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 bool os::enable_vtime() { return false; }
1368 bool os::vtime_enabled() { return false; }
1369
1370 double os::elapsedVTime() {
1371 struct rusage usage;
1372 int retval = getrusage(RUSAGE_THREAD, &usage);
1373 if (retval == 0) {
1374 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);
1375 } else {
1376 // better than nothing, but not much
1377 return elapsedTime();
1378 }
1379 }
1380
1381 jlong os::javaTimeMillis() {
1382 timeval time;
1383 int status = gettimeofday(&time, NULL);
1384 assert(status != -1, "linux error");
1385 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1386 }
1387
1388 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1389 timeval time;
1390 int status = gettimeofday(&time, NULL);
1391 assert(status != -1, "linux error");
1392 seconds = jlong(time.tv_sec);
1393 nanos = jlong(time.tv_usec) * 1000;
1394 }
1395
1396 void os::Linux::fast_thread_clock_init() {
1397 if (!UseLinuxPosixThreadCPUClocks) {
1398 return;
1399 }
1400 clockid_t clockid;
1401 struct timespec tp;
1402 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1403 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1404
1405 // Switch to using fast clocks for thread cpu time if
1406 // the clock_getres() returns 0 error code.
1407 // Note, that some kernels may support the current thread
1408 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1409 // returned by the pthread_getcpuclockid().
1410 // If the fast Posix clocks are supported then the clock_getres()
1411 // must return at least tp.tv_sec == 0 which means a resolution
1412 // better than 1 sec. This is extra check for reliability.
1413
1414 if (pthread_getcpuclockid_func &&
1415 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1416 os::Posix::clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1417 _supports_fast_thread_cpu_time = true;
1418 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1419 }
1420 }
1421
1422 jlong os::javaTimeNanos() {
1423 if (os::supports_monotonic_clock()) {
1424 struct timespec tp;
1425 int status = os::Posix::clock_gettime(CLOCK_MONOTONIC, &tp);
1426 assert(status == 0, "gettime error");
1427 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1428 return result;
1429 } else {
1430 timeval time;
1431 int status = gettimeofday(&time, NULL);
1432 assert(status != -1, "linux error");
1433 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1434 return 1000 * usecs;
1435 }
1436 }
1437
1438 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1439 if (os::supports_monotonic_clock()) {
1440 info_ptr->max_value = ALL_64_BITS;
1441
1442 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1443 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1444 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1445 } else {
1446 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1447 info_ptr->max_value = ALL_64_BITS;
1448
1449 // gettimeofday is a real time clock so it skips
1450 info_ptr->may_skip_backward = true;
1451 info_ptr->may_skip_forward = true;
1452 }
1453
1454 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1455 }
1456
1457 // Return the real, user, and system times in seconds from an
1458 // arbitrary fixed point in the past.
1459 bool os::getTimesSecs(double* process_real_time,
1460 double* process_user_time,
1461 double* process_system_time) {
1462 struct tms ticks;
1463 clock_t real_ticks = times(&ticks);
1464
1465 if (real_ticks == (clock_t) (-1)) {
1466 return false;
1467 } else {
1468 double ticks_per_second = (double) clock_tics_per_sec;
1469 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1470 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1471 *process_real_time = ((double) real_ticks) / ticks_per_second;
1472
1473 return true;
1474 }
1475 }
1476
1477
1478 char * os::local_time_string(char *buf, size_t buflen) {
1479 struct tm t;
1480 time_t long_time;
1481 time(&long_time);
1482 localtime_r(&long_time, &t);
1483 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1484 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1485 t.tm_hour, t.tm_min, t.tm_sec);
1486 return buf;
1487 }
1488
1489 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1490 return localtime_r(clock, res);
1491 }
1492
1493 ////////////////////////////////////////////////////////////////////////////////
1494 // runtime exit support
1495
1496 // Note: os::shutdown() might be called very early during initialization, or
1497 // called from signal handler. Before adding something to os::shutdown(), make
1498 // sure it is async-safe and can handle partially initialized VM.
1499 void os::shutdown() {
1500
1501 // allow PerfMemory to attempt cleanup of any persistent resources
1502 perfMemory_exit();
1503
1504 // needs to remove object in file system
1505 AttachListener::abort();
1506
1507 // flush buffered output, finish log files
1508 ostream_abort();
1509
1510 // Check for abort hook
1511 abort_hook_t abort_hook = Arguments::abort_hook();
1512 if (abort_hook != NULL) {
1513 abort_hook();
1514 }
1515
1516 }
1517
1518 // Note: os::abort() might be called very early during initialization, or
1519 // called from signal handler. Before adding something to os::abort(), make
1520 // sure it is async-safe and can handle partially initialized VM.
1521 void os::abort(bool dump_core, void* siginfo, const void* context) {
1522 os::shutdown();
1523 if (dump_core) {
1524 if (DumpPrivateMappingsInCore) {
1525 ClassLoader::close_jrt_image();
1526 }
1527 #ifndef PRODUCT
1528 fdStream out(defaultStream::output_fd());
1529 out.print_raw("Current thread is ");
1530 char buf[16];
1531 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1532 out.print_raw_cr(buf);
1533 out.print_raw_cr("Dumping core ...");
1534 #endif
1535 ::abort(); // dump core
1536 }
1537
1538 ::exit(1);
1539 }
1540
1541 // Die immediately, no exit hook, no abort hook, no cleanup.
1542 // Dump a core file, if possible, for debugging.
1543 void os::die() {
1544 if (TestUnresponsiveErrorHandler && !CreateCoredumpOnCrash) {
1545 // For TimeoutInErrorHandlingTest.java, we just kill the VM
1546 // and don't take the time to generate a core file.
1547 os::signal_raise(SIGKILL);
1548 } else {
1549 ::abort();
1550 }
1551 }
1552
1553 // thread_id is kernel thread id (similar to Solaris LWP id)
1554 intx os::current_thread_id() { return os::Linux::gettid(); }
1555 int os::current_process_id() {
1556 return ::getpid();
1557 }
1558
1559 // DLL functions
1560
1561 const char* os::dll_file_extension() { return ".so"; }
1562
1563 // This must be hard coded because it's the system's temporary
1564 // directory not the java application's temp directory, ala java.io.tmpdir.
1565 const char* os::get_temp_directory() { return "/tmp"; }
1566
1567 static bool file_exists(const char* filename) {
1568 struct stat statbuf;
1569 if (filename == NULL || strlen(filename) == 0) {
1570 return false;
1571 }
1572 return os::stat(filename, &statbuf) == 0;
1573 }
1574
1575 // check if addr is inside libjvm.so
1576 bool os::address_is_in_vm(address addr) {
1577 static address libjvm_base_addr;
1578 Dl_info dlinfo;
1579
1580 if (libjvm_base_addr == NULL) {
1581 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1582 libjvm_base_addr = (address)dlinfo.dli_fbase;
1583 }
1584 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1585 }
1586
1587 if (dladdr((void *)addr, &dlinfo) != 0) {
1588 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1589 }
1590
1591 return false;
1592 }
1593
1594 bool os::dll_address_to_function_name(address addr, char *buf,
1595 int buflen, int *offset,
1596 bool demangle) {
1597 // buf is not optional, but offset is optional
1598 assert(buf != NULL, "sanity check");
1599
1600 Dl_info dlinfo;
1601
1602 if (dladdr((void*)addr, &dlinfo) != 0) {
1603 // see if we have a matching symbol
1604 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1605 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1606 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1607 }
1608 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1609 return true;
1610 }
1611 // no matching symbol so try for just file info
1612 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1613 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1614 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1615 return true;
1616 }
1617 }
1618 }
1619
1620 buf[0] = '\0';
1621 if (offset != NULL) *offset = -1;
1622 return false;
1623 }
1624
1625 struct _address_to_library_name {
1626 address addr; // input : memory address
1627 size_t buflen; // size of fname
1628 char* fname; // output: library name
1629 address base; // library base addr
1630 };
1631
1632 static int address_to_library_name_callback(struct dl_phdr_info *info,
1633 size_t size, void *data) {
1634 int i;
1635 bool found = false;
1636 address libbase = NULL;
1637 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1638
1639 // iterate through all loadable segments
1640 for (i = 0; i < info->dlpi_phnum; i++) {
1641 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1642 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1643 // base address of a library is the lowest address of its loaded
1644 // segments.
1645 if (libbase == NULL || libbase > segbase) {
1646 libbase = segbase;
1647 }
1648 // see if 'addr' is within current segment
1649 if (segbase <= d->addr &&
1650 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1651 found = true;
1652 }
1653 }
1654 }
1655
1656 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1657 // so dll_address_to_library_name() can fall through to use dladdr() which
1658 // can figure out executable name from argv[0].
1659 if (found && info->dlpi_name && info->dlpi_name[0]) {
1660 d->base = libbase;
1661 if (d->fname) {
1662 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1663 }
1664 return 1;
1665 }
1666 return 0;
1667 }
1668
1669 bool os::dll_address_to_library_name(address addr, char* buf,
1670 int buflen, int* offset) {
1671 // buf is not optional, but offset is optional
1672 assert(buf != NULL, "sanity check");
1673
1674 Dl_info dlinfo;
1675 struct _address_to_library_name data;
1676
1677 // There is a bug in old glibc dladdr() implementation that it could resolve
1678 // to wrong library name if the .so file has a base address != NULL. Here
1679 // we iterate through the program headers of all loaded libraries to find
1680 // out which library 'addr' really belongs to. This workaround can be
1681 // removed once the minimum requirement for glibc is moved to 2.3.x.
1682 data.addr = addr;
1683 data.fname = buf;
1684 data.buflen = buflen;
1685 data.base = NULL;
1686 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1687
1688 if (rslt) {
1689 // buf already contains library name
1690 if (offset) *offset = addr - data.base;
1691 return true;
1692 }
1693 if (dladdr((void*)addr, &dlinfo) != 0) {
1694 if (dlinfo.dli_fname != NULL) {
1695 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1696 }
1697 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1698 *offset = addr - (address)dlinfo.dli_fbase;
1699 }
1700 return true;
1701 }
1702
1703 buf[0] = '\0';
1704 if (offset) *offset = -1;
1705 return false;
1706 }
1707
1708 // Loads .dll/.so and
1709 // in case of error it checks if .dll/.so was built for the
1710 // same architecture as Hotspot is running on
1711
1712
1713 // Remember the stack's state. The Linux dynamic linker will change
1714 // the stack to 'executable' at most once, so we must safepoint only once.
1715 bool os::Linux::_stack_is_executable = false;
1716
1717 // VM operation that loads a library. This is necessary if stack protection
1718 // of the Java stacks can be lost during loading the library. If we
1719 // do not stop the Java threads, they can stack overflow before the stacks
1720 // are protected again.
1721 class VM_LinuxDllLoad: public VM_Operation {
1722 private:
1723 const char *_filename;
1724 char *_ebuf;
1725 int _ebuflen;
1726 void *_lib;
1727 public:
1728 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1729 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1730 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1731 void doit() {
1732 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1733 os::Linux::_stack_is_executable = true;
1734 }
1735 void* loaded_library() { return _lib; }
1736 };
1737
1738 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1739 void * result = NULL;
1740 bool load_attempted = false;
1741
1742 log_info(os)("attempting shared library load of %s", filename);
1743
1744 // Check whether the library to load might change execution rights
1745 // of the stack. If they are changed, the protection of the stack
1746 // guard pages will be lost. We need a safepoint to fix this.
1747 //
1748 // See Linux man page execstack(8) for more info.
1749 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1750 if (!ElfFile::specifies_noexecstack(filename)) {
1751 if (!is_init_completed()) {
1752 os::Linux::_stack_is_executable = true;
1753 // This is OK - No Java threads have been created yet, and hence no
1754 // stack guard pages to fix.
1755 //
1756 // Dynamic loader will make all stacks executable after
1757 // this function returns, and will not do that again.
1758 assert(Threads::number_of_threads() == 0, "no Java threads should exist yet.");
1759 } else {
1760 warning("You have loaded library %s which might have disabled stack guard. "
1761 "The VM will try to fix the stack guard now.\n"
1762 "It's highly recommended that you fix the library with "
1763 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1764 filename);
1765
1766 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1767 JavaThread *jt = JavaThread::current();
1768 if (jt->thread_state() != _thread_in_native) {
1769 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1770 // that requires ExecStack. Cannot enter safe point. Let's give up.
1771 warning("Unable to fix stack guard. Giving up.");
1772 } else {
1773 if (!LoadExecStackDllInVMThread) {
1774 // This is for the case where the DLL has an static
1775 // constructor function that executes JNI code. We cannot
1776 // load such DLLs in the VMThread.
1777 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1778 }
1779
1780 ThreadInVMfromNative tiv(jt);
1781 debug_only(VMNativeEntryWrapper vew;)
1782
1783 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1784 VMThread::execute(&op);
1785 if (LoadExecStackDllInVMThread) {
1786 result = op.loaded_library();
1787 }
1788 load_attempted = true;
1789 }
1790 }
1791 }
1792 }
1793
1794 if (!load_attempted) {
1795 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1796 }
1797
1798 if (result != NULL) {
1799 // Successful loading
1800 return result;
1801 }
1802
1803 Elf32_Ehdr elf_head;
1804 int diag_msg_max_length=ebuflen-strlen(ebuf);
1805 char* diag_msg_buf=ebuf+strlen(ebuf);
1806
1807 if (diag_msg_max_length==0) {
1808 // No more space in ebuf for additional diagnostics message
1809 return NULL;
1810 }
1811
1812
1813 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1814
1815 if (file_descriptor < 0) {
1816 // Can't open library, report dlerror() message
1817 return NULL;
1818 }
1819
1820 bool failed_to_read_elf_head=
1821 (sizeof(elf_head)!=
1822 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1823
1824 ::close(file_descriptor);
1825 if (failed_to_read_elf_head) {
1826 // file i/o error - report dlerror() msg
1827 return NULL;
1828 }
1829
1830 if (elf_head.e_ident[EI_DATA] != LITTLE_ENDIAN_ONLY(ELFDATA2LSB) BIG_ENDIAN_ONLY(ELFDATA2MSB)) {
1831 // handle invalid/out of range endianness values
1832 if (elf_head.e_ident[EI_DATA] == 0 || elf_head.e_ident[EI_DATA] > 2) {
1833 return NULL;
1834 }
1835
1836 #if defined(VM_LITTLE_ENDIAN)
1837 // VM is LE, shared object BE
1838 elf_head.e_machine = be16toh(elf_head.e_machine);
1839 #else
1840 // VM is BE, shared object LE
1841 elf_head.e_machine = le16toh(elf_head.e_machine);
1842 #endif
1843 }
1844
1845 typedef struct {
1846 Elf32_Half code; // Actual value as defined in elf.h
1847 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1848 unsigned char elf_class; // 32 or 64 bit
1849 unsigned char endianness; // MSB or LSB
1850 char* name; // String representation
1851 } arch_t;
1852
1853 #ifndef EM_486
1854 #define EM_486 6 /* Intel 80486 */
1855 #endif
1856 #ifndef EM_AARCH64
1857 #define EM_AARCH64 183 /* ARM AARCH64 */
1858 #endif
1859
1860 static const arch_t arch_array[]={
1861 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1862 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1863 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1864 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1865 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1866 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1867 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1868 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1869 #if defined(VM_LITTLE_ENDIAN)
1870 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1871 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1872 #else
1873 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1874 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1875 #endif
1876 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1877 // we only support 64 bit z architecture
1878 {EM_S390, EM_S390, ELFCLASS64, ELFDATA2MSB, (char*)"IBM System/390"},
1879 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1880 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1881 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1882 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1883 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1884 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1885 };
1886
1887 #if (defined IA32)
1888 static Elf32_Half running_arch_code=EM_386;
1889 #elif (defined AMD64) || (defined X32)
1890 static Elf32_Half running_arch_code=EM_X86_64;
1891 #elif (defined IA64)
1892 static Elf32_Half running_arch_code=EM_IA_64;
1893 #elif (defined __sparc) && (defined _LP64)
1894 static Elf32_Half running_arch_code=EM_SPARCV9;
1895 #elif (defined __sparc) && (!defined _LP64)
1896 static Elf32_Half running_arch_code=EM_SPARC;
1897 #elif (defined __powerpc64__)
1898 static Elf32_Half running_arch_code=EM_PPC64;
1899 #elif (defined __powerpc__)
1900 static Elf32_Half running_arch_code=EM_PPC;
1901 #elif (defined AARCH64)
1902 static Elf32_Half running_arch_code=EM_AARCH64;
1903 #elif (defined ARM)
1904 static Elf32_Half running_arch_code=EM_ARM;
1905 #elif (defined S390)
1906 static Elf32_Half running_arch_code=EM_S390;
1907 #elif (defined ALPHA)
1908 static Elf32_Half running_arch_code=EM_ALPHA;
1909 #elif (defined MIPSEL)
1910 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1911 #elif (defined PARISC)
1912 static Elf32_Half running_arch_code=EM_PARISC;
1913 #elif (defined MIPS)
1914 static Elf32_Half running_arch_code=EM_MIPS;
1915 #elif (defined M68K)
1916 static Elf32_Half running_arch_code=EM_68K;
1917 #elif (defined SH)
1918 static Elf32_Half running_arch_code=EM_SH;
1919 #else
1920 #error Method os::dll_load requires that one of following is defined:\
1921 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1922 #endif
1923
1924 // Identify compatibility class for VM's architecture and library's architecture
1925 // Obtain string descriptions for architectures
1926
1927 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1928 int running_arch_index=-1;
1929
1930 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1931 if (running_arch_code == arch_array[i].code) {
1932 running_arch_index = i;
1933 }
1934 if (lib_arch.code == arch_array[i].code) {
1935 lib_arch.compat_class = arch_array[i].compat_class;
1936 lib_arch.name = arch_array[i].name;
1937 }
1938 }
1939
1940 assert(running_arch_index != -1,
1941 "Didn't find running architecture code (running_arch_code) in arch_array");
1942 if (running_arch_index == -1) {
1943 // Even though running architecture detection failed
1944 // we may still continue with reporting dlerror() message
1945 return NULL;
1946 }
1947
1948 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1949 if (lib_arch.name != NULL) {
1950 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1951 " (Possible cause: can't load %s .so on a %s platform)",
1952 lib_arch.name, arch_array[running_arch_index].name);
1953 } else {
1954 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1955 " (Possible cause: can't load this .so (machine code=0x%x) on a %s platform)",
1956 lib_arch.code, arch_array[running_arch_index].name);
1957 }
1958 return NULL;
1959 }
1960
1961 if (lib_arch.endianness != arch_array[running_arch_index].endianness) {
1962 ::snprintf(diag_msg_buf, diag_msg_max_length-1, " (Possible cause: endianness mismatch)");
1963 return NULL;
1964 }
1965
1966 // ELF file class/capacity : 0 - invalid, 1 - 32bit, 2 - 64bit
1967 if (lib_arch.elf_class > 2 || lib_arch.elf_class < 1) {
1968 ::snprintf(diag_msg_buf, diag_msg_max_length-1, " (Possible cause: invalid ELF file class)");
1969 return NULL;
1970 }
1971
1972 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1973 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1974 " (Possible cause: architecture word width mismatch, can't load %d-bit .so on a %d-bit platform)",
1975 (int) lib_arch.elf_class * 32, arch_array[running_arch_index].elf_class * 32);
1976 return NULL;
1977 }
1978
1979 return NULL;
1980 }
1981
1982 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1983 int ebuflen) {
1984 void * result = ::dlopen(filename, RTLD_LAZY);
1985 if (result == NULL) {
1986 const char* error_report = ::dlerror();
1987 if (error_report == NULL) {
1988 error_report = "dlerror returned no error description";
1989 }
1990 if (ebuf != NULL && ebuflen > 0) {
1991 ::strncpy(ebuf, error_report, ebuflen-1);
1992 ebuf[ebuflen-1]='\0';
1993 }
1994 Events::log(NULL, "Loading shared library %s failed, %s", filename, error_report);
1995 log_info(os)("shared library load of %s failed, %s", filename, error_report);
1996 } else {
1997 Events::log(NULL, "Loaded shared library %s", filename);
1998 log_info(os)("shared library load of %s was successful", filename);
1999 }
2000 return result;
2001 }
2002
2003 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
2004 int ebuflen) {
2005 void * result = NULL;
2006 if (LoadExecStackDllInVMThread) {
2007 result = dlopen_helper(filename, ebuf, ebuflen);
2008 }
2009
2010 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2011 // library that requires an executable stack, or which does not have this
2012 // stack attribute set, dlopen changes the stack attribute to executable. The
2013 // read protection of the guard pages gets lost.
