn might:
5587 //
5588 // - cause end-of-file to fail to be detected on some file
5589 // descriptors, resulting in mysterious hangs, or
5590 //
5591 // - might cause an fopen in the subprocess to fail on a system
5592 // suffering from bug 1085341.
5593 //
5594 // (Yes, the default setting of the close-on-exec flag is a Unix
5595 // design flaw)
5596 //
5597 // See:
5598 // 1085341: 32-bit stdio routines should support file descriptors >255
5599 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5600 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5601 //
5602 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5603 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5604 // because it saves a system call and removes a small window where the flag
5605 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5606 // and we fall back to using FD_CLOEXEC (see below).
5607 #ifdef O_CLOEXEC
5608 oflag |= O_CLOEXEC;
5609 #endif
5610
5611 int fd = ::open64(path, oflag, mode);
5612 if (fd == -1) return -1;
5613
5614 //If the open succeeded, the file might still be a directory
5615 {
5616 struct stat64 buf64;
5617 int ret = ::fstat64(fd, &buf64);
5618 int st_mode = buf64.st_mode;
5619
5620 if (ret != -1) {
5621 if ((st_mode & S_IFMT) == S_IFDIR) {
5622 errno = EISDIR;
5623 ::close(fd);
5624 return -1;
5625 }
5626 } else {
5627 ::close(fd);
5628 return -1;
5629 }
5630 }
5631
5632 #ifdef FD_CLOEXEC
5633 // Validate that the use of the O_CLOEXEC flag on open above worked.
5634 // With recent kernels, we will perform this check exactly once.
5635 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5636 if (!O_CLOEXEC_is_known_to_work) {
5637 int flags = ::fcntl(fd, F_GETFD);
5638 if (flags != -1) {
5639 if ((flags & FD_CLOEXEC) != 0)
5640 O_CLOEXEC_is_known_to_work = 1;
5641 else
5642 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5643 }
5644 }
5645 #endif
5646
5647 return fd;
5648 }
5649
5650
5651 // create binary file, rewriting existing file if required
5652 int os::create_binary_file(const char* path, bool rewrite_existing) {
5653 int oflags = O_WRONLY | O_CREAT;
5654 if (!rewrite_existing) {
5655 oflags |= O_EXCL;
5656 }
5657 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5658 }
5659
5660 // return current position of file pointer
5661 jlong os::current_file_offset(int fd) {
5662 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5663 }
5664
5665 // move file pointer to the specified offset
5666 jlong os::seek_to_file_offset(int fd, jlong offset) {
5667 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5668 }
5669
5670 // This code originates from JDK's sysAvailable
5671 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5672
5673 int os::available(int fd, jlong *bytes) {
5674 jlong cur, end;
5675 int mode;
5676 struct stat64 buf64;
5677
5678 if (::fstat64(fd, &buf64) >= 0) {
5679 mode = buf64.st_mode;
5680 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5681 int n;
5682 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5683 *bytes = n;
5684 return 1;
5685 }
5686 }
5687 }
5688 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5689 return 0;
5690 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5691 return 0;
5692 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5693 return 0;
5694 }
5695 *bytes = end - cur;
5696 return 1;
5697 }
5698
5699 // Map a block of memory.
5700 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5701 char *addr, size_t bytes, bool read_only,
5702 bool allow_exec) {
5703 int prot;
5704 int flags = MAP_PRIVATE;
5705
5706 if (read_only) {
5707 prot = PROT_READ;
5708 } else {
5709 prot = PROT_READ | PROT_WRITE;
5710 }
5711
5712 if (allow_exec) {
5713 prot |= PROT_EXEC;
5714 }
5715
5716 if (addr != NULL) {
5717 flags |= MAP_FIXED;
5718 }
5719
5720 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5721 fd, file_offset);
5722 if (mapped_address == MAP_FAILED) {
5723 return NULL;
5724 }
5725 return mapped_address;
5726 }
5727
5728
5729 // Remap a block of memory.
5730 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5731 char *addr, size_t bytes, bool read_only,
5732 bool allow_exec) {
5733 // same as map_memory() on this OS
5734 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5735 allow_exec);
5736 }
5737
5738
5739 // Unmap a block of memory.
5740 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5741 return munmap(addr, bytes) == 0;
5742 }
5743
5744 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5745
5746 static jlong fast_cpu_time(Thread *thread) {
5747 clockid_t clockid;
5748 int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(),
5749 &clockid);
5750 if (rc == 0) {
5751 return os::Linux::fast_thread_cpu_time(clockid);
5752 } else {
5753 // It's possible to encounter a terminated native thread that failed
5754 // to detach itself from the VM - which should result in ESRCH.