2014 //
2015 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2016 // may have been queued at the same time.
2017
2018 if (!_stack_is_executable) {
2019 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
2020 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2021 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
2022 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
2023 warning("Attempt to reguard stack yellow zone failed.");
2024 }
2025 }
2026 }
2027 }
2028
2029 return result;
2030 }
2031
2032 void* os::dll_lookup(void* handle, const char* name) {
2033 void* res = dlsym(handle, name);
2034 return res;
2035 }
2036
2037 void* os::get_default_process_handle() {
2038 return (void*)::dlopen(NULL, RTLD_LAZY);
2039 }
2040
2041 static bool _print_ascii_file(const char* filename, outputStream* st, const char* hdr = NULL) {
2042 int fd = ::open(filename, O_RDONLY);
2043 if (fd == -1) {
2044 return false;
2045 }
2046
2047 if (hdr != NULL) {
2048 st->print_cr("%s", hdr);
2049 }
2050
2051 char buf[33];
2052 int bytes;
2053 buf[32] = '\0';
2054 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
2055 st->print_raw(buf, bytes);
2056 }
2057
2058 ::close(fd);
2059
2060 return true;
2061 }
2062
2063 void os::print_dll_info(outputStream *st) {
2064 st->print_cr("Dynamic libraries:");
2065
2066 char fname[32];
2067 pid_t pid = os::Linux::gettid();
2068
2069 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2070
2071 if (!_print_ascii_file(fname, st)) {
2072 st->print("Can not get library information for pid = %d\n", pid);
2073 }
2074 }
2075
2076 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2077 FILE *procmapsFile = NULL;
2078
2079 // Open the procfs maps file for the current process
2080 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2081 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2082 char line[PATH_MAX + 100];
2083
2084 // Read line by line from 'file'
2085 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2086 u8 base, top, offset, inode;
2087 char permissions[5];
2088 char device[6];
2089 char name[PATH_MAX + 1];
2090
2091 // Parse fields from line
2092 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %7s " INT64_FORMAT " %s",
2093 &base, &top, permissions, &offset, device, &inode, name);
2094
2095 // Filter by device id '00:00' so that we only get file system mapped files.
2096 if (strcmp(device, "00:00") != 0) {
2097
2098 // Call callback with the fields of interest
2099 if(callback(name, (address)base, (address)top, param)) {
2100 // Oops abort, callback aborted
2101 fclose(procmapsFile);
2102 return 1;
2103 }
2104 }
2105 }
2106 fclose(procmapsFile);
2107 }
2108 return 0;
2109 }
2110
2111 void os::print_os_info_brief(outputStream* st) {
2112 os::Linux::print_distro_info(st);
2113
2114 os::Posix::print_uname_info(st);
2115
2116 os::Linux::print_libversion_info(st);
2117
2118 }
2119
2120 void os::print_os_info(outputStream* st) {
2121 st->print("OS:");
2122
2123 os::Linux::print_distro_info(st);
2124
2125 os::Posix::print_uname_info(st);
2126
2127 // Print warning if unsafe chroot environment detected
2128 if (unsafe_chroot_detected) {
2129 st->print("WARNING!! ");
2130 st->print_cr("%s", unstable_chroot_error);
2131 }
2132
2133 os::Linux::print_libversion_info(st);
2134
2135 os::Posix::print_rlimit_info(st);
2136
2137 os::Posix::print_load_average(st);
2138
2139 os::Linux::print_full_memory_info(st);
2140
2141 os::Linux::print_proc_sys_info(st);
2142
2143 os::Linux::print_ld_preload_file(st);
2144
2145 os::Linux::print_container_info(st);
2146
2147 VM_Version::print_platform_virtualization_info(st);
2148
2149 os::Linux::print_steal_info(st);
2150 }
2151
2152 // Try to identify popular distros.
2153 // Most Linux distributions have a /etc/XXX-release file, which contains
2154 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2155 // file that also contains the OS version string. Some have more than one
2156 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2157 // /etc/redhat-release.), so the order is important.
2158 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2159 // their own specific XXX-release file as well as a redhat-release file.
2160 // Because of this the XXX-release file needs to be searched for before the
2161 // redhat-release file.
2162 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2163 // search for redhat-release / SuSE-release needs to be before lsb-release.
2164 // Since the lsb-release file is the new standard it needs to be searched
2165 // before the older style release files.
2166 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2167 // next to last resort. The os-release file is a new standard that contains
2168 // distribution information and the system-release file seems to be an old
2169 // standard that has been replaced by the lsb-release and os-release files.
2170 // Searching for the debian_version file is the last resort. It contains
2171 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2172 // "Debian " is printed before the contents of the debian_version file.
2173
2174 const char* distro_files[] = {
2175 "/etc/oracle-release",
2176 "/etc/mandriva-release",
2177 "/etc/mandrake-release",
2178 "/etc/sun-release",
2179 "/etc/redhat-release",
2180 "/etc/SuSE-release",
2181 "/etc/lsb-release",
2182 "/etc/turbolinux-release",
2183 "/etc/gentoo-release",
2184 "/etc/ltib-release",
2185 "/etc/angstrom-version",
2186 "/etc/system-release",
2187 "/etc/os-release",
2188 NULL };
2189
2190 void os::Linux::print_distro_info(outputStream* st) {
2191 for (int i = 0;; i++) {
2192 const char* file = distro_files[i];
2193 if (file == NULL) {
2194 break; // done
2195 }
2196 // If file prints, we found it.
2197 if (_print_ascii_file(file, st)) {
2198 return;
2199 }
2200 }
2201
2202 if (file_exists("/etc/debian_version")) {
2203 st->print("Debian ");
2204 _print_ascii_file("/etc/debian_version", st);
2205 } else {
2206 st->print("Linux");
2207 }
2208 st->cr();
2209 }
2210
2211 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2212 char buf[256];
2213 while (fgets(buf, sizeof(buf), fp)) {
2214 // Edit out extra stuff in expected format
2215 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2216 char* ptr = strstr(buf, "\""); // the name is in quotes
2217 if (ptr != NULL) {
2218 ptr++; // go beyond first quote
2219 char* nl = strchr(ptr, '\"');
2220 if (nl != NULL) *nl = '\0';
2221 strncpy(distro, ptr, length);
2222 } else {
2223 ptr = strstr(buf, "=");
2224 ptr++; // go beyond equals then
2225 char* nl = strchr(ptr, '\n');
2226 if (nl != NULL) *nl = '\0';
2227 strncpy(distro, ptr, length);
2228 }
2229 return;
2230 } else if (get_first_line) {
2231 char* nl = strchr(buf, '\n');
2232 if (nl != NULL) *nl = '\0';
2233 strncpy(distro, buf, length);
2234 return;
2235 }
2236 }
2237 // print last line and close
2238 char* nl = strchr(buf, '\n');
2239 if (nl != NULL) *nl = '\0';
2240 strncpy(distro, buf, length);
2241 }
2242
2243 static void parse_os_info(char* distro, size_t length, const char* file) {
2244 FILE* fp = fopen(file, "r");
2245 if (fp != NULL) {
2246 // if suse format, print out first line
2247 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2248 parse_os_info_helper(fp, distro, length, get_first_line);
2249 fclose(fp);
2250 }
2251 }
2252
2253 void os::get_summary_os_info(char* buf, size_t buflen) {
2254 for (int i = 0;; i++) {
2255 const char* file = distro_files[i];
2256 if (file == NULL) {
2257 break; // ran out of distro_files
2258 }
2259 if (file_exists(file)) {
2260 parse_os_info(buf, buflen, file);
2261 return;
2262 }
2263 }
2264 // special case for debian
2265 if (file_exists("/etc/debian_version")) {
2266 strncpy(buf, "Debian ", buflen);
2267 if (buflen > 7) {
2268 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2269 }
2270 } else {
2271 strncpy(buf, "Linux", buflen);
2272 }
2273 }
2274
2275 void os::Linux::print_libversion_info(outputStream* st) {
2276 // libc, pthread
2277 st->print("libc:");
2278 st->print("%s ", os::Linux::glibc_version());
2279 st->print("%s ", os::Linux::libpthread_version());
2280 st->cr();
2281 }
2282
2283 void os::Linux::print_proc_sys_info(outputStream* st) {
2284 st->cr();
2285 st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):");
2286 _print_ascii_file("/proc/sys/kernel/threads-max", st);
2287 st->cr();
2288 st->cr();
2289
2290 st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):");
2291 _print_ascii_file("/proc/sys/vm/max_map_count", st);
2292 st->cr();
2293 st->cr();
2294
2295 st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):");
2296 _print_ascii_file("/proc/sys/kernel/pid_max", st);
2297 st->cr();
2298 st->cr();
2299 }
2300
2301 void os::Linux::print_full_memory_info(outputStream* st) {
2302 st->print("\n/proc/meminfo:\n");
2303 _print_ascii_file("/proc/meminfo", st);
2304 st->cr();
2305 }
2306
2307 void os::Linux::print_ld_preload_file(outputStream* st) {
2308 _print_ascii_file("/etc/ld.so.preload", st, "\n/etc/ld.so.preload:");
2309 st->cr();
2310 }
2311
2312 void os::Linux::print_container_info(outputStream* st) {
2313 if (!OSContainer::is_containerized()) {
2314 return;
2315 }
2316
2317 st->print("container (cgroup) information:\n");
2318
2319 const char *p_ct = OSContainer::container_type();
2320 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "not supported");
2321
2322 char *p = OSContainer::cpu_cpuset_cpus();
2323 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "not supported");
2324 free(p);
2325
2326 p = OSContainer::cpu_cpuset_memory_nodes();
2327 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "not supported");
2328 free(p);
2329
2330 int i = OSContainer::active_processor_count();
2331 st->print("active_processor_count: ");
2332 if (i > 0) {
2333 st->print("%d\n", i);
2334 } else {
2335 st->print("not supported\n");
2336 }
2337
2338 i = OSContainer::cpu_quota();
2339 st->print("cpu_quota: ");
2340 if (i > 0) {
2341 st->print("%d\n", i);
2342 } else {
2343 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no quota");
2344 }
2345
2346 i = OSContainer::cpu_period();
2347 st->print("cpu_period: ");
2348 if (i > 0) {
2349 st->print("%d\n", i);
2350 } else {
2351 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no period");
2352 }
2353
2354 i = OSContainer::cpu_shares();
2355 st->print("cpu_shares: ");
2356 if (i > 0) {
2357 st->print("%d\n", i);
2358 } else {
2359 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no shares");
2360 }
2361
2362 jlong j = OSContainer::memory_limit_in_bytes();
2363 st->print("memory_limit_in_bytes: ");
2364 if (j > 0) {
2365 st->print(JLONG_FORMAT "\n", j);
2366 } else {
2367 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2368 }
2369
2370 j = OSContainer::memory_and_swap_limit_in_bytes();
2371 st->print("memory_and_swap_limit_in_bytes: ");
2372 if (j > 0) {
2373 st->print(JLONG_FORMAT "\n", j);
2374 } else {
2375 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2376 }
2377
2378 j = OSContainer::memory_soft_limit_in_bytes();
2379 st->print("memory_soft_limit_in_bytes: ");
2380 if (j > 0) {
2381 st->print(JLONG_FORMAT "\n", j);
2382 } else {
2383 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2384 }
2385
2386 j = OSContainer::OSContainer::memory_usage_in_bytes();
2387 st->print("memory_usage_in_bytes: ");
2388 if (j > 0) {
2389 st->print(JLONG_FORMAT "\n", j);
2390 } else {
2391 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2392 }
2393
2394 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2395 st->print("memory_max_usage_in_bytes: ");
2396 if (j > 0) {
2397 st->print(JLONG_FORMAT "\n", j);
2398 } else {
2399 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2400 }
2401 st->cr();
2402 }
2403
2404 void os::Linux::print_steal_info(outputStream* st) {
2405 if (has_initial_tick_info) {
2406 CPUPerfTicks pticks;
2407 bool res = os::Linux::get_tick_information(&pticks, -1);
2408
2409 if (res && pticks.has_steal_ticks) {
2410 uint64_t steal_ticks_difference = pticks.steal - initial_steal_ticks;
2411 uint64_t total_ticks_difference = pticks.total - initial_total_ticks;
2412 double steal_ticks_perc = 0.0;
2413 if (total_ticks_difference != 0) {
2414 steal_ticks_perc = (double) steal_ticks_difference / total_ticks_difference;
2415 }
2416 st->print_cr("Steal ticks since vm start: " UINT64_FORMAT, steal_ticks_difference);
2417 st->print_cr("Steal ticks percentage since vm start:%7.3f", steal_ticks_perc);
2418 }
2419 }
2420 }
2421
2422 void os::print_memory_info(outputStream* st) {
2423
2424 st->print("Memory:");
2425 st->print(" %dk page", os::vm_page_size()>>10);
2426
2427 // values in struct sysinfo are "unsigned long"
2428 struct sysinfo si;
2429 sysinfo(&si);
2430
2431 st->print(", physical " UINT64_FORMAT "k",
2432 os::physical_memory() >> 10);
2433 st->print("(" UINT64_FORMAT "k free)",
2434 os::available_memory() >> 10);
2435 st->print(", swap " UINT64_FORMAT "k",
2436 ((jlong)si.totalswap * si.mem_unit) >> 10);
2437 st->print("(" UINT64_FORMAT "k free)",
2438 ((jlong)si.freeswap * si.mem_unit) >> 10);
2439 st->cr();
2440 }
2441
2442 // Print the first "model name" line and the first "flags" line
2443 // that we find and nothing more. We assume "model name" comes
2444 // before "flags" so if we find a second "model name", then the
2445 // "flags" field is considered missing.
2446 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2447 #if defined(IA32) || defined(AMD64)
2448 // Other platforms have less repetitive cpuinfo files
2449 FILE *fp = fopen("/proc/cpuinfo", "r");
2450 if (fp) {
2451 while (!feof(fp)) {
2452 if (fgets(buf, buflen, fp)) {
2453 // Assume model name comes before flags
2454 bool model_name_printed = false;
2455 if (strstr(buf, "model name") != NULL) {
2456 if (!model_name_printed) {
2457 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2458 st->print_raw(buf);
2459 model_name_printed = true;
2460 } else {
2461 // model name printed but not flags? Odd, just return
2462 fclose(fp);
2463 return true;
2464 }
2465 }
2466 // print the flags line too
2467 if (strstr(buf, "flags") != NULL) {
2468 st->print_raw(buf);
2469 fclose(fp);
2470 return true;
2471 }
2472 }
2473 }
2474 fclose(fp);
2475 }
2476 #endif // x86 platforms
2477 return false;
2478 }
2479
2480 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2481 // Only print the model name if the platform provides this as a summary
2482 if (!print_model_name_and_flags(st, buf, buflen)) {
2483 st->print("\n/proc/cpuinfo:\n");
2484 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2485 st->print_cr(" <Not Available>");
2486 }
2487 }
2488 }
2489
2490 #if defined(AMD64) || defined(IA32) || defined(X32)
2491 const char* search_string = "model name";
2492 #elif defined(M68K)
2493 const char* search_string = "CPU";
2494 #elif defined(PPC64)
2495 const char* search_string = "cpu";
2496 #elif defined(S390)
2497 const char* search_string = "machine =";
2498 #elif defined(SPARC)
2499 const char* search_string = "cpu";
2500 #else
2501 const char* search_string = "Processor";
2502 #endif
2503
2504 // Parses the cpuinfo file for string representing the model name.
2505 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2506 FILE* fp = fopen("/proc/cpuinfo", "r");
2507 if (fp != NULL) {
2508 while (!feof(fp)) {
2509 char buf[256];
2510 if (fgets(buf, sizeof(buf), fp)) {
2511 char* start = strstr(buf, search_string);
2512 if (start != NULL) {
2513 char *ptr = start + strlen(search_string);
2514 char *end = buf + strlen(buf);
2515 while (ptr != end) {
2516 // skip whitespace and colon for the rest of the name.
2517 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2518 break;
2519 }
2520 ptr++;
2521 }
2522 if (ptr != end) {
2523 // reasonable string, get rid of newline and keep the rest
2524 char* nl = strchr(buf, '\n');
2525 if (nl != NULL) *nl = '\0';
2526 strncpy(cpuinfo, ptr, length);
2527 fclose(fp);
2528 return;
2529 }
2530 }
2531 }
2532 }
2533 fclose(fp);
2534 }
2535 // cpuinfo not found or parsing failed, just print generic string. The entire
2536 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2537 #if defined(AARCH64)
2538 strncpy(cpuinfo, "AArch64", length);
2539 #elif defined(AMD64)
2540 strncpy(cpuinfo, "x86_64", length);
2541 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2542 strncpy(cpuinfo, "ARM", length);
2543 #elif defined(IA32)
2544 strncpy(cpuinfo, "x86_32", length);
2545 #elif defined(IA64)
2546 strncpy(cpuinfo, "IA64", length);
2547 #elif defined(PPC)
2548 strncpy(cpuinfo, "PPC64", length);
2549 #elif defined(S390)
2550 strncpy(cpuinfo, "S390", length);
2551 #elif defined(SPARC)
2552 strncpy(cpuinfo, "sparcv9", length);
2553 #elif defined(ZERO_LIBARCH)
2554 strncpy(cpuinfo, ZERO_LIBARCH, length);
2555 #else
2556 strncpy(cpuinfo, "unknown", length);
2557 #endif
2558 }
2559
2560 static void print_signal_handler(outputStream* st, int sig,
2561 char* buf, size_t buflen);
2562
2563 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2564 st->print_cr("Signal Handlers:");
2565 print_signal_handler(st, SIGSEGV, buf, buflen);
2566 print_signal_handler(st, SIGBUS , buf, buflen);
2567 print_signal_handler(st, SIGFPE , buf, buflen);
2568 print_signal_handler(st, SIGPIPE, buf, buflen);
2569 print_signal_handler(st, SIGXFSZ, buf, buflen);
2570 print_signal_handler(st, SIGILL , buf, buflen);
2571 print_signal_handler(st, SR_signum, buf, buflen);
2572 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2573 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2574 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2575 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2576 #if defined(PPC64)
2577 print_signal_handler(st, SIGTRAP, buf, buflen);
2578 #endif
2579 }
2580
2581 static char saved_jvm_path[MAXPATHLEN] = {0};
2582
2583 // Find the full path to the current module, libjvm.so
2584 void os::jvm_path(char *buf, jint buflen) {
2585 // Error checking.
2586 if (buflen < MAXPATHLEN) {
2587 assert(false, "must use a large-enough buffer");
2588 buf[0] = '\0';
2589 return;
2590 }
2591 // Lazy resolve the path to current module.
2592 if (saved_jvm_path[0] != 0) {
2593 strcpy(buf, saved_jvm_path);
2594 return;
2595 }
2596
2597 char dli_fname[MAXPATHLEN];
2598 bool ret = dll_address_to_library_name(
2599 CAST_FROM_FN_PTR(address, os::jvm_path),
2600 dli_fname, sizeof(dli_fname), NULL);
2601 assert(ret, "cannot locate libjvm");
2602 char *rp = NULL;
2603 if (ret && dli_fname[0] != '\0') {
2604 rp = os::Posix::realpath(dli_fname, buf, buflen);
2605 }
2606 if (rp == NULL) {
2607 return;
2608 }
2609
2610 if (Arguments::sun_java_launcher_is_altjvm()) {
2611 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2612 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2613 // If "/jre/lib/" appears at the right place in the string, then
2614 // assume we are installed in a JDK and we're done. Otherwise, check
2615 // for a JAVA_HOME environment variable and fix up the path so it
2616 // looks like libjvm.so is installed there (append a fake suffix
2617 // hotspot/libjvm.so).
2618 const char *p = buf + strlen(buf) - 1;
2619 for (int count = 0; p > buf && count < 5; ++count) {
2620 for (--p; p > buf && *p != '/'; --p)
2621 /* empty */ ;
2622 }
2623
2624 if (strncmp(p, "/jre/lib/", 9) != 0) {
2625 // Look for JAVA_HOME in the environment.
2626 char* java_home_var = ::getenv("JAVA_HOME");
2627 if (java_home_var != NULL && java_home_var[0] != 0) {
2628 char* jrelib_p;
2629 int len;
2630
2631 // Check the current module name "libjvm.so".
2632 p = strrchr(buf, '/');
2633 if (p == NULL) {
2634 return;
2635 }
2636 assert(strstr(p, "/libjvm") == p, "invalid library name");
2637
2638 rp = os::Posix::realpath(java_home_var, buf, buflen);
2639 if (rp == NULL) {
2640 return;
2641 }
2642
2643 // determine if this is a legacy image or modules image
2644 // modules image doesn't have "jre" subdirectory
2645 len = strlen(buf);
2646 assert(len < buflen, "Ran out of buffer room");
2647 jrelib_p = buf + len;
2648 snprintf(jrelib_p, buflen-len, "/jre/lib");
2649 if (0 != access(buf, F_OK)) {
2650 snprintf(jrelib_p, buflen-len, "/lib");
2651 }
2652
2653 if (0 == access(buf, F_OK)) {
2654 // Use current module name "libjvm.so"
2655 len = strlen(buf);
2656 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2657 } else {
2658 // Go back to path of .so
2659 rp = os::Posix::realpath(dli_fname, buf, buflen);
2660 if (rp == NULL) {
2661 return;
2662 }
2663 }
2664 }
2665 }
2666 }
2667
2668 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2669 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2670 }
2671
2672 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2673 // no prefix required, not even "_"
2674 }
2675
2676 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2677 // no suffix required
2678 }
2679
2680 ////////////////////////////////////////////////////////////////////////////////
2681 // sun.misc.Signal support
2682
2683 static void UserHandler(int sig, void *siginfo, void *context) {
2684 // Ctrl-C is pressed during error reporting, likely because the error
2685 // handler fails to abort. Let VM die immediately.
2686 if (sig == SIGINT && VMError::is_error_reported()) {
2687 os::die();
2688 }
2689
2690 os::signal_notify(sig);
2691 }
2692
2693 void* os::user_handler() {
2694 return CAST_FROM_FN_PTR(void*, UserHandler);
2695 }
2696
2697 extern "C" {
2698 typedef void (*sa_handler_t)(int);
2699 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2700 }
2701
2702 void* os::signal(int signal_number, void* handler) {
2703 struct sigaction sigAct, oldSigAct;
2704
2705 sigfillset(&(sigAct.sa_mask));
2706 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2707 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2708
2709 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2710 // -1 means registration failed
2711 return (void *)-1;
2712 }
2713
2714 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2715 }
2716
2717 void os::signal_raise(int signal_number) {
2718 ::raise(signal_number);
2719 }
2720
2721 // The following code is moved from os.cpp for making this
2722 // code platform specific, which it is by its very nature.
2723
2724 // Will be modified when max signal is changed to be dynamic
2725 int os::sigexitnum_pd() {
2726 return NSIG;
2727 }
2728
2729 // a counter for each possible signal value
2730 static volatile jint pending_signals[NSIG+1] = { 0 };
2731
2732 // Linux(POSIX) specific hand shaking semaphore.
2733 static Semaphore* sig_sem = NULL;
2734 static PosixSemaphore sr_semaphore;
2735
2736 static void jdk_misc_signal_init() {
2737 // Initialize signal structures
2738 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2739
2740 // Initialize signal semaphore
2741 sig_sem = new Semaphore();
2742 }
2743
2744 void os::signal_notify(int sig) {
2745 if (sig_sem != NULL) {
2746 Atomic::inc(&pending_signals[sig]);
2747 sig_sem->signal();
2748 } else {
2749 // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2750 // initialization isn't called.
2751 assert(ReduceSignalUsage, "signal semaphore should be created");
2752 }
2753 }
2754
2755 static int check_pending_signals() {
2756 for (;;) {
2757 for (int i = 0; i < NSIG + 1; i++) {
2758 jint n = pending_signals[i];
2759 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2760 return i;
2761 }
2762 }
2763 JavaThread *thread = JavaThread::current();
2764 ThreadBlockInVM tbivm(thread);
2765
2766 bool threadIsSuspended;
2767 do {
2768 thread->set_suspend_equivalent();
2769 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2770 sig_sem->wait();
2771
2772 // were we externally suspended while we were waiting?
2773 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2774 if (threadIsSuspended) {
2775 // The semaphore has been incremented, but while we were waiting
2776 // another thread suspended us. We don't want to continue running
2777 // while suspended because that would surprise the thread that
2778 // suspended us.
2779 sig_sem->signal();
2780
2781 thread->java_suspend_self();
2782 }
2783 } while (threadIsSuspended);
2784 }
2785 }
2786
2787 int os::signal_wait() {
2788 return check_pending_signals();
2789 }
2790
2791 ////////////////////////////////////////////////////////////////////////////////
2792 // Virtual Memory
2793
2794 int os::vm_page_size() {
2795 // Seems redundant as all get out
2796 assert(os::Linux::page_size() != -1, "must call os::init");
2797 return os::Linux::page_size();
2798 }
2799
2800 // Solaris allocates memory by pages.
2801 int os::vm_allocation_granularity() {
2802 assert(os::Linux::page_size() != -1, "must call os::init");
2803 return os::Linux::page_size();
2804 }
2805
2806 // Rationale behind this function:
2807 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2808 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2809 // samples for JITted code. Here we create private executable mapping over the code cache
2810 // and then we can use standard (well, almost, as mapping can change) way to provide
2811 // info for the reporting script by storing timestamp and location of symbol
2812 void linux_wrap_code(char* base, size_t size) {
2813 static volatile jint cnt = 0;
2814
2815 if (!UseOprofile) {
2816 return;
2817 }
2818
2819 char buf[PATH_MAX+1];
2820 int num = Atomic::add(1, &cnt);
2821
2822 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2823 os::get_temp_directory(), os::current_process_id(), num);
2824 unlink(buf);
2825
2826 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2827
2828 if (fd != -1) {
2829 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2830 if (rv != (off_t)-1) {
2831 if (::write(fd, "", 1) == 1) {
2832 mmap(base, size,
2833 PROT_READ|PROT_WRITE|PROT_EXEC,
2834 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2835 }
2836 }
2837 ::close(fd);
2838 unlink(buf);
2839 }
2840 }
2841
2842 static bool recoverable_mmap_error(int err) {
2843 // See if the error is one we can let the caller handle. This
2844 // list of errno values comes from JBS-6843484. I can't find a
2845 // Linux man page that documents this specific set of errno
2846 // values so while this list currently matches Solaris, it may
2847 // change as we gain experience with this failure mode.
2848 switch (err) {
2849 case EBADF:
2850 case EINVAL:
2851 case ENOTSUP:
2852 // let the caller deal with these errors
2853 return true;
2854
2855 default:
2856 // Any remaining errors on this OS can cause our reserved mapping
2857 // to be lost. That can cause confusion where different data
2858 // structures think they have the same memory mapped. The worst
2859 // scenario is if both the VM and a library think they have the
2860 // same memory mapped.
2861 return false;
2862 }
2863 }
2864
2865 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2866 int err) {
2867 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2868 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2869 os::strerror(err), err);
2870 }
2871
2872 static void warn_fail_commit_memory(char* addr, size_t size,
2873 size_t alignment_hint, bool exec,
2874 int err) {
2875 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2876 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2877 alignment_hint, exec, os::strerror(err), err);
2878 }
2879
2880 // NOTE: Linux kernel does not really reserve the pages for us.
2881 // All it does is to check if there are enough free pages
2882 // left at the time of mmap(). This could be a potential
2883 // problem.
2884 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2885 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2886 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2887 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2888 if (res != (uintptr_t) MAP_FAILED) {
2889 if (UseNUMAInterleaving) {
2890 numa_make_global(addr, size);
2891 }
2892 return 0;
2893 }
2894
2895 int err = errno; // save errno from mmap() call above
2896
2897 if (!recoverable_mmap_error(err)) {
2898 warn_fail_commit_memory(addr, size, exec, err);
2899 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2900 }
2901
2902 return err;
2903 }
2904
2905 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2906 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2907 }
2908
2909 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2910 const char* mesg) {
2911 assert(mesg != NULL, "mesg must be specified");
2912 int err = os::Linux::commit_memory_impl(addr, size, exec);
2913 if (err != 0) {
2914 // the caller wants all commit errors to exit with the specified mesg:
2915 warn_fail_commit_memory(addr, size, exec, err);
2916 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2917 }
2918 }
2919
2920 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2921 #ifndef MAP_HUGETLB
2922 #define MAP_HUGETLB 0x40000
2923 #endif
2924
2925 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2926 #ifndef MADV_HUGEPAGE
2927 #define MADV_HUGEPAGE 14
2928 #endif
2929
2930 int os::Linux::commit_memory_impl(char* addr, size_t size,
2931 size_t alignment_hint, bool exec) {
2932 int err = os::Linux::commit_memory_impl(addr, size, exec);
2933 if (err == 0) {
2934 realign_memory(addr, size, alignment_hint);
2935 }
2936 return err;
2937 }
2938
2939 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2940 bool exec) {
2941 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2942 }
2943
2944 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2945 size_t alignment_hint, bool exec,
2946 const char* mesg) {
2947 assert(mesg != NULL, "mesg must be specified");
2948 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2949 if (err != 0) {
2950 // the caller wants all commit errors to exit with the specified mesg:
2951 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2952 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2953 }
2954 }
2955
2956 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2957 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2958 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2959 // be supported or the memory may already be backed by huge pages.
2960 ::madvise(addr, bytes, MADV_HUGEPAGE);
2961 }
2962 }
2963
2964 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2965 // This method works by doing an mmap over an existing mmaping and effectively discarding
2966 // the existing pages. However it won't work for SHM-based large pages that cannot be
2967 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2968 // small pages on top of the SHM segment. This method always works for small pages, so we
2969 // allow that in any case.
2970 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2971 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2972 }
2973 }
2974
2975 void os::numa_make_global(char *addr, size_t bytes) {
2976 Linux::numa_interleave_memory(addr, bytes);
2977 }
2978
2979 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2980 // bind policy to MPOL_PREFERRED for the current thread.
2981 #define USE_MPOL_PREFERRED 0
2982
2983 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2984 // To make NUMA and large pages more robust when both enabled, we need to ease
2985 // the requirements on where the memory should be allocated. MPOL_BIND is the
2986 // default policy and it will force memory to be allocated on the specified
2987 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2988 // the specified node, but will not force it. Using this policy will prevent
2989 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2990 // free large pages.
2991 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2992 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2993 }
2994
2995 bool os::numa_topology_changed() { return false; }
2996
2997 size_t os::numa_get_groups_num() {
2998 // Return just the number of nodes in which it's possible to allocate memory
2999 // (in numa terminology, configured nodes).
3000 return Linux::numa_num_configured_nodes();
3001 }
3002
3003 int os::numa_get_group_id() {
3004 int cpu_id = Linux::sched_getcpu();
3005 if (cpu_id != -1) {
3006 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
3007 if (lgrp_id != -1) {
3008 return lgrp_id;
3009 }
3010 }
3011 return 0;
3012 }
3013
3014 int os::Linux::get_existing_num_nodes() {
3015 int node;
3016 int highest_node_number = Linux::numa_max_node();
3017 int num_nodes = 0;
3018
3019 // Get the total number of nodes in the system including nodes without memory.