5755 assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed");
5756 return -1;
5757 }
5758 }
5759
5760 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5761 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5762 // of a thread.
5763 //
5764 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5765 // the fast estimate available on the platform.
5766
5767 jlong os::current_thread_cpu_time() {
5768 if (os::Linux::supports_fast_thread_cpu_time()) {
5769 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5770 } else {
5771 // return user + sys since the cost is the same
5772 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5773 }
5774 }
5775
5776 jlong os::thread_cpu_time(Thread* thread) {
5777 // consistent with what current_thread_cpu_time() returns
5778 if (os::Linux::supports_fast_thread_cpu_time()) {
5779 return fast_cpu_time(thread);
5780 } else {
5781 return slow_thread_cpu_time(thread, true /* user + sys */);
5782 }
5783 }
5784
5785 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5786 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5787 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5788 } else {
5789 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5790 }
5791 }
5792
5793 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5794 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5795 return fast_cpu_time(thread);
5796 } else {
5797 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5798 }
5799 }
5800
5801 // -1 on error.
5802 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5803 pid_t tid = thread->osthread()->thread_id();
5804 char *s;
5805 char stat[2048];
5806 int statlen;
5807 char proc_name[64];
5808 int count;
5809 long sys_time, user_time;
5810 char cdummy;
5811 int idummy;
5812 long ldummy;
5813 FILE *fp;
5814
5815 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5816 fp = fopen(proc_name, "r");
5817 if (fp == NULL) return -1;
5818 statlen = fread(stat, 1, 2047, fp);
5819 stat[statlen] = '\0';
5820 fclose(fp);
5821
5822 // Skip pid and the command string. Note that we could be dealing with
5823 // weird command names, e.g. user could decide to rename java launcher
5824 // to "java 1.4.2 :)", then the stat file would look like
5825 // 1234 (java 1.4.2 :)) R ... ...
5826 // We don't really need to know the command string, just find the last
5827 // occurrence of ")" and then start parsing from there. See bug 4726580.
5828 s = strrchr(stat, ')');
5829 if (s == NULL) return -1;
5830
5831 // Skip blank chars
5832 do { s++; } while (s && isspace(*s));
5833
5834 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5835 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5836 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5837 &user_time, &sys_time);
5838 if (count != 13) return -1;
5839 if (user_sys_cpu_time) {
5840 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5841 } else {
5842 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5843 }
5844 }
5845
5846 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5847 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5848 info_ptr->may_skip_backward = false; // elapsed time not wall time
5849 info_ptr->may_skip_forward = false; // elapsed time not wall time
5850 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5851 }
5852
5853 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5854 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5855 info_ptr->may_skip_backward = false; // elapsed time not wall time
5856 info_ptr->may_skip_forward = false; // elapsed time not wall time
5857 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5858 }
5859
5860 bool os::is_thread_cpu_time_supported() {
5861 return true;
5862 }
5863
5864 // System loadavg support. Returns -1 if load average cannot be obtained.
5865 // Linux doesn't yet have a (official) notion of processor sets,
5866 // so just return the system wide load average.
5867 int os::loadavg(double loadavg[], int nelem) {
5868 return ::getloadavg(loadavg, nelem);
5869 }
5870
5871 void os::pause() {
5872 char filename[MAX_PATH];
5873 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5874 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5875 } else {
5876 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5877 }
5878
5879 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5880 if (fd != -1) {
5881 struct stat buf;
5882 ::close(fd);
5883 while (::stat(filename, &buf) == 0) {
5884 (void)::poll(NULL, 0, 100);
5885 }
5886 } else {
5887 jio_fprintf(stderr,
5888 "Could not open pause file '%s', continuing immediately.\n", filename);
5889 }
5890 }
5891
5892 extern char** environ;
5893
5894 // Run the specified command in a separate process. Return its exit value,
5895 // or -1 on failure (e.g. can't fork a new process).
5896 // Unlike system(), this function can be called from signal handler. It
5897 // doesn't block SIGINT et al.