3020 for (node = 0; node <= highest_node_number; node++) {
3021 if (is_node_in_existing_nodes(node)) {
3022 num_nodes++;
3023 }
3024 }
3025 return num_nodes;
3026 }
3027
3028 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
3029 int highest_node_number = Linux::numa_max_node();
3030 size_t i = 0;
3031
3032 // Map all node ids in which it is possible to allocate memory. Also nodes are
3033 // not always consecutively available, i.e. available from 0 to the highest
3034 // node number. If the nodes have been bound explicitly using numactl membind,
3035 // then allocate memory from those nodes only.
3036 for (int node = 0; node <= highest_node_number; node++) {
3037 if (Linux::is_node_in_bound_nodes((unsigned int)node)) {
3038 ids[i++] = node;
3039 }
3040 }
3041 return i;
3042 }
3043
3044 bool os::get_page_info(char *start, page_info* info) {
3045 return false;
3046 }
3047
3048 char *os::scan_pages(char *start, char* end, page_info* page_expected,
3049 page_info* page_found) {
3050 return end;
3051 }
3052
3053
3054 int os::Linux::sched_getcpu_syscall(void) {
3055 unsigned int cpu = 0;
3056 int retval = -1;
3057
3058 #if defined(IA32)
3059 #ifndef SYS_getcpu
3060 #define SYS_getcpu 318
3061 #endif
3062 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
3063 #elif defined(AMD64)
3064 // Unfortunately we have to bring all these macros here from vsyscall.h
3065 // to be able to compile on old linuxes.
3066 #define __NR_vgetcpu 2
3067 #define VSYSCALL_START (-10UL << 20)
3068 #define VSYSCALL_SIZE 1024
3069 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
3070 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
3071 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
3072 retval = vgetcpu(&cpu, NULL, NULL);
3073 #endif
3074
3075 return (retval == -1) ? retval : cpu;
3076 }
3077
3078 void os::Linux::sched_getcpu_init() {
3079 // sched_getcpu() should be in libc.
3080 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
3081 dlsym(RTLD_DEFAULT, "sched_getcpu")));
3082
3083 // If it's not, try a direct syscall.
3084 if (sched_getcpu() == -1) {
3085 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
3086 (void*)&sched_getcpu_syscall));
3087 }
3088
3089 if (sched_getcpu() == -1) {
3090 vm_exit_during_initialization("getcpu(2) system call not supported by kernel");
3091 }
3092 }
3093
3094 // Something to do with the numa-aware allocator needs these symbols
3095 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
3096 extern "C" JNIEXPORT void numa_error(char *where) { }
3097
3098 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
3099 // load symbol from base version instead.
3100 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
3101 void *f = dlvsym(handle, name, "libnuma_1.1");
3102 if (f == NULL) {
3103 f = dlsym(handle, name);
3104 }
3105 return f;
3106 }
3107
3108 // Handle request to load libnuma symbol version 1.2 (API v2) only.
3109 // Return NULL if the symbol is not defined in this particular version.
3110 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
3111 return dlvsym(handle, name, "libnuma_1.2");
3112 }
3113
3114 bool os::Linux::libnuma_init() {
3115 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
3116 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
3117 if (handle != NULL) {
3118 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
3119 libnuma_dlsym(handle, "numa_node_to_cpus")));
3120 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
3121 libnuma_dlsym(handle, "numa_max_node")));
3122 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
3123 libnuma_dlsym(handle, "numa_num_configured_nodes")));
3124 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
3125 libnuma_dlsym(handle, "numa_available")));
3126 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
3127 libnuma_dlsym(handle, "numa_tonode_memory")));
3128 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
3129 libnuma_dlsym(handle, "numa_interleave_memory")));
3130 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
3131 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
3132 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
3133 libnuma_dlsym(handle, "numa_set_bind_policy")));
3134 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
3135 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
3136 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
3137 libnuma_dlsym(handle, "numa_distance")));
3138 set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t,
3139 libnuma_v2_dlsym(handle, "numa_get_membind")));
3140 set_numa_get_interleave_mask(CAST_TO_FN_PTR(numa_get_interleave_mask_func_t,
3141 libnuma_v2_dlsym(handle, "numa_get_interleave_mask")));
3142
3143 if (numa_available() != -1) {
3144 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
3145 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
3146 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
3147 set_numa_interleave_bitmask(_numa_get_interleave_mask());
3148 set_numa_membind_bitmask(_numa_get_membind());
3149 // Create an index -> node mapping, since nodes are not always consecutive
3150 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3151 rebuild_nindex_to_node_map();
3152 // Create a cpu -> node mapping
3153 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3154 rebuild_cpu_to_node_map();
3155 return true;
3156 }
3157 }
3158 }
3159 return false;
3160 }
3161
3162 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
3163 // Creating guard page is very expensive. Java thread has HotSpot
3164 // guard pages, only enable glibc guard page for non-Java threads.
3165 // (Remember: compiler thread is a Java thread, too!)
3166 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
3167 }
3168
3169 void os::Linux::rebuild_nindex_to_node_map() {
3170 int highest_node_number = Linux::numa_max_node();
3171
3172 nindex_to_node()->clear();
3173 for (int node = 0; node <= highest_node_number; node++) {
3174 if (Linux::is_node_in_existing_nodes(node)) {
3175 nindex_to_node()->append(node);
3176 }
3177 }
3178 }
3179
3180 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
3181 // The table is later used in get_node_by_cpu().
3182 void os::Linux::rebuild_cpu_to_node_map() {
3183 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
3184 // in libnuma (possible values are starting from 16,
3185 // and continuing up with every other power of 2, but less
3186 // than the maximum number of CPUs supported by kernel), and
3187 // is a subject to change (in libnuma version 2 the requirements
3188 // are more reasonable) we'll just hardcode the number they use
3189 // in the library.
3190 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
3191
3192 size_t cpu_num = processor_count();
3193 size_t cpu_map_size = NCPUS / BitsPerCLong;
3194 size_t cpu_map_valid_size =
3195 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
3196
3197 cpu_to_node()->clear();
3198 cpu_to_node()->at_grow(cpu_num - 1);
3199
3200 size_t node_num = get_existing_num_nodes();
3201
3202 int distance = 0;
3203 int closest_distance = INT_MAX;
3204 int closest_node = 0;
3205 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
3206 for (size_t i = 0; i < node_num; i++) {
3207 // Check if node is configured (not a memory-less node). If it is not, find
3208 // the closest configured node. Check also if node is bound, i.e. it's allowed
3209 // to allocate memory from the node. If it's not allowed, map cpus in that node
3210 // to the closest node from which memory allocation is allowed.
3211 if (!is_node_in_configured_nodes(nindex_to_node()->at(i)) ||
3212 !is_node_in_bound_nodes(nindex_to_node()->at(i))) {
3213 closest_distance = INT_MAX;
3214 // Check distance from all remaining nodes in the system. Ignore distance
3215 // from itself, from another non-configured node, and from another non-bound
3216 // node.
3217 for (size_t m = 0; m < node_num; m++) {
3218 if (m != i &&
3219 is_node_in_configured_nodes(nindex_to_node()->at(m)) &&
3220 is_node_in_bound_nodes(nindex_to_node()->at(m))) {
3221 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3222 // If a closest node is found, update. There is always at least one
3223 // configured and bound node in the system so there is always at least
3224 // one node close.
3225 if (distance != 0 && distance < closest_distance) {
3226 closest_distance = distance;
3227 closest_node = nindex_to_node()->at(m);
3228 }
3229 }
3230 }
3231 } else {
3232 // Current node is already a configured node.
3233 closest_node = nindex_to_node()->at(i);
3234 }
3235
3236 // Get cpus from the original node and map them to the closest node. If node
3237 // is a configured node (not a memory-less node), then original node and
3238 // closest node are the same.
3239 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3240 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3241 if (cpu_map[j] != 0) {
3242 for (size_t k = 0; k < BitsPerCLong; k++) {
3243 if (cpu_map[j] & (1UL << k)) {
3244 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3245 }
3246 }
3247 }
3248 }
3249 }
3250 }
3251 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3252 }
3253
3254 int os::Linux::get_node_by_cpu(int cpu_id) {
3255 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3256 return cpu_to_node()->at(cpu_id);
3257 }
3258 return -1;
3259 }
3260
3261 GrowableArray<int>* os::Linux::_cpu_to_node;
3262 GrowableArray<int>* os::Linux::_nindex_to_node;
3263 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3264 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3265 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3266 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3267 os::Linux::numa_available_func_t os::Linux::_numa_available;
3268 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3269 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3270 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3271 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3272 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3273 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3274 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind;
3275 os::Linux::numa_get_interleave_mask_func_t os::Linux::_numa_get_interleave_mask;
3276 os::Linux::NumaAllocationPolicy os::Linux::_current_numa_policy;
3277 unsigned long* os::Linux::_numa_all_nodes;
3278 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3279 struct bitmask* os::Linux::_numa_nodes_ptr;
3280 struct bitmask* os::Linux::_numa_interleave_bitmask;
3281 struct bitmask* os::Linux::_numa_membind_bitmask;
3282
3283 bool os::pd_uncommit_memory(char* addr, size_t size) {
3284 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3285 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3286 return res != (uintptr_t) MAP_FAILED;
3287 }
3288
3289 static address get_stack_commited_bottom(address bottom, size_t size) {
3290 address nbot = bottom;
3291 address ntop = bottom + size;
3292
3293 size_t page_sz = os::vm_page_size();
3294 unsigned pages = size / page_sz;
3295
3296 unsigned char vec[1];
3297 unsigned imin = 1, imax = pages + 1, imid;
3298 int mincore_return_value = 0;
3299
3300 assert(imin <= imax, "Unexpected page size");
3301
3302 while (imin < imax) {
3303 imid = (imax + imin) / 2;
3304 nbot = ntop - (imid * page_sz);
3305
3306 // Use a trick with mincore to check whether the page is mapped or not.
3307 // mincore sets vec to 1 if page resides in memory and to 0 if page
3308 // is swapped output but if page we are asking for is unmapped
3309 // it returns -1,ENOMEM
3310 mincore_return_value = mincore(nbot, page_sz, vec);
3311
3312 if (mincore_return_value == -1) {
3313 // Page is not mapped go up
3314 // to find first mapped page
3315 if (errno != EAGAIN) {
3316 assert(errno == ENOMEM, "Unexpected mincore errno");
3317 imax = imid;
3318 }
3319 } else {
3320 // Page is mapped go down
3321 // to find first not mapped page
3322 imin = imid + 1;
3323 }
3324 }
3325
3326 nbot = nbot + page_sz;
3327
3328 // Adjust stack bottom one page up if last checked page is not mapped
3329 if (mincore_return_value == -1) {
3330 nbot = nbot + page_sz;
3331 }
3332
3333 return nbot;
3334 }
3335
3336 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3337 int mincore_return_value;
3338 const size_t stripe = 1024; // query this many pages each time
3339 unsigned char vec[stripe + 1];
3340 // set a guard
3341 vec[stripe] = 'X';
3342
3343 const size_t page_sz = os::vm_page_size();
3344 size_t pages = size / page_sz;
3345
3346 assert(is_aligned(start, page_sz), "Start address must be page aligned");
3347 assert(is_aligned(size, page_sz), "Size must be page aligned");
3348
3349 committed_start = NULL;
3350
3351 int loops = (pages + stripe - 1) / stripe;
3352 int committed_pages = 0;
3353 address loop_base = start;
3354 bool found_range = false;
3355
3356 for (int index = 0; index < loops && !found_range; index ++) {
3357 assert(pages > 0, "Nothing to do");
3358 int pages_to_query = (pages >= stripe) ? stripe : pages;
3359 pages -= pages_to_query;
3360
3361 // Get stable read
3362 while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3363
3364 // During shutdown, some memory goes away without properly notifying NMT,
3365 // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3366 // Bailout and return as not committed for now.
3367 if (mincore_return_value == -1 && errno == ENOMEM) {
3368 return false;
3369 }
3370
3371 assert(vec[stripe] == 'X', "overflow guard");
3372 assert(mincore_return_value == 0, "Range must be valid");
3373 // Process this stripe
3374 for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3375 if ((vec[vecIdx] & 0x01) == 0) { // not committed
3376 // End of current contiguous region
3377 if (committed_start != NULL) {
3378 found_range = true;
3379 break;
3380 }
3381 } else { // committed
3382 // Start of region
3383 if (committed_start == NULL) {
3384 committed_start = loop_base + page_sz * vecIdx;
3385 }
3386 committed_pages ++;
3387 }
3388 }
3389
3390 loop_base += pages_to_query * page_sz;
3391 }
3392
3393 if (committed_start != NULL) {
3394 assert(committed_pages > 0, "Must have committed region");
3395 assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3396 assert(committed_start >= start && committed_start < start + size, "Out of range");
3397 committed_size = page_sz * committed_pages;
3398 return true;
3399 } else {
3400 assert(committed_pages == 0, "Should not have committed region");
3401 return false;
3402 }
3403 }
3404
3405
3406 // Linux uses a growable mapping for the stack, and if the mapping for
3407 // the stack guard pages is not removed when we detach a thread the
3408 // stack cannot grow beyond the pages where the stack guard was
3409 // mapped. If at some point later in the process the stack expands to
3410 // that point, the Linux kernel cannot expand the stack any further
3411 // because the guard pages are in the way, and a segfault occurs.
3412 //
3413 // However, it's essential not to split the stack region by unmapping
3414 // a region (leaving a hole) that's already part of the stack mapping,
3415 // so if the stack mapping has already grown beyond the guard pages at
3416 // the time we create them, we have to truncate the stack mapping.
3417 // So, we need to know the extent of the stack mapping when
3418 // create_stack_guard_pages() is called.
3419
3420 // We only need this for stacks that are growable: at the time of
3421 // writing thread stacks don't use growable mappings (i.e. those
3422 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3423 // only applies to the main thread.
3424
3425 // If the (growable) stack mapping already extends beyond the point
3426 // where we're going to put our guard pages, truncate the mapping at
3427 // that point by munmap()ping it. This ensures that when we later
3428 // munmap() the guard pages we don't leave a hole in the stack
3429 // mapping. This only affects the main/primordial thread
3430
3431 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3432 if (os::is_primordial_thread()) {
3433 // As we manually grow stack up to bottom inside create_attached_thread(),
3434 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3435 // we don't need to do anything special.
3436 // Check it first, before calling heavy function.
3437 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3438 unsigned char vec[1];
3439
3440 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3441 // Fallback to slow path on all errors, including EAGAIN
3442 stack_extent = (uintptr_t) get_stack_commited_bottom(
3443 os::Linux::initial_thread_stack_bottom(),
3444 (size_t)addr - stack_extent);
3445 }
3446
3447 if (stack_extent < (uintptr_t)addr) {
3448 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3449 }
3450 }
3451
3452 return os::commit_memory(addr, size, !ExecMem);
3453 }
3454
3455 // If this is a growable mapping, remove the guard pages entirely by
3456 // munmap()ping them. If not, just call uncommit_memory(). This only
3457 // affects the main/primordial thread, but guard against future OS changes.
3458 // It's safe to always unmap guard pages for primordial thread because we
3459 // always place it right after end of the mapped region.
3460
3461 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3462 uintptr_t stack_extent, stack_base;
3463
3464 if (os::is_primordial_thread()) {
3465 return ::munmap(addr, size) == 0;
3466 }
3467
3468 return os::uncommit_memory(addr, size);
3469 }
3470
3471 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3472 // at 'requested_addr'. If there are existing memory mappings at the same
3473 // location, however, they will be overwritten. If 'fixed' is false,
3474 // 'requested_addr' is only treated as a hint, the return value may or
3475 // may not start from the requested address. Unlike Linux mmap(), this
3476 // function returns NULL to indicate failure.
3477 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3478 char * addr;
3479 int flags;
3480
3481 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3482 if (fixed) {
3483 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3484 flags |= MAP_FIXED;
3485 }
3486
3487 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3488 // touch an uncommitted page. Otherwise, the read/write might
3489 // succeed if we have enough swap space to back the physical page.
3490 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3491 flags, -1, 0);
3492
3493 return addr == MAP_FAILED ? NULL : addr;
3494 }
3495
3496 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3497 // (req_addr != NULL) or with a given alignment.
3498 // - bytes shall be a multiple of alignment.
3499 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3500 // - alignment sets the alignment at which memory shall be allocated.
3501 // It must be a multiple of allocation granularity.
3502 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3503 // req_addr or NULL.
3504 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3505
3506 size_t extra_size = bytes;
3507 if (req_addr == NULL && alignment > 0) {
3508 extra_size += alignment;
3509 }
3510
3511 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3512 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3513 -1, 0);
3514 if (start == MAP_FAILED) {
3515 start = NULL;
3516 } else {
3517 if (req_addr != NULL) {
3518 if (start != req_addr) {
3519 ::munmap(start, extra_size);
3520 start = NULL;
3521 }
3522 } else {
3523 char* const start_aligned = align_up(start, alignment);
3524 char* const end_aligned = start_aligned + bytes;
3525 char* const end = start + extra_size;
3526 if (start_aligned > start) {
3527 ::munmap(start, start_aligned - start);
3528 }
3529 if (end_aligned < end) {
3530 ::munmap(end_aligned, end - end_aligned);
3531 }
3532 start = start_aligned;
3533 }
3534 }
3535 return start;
3536 }
3537
3538 static int anon_munmap(char * addr, size_t size) {
3539 return ::munmap(addr, size) == 0;
3540 }
3541
3542 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3543 size_t alignment_hint) {
3544 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3545 }
3546
3547 bool os::pd_release_memory(char* addr, size_t size) {
3548 return anon_munmap(addr, size);
3549 }
3550
3551 static bool linux_mprotect(char* addr, size_t size, int prot) {
3552 // Linux wants the mprotect address argument to be page aligned.
3553 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3554
3555 // According to SUSv3, mprotect() should only be used with mappings
3556 // established by mmap(), and mmap() always maps whole pages. Unaligned
3557 // 'addr' likely indicates problem in the VM (e.g. trying to change
3558 // protection of malloc'ed or statically allocated memory). Check the
3559 // caller if you hit this assert.
3560 assert(addr == bottom, "sanity check");
3561
3562 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3563 Events::log(NULL, "Protecting memory [" INTPTR_FORMAT "," INTPTR_FORMAT "] with protection modes %x", p2i(bottom), p2i(bottom+size), prot);
3564 return ::mprotect(bottom, size, prot) == 0;
3565 }
3566
3567 // Set protections specified
3568 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3569 bool is_committed) {
3570 unsigned int p = 0;
3571 switch (prot) {
3572 case MEM_PROT_NONE: p = PROT_NONE; break;
3573 case MEM_PROT_READ: p = PROT_READ; break;
3574 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3575 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3576 default:
3577 ShouldNotReachHere();
3578 }
3579 // is_committed is unused.
3580 return linux_mprotect(addr, bytes, p);
3581 }
3582
3583 bool os::guard_memory(char* addr, size_t size) {
3584 return linux_mprotect(addr, size, PROT_NONE);
3585 }
3586
3587 bool os::unguard_memory(char* addr, size_t size) {
3588 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3589 }
3590
3591 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3592 size_t page_size) {
3593 bool result = false;
3594 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3595 MAP_ANONYMOUS|MAP_PRIVATE,
3596 -1, 0);
3597 if (p != MAP_FAILED) {
3598 void *aligned_p = align_up(p, page_size);
3599
3600 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3601
3602 munmap(p, page_size * 2);
3603 }
3604
3605 if (warn && !result) {
3606 warning("TransparentHugePages is not supported by the operating system.");
3607 }
3608
3609 return result;
3610 }
3611
3612 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3613 bool result = false;
3614 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3615 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3616 -1, 0);
3617
3618 if (p != MAP_FAILED) {
3619 // We don't know if this really is a huge page or not.
3620 FILE *fp = fopen("/proc/self/maps", "r");
3621 if (fp) {
3622 while (!feof(fp)) {
3623 char chars[257];
3624 long x = 0;
3625 if (fgets(chars, sizeof(chars), fp)) {
3626 if (sscanf(chars, "%lx-%*x", &x) == 1
3627 && x == (long)p) {
3628 if (strstr (chars, "hugepage")) {
3629 result = true;
3630 break;
3631 }
3632 }
3633 }
3634 }
3635 fclose(fp);
3636 }
3637 munmap(p, page_size);
3638 }
3639
3640 if (warn && !result) {
3641 warning("HugeTLBFS is not supported by the operating system.");
3642 }
3643
3644 return result;
3645 }
3646
3647 // From the coredump_filter documentation:
3648 //
3649 // - (bit 0) anonymous private memory
3650 // - (bit 1) anonymous shared memory
3651 // - (bit 2) file-backed private memory
3652 // - (bit 3) file-backed shared memory
3653 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3654 // effective only if the bit 2 is cleared)
3655 // - (bit 5) hugetlb private memory
3656 // - (bit 6) hugetlb shared memory
3657 // - (bit 7) dax private memory
3658 // - (bit 8) dax shared memory
3659 //
3660 static void set_coredump_filter(CoredumpFilterBit bit) {
3661 FILE *f;
3662 long cdm;
3663
3664 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3665 return;
3666 }
3667
3668 if (fscanf(f, "%lx", &cdm) != 1) {
3669 fclose(f);
3670 return;
3671 }
3672
3673 long saved_cdm = cdm;
3674 rewind(f);
3675 cdm |= bit;
3676
3677 if (cdm != saved_cdm) {
3678 fprintf(f, "%#lx", cdm);
3679 }
3680
3681 fclose(f);
3682 }
3683
3684 // Large page support
3685
3686 static size_t _large_page_size = 0;
3687
3688 size_t os::Linux::find_large_page_size() {
3689 size_t large_page_size = 0;
3690
3691 // large_page_size on Linux is used to round up heap size. x86 uses either
3692 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3693 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3694 // page as large as 256M.
3695 //
3696 // Here we try to figure out page size by parsing /proc/meminfo and looking
3697 // for a line with the following format:
3698 // Hugepagesize: 2048 kB
3699 //
3700 // If we can't determine the value (e.g. /proc is not mounted, or the text
3701 // format has been changed), we'll use the largest page size supported by
3702 // the processor.
3703
3704 #ifndef ZERO
3705 large_page_size =
3706 AARCH64_ONLY(2 * M)
3707 AMD64_ONLY(2 * M)
3708 ARM32_ONLY(2 * M)
3709 IA32_ONLY(4 * M)
3710 IA64_ONLY(256 * M)
3711 PPC_ONLY(4 * M)
3712 S390_ONLY(1 * M)
3713 SPARC_ONLY(4 * M);
3714 #endif // ZERO
3715
3716 FILE *fp = fopen("/proc/meminfo", "r");
3717 if (fp) {
3718 while (!feof(fp)) {
3719 int x = 0;
3720 char buf[16];
3721 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3722 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3723 large_page_size = x * K;
3724 break;
3725 }
3726 } else {
3727 // skip to next line
3728 for (;;) {
3729 int ch = fgetc(fp);
3730 if (ch == EOF || ch == (int)'\n') break;
3731 }
3732 }
3733 }
3734 fclose(fp);
3735 }
3736
3737 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3738 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3739 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3740 proper_unit_for_byte_size(large_page_size));
3741 }
3742
3743 return large_page_size;
3744 }
3745
3746 size_t os::Linux::setup_large_page_size() {
3747 _large_page_size = Linux::find_large_page_size();
3748 const size_t default_page_size = (size_t)Linux::page_size();
3749 if (_large_page_size > default_page_size) {
3750 _page_sizes[0] = _large_page_size;
3751 _page_sizes[1] = default_page_size;
3752 _page_sizes[2] = 0;
3753 }
3754
3755 return _large_page_size;
3756 }
3757
3758 bool os::Linux::setup_large_page_type(size_t page_size) {
3759 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3760 FLAG_IS_DEFAULT(UseSHM) &&
3761 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3762
3763 // The type of large pages has not been specified by the user.
3764
3765 // Try UseHugeTLBFS and then UseSHM.
3766 UseHugeTLBFS = UseSHM = true;
3767
3768 // Don't try UseTransparentHugePages since there are known
3769 // performance issues with it turned on. This might change in the future.
3770 UseTransparentHugePages = false;
3771 }
3772
3773 if (UseTransparentHugePages) {
3774 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3775 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3776 UseHugeTLBFS = false;
3777 UseSHM = false;
3778 return true;
3779 }
3780 UseTransparentHugePages = false;
3781 }
3782
3783 if (UseHugeTLBFS) {
3784 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3785 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3786 UseSHM = false;
3787 return true;
3788 }
3789 UseHugeTLBFS = false;
3790 }
3791
3792 return UseSHM;
3793 }
3794
3795 void os::large_page_init() {
3796 if (!UseLargePages &&
3797 !UseTransparentHugePages &&
3798 !UseHugeTLBFS &&
3799 !UseSHM) {
3800 // Not using large pages.
3801 return;
3802 }
3803
3804 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3805 // The user explicitly turned off large pages.
3806 // Ignore the rest of the large pages flags.
3807 UseTransparentHugePages = false;
3808 UseHugeTLBFS = false;
3809 UseSHM = false;
3810 return;
3811 }
3812
3813 size_t large_page_size = Linux::setup_large_page_size();
3814 UseLargePages = Linux::setup_large_page_type(large_page_size);
3815
3816 set_coredump_filter(LARGEPAGES_BIT);
3817 }
3818
3819 #ifndef SHM_HUGETLB
3820 #define SHM_HUGETLB 04000
3821 #endif
3822
3823 #define shm_warning_format(format, ...) \
3824 do { \
3825 if (UseLargePages && \
3826 (!FLAG_IS_DEFAULT(UseLargePages) || \
3827 !FLAG_IS_DEFAULT(UseSHM) || \
3828 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3829 warning(format, __VA_ARGS__); \
3830 } \
3831 } while (0)
3832
3833 #define shm_warning(str) shm_warning_format("%s", str)
3834
3835 #define shm_warning_with_errno(str) \
3836 do { \
3837 int err = errno; \
3838 shm_warning_format(str " (error = %d)", err); \
3839 } while (0)
3840
3841 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3842 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3843
3844 if (!is_aligned(alignment, SHMLBA)) {
3845 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3846 return NULL;
3847 }
3848
3849 // To ensure that we get 'alignment' aligned memory from shmat,
3850 // we pre-reserve aligned virtual memory and then attach to that.
3851
3852 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3853 if (pre_reserved_addr == NULL) {
3854 // Couldn't pre-reserve aligned memory.
3855 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3856 return NULL;
3857 }
3858
3859 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3860 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3861
3862 if ((intptr_t)addr == -1) {
3863 int err = errno;
3864 shm_warning_with_errno("Failed to attach shared memory.");
3865
3866 assert(err != EACCES, "Unexpected error");
3867 assert(err != EIDRM, "Unexpected error");
3868 assert(err != EINVAL, "Unexpected error");
3869
3870 // Since we don't know if the kernel unmapped the pre-reserved memory area
3871 // we can't unmap it, since that would potentially unmap memory that was
3872 // mapped from other threads.
3873 return NULL;
3874 }
3875
3876 return addr;
3877 }
3878
3879 static char* shmat_at_address(int shmid, char* req_addr) {
3880 if (!is_aligned(req_addr, SHMLBA)) {
3881 assert(false, "Requested address needs to be SHMLBA aligned");
3882 return NULL;
3883 }
3884
3885 char* addr = (char*)shmat(shmid, req_addr, 0);
3886
3887 if ((intptr_t)addr == -1) {
3888 shm_warning_with_errno("Failed to attach shared memory.");
3889 return NULL;
3890 }
3891
3892 return addr;
3893 }
3894
3895 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3896 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3897 if (req_addr != NULL) {
3898 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3899 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3900 return shmat_at_address(shmid, req_addr);
3901 }
3902
3903 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3904 // return large page size aligned memory addresses when req_addr == NULL.
3905 // However, if the alignment is larger than the large page size, we have
3906 // to manually ensure that the memory returned is 'alignment' aligned.
3907 if (alignment > os::large_page_size()) {
3908 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3909 return shmat_with_alignment(shmid, bytes, alignment);
3910 } else {
3911 return shmat_at_address(shmid, NULL);
3912 }
3913 }
3914
3915 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3916 char* req_addr, bool exec) {
3917 // "exec" is passed in but not used. Creating the shared image for
3918 // the code cache doesn't have an SHM_X executable permission to check.
3919 assert(UseLargePages && UseSHM, "only for SHM large pages");
3920 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3921 assert(is_aligned(req_addr, alignment), "Unaligned address");
3922
3923 if (!is_aligned(bytes, os::large_page_size())) {
3924 return NULL; // Fallback to small pages.
3925 }
3926
3927 // Create a large shared memory region to attach to based on size.
3928 // Currently, size is the total size of the heap.
3929 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3930 if (shmid == -1) {
3931 // Possible reasons for shmget failure:
3932 // 1. shmmax is too small for Java heap.
3933 // > check shmmax value: cat /proc/sys/kernel/shmmax
3934 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3935 // 2. not enough large page memory.
3936 // > check available large pages: cat /proc/meminfo
3937 // > increase amount of large pages:
3938 // echo new_value > /proc/sys/vm/nr_hugepages
3939 // Note 1: different Linux may use different name for this property,
3940 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3941 // Note 2: it's possible there's enough physical memory available but
3942 // they are so fragmented after a long run that they can't
3943 // coalesce into large pages. Try to reserve large pages when
3944 // the system is still "fresh".
3945 shm_warning_with_errno("Failed to reserve shared memory.");
3946 return NULL;
3947 }
3948
3949 // Attach to the region.
3950 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3951
3952 // Remove shmid. If shmat() is successful, the actual shared memory segment
3953 // will be deleted when it's detached by shmdt() or when the process
3954 // terminates. If shmat() is not successful this will remove the shared
3955 // segment immediately.
3956 shmctl(shmid, IPC_RMID, NULL);
3957
3958 return addr;
3959 }
3960
3961 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3962 int error) {
3963 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3964
3965 bool warn_on_failure = UseLargePages &&
3966 (!FLAG_IS_DEFAULT(UseLargePages) ||
3967 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3968 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3969
3970 if (warn_on_failure) {
3971 char msg[128];
3972 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3973 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3974 warning("%s", msg);
3975 }
3976 }
3977
3978 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3979 char* req_addr,
3980 bool exec) {
3981 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3982 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3983 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3984
3985 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3986 char* addr = (char*)::mmap(req_addr, bytes, prot,
3987 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3988 -1, 0);
3989
3990 if (addr == MAP_FAILED) {
3991 warn_on_large_pages_failure(req_addr, bytes, errno);
3992 return NULL;
3993 }
3994
3995 assert(is_aligned(addr, os::large_page_size()), "Must be");
3996
3997 return addr;
3998 }
3999
4000 // Reserve memory using mmap(MAP_HUGETLB).
4001 // - bytes shall be a multiple of alignment.
4002 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
4003 // - alignment sets the alignment at which memory shall be allocated.
4004 // It must be a multiple of allocation granularity.
4005 // Returns address of memory or NULL. If req_addr was not NULL, will only return
4006 // req_addr or NULL.
4007 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
4008 size_t alignment,
4009 char* req_addr,
4010 bool exec) {
4011 size_t large_page_size = os::large_page_size();
4012 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
4013
4014 assert(is_aligned(req_addr, alignment), "Must be");
4015 assert(is_aligned(bytes, alignment), "Must be");
4016
4017 // First reserve - but not commit - the address range in small pages.
4018 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
4019
4020 if (start == NULL) {
4021 return NULL;
4022 }
4023
4024 assert(is_aligned(start, alignment), "Must be");
4025
4026 char* end = start + bytes;
4027
4028 // Find the regions of the allocated chunk that can be promoted to large pages.
4029 char* lp_start = align_up(start, large_page_size);
4030 char* lp_end = align_down(end, large_page_size);
4031
4032 size_t lp_bytes = lp_end - lp_start;
4033
4034 assert(is_aligned(lp_bytes, large_page_size), "Must be");
4035
4036 if (lp_bytes == 0) {
4037 // The mapped region doesn't even span the start and the end of a large page.
4038 // Fall back to allocate a non-special area.
4039 ::munmap(start, end - start);
4040 return NULL;
4041 }
4042
4043 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
4044
4045 void* result;
4046
4047 // Commit small-paged leading area.
4048 if (start != lp_start) {
4049 result = ::mmap(start, lp_start - start, prot,
4050 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
4051 -1, 0);
4052 if (result == MAP_FAILED) {
4053 ::munmap(lp_start, end - lp_start);
4054 return NULL;
4055 }
4056 }
4057
4058 // Commit large-paged area.
4059 result = ::mmap(lp_start, lp_bytes, prot,
4060 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
4061 -1, 0);
4062 if (result == MAP_FAILED) {
4063 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
4064 // If the mmap above fails, the large pages region will be unmapped and we
4065 // have regions before and after with small pages. Release these regions.
4066 //
4067 // | mapped | unmapped | mapped |
4068 // ^ ^ ^ ^
4069 // start lp_start lp_end end
4070 //
4071 ::munmap(start, lp_start - start);
4072 ::munmap(lp_end, end - lp_end);
4073 return NULL;
4074 }
4075
4076 // Commit small-paged trailing area.
4077 if (lp_end != end) {
4078 result = ::mmap(lp_end, end - lp_end, prot,
4079 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
4080 -1, 0);
4081 if (result == MAP_FAILED) {
4082 ::munmap(start, lp_end - start);
4083 return NULL;
4084 }
4085 }
4086
4087 return start;
4088 }
4089
4090 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
4091 size_t alignment,
4092 char* req_addr,
4093 bool exec) {
4094 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
4095 assert(is_aligned(req_addr, alignment), "Must be");
4096 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
4097 assert(is_power_of_2(os::large_page_size()), "Must be");
4098 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
4099
4100 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
4101 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
4102 } else {
4103 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
4104 }
4105 }
4106
4107 char* os::reserve_memory_special(size_t bytes, size_t alignment,
4108 char* req_addr, bool exec) {
4109 assert(UseLargePages, "only for large pages");
4110
4111 char* addr;
4112 if (UseSHM) {
4113 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
4114 } else {
4115 assert(UseHugeTLBFS, "must be");
4116 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
4117 }
4118
4119 if (addr != NULL) {
4120 if (UseNUMAInterleaving) {
4121 numa_make_global(addr, bytes);
4122 }
4123
4124 // The memory is committed
4125 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
4126 }
4127
4128 return addr;
4129 }
4130
4131 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
4132 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
4133 return shmdt(base) == 0;
4134 }
4135
4136 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
4137 return pd_release_memory(base, bytes);
4138 }
4139
4140 bool os::release_memory_special(char* base, size_t bytes) {
4141 bool res;
4142 if (MemTracker::tracking_level() > NMT_minimal) {
4143 Tracker tkr(Tracker::release);
4144 res = os::Linux::release_memory_special_impl(base, bytes);
4145 if (res) {
4146 tkr.record((address)base, bytes);
4147 }
4148
4149 } else {
4150 res = os::Linux::release_memory_special_impl(base, bytes);
4151 }
4152 return res;
4153 }
4154
4155 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
4156 assert(UseLargePages, "only for large pages");
4157 bool res;
4158
4159 if (UseSHM) {
4160 res = os::Linux::release_memory_special_shm(base, bytes);
4161 } else {
4162 assert(UseHugeTLBFS, "must be");
4163 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
4164 }
4165 return res;
4166 }
4167
4168 size_t os::large_page_size() {
4169 return _large_page_size;
4170 }
4171
4172 // With SysV SHM the entire memory region must be allocated as shared
4173 // memory.