5898 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) {
5899 const char * argv[4] = {"sh", "-c", cmd, NULL};
5900
5901 pid_t pid ;
5902
5903 if (use_vfork_if_available) {
5904 pid = vfork();
5905 } else {
5906 pid = fork();
5907 }
5908
5909 if (pid < 0) {
5910 // fork failed
5911 return -1;
5912
5913 } else if (pid == 0) {
5914 // child process
5915
5916 execve("/bin/sh", (char* const*)argv, environ);
5917
5918 // execve failed
5919 _exit(-1);
5920
5921 } else {
5922 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5923 // care about the actual exit code, for now.
5924
5925 int status;
5926
5927 // Wait for the child process to exit. This returns immediately if
5928 // the child has already exited. */
5929 while (waitpid(pid, &status, 0) < 0) {
5930 switch (errno) {
5931 case ECHILD: return 0;
5932 case EINTR: break;
5933 default: return -1;
5934 }
5935 }
5936
5937 if (WIFEXITED(status)) {
5938 // The child exited normally; get its exit code.
5939 return WEXITSTATUS(status);
5940 } else if (WIFSIGNALED(status)) {
5941 // The child exited because of a signal
5942 // The best value to return is 0x80 + signal number,
5943 // because that is what all Unix shells do, and because
5944 // it allows callers to distinguish between process exit and
5945 // process death by signal.
5946 return 0x80 + WTERMSIG(status);
5947 } else {
5948 // Unknown exit code; pass it through
5949 return status;
5950 }
5951 }
5952 }
5953
5954 // Get the default path to the core file
5955 // Returns the length of the string
5956 int os::get_core_path(char* buffer, size_t bufferSize) {
5957 /*
5958 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5959 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5960 */
5961 const int core_pattern_len = 129;
5962 char core_pattern[core_pattern_len] = {0};
5963
5964 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5965 if (core_pattern_file == -1) {
5966 return -1;
5967 }
5968
5969 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5970 ::close(core_pattern_file);
5971 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5972 return -1;
5973 }
5974 if (core_pattern[ret-1] == '\n') {
5975 core_pattern[ret-1] = '\0';
5976 } else {
5977 core_pattern[ret] = '\0';
5978 }
5979
5980 // Replace the %p in the core pattern with the process id. NOTE: we do this
5981 // only if the pattern doesn't start with "|", and we support only one %p in
5982 // the pattern.
5983 char *pid_pos = strstr(core_pattern, "%p");
5984 const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : ""; // skip over the "%p"
5985 int written;
5986
5987 if (core_pattern[0] == '/') {
5988 if (pid_pos != NULL) {
5989 *pid_pos = '\0';
5990 written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern,
5991 current_process_id(), tail);
5992 } else {
5993 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5994 }
5995 } else {
5996 char cwd[PATH_MAX];
5997
5998 const char* p = get_current_directory(cwd, PATH_MAX);
5999 if (p == NULL) {
6000 return -1;
6001 }
6002
6003 if (core_pattern[0] == '|') {
6004 written = jio_snprintf(buffer, bufferSize,
6005 "\"%s\" (or dumping to %s/core.%d)",
6006 &core_pattern[1], p, current_process_id());
6007 } else if (pid_pos != NULL) {
6008 *pid_pos = '\0';
6009 written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern,
6010 current_process_id(), tail);
6011 } else {
6012 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6013 }
6014 }
6015
6016 if (written < 0) {
6017 return -1;
6018 }
6019
6020 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6021 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6022
6023 if (core_uses_pid_file != -1) {
6024 char core_uses_pid = 0;
6025 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6026 ::close(core_uses_pid_file);
6027
6028 if (core_uses_pid == '1') {
6029 jio_snprintf(buffer + written, bufferSize - written,
6030 ".%d", current_process_id());
6031 }
6032 }
6033 }
6034
6035 return strlen(buffer);
6036 }
6037
6038 bool os::start_debugging(char *buf, int buflen) {
6039 int len = (int)strlen(buf);
6040 char *p = &buf[len];
6041
6042 jio_snprintf(p, buflen-len,
6043 "\n\n"
6044 "Do you want to debug the problem?\n\n"
6045 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
6046 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
6047 "Otherwise, press RETURN to abort...",
6048 os::current_process_id(), os::current_process_id(),
6049 os::current_thread_id(), os::current_thread_id());
6050
6051 bool yes = os::message_box("Unexpected Error", buf);
6052
6053 if (yes) {
6054 // yes, user asked VM to launch debugger
6055 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
6056 os::current_process_id(), os::current_process_id());
6057
6058 os::fork_and_exec(buf);
6059 yes = false;
6060 }
6061 return yes;
6062 }
6063
6064
6065 // Java/Compiler thread:
6066 //
6067 // Low memory addresses
6068 // P0 +------------------------+
6069 // | |\ Java thread created by VM does not have glibc
6070 // | glibc guard page | - guard page, attached Java thread usually has
6071 // | |/ 1 glibc guard page.