4174 // HugeTLBFS allows application to commit large page memory on demand.
4175 // However, when committing memory with HugeTLBFS fails, the region
4176 // that was supposed to be committed will lose the old reservation
4177 // and allow other threads to steal that memory region. Because of this
4178 // behavior we can't commit HugeTLBFS memory.
4179 bool os::can_commit_large_page_memory() {
4180 return UseTransparentHugePages;
4181 }
4182
4183 bool os::can_execute_large_page_memory() {
4184 return UseTransparentHugePages || UseHugeTLBFS;
4185 }
4186
4187 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
4188 assert(file_desc >= 0, "file_desc is not valid");
4189 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
4190 if (result != NULL) {
4191 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
4192 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
4193 }
4194 }
4195 return result;
4196 }
4197
4198 // Reserve memory at an arbitrary address, only if that area is
4199 // available (and not reserved for something else).
4200
4201 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
4202 // Assert only that the size is a multiple of the page size, since
4203 // that's all that mmap requires, and since that's all we really know
4204 // about at this low abstraction level. If we need higher alignment,
4205 // we can either pass an alignment to this method or verify alignment
4206 // in one of the methods further up the call chain. See bug 5044738.
4207 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
4208
4209 // Repeatedly allocate blocks until the block is allocated at the
4210 // right spot.
4211
4212 // Linux mmap allows caller to pass an address as hint; give it a try first,
4213 // if kernel honors the hint then we can return immediately.
4214 char * addr = anon_mmap(requested_addr, bytes, false);
4215 if (addr == requested_addr) {
4216 return requested_addr;
4217 }
4218
4219 if (addr != NULL) {
4220 // mmap() is successful but it fails to reserve at the requested address
4221 anon_munmap(addr, bytes);
4222 }
4223
4224 return NULL;
4225 }
4226
4227 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4228 void os::infinite_sleep() {
4229 while (true) { // sleep forever ...
4230 ::sleep(100); // ... 100 seconds at a time
4231 }
4232 }
4233
4234 // Used to convert frequent JVM_Yield() to nops
4235 bool os::dont_yield() {
4236 return DontYieldALot;
4237 }
4238
4239 // Linux CFS scheduler (since 2.6.23) does not guarantee sched_yield(2) will
4240 // actually give up the CPU. Since skip buddy (v2.6.28):
4241 //
4242 // * Sets the yielding task as skip buddy for current CPU's run queue.
4243 // * Picks next from run queue, if empty, picks a skip buddy (can be the yielding task).
4244 // * Clears skip buddies for this run queue (yielding task no longer a skip buddy).
4245 //
4246 // An alternative is calling os::naked_short_nanosleep with a small number to avoid
4247 // getting re-scheduled immediately.
4248 //
4249 void os::naked_yield() {
4250 sched_yield();
4251 }
4252
4253 ////////////////////////////////////////////////////////////////////////////////
4254 // thread priority support
4255
4256 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4257 // only supports dynamic priority, static priority must be zero. For real-time
4258 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4259 // However, for large multi-threaded applications, SCHED_RR is not only slower
4260 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4261 // of 5 runs - Sep 2005).
4262 //
4263 // The following code actually changes the niceness of kernel-thread/LWP. It
4264 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4265 // not the entire user process, and user level threads are 1:1 mapped to kernel
4266 // threads. It has always been the case, but could change in the future. For
4267 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4268 // It is only used when ThreadPriorityPolicy=1 and may require system level permission
4269 // (e.g., root privilege or CAP_SYS_NICE capability).
4270
4271 int os::java_to_os_priority[CriticalPriority + 1] = {
4272 19, // 0 Entry should never be used
4273
4274 4, // 1 MinPriority
4275 3, // 2
4276 2, // 3
4277
4278 1, // 4
4279 0, // 5 NormPriority
4280 -1, // 6
4281
4282 -2, // 7
4283 -3, // 8
4284 -4, // 9 NearMaxPriority
4285
4286 -5, // 10 MaxPriority
4287
4288 -5 // 11 CriticalPriority
4289 };
4290
4291 static int prio_init() {
4292 if (ThreadPriorityPolicy == 1) {
4293 if (geteuid() != 0) {
4294 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4295 warning("-XX:ThreadPriorityPolicy=1 may require system level permission, " \
4296 "e.g., being the root user. If the necessary permission is not " \
4297 "possessed, changes to priority will be silently ignored.");
4298 }
4299 }
4300 }
4301 if (UseCriticalJavaThreadPriority) {
4302 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4303 }
4304 return 0;
4305 }
4306
4307 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4308 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4309
4310 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4311 return (ret == 0) ? OS_OK : OS_ERR;
4312 }
4313
4314 OSReturn os::get_native_priority(const Thread* const thread,
4315 int *priority_ptr) {
4316 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4317 *priority_ptr = java_to_os_priority[NormPriority];
4318 return OS_OK;
4319 }
4320
4321 errno = 0;
4322 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4323 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4324 }
4325
4326 ////////////////////////////////////////////////////////////////////////////////
4327 // suspend/resume support
4328
4329 // The low-level signal-based suspend/resume support is a remnant from the
4330 // old VM-suspension that used to be for java-suspension, safepoints etc,
4331 // within hotspot. Currently used by JFR's OSThreadSampler
4332 //
4333 // The remaining code is greatly simplified from the more general suspension
4334 // code that used to be used.
4335 //
4336 // The protocol is quite simple:
4337 // - suspend:
4338 // - sends a signal to the target thread
4339 // - polls the suspend state of the osthread using a yield loop
4340 // - target thread signal handler (SR_handler) sets suspend state
4341 // and blocks in sigsuspend until continued
4342 // - resume:
4343 // - sets target osthread state to continue
4344 // - sends signal to end the sigsuspend loop in the SR_handler
4345 //
4346 // Note that the SR_lock plays no role in this suspend/resume protocol,
4347 // but is checked for NULL in SR_handler as a thread termination indicator.
4348 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4349 //
4350 // Note that resume_clear_context() and suspend_save_context() are needed
4351 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4352 // which in part is used by:
4353 // - Forte Analyzer: AsyncGetCallTrace()
4354 // - StackBanging: get_frame_at_stack_banging_point()
4355
4356 static void resume_clear_context(OSThread *osthread) {
4357 osthread->set_ucontext(NULL);
4358 osthread->set_siginfo(NULL);
4359 }
4360
4361 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4362 ucontext_t* context) {
4363 osthread->set_ucontext(context);
4364 osthread->set_siginfo(siginfo);
4365 }
4366
4367 // Handler function invoked when a thread's execution is suspended or
4368 // resumed. We have to be careful that only async-safe functions are
4369 // called here (Note: most pthread functions are not async safe and
4370 // should be avoided.)
4371 //
4372 // Note: sigwait() is a more natural fit than sigsuspend() from an
4373 // interface point of view, but sigwait() prevents the signal hander
4374 // from being run. libpthread would get very confused by not having
4375 // its signal handlers run and prevents sigwait()'s use with the
4376 // mutex granting granting signal.
4377 //
4378 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4379 //
4380 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4381 // Save and restore errno to avoid confusing native code with EINTR
4382 // after sigsuspend.
4383 int old_errno = errno;
4384
4385 Thread* thread = Thread::current_or_null_safe();
4386 assert(thread != NULL, "Missing current thread in SR_handler");
4387
4388 // On some systems we have seen signal delivery get "stuck" until the signal
4389 // mask is changed as part of thread termination. Check that the current thread
4390 // has not already terminated (via SR_lock()) - else the following assertion
4391 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4392 // destructor has completed.
4393
4394 if (thread->SR_lock() == NULL) {
4395 return;
4396 }
4397
4398 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4399
4400 OSThread* osthread = thread->osthread();
4401
4402 os::SuspendResume::State current = osthread->sr.state();
4403 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4404 suspend_save_context(osthread, siginfo, context);
4405
4406 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4407 os::SuspendResume::State state = osthread->sr.suspended();
4408 if (state == os::SuspendResume::SR_SUSPENDED) {
4409 sigset_t suspend_set; // signals for sigsuspend()
4410 sigemptyset(&suspend_set);
4411 // get current set of blocked signals and unblock resume signal
4412 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4413 sigdelset(&suspend_set, SR_signum);
4414
4415 sr_semaphore.signal();
4416 // wait here until we are resumed
4417 while (1) {
4418 sigsuspend(&suspend_set);
4419
4420 os::SuspendResume::State result = osthread->sr.running();
4421 if (result == os::SuspendResume::SR_RUNNING) {
4422 sr_semaphore.signal();
4423 break;
4424 }
4425 }
4426
4427 } else if (state == os::SuspendResume::SR_RUNNING) {
4428 // request was cancelled, continue
4429 } else {
4430 ShouldNotReachHere();
4431 }
4432
4433 resume_clear_context(osthread);
4434 } else if (current == os::SuspendResume::SR_RUNNING) {
4435 // request was cancelled, continue
4436 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4437 // ignore
4438 } else {
4439 // ignore
4440 }
4441
4442 errno = old_errno;
4443 }
4444
4445 static int SR_initialize() {
4446 struct sigaction act;
4447 char *s;
4448
4449 // Get signal number to use for suspend/resume
4450 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4451 int sig = ::strtol(s, 0, 10);
4452 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4453 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4454 SR_signum = sig;
4455 } else {
4456 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4457 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4458 }
4459 }
4460
4461 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4462 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4463
4464 sigemptyset(&SR_sigset);
4465 sigaddset(&SR_sigset, SR_signum);
4466
4467 // Set up signal handler for suspend/resume
4468 act.sa_flags = SA_RESTART|SA_SIGINFO;
4469 act.sa_handler = (void (*)(int)) SR_handler;
4470
4471 // SR_signum is blocked by default.
4472 // 4528190 - We also need to block pthread restart signal (32 on all
4473 // supported Linux platforms). Note that LinuxThreads need to block
4474 // this signal for all threads to work properly. So we don't have
4475 // to use hard-coded signal number when setting up the mask.
4476 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4477
4478 if (sigaction(SR_signum, &act, 0) == -1) {
4479 return -1;
4480 }
4481
4482 // Save signal flag
4483 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4484 return 0;
4485 }
4486
4487 static int sr_notify(OSThread* osthread) {
4488 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4489 assert_status(status == 0, status, "pthread_kill");
4490 return status;
4491 }
4492
4493 // "Randomly" selected value for how long we want to spin
4494 // before bailing out on suspending a thread, also how often
4495 // we send a signal to a thread we want to resume
4496 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4497 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4498
4499 // returns true on success and false on error - really an error is fatal
4500 // but this seems the normal response to library errors
4501 static bool do_suspend(OSThread* osthread) {
4502 assert(osthread->sr.is_running(), "thread should be running");
4503 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4504
4505 // mark as suspended and send signal
4506 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4507 // failed to switch, state wasn't running?
4508 ShouldNotReachHere();
4509 return false;
4510 }
4511
4512 if (sr_notify(osthread) != 0) {
4513 ShouldNotReachHere();
4514 }
4515
4516 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4517 while (true) {
4518 if (sr_semaphore.timedwait(2)) {
4519 break;
4520 } else {
4521 // timeout
4522 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4523 if (cancelled == os::SuspendResume::SR_RUNNING) {
4524 return false;
4525 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4526 // make sure that we consume the signal on the semaphore as well
4527 sr_semaphore.wait();
4528 break;
4529 } else {
4530 ShouldNotReachHere();
4531 return false;
4532 }
4533 }
4534 }
4535
4536 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4537 return true;
4538 }
4539
4540 static void do_resume(OSThread* osthread) {
4541 assert(osthread->sr.is_suspended(), "thread should be suspended");
4542 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4543
4544 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4545 // failed to switch to WAKEUP_REQUEST
4546 ShouldNotReachHere();
4547 return;
4548 }
4549
4550 while (true) {
4551 if (sr_notify(osthread) == 0) {
4552 if (sr_semaphore.timedwait(2)) {
4553 if (osthread->sr.is_running()) {
4554 return;
4555 }
4556 }
4557 } else {
4558 ShouldNotReachHere();
4559 }
4560 }
4561
4562 guarantee(osthread->sr.is_running(), "Must be running!");
4563 }
4564
4565 ///////////////////////////////////////////////////////////////////////////////////
4566 // signal handling (except suspend/resume)
4567
4568 // This routine may be used by user applications as a "hook" to catch signals.
4569 // The user-defined signal handler must pass unrecognized signals to this
4570 // routine, and if it returns true (non-zero), then the signal handler must
4571 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4572 // routine will never retun false (zero), but instead will execute a VM panic
4573 // routine kill the process.
4574 //
4575 // If this routine returns false, it is OK to call it again. This allows
4576 // the user-defined signal handler to perform checks either before or after
4577 // the VM performs its own checks. Naturally, the user code would be making
4578 // a serious error if it tried to handle an exception (such as a null check
4579 // or breakpoint) that the VM was generating for its own correct operation.
4580 //
4581 // This routine may recognize any of the following kinds of signals:
4582 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4583 // It should be consulted by handlers for any of those signals.
4584 //
4585 // The caller of this routine must pass in the three arguments supplied
4586 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4587 // field of the structure passed to sigaction(). This routine assumes that
4588 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4589 //
4590 // Note that the VM will print warnings if it detects conflicting signal
4591 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4592 //
4593 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4594 siginfo_t* siginfo,
4595 void* ucontext,
4596 int abort_if_unrecognized);
4597
4598 static void signalHandler(int sig, siginfo_t* info, void* uc) {
4599 assert(info != NULL && uc != NULL, "it must be old kernel");
4600 int orig_errno = errno; // Preserve errno value over signal handler.
4601 JVM_handle_linux_signal(sig, info, uc, true);
4602 errno = orig_errno;
4603 }
4604
4605
4606 // This boolean allows users to forward their own non-matching signals
4607 // to JVM_handle_linux_signal, harmlessly.
4608 bool os::Linux::signal_handlers_are_installed = false;
4609
4610 // For signal-chaining
4611 bool os::Linux::libjsig_is_loaded = false;
4612 typedef struct sigaction *(*get_signal_t)(int);
4613 get_signal_t os::Linux::get_signal_action = NULL;
4614
4615 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4616 struct sigaction *actp = NULL;
4617
4618 if (libjsig_is_loaded) {
4619 // Retrieve the old signal handler from libjsig
4620 actp = (*get_signal_action)(sig);
4621 }
4622 if (actp == NULL) {
4623 // Retrieve the preinstalled signal handler from jvm
4624 actp = os::Posix::get_preinstalled_handler(sig);
4625 }
4626
4627 return actp;
4628 }
4629
4630 static bool call_chained_handler(struct sigaction *actp, int sig,
4631 siginfo_t *siginfo, void *context) {
4632 // Call the old signal handler
4633 if (actp->sa_handler == SIG_DFL) {
4634 // It's more reasonable to let jvm treat it as an unexpected exception
4635 // instead of taking the default action.
4636 return false;
4637 } else if (actp->sa_handler != SIG_IGN) {
4638 if ((actp->sa_flags & SA_NODEFER) == 0) {
4639 // automaticlly block the signal
4640 sigaddset(&(actp->sa_mask), sig);
4641 }
4642
4643 sa_handler_t hand = NULL;
4644 sa_sigaction_t sa = NULL;
4645 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4646 // retrieve the chained handler
4647 if (siginfo_flag_set) {
4648 sa = actp->sa_sigaction;
4649 } else {
4650 hand = actp->sa_handler;
4651 }
4652
4653 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4654 actp->sa_handler = SIG_DFL;
4655 }
4656
4657 // try to honor the signal mask
4658 sigset_t oset;
4659 sigemptyset(&oset);
4660 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4661
4662 // call into the chained handler
4663 if (siginfo_flag_set) {
4664 (*sa)(sig, siginfo, context);
4665 } else {
4666 (*hand)(sig);
4667 }
4668
4669 // restore the signal mask
4670 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4671 }
4672 // Tell jvm's signal handler the signal is taken care of.
4673 return true;
4674 }
4675
4676 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4677 bool chained = false;
4678 // signal-chaining
4679 if (UseSignalChaining) {
4680 struct sigaction *actp = get_chained_signal_action(sig);
4681 if (actp != NULL) {
4682 chained = call_chained_handler(actp, sig, siginfo, context);
4683 }
4684 }
4685 return chained;
4686 }
4687
4688 // for diagnostic
4689 int sigflags[NSIG];
4690
4691 int os::Linux::get_our_sigflags(int sig) {
4692 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4693 return sigflags[sig];
4694 }
4695
4696 void os::Linux::set_our_sigflags(int sig, int flags) {
4697 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4698 if (sig > 0 && sig < NSIG) {
4699 sigflags[sig] = flags;
4700 }
4701 }
4702
4703 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4704 // Check for overwrite.
4705 struct sigaction oldAct;
4706 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4707
4708 void* oldhand = oldAct.sa_sigaction
4709 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4710 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4711 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4712 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4713 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4714 if (AllowUserSignalHandlers || !set_installed) {
4715 // Do not overwrite; user takes responsibility to forward to us.
4716 return;
4717 } else if (UseSignalChaining) {
4718 // save the old handler in jvm
4719 os::Posix::save_preinstalled_handler(sig, oldAct);
4720 // libjsig also interposes the sigaction() call below and saves the
4721 // old sigaction on it own.
4722 } else {
4723 fatal("Encountered unexpected pre-existing sigaction handler "
4724 "%#lx for signal %d.", (long)oldhand, sig);
4725 }
4726 }
4727
4728 struct sigaction sigAct;
4729 sigfillset(&(sigAct.sa_mask));
4730 sigAct.sa_handler = SIG_DFL;
4731 if (!set_installed) {
4732 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4733 } else {
4734 sigAct.sa_sigaction = signalHandler;
4735 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4736 }
4737 // Save flags, which are set by ours
4738 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4739 sigflags[sig] = sigAct.sa_flags;
4740
4741 int ret = sigaction(sig, &sigAct, &oldAct);
4742 assert(ret == 0, "check");
4743
4744 void* oldhand2 = oldAct.sa_sigaction
4745 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4746 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4747 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4748 }
4749
4750 // install signal handlers for signals that HotSpot needs to
4751 // handle in order to support Java-level exception handling.
4752
4753 void os::Linux::install_signal_handlers() {
4754 if (!signal_handlers_are_installed) {
4755 signal_handlers_are_installed = true;
4756
4757 // signal-chaining
4758 typedef void (*signal_setting_t)();
4759 signal_setting_t begin_signal_setting = NULL;
4760 signal_setting_t end_signal_setting = NULL;
4761 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4762 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4763 if (begin_signal_setting != NULL) {
4764 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4765 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4766 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4767 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4768 libjsig_is_loaded = true;
4769 assert(UseSignalChaining, "should enable signal-chaining");
4770 }
4771 if (libjsig_is_loaded) {
4772 // Tell libjsig jvm is setting signal handlers
4773 (*begin_signal_setting)();
4774 }
4775
4776 set_signal_handler(SIGSEGV, true);
4777 set_signal_handler(SIGPIPE, true);
4778 set_signal_handler(SIGBUS, true);
4779 set_signal_handler(SIGILL, true);
4780 set_signal_handler(SIGFPE, true);
4781 #if defined(PPC64)
4782 set_signal_handler(SIGTRAP, true);
4783 #endif
4784 set_signal_handler(SIGXFSZ, true);
4785
4786 if (libjsig_is_loaded) {
4787 // Tell libjsig jvm finishes setting signal handlers
4788 (*end_signal_setting)();
4789 }
4790
4791 // We don't activate signal checker if libjsig is in place, we trust ourselves
4792 // and if UserSignalHandler is installed all bets are off.
4793 // Log that signal checking is off only if -verbose:jni is specified.
4794 if (CheckJNICalls) {
4795 if (libjsig_is_loaded) {
4796 if (PrintJNIResolving) {
4797 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4798 }
4799 check_signals = false;
4800 }
4801 if (AllowUserSignalHandlers) {
4802 if (PrintJNIResolving) {
4803 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4804 }
4805 check_signals = false;
4806 }
4807 }
4808 }
4809 }
4810
4811 // This is the fastest way to get thread cpu time on Linux.
4812 // Returns cpu time (user+sys) for any thread, not only for current.
4813 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4814 // It might work on 2.6.10+ with a special kernel/glibc patch.
4815 // For reference, please, see IEEE Std 1003.1-2004:
4816 // http://www.unix.org/single_unix_specification
4817
4818 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4819 struct timespec tp;
4820 int rc = os::Posix::clock_gettime(clockid, &tp);
4821 assert(rc == 0, "clock_gettime is expected to return 0 code");
4822
4823 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4824 }
4825
4826 void os::Linux::initialize_os_info() {
4827 assert(_os_version == 0, "OS info already initialized");
4828
4829 struct utsname _uname;
4830
4831 uint32_t major;
4832 uint32_t minor;
4833 uint32_t fix;
4834
4835 int rc;
4836
4837 // Kernel version is unknown if
4838 // verification below fails.
4839 _os_version = 0x01000000;
4840
4841 rc = uname(&_uname);
4842 if (rc != -1) {
4843
4844 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4845 if (rc == 3) {
4846
4847 if (major < 256 && minor < 256 && fix < 256) {
4848 // Kernel version format is as expected,
4849 // set it overriding unknown state.
4850 _os_version = (major << 16) |
4851 (minor << 8 ) |
4852 (fix << 0 ) ;
4853 }
4854 }
4855 }
4856 }
4857
4858 uint32_t os::Linux::os_version() {
4859 assert(_os_version != 0, "not initialized");
4860 return _os_version & 0x00FFFFFF;
4861 }
4862
4863 bool os::Linux::os_version_is_known() {
4864 assert(_os_version != 0, "not initialized");
4865 return _os_version & 0x01000000 ? false : true;
4866 }
4867
4868 /////
4869 // glibc on Linux platform uses non-documented flag
4870 // to indicate, that some special sort of signal
4871 // trampoline is used.
4872 // We will never set this flag, and we should
4873 // ignore this flag in our diagnostic
4874 #ifdef SIGNIFICANT_SIGNAL_MASK
4875 #undef SIGNIFICANT_SIGNAL_MASK
4876 #endif
4877 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4878
4879 static const char* get_signal_handler_name(address handler,
4880 char* buf, int buflen) {
4881 int offset = 0;
4882 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4883 if (found) {
4884 // skip directory names
4885 const char *p1, *p2;
4886 p1 = buf;
4887 size_t len = strlen(os::file_separator());
4888 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4889 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4890 } else {
4891 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4892 }
4893 return buf;
4894 }
4895
4896 static void print_signal_handler(outputStream* st, int sig,
4897 char* buf, size_t buflen) {
4898 struct sigaction sa;
4899
4900 sigaction(sig, NULL, &sa);
4901
4902 // See comment for SIGNIFICANT_SIGNAL_MASK define
4903 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4904
4905 st->print("%s: ", os::exception_name(sig, buf, buflen));
4906
4907 address handler = (sa.sa_flags & SA_SIGINFO)
4908 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4909 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4910
4911 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4912 st->print("SIG_DFL");
4913 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4914 st->print("SIG_IGN");
4915 } else {
4916 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4917 }
4918
4919 st->print(", sa_mask[0]=");
4920 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4921
4922 address rh = VMError::get_resetted_sighandler(sig);
4923 // May be, handler was resetted by VMError?
4924 if (rh != NULL) {
4925 handler = rh;
4926 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4927 }
4928
4929 st->print(", sa_flags=");
4930 os::Posix::print_sa_flags(st, sa.sa_flags);
4931
4932 // Check: is it our handler?
4933 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4934 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4935 // It is our signal handler
4936 // check for flags, reset system-used one!
4937 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4938 st->print(
4939 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4940 os::Linux::get_our_sigflags(sig));
4941 }
4942 }
4943 st->cr();
4944 }
4945
4946
4947 #define DO_SIGNAL_CHECK(sig) \
4948 do { \
4949 if (!sigismember(&check_signal_done, sig)) { \
4950 os::Linux::check_signal_handler(sig); \
4951 } \
4952 } while (0)
4953
4954 // This method is a periodic task to check for misbehaving JNI applications
4955 // under CheckJNI, we can add any periodic checks here
4956
4957 void os::run_periodic_checks() {
4958 if (check_signals == false) return;
4959
4960 // SEGV and BUS if overridden could potentially prevent
4961 // generation of hs*.log in the event of a crash, debugging
4962 // such a case can be very challenging, so we absolutely
4963 // check the following for a good measure:
4964 DO_SIGNAL_CHECK(SIGSEGV);
4965 DO_SIGNAL_CHECK(SIGILL);
4966 DO_SIGNAL_CHECK(SIGFPE);
4967 DO_SIGNAL_CHECK(SIGBUS);
4968 DO_SIGNAL_CHECK(SIGPIPE);
4969 DO_SIGNAL_CHECK(SIGXFSZ);
4970 #if defined(PPC64)
4971 DO_SIGNAL_CHECK(SIGTRAP);
4972 #endif
4973
4974 // ReduceSignalUsage allows the user to override these handlers
4975 // see comments at the very top and jvm_md.h
4976 if (!ReduceSignalUsage) {
4977 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4978 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4979 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4980 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4981 }
4982
4983 DO_SIGNAL_CHECK(SR_signum);
4984 }
4985
4986 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4987
4988 static os_sigaction_t os_sigaction = NULL;
4989
4990 void os::Linux::check_signal_handler(int sig) {
4991 char buf[O_BUFLEN];
4992 address jvmHandler = NULL;
4993
4994
4995 struct sigaction act;
4996 if (os_sigaction == NULL) {
4997 // only trust the default sigaction, in case it has been interposed
4998 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4999 if (os_sigaction == NULL) return;
5000 }
5001
5002 os_sigaction(sig, (struct sigaction*)NULL, &act);
5003
5004
5005 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
5006
5007 address thisHandler = (act.sa_flags & SA_SIGINFO)
5008 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
5009 : CAST_FROM_FN_PTR(address, act.sa_handler);
5010
5011
5012 switch (sig) {
5013 case SIGSEGV:
5014 case SIGBUS:
5015 case SIGFPE:
5016 case SIGPIPE:
5017 case SIGILL:
5018 case SIGXFSZ:
5019 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
5020 break;
5021
5022 case SHUTDOWN1_SIGNAL:
5023 case SHUTDOWN2_SIGNAL:
5024 case SHUTDOWN3_SIGNAL:
5025 case BREAK_SIGNAL:
5026 jvmHandler = (address)user_handler();
5027 break;
5028
5029 default:
5030 if (sig == SR_signum) {
5031 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
5032 } else {
5033 return;
5034 }
5035 break;
5036 }
5037
5038 if (thisHandler != jvmHandler) {
5039 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
5040 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
5041 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
5042 // No need to check this sig any longer
5043 sigaddset(&check_signal_done, sig);
5044 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
5045 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
5046 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
5047 exception_name(sig, buf, O_BUFLEN));
5048 }
5049 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
5050 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
5051 tty->print("expected:");
5052 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
5053 tty->cr();
5054 tty->print(" found:");
5055 os::Posix::print_sa_flags(tty, act.sa_flags);
5056 tty->cr();
5057 // No need to check this sig any longer
5058 sigaddset(&check_signal_done, sig);
5059 }
5060
5061 // Dump all the signal
5062 if (sigismember(&check_signal_done, sig)) {
5063 print_signal_handlers(tty, buf, O_BUFLEN);
5064 }
5065 }
5066
5067 extern void report_error(char* file_name, int line_no, char* title,
5068 char* format, ...);
5069
5070 // this is called _before_ most of the global arguments have been parsed
5071 void os::init(void) {
5072 char dummy; // used to get a guess on initial stack address
5073
5074 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
5075
5076 init_random(1234567);
5077
5078 Linux::set_page_size(sysconf(_SC_PAGESIZE));
5079 if (Linux::page_size() == -1) {
5080 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
5081 os::strerror(errno));
5082 }
5083 init_page_sizes((size_t) Linux::page_size());
5084
5085 Linux::initialize_system_info();
5086
5087 Linux::initialize_os_info();
5088
5089 os::Linux::CPUPerfTicks pticks;
5090 bool res = os::Linux::get_tick_information(&pticks, -1);
5091
5092 if (res && pticks.has_steal_ticks) {
5093 has_initial_tick_info = true;
5094 initial_total_ticks = pticks.total;
5095 initial_steal_ticks = pticks.steal;
5096 }
5097
5098 // _main_thread points to the thread that created/loaded the JVM.
5099 Linux::_main_thread = pthread_self();
5100
5101 // retrieve entry point for pthread_setname_np
5102 Linux::_pthread_setname_np =
5103 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
5104
5105 os::Posix::init();
5106
5107 initial_time_count = javaTimeNanos();
5108
5109 // Always warn if no monotonic clock available
5110 if (!os::Posix::supports_monotonic_clock()) {
5111 warning("No monotonic clock was available - timed services may " \
5112 "be adversely affected if the time-of-day clock changes");
5113 }
5114 }
5115
5116 // To install functions for atexit system call
5117 extern "C" {
5118 static void perfMemory_exit_helper() {
5119 perfMemory_exit();
5120 }
5121 }
5122
5123 void os::pd_init_container_support() {
5124 OSContainer::init();
5125 }
5126
5127 void os::Linux::numa_init() {
5128
5129 // Java can be invoked as
5130 // 1. Without numactl and heap will be allocated/configured on all nodes as
5131 // per the system policy.
5132 // 2. With numactl --interleave:
5133 // Use numa_get_interleave_mask(v2) API to get nodes bitmask. The same
5134 // API for membind case bitmask is reset.
5135 // Interleave is only hint and Kernel can fallback to other nodes if
5136 // no memory is available on the target nodes.
5137 // 3. With numactl --membind:
5138 // Use numa_get_membind(v2) API to get nodes bitmask. The same API for
5139 // interleave case returns bitmask of all nodes.
5140 // numa_all_nodes_ptr holds bitmask of all nodes.
5141 // numa_get_interleave_mask(v2) and numa_get_membind(v2) APIs returns correct
5142 // bitmask when externally configured to run on all or fewer nodes.
5143
5144 if (!Linux::libnuma_init()) {
5145 UseNUMA = false;
5146 } else {
5147 if ((Linux::numa_max_node() < 1) || Linux::is_bound_to_single_node()) {
5148 // If there's only one node (they start from 0) or if the process
5149 // is bound explicitly to a single node using membind, disable NUMA.
5150 UseNUMA = false;
5151 } else {
5152
5153 LogTarget(Info,os) log;
5154 LogStream ls(log);
5155
5156 Linux::set_configured_numa_policy(Linux::identify_numa_policy());
5157
5158 struct bitmask* bmp = Linux::_numa_membind_bitmask;
5159 const char* numa_mode = "membind";
5160
5161 if (Linux::is_running_in_interleave_mode()) {
5162 bmp = Linux::_numa_interleave_bitmask;
5163 numa_mode = "interleave";
5164 }
5165
5166 ls.print("UseNUMA is enabled and invoked in '%s' mode."
5167 " Heap will be configured using NUMA memory nodes:", numa_mode);
5168
5169 for (int node = 0; node <= Linux::numa_max_node(); node++) {
5170 if (Linux::_numa_bitmask_isbitset(bmp, node)) {
5171 ls.print(" %d", node);
5172 }
5173 }
5174 }
5175 }
5176
5177 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5178 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5179 // we can make the adaptive lgrp chunk resizing work. If the user specified both
5180 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5181 // and disable adaptive resizing.
5182 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5183 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5184 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5185 UseAdaptiveSizePolicy = false;
5186 UseAdaptiveNUMAChunkSizing = false;
5187 }
5188 }
5189
5190 if (!UseNUMA && ForceNUMA) {
5191 UseNUMA = true;
5192 }
5193 }
5194
5195 // this is called _after_ the global arguments have been parsed
5196 jint os::init_2(void) {
5197
5198 // This could be set after os::Posix::init() but all platforms
5199 // have to set it the same so we have to mirror Solaris.
5200 DEBUG_ONLY(os::set_mutex_init_done();)
5201
5202 os::Posix::init_2();
5203
5204 Linux::fast_thread_clock_init();
5205
5206 // initialize suspend/resume support - must do this before signal_sets_init()
5207 if (SR_initialize() != 0) {
5208 perror("SR_initialize failed");
5209 return JNI_ERR;
5210 }
5211
5212 Linux::signal_sets_init();
5213 Linux::install_signal_handlers();
5214 // Initialize data for jdk.internal.misc.Signal
5215 if (!ReduceSignalUsage) {
5216 jdk_misc_signal_init();
5217 }
5218
5219 if (AdjustStackSizeForTLS) {
5220 get_minstack_init();
5221 }
5222
5223 // Check and sets minimum stack sizes against command line options
5224 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
5225 return JNI_ERR;
5226 }
5227
5228 #if defined(IA32)
5229 // Need to ensure we've determined the process's initial stack to
5230 // perform the workaround
5231 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5232 workaround_expand_exec_shield_cs_limit();
5233 #else
5234 suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
5235 if (!suppress_primordial_thread_resolution) {
5236 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5237 }
5238 #endif
5239
5240 Linux::libpthread_init();
5241 Linux::sched_getcpu_init();
5242 log_info(os)("HotSpot is running with %s, %s",
5243 Linux::glibc_version(), Linux::libpthread_version());
5244
5245 if (UseNUMA) {
5246 Linux::numa_init();
5247 }
5248
5249 if (MaxFDLimit) {
5250 // set the number of file descriptors to max. print out error
5251 // if getrlimit/setrlimit fails but continue regardless.
5252 struct rlimit nbr_files;
5253 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5254 if (status != 0) {
5255 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5256 } else {
5257 nbr_files.rlim_cur = nbr_files.rlim_max;
5258 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5259 if (status != 0) {
5260 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5261 }
5262 }
5263 }
5264
5265 // Initialize lock used to serialize thread creation (see os::create_thread)
5266 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5267
5268 // at-exit methods are called in the reverse order of their registration.