6072 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6073 // | |\
6074 // | HotSpot Guard Pages | - red, yellow and reserved pages
6075 // | |/
6076 // +------------------------+ JavaThread::stack_reserved_zone_base()
6077 // | |\
6078 // | Normal Stack | -
6079 // | |/
6080 // P2 +------------------------+ Thread::stack_base()
6081 //
6082 // Non-Java thread:
6083 //
6084 // Low memory addresses
6085 // P0 +------------------------+
6086 // | |\
6087 // | glibc guard page | - usually 1 page
6088 // | |/
6089 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6090 // | |\
6091 // | Normal Stack | -
6092 // | |/
6093 // P2 +------------------------+ Thread::stack_base()
6094 //
6095 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
6096 // returned from pthread_attr_getstack().
6097 // ** Due to NPTL implementation error, linux takes the glibc guard page out
6098 // of the stack size given in pthread_attr. We work around this for
6099 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
6100 //
6101 #ifndef ZERO
6102 static void current_stack_region(address * bottom, size_t * size) {
6103 if (os::is_primordial_thread()) {
6104 // primordial thread needs special handling because pthread_getattr_np()
6105 // may return bogus value.
6106 *bottom = os::Linux::initial_thread_stack_bottom();
6107 *size = os::Linux::initial_thread_stack_size();
6108 } else {
6109 pthread_attr_t attr;
6110
6111 int rslt = pthread_getattr_np(pthread_self(), &attr);
6112
6113 // JVM needs to know exact stack location, abort if it fails
6114 if (rslt != 0) {
6115 if (rslt == ENOMEM) {
6116 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
6117 } else {
6118 fatal("pthread_getattr_np failed with error = %d", rslt);
6119 }
6120 }
6121
6122 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
6123 fatal("Cannot locate current stack attributes!");
6124 }
6125
6126 // Work around NPTL stack guard error.
6127 size_t guard_size = 0;
6128 rslt = pthread_attr_getguardsize(&attr, &guard_size);
6129 if (rslt != 0) {
6130 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
6131 }
6132 *bottom += guard_size;
6133 *size -= guard_size;
6134
6135 pthread_attr_destroy(&attr);
6136
6137 }
6138 assert(os::current_stack_pointer() >= *bottom &&
6139 os::current_stack_pointer() < *bottom + *size, "just checking");
6140 }
6141
6142 address os::current_stack_base() {
6143 address bottom;
6144 size_t size;
6145 current_stack_region(&bottom, &size);
6146 return (bottom + size);
6147 }
6148
6149 size_t os::current_stack_size() {
6150 // This stack size includes the usable stack and HotSpot guard pages
6151 // (for the threads that have Hotspot guard pages).
6152 address bottom;
6153 size_t size;
6154 current_stack_region(&bottom, &size);
6155 return size;
6156 }
6157 #endif
6158
6159 static inline struct timespec get_mtime(const char* filename) {
6160 struct stat st;
6161 int ret = os::stat(filename, &st);
6162 assert(ret == 0, "failed to stat() file '%s': %s", filename, os::strerror(errno));
6163 return st.st_mtim;
6164 }
6165
6166 int os::compare_file_modified_times(const char* file1, const char* file2) {
6167 struct timespec filetime1 = get_mtime(file1);
6168 struct timespec filetime2 = get_mtime(file2);
6169 int diff = filetime1.tv_sec - filetime2.tv_sec;
6170 if (diff == 0) {
6171 return filetime1.tv_nsec - filetime2.tv_nsec;
6172 }
6173 return diff;
6174 }
6175
6176 bool os::supports_map_sync() {
6177 return true;
6178 }
6179
6180 /////////////// Unit tests ///////////////
6181
6182 #ifndef PRODUCT
6183
6184 class TestReserveMemorySpecial : AllStatic {
6185 public:
6186 static void small_page_write(void* addr, size_t size) {
6187 size_t page_size = os::vm_page_size();
6188
6189 char* end = (char*)addr + size;
6190 for (char* p = (char*)addr; p < end; p += page_size) {
6191 *p = 1;
6192 }
6193 }
6194
6195 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6196 if (!UseHugeTLBFS) {
6197 return;
6198 }
6199
6200 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6201
6202 if (addr != NULL) {
6203 small_page_write(addr, size);
6204
6205 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6206 }
6207 }
6208
6209 static void test_reserve_memory_special_huge_tlbfs_only() {
6210 if (!UseHugeTLBFS) {
6211 return;
6212 }
6213
6214 size_t lp = os::large_page_size();
6215
6216 for (size_t size = lp; size <= lp * 10; size += lp) {
6217 test_reserve_memory_special_huge_tlbfs_only(size);
6218 }
6219 }
6220
6221 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6222 size_t lp = os::large_page_size();
6223 size_t ag = os::vm_allocation_granularity();
6224
6225 // sizes to test
6226 const size_t sizes[] = {
6227 lp, lp + ag, lp + lp / 2, lp * 2,
6228 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6229 lp * 10, lp * 10 + lp / 2
6230 };
6231 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6232
6233 // For each size/alignment combination, we test three scenarios:
6234 // 1) with req_addr == NULL
6235 // 2) with a non-null req_addr at which we expect to successfully allocate
6236 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6237 // expect the allocation to either fail or to ignore req_addr
6238
6239 // Pre-allocate two areas; they shall be as large as the largest allocation
6240 // and aligned to the largest alignment we will be testing.
6241 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6242 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6243 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6244 -1, 0);
6245 assert(mapping1 != MAP_FAILED, "should work");
6246
6247 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6248 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6249 -1, 0);
6250 assert(mapping2 != MAP_FAILED, "should work");
6251
6252 // Unmap the first mapping, but leave the second mapping intact: the first
6253 // mapping will serve as a value for a "good" req_addr (case 2). The second
6254 // mapping, still intact, as "bad" req_addr (case 3).
6255 ::munmap(mapping1, mapping_size);
6256
6257 // Case 1
6258 for (int i = 0; i < num_sizes; i++) {
6259 const size_t size = sizes[i];
6260 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6261 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6262 if (p != NULL) {
6263 assert(is_aligned(p, alignment), "must be");
6264 small_page_write(p, size);
6265 os::Linux::release_memory_special_huge_tlbfs(p, size);
6266 }
6267 }
6268 }
6269
6270 // Case 2
6271 for (int i = 0; i < num_sizes; i++) {
6272 const size_t size = sizes[i];
6273 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6274 char* const req_addr = align_up(mapping1, alignment);
6275 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6276 if (p != NULL) {
6277 assert(p == req_addr, "must be");
6278 small_page_write(p, size);
6279 os::Linux::release_memory_special_huge_tlbfs(p, size);
6280 }
6281 }
6282 }
6283
6284 // Case 3
6285 for (int i = 0; i < num_sizes; i++) {
6286 const size_t size = sizes[i];
6287 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6288 char* const req_addr = align_up(mapping2, alignment);
6289 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6290 // as the area around req_addr contains already existing mappings, the API should always
6291 // return NULL (as per contract, it cannot return another address)
6292 assert(p == NULL, "must be");
6293 }
6294 }
6295
6296 ::munmap(mapping2, mapping_size);
6297
6298 }
6299
6300 static void test_reserve_memory_special_huge_tlbfs() {
6301 if (!UseHugeTLBFS) {
6302 return;
6303 }
6304
6305 test_reserve_memory_special_huge_tlbfs_only();
6306 test_reserve_memory_special_huge_tlbfs_mixed();
6307 }
6308
6309 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6310 if (!UseSHM) {
6311 return;
6312 }
6313
6314 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6315
6316 if (addr != NULL) {
6317 assert(is_aligned(addr, alignment), "Check");
6318 assert(is_aligned(addr, os::large_page_size()), "Check");
6319
6320 small_page_write(addr, size);
6321
6322 os::Linux::release_memory_special_shm(addr, size);
6323 }
6324 }
6325
6326 static void test_reserve_memory_special_shm() {
6327 size_t lp = os::large_page_size();
6328 size_t ag = os::vm_allocation_granularity();
6329
6330 for (size_t size = ag; size < lp * 3; size += ag) {
6331 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6332 test_reserve_memory_special_shm(size, alignment);
6333 }
6334 }
6335 }
6336
6337 static void test() {
6338 test_reserve_memory_special_huge_tlbfs();
6339 test_reserve_memory_special_shm();
6340 }
6341 };
6342
6343 void TestReserveMemorySpecial_test() {
6344 TestReserveMemorySpecial::test();
6345 }
6346
6347 #endif