5269 // atexit functions are called on return from main or as a result of a
5270 // call to exit(3C). There can be only 32 of these functions registered
5271 // and atexit() does not set errno.
5272
5273 if (PerfAllowAtExitRegistration) {
5274 // only register atexit functions if PerfAllowAtExitRegistration is set.
5275 // atexit functions can be delayed until process exit time, which
5276 // can be problematic for embedded VM situations. Embedded VMs should
5277 // call DestroyJavaVM() to assure that VM resources are released.
5278
5279 // note: perfMemory_exit_helper atexit function may be removed in
5280 // the future if the appropriate cleanup code can be added to the
5281 // VM_Exit VMOperation's doit method.
5282 if (atexit(perfMemory_exit_helper) != 0) {
5283 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5284 }
5285 }
5286
5287 // initialize thread priority policy
5288 prio_init();
5289
5290 if (!FLAG_IS_DEFAULT(AllocateHeapAt) || !FLAG_IS_DEFAULT(AllocateOldGenAt)) {
5291 set_coredump_filter(DAX_SHARED_BIT);
5292 }
5293
5294 if (DumpPrivateMappingsInCore) {
5295 set_coredump_filter(FILE_BACKED_PVT_BIT);
5296 }
5297
5298 if (DumpSharedMappingsInCore) {
5299 set_coredump_filter(FILE_BACKED_SHARED_BIT);
5300 }
5301
5302 return JNI_OK;
5303 }
5304
5305 // Mark the polling page as unreadable
5306 void os::make_polling_page_unreadable(void) {
5307 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5308 fatal("Could not disable polling page");
5309 }
5310 }
5311
5312 // Mark the polling page as readable
5313 void os::make_polling_page_readable(void) {
5314 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5315 fatal("Could not enable polling page");
5316 }
5317 }
5318
5319 // older glibc versions don't have this macro (which expands to
5320 // an optimized bit-counting function) so we have to roll our own
5321 #ifndef CPU_COUNT
5322
5323 static int _cpu_count(const cpu_set_t* cpus) {
5324 int count = 0;
5325 // only look up to the number of configured processors
5326 for (int i = 0; i < os::processor_count(); i++) {
5327 if (CPU_ISSET(i, cpus)) {
5328 count++;
5329 }
5330 }
5331 return count;
5332 }
5333
5334 #define CPU_COUNT(cpus) _cpu_count(cpus)
5335
5336 #endif // CPU_COUNT
5337
5338 // Get the current number of available processors for this process.
5339 // This value can change at any time during a process's lifetime.
5340 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5341 // If it appears there may be more than 1024 processors then we do a
5342 // dynamic check - see 6515172 for details.
5343 // If anything goes wrong we fallback to returning the number of online
5344 // processors - which can be greater than the number available to the process.
5345 int os::Linux::active_processor_count() {
5346 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5347 cpu_set_t* cpus_p = &cpus;
5348 int cpus_size = sizeof(cpu_set_t);
5349
5350 int configured_cpus = os::processor_count(); // upper bound on available cpus
5351 int cpu_count = 0;
5352
5353 // old build platforms may not support dynamic cpu sets
5354 #ifdef CPU_ALLOC
5355
5356 // To enable easy testing of the dynamic path on different platforms we
5357 // introduce a diagnostic flag: UseCpuAllocPath
5358 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5359 // kernel may use a mask bigger than cpu_set_t
5360 log_trace(os)("active_processor_count: using dynamic path %s"
5361 "- configured processors: %d",
5362 UseCpuAllocPath ? "(forced) " : "",
5363 configured_cpus);
5364 cpus_p = CPU_ALLOC(configured_cpus);
5365 if (cpus_p != NULL) {
5366 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5367 // zero it just to be safe
5368 CPU_ZERO_S(cpus_size, cpus_p);
5369 }
5370 else {
5371 // failed to allocate so fallback to online cpus
5372 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5373 log_trace(os)("active_processor_count: "
5374 "CPU_ALLOC failed (%s) - using "
5375 "online processor count: %d",
5376 os::strerror(errno), online_cpus);
5377 return online_cpus;
5378 }
5379 }
5380 else {
5381 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5382 configured_cpus);
5383 }
5384 #else // CPU_ALLOC
5385 // these stubs won't be executed
5386 #define CPU_COUNT_S(size, cpus) -1
5387 #define CPU_FREE(cpus)
5388
5389 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5390 configured_cpus);
5391 #endif // CPU_ALLOC
5392
5393 // pid 0 means the current thread - which we have to assume represents the process
5394 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5395 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5396 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5397 }
5398 else {
5399 cpu_count = CPU_COUNT(cpus_p);
5400 }
5401 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5402 }
5403 else {
5404 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5405 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5406 "which may exceed available processors", os::strerror(errno), cpu_count);
5407 }
5408
5409 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5410 CPU_FREE(cpus_p);
5411 }
5412
5413 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5414 return cpu_count;
5415 }
5416
5417 // Determine the active processor count from one of
5418 // three different sources:
5419 //
5420 // 1. User option -XX:ActiveProcessorCount
5421 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5422 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5423 //
5424 // Option 1, if specified, will always override.
5425 // If the cgroup subsystem is active and configured, we
5426 // will return the min of the cgroup and option 2 results.
5427 // This is required since tools, such as numactl, that
5428 // alter cpu affinity do not update cgroup subsystem
5429 // cpuset configuration files.
5430 int os::active_processor_count() {
5431 // User has overridden the number of active processors
5432 if (ActiveProcessorCount > 0) {
5433 log_trace(os)("active_processor_count: "
5434 "active processor count set by user : %d",
5435 ActiveProcessorCount);
5436 return ActiveProcessorCount;
5437 }
5438
5439 int active_cpus;
5440 if (OSContainer::is_containerized()) {
5441 active_cpus = OSContainer::active_processor_count();
5442 log_trace(os)("active_processor_count: determined by OSContainer: %d",
5443 active_cpus);
5444 } else {
5445 active_cpus = os::Linux::active_processor_count();
5446 }
5447
5448 return active_cpus;
5449 }
5450
5451 uint os::processor_id() {
5452 const int id = Linux::sched_getcpu();
5453 assert(id >= 0 && id < _processor_count, "Invalid processor id");
5454 return (uint)id;
5455 }
5456
5457 void os::set_native_thread_name(const char *name) {
5458 if (Linux::_pthread_setname_np) {
5459 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5460 snprintf(buf, sizeof(buf), "%s", name);
5461 buf[sizeof(buf) - 1] = '\0';
5462 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5463 // ERANGE should not happen; all other errors should just be ignored.
5464 assert(rc != ERANGE, "pthread_setname_np failed");
5465 }
5466 }
5467
5468 bool os::distribute_processes(uint length, uint* distribution) {
5469 // Not yet implemented.
5470 return false;
5471 }
5472
5473 bool os::bind_to_processor(uint processor_id) {
5474 // Not yet implemented.
5475 return false;
5476 }
5477
5478 ///
5479
5480 void os::SuspendedThreadTask::internal_do_task() {
5481 if (do_suspend(_thread->osthread())) {
5482 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5483 do_task(context);
5484 do_resume(_thread->osthread());
5485 }
5486 }
5487
5488 ////////////////////////////////////////////////////////////////////////////////
5489 // debug support
5490
5491 bool os::find(address addr, outputStream* st) {
5492 Dl_info dlinfo;
5493 memset(&dlinfo, 0, sizeof(dlinfo));
5494 if (dladdr(addr, &dlinfo) != 0) {
5495 st->print(PTR_FORMAT ": ", p2i(addr));
5496 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5497 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5498 p2i(addr) - p2i(dlinfo.dli_saddr));
5499 } else if (dlinfo.dli_fbase != NULL) {
5500 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5501 } else {
5502 st->print("<absolute address>");
5503 }
5504 if (dlinfo.dli_fname != NULL) {
5505 st->print(" in %s", dlinfo.dli_fname);
5506 }
5507 if (dlinfo.dli_fbase != NULL) {
5508 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5509 }
5510 st->cr();
5511
5512 if (Verbose) {
5513 // decode some bytes around the PC
5514 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5515 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5516 address lowest = (address) dlinfo.dli_sname;
5517 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5518 if (begin < lowest) begin = lowest;
5519 Dl_info dlinfo2;
5520 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5521 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5522 end = (address) dlinfo2.dli_saddr;
5523 }
5524 Disassembler::decode(begin, end, st);
5525 }
5526 return true;
5527 }
5528 return false;
5529 }
5530
5531 ////////////////////////////////////////////////////////////////////////////////
5532 // misc
5533
5534 // This does not do anything on Linux. This is basically a hook for being
5535 // able to use structured exception handling (thread-local exception filters)
5536 // on, e.g., Win32.
5537 void
5538 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5539 JavaCallArguments* args, Thread* thread) {
5540 f(value, method, args, thread);
5541 }
5542
5543 void os::print_statistics() {
5544 }
5545
5546 bool os::message_box(const char* title, const char* message) {
5547 int i;
5548 fdStream err(defaultStream::error_fd());
5549 for (i = 0; i < 78; i++) err.print_raw("=");
5550 err.cr();
5551 err.print_raw_cr(title);
5552 for (i = 0; i < 78; i++) err.print_raw("-");
5553 err.cr();
5554 err.print_raw_cr(message);
5555 for (i = 0; i < 78; i++) err.print_raw("=");
5556 err.cr();
5557
5558 char buf[16];
5559 // Prevent process from exiting upon "read error" without consuming all CPU
5560 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5561
5562 return buf[0] == 'y' || buf[0] == 'Y';
5563 }
5564
5565 // Is a (classpath) directory empty?
5566 bool os::dir_is_empty(const char* path) {
5567 DIR *dir = NULL;
5568 struct dirent *ptr;
5569
5570 dir = opendir(path);
5571 if (dir == NULL) return true;
5572
5573 // Scan the directory
5574 bool result = true;
5575 while (result && (ptr = readdir(dir)) != NULL) {
5576 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5577 result = false;
5578 }
5579 }
5580 closedir(dir);
5581 return result;
5582 }
5583
5584 // This code originates from JDK's sysOpen and open64_w
5585 // from src/solaris/hpi/src/system_md.c
5586
5587 int os::open(const char *path, int oflag, int mode) {
5588 if (strlen(path) > MAX_PATH - 1) {
5589 errno = ENAMETOOLONG;
5590 return -1;
5591 }
5592
5593 // All file descriptors that are opened in the Java process and not
5594 // specifically destined for a subprocess should have the close-on-exec
5595 // flag set. If we don't set it, then careless 3rd party native code
5596 // might fork and exec without closing all appropriate file descriptors
5597 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5598 // turn might:
5599 //
5600 // - cause end-of-file to fail to be detected on some file
5601 // descriptors, resulting in mysterious hangs, or
5602 //
5603 // - might cause an fopen in the subprocess to fail on a system
5604 // suffering from bug 1085341.
5605 //
5606 // (Yes, the default setting of the close-on-exec flag is a Unix
5607 // design flaw)
5608 //
5609 // See:
5610 // 1085341: 32-bit stdio routines should support file descriptors >255
5611 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5612 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5613 //
5614 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5615 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5616 // because it saves a system call and removes a small window where the flag
5617 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5618 // and we fall back to using FD_CLOEXEC (see below).
5619 #ifdef O_CLOEXEC
5620 oflag |= O_CLOEXEC;
5621 #endif
5622
5623 int fd = ::open64(path, oflag, mode);
5624 if (fd == -1) return -1;
5625
5626 //If the open succeeded, the file might still be a directory
5627 {
5628 struct stat64 buf64;
5629 int ret = ::fstat64(fd, &buf64);
5630 int st_mode = buf64.st_mode;
5631
5632 if (ret != -1) {
5633 if ((st_mode & S_IFMT) == S_IFDIR) {
5634 errno = EISDIR;
5635 ::close(fd);
5636 return -1;
5637 }
5638 } else {
5639 ::close(fd);
5640 return -1;
5641 }
5642 }
5643
5644 #ifdef FD_CLOEXEC
5645 // Validate that the use of the O_CLOEXEC flag on open above worked.
5646 // With recent kernels, we will perform this check exactly once.
5647 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5648 if (!O_CLOEXEC_is_known_to_work) {
5649 int flags = ::fcntl(fd, F_GETFD);
5650 if (flags != -1) {
5651 if ((flags & FD_CLOEXEC) != 0)
5652 O_CLOEXEC_is_known_to_work = 1;
5653 else
5654 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5655 }
5656 }
5657 #endif
5658
5659 return fd;
5660 }
5661
5662
5663 // create binary file, rewriting existing file if required
5664 int os::create_binary_file(const char* path, bool rewrite_existing) {
5665 int oflags = O_WRONLY | O_CREAT;
5666 if (!rewrite_existing) {
5667 oflags |= O_EXCL;
5668 }
5669 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5670 }
5671
5672 // return current position of file pointer
5673 jlong os::current_file_offset(int fd) {
5674 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5675 }
5676
5677 // move file pointer to the specified offset
5678 jlong os::seek_to_file_offset(int fd, jlong offset) {
5679 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5680 }
5681
5682 // This code originates from JDK's sysAvailable
5683 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5684
5685 int os::available(int fd, jlong *bytes) {
5686 jlong cur, end;
5687 int mode;
5688 struct stat64 buf64;
5689
5690 if (::fstat64(fd, &buf64) >= 0) {
5691 mode = buf64.st_mode;
5692 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5693 int n;
5694 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5695 *bytes = n;
5696 return 1;
5697 }
5698 }
5699 }
5700 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5701 return 0;
5702 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5703 return 0;
5704 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5705 return 0;
5706 }
5707 *bytes = end - cur;
5708 return 1;
5709 }
5710
5711 // Map a block of memory.
5712 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5713 char *addr, size_t bytes, bool read_only,
5714 bool allow_exec) {
5715 int prot;
5716 int flags = MAP_PRIVATE;
5717
5718 if (read_only) {
5719 prot = PROT_READ;
5720 } else {
5721 prot = PROT_READ | PROT_WRITE;
5722 }
5723
5724 if (allow_exec) {
5725 prot |= PROT_EXEC;
5726 }
5727
5728 if (addr != NULL) {
5729 flags |= MAP_FIXED;
5730 }
5731
5732 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5733 fd, file_offset);
5734 if (mapped_address == MAP_FAILED) {
5735 return NULL;
5736 }
5737 return mapped_address;
5738 }
5739
5740
5741 // Remap a block of memory.
5742 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5743 char *addr, size_t bytes, bool read_only,
5744 bool allow_exec) {
5745 // same as map_memory() on this OS
5746 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5747 allow_exec);
5748 }
5749
5750
5751 // Unmap a block of memory.
5752 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5753 return munmap(addr, bytes) == 0;
5754 }
5755
5756 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5757
5758 static jlong fast_cpu_time(Thread *thread) {
5759 clockid_t clockid;
5760 int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(),
5761 &clockid);
5762 if (rc == 0) {
5763 return os::Linux::fast_thread_cpu_time(clockid);
5764 } else {
5765 // It's possible to encounter a terminated native thread that failed
5766 // to detach itself from the VM - which should result in ESRCH.
5767 assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed");
5768 return -1;
5769 }
5770 }
5771
5772 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5773 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5774 // of a thread.
5775 //
5776 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5777 // the fast estimate available on the platform.
5778
5779 jlong os::current_thread_cpu_time() {
5780 if (os::Linux::supports_fast_thread_cpu_time()) {
5781 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5782 } else {
5783 // return user + sys since the cost is the same
5784 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5785 }
5786 }
5787
5788 jlong os::thread_cpu_time(Thread* thread) {
5789 // consistent with what current_thread_cpu_time() returns
5790 if (os::Linux::supports_fast_thread_cpu_time()) {
5791 return fast_cpu_time(thread);
5792 } else {
5793 return slow_thread_cpu_time(thread, true /* user + sys */);
5794 }
5795 }
5796
5797 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5798 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5799 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5800 } else {
5801 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5802 }
5803 }
5804
5805 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5806 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5807 return fast_cpu_time(thread);
5808 } else {
5809 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5810 }
5811 }
5812
5813 // -1 on error.
5814 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5815 pid_t tid = thread->osthread()->thread_id();
5816 char *s;
5817 char stat[2048];
5818 int statlen;
5819 char proc_name[64];
5820 int count;
5821 long sys_time, user_time;
5822 char cdummy;
5823 int idummy;
5824 long ldummy;
5825 FILE *fp;
5826
5827 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5828 fp = fopen(proc_name, "r");
5829 if (fp == NULL) return -1;
5830 statlen = fread(stat, 1, 2047, fp);
5831 stat[statlen] = '\0';
5832 fclose(fp);
5833
5834 // Skip pid and the command string. Note that we could be dealing with
5835 // weird command names, e.g. user could decide to rename java launcher
5836 // to "java 1.4.2 :)", then the stat file would look like
5837 // 1234 (java 1.4.2 :)) R ... ...
5838 // We don't really need to know the command string, just find the last
5839 // occurrence of ")" and then start parsing from there. See bug 4726580.
5840 s = strrchr(stat, ')');
5841 if (s == NULL) return -1;
5842
5843 // Skip blank chars
5844 do { s++; } while (s && isspace(*s));
5845
5846 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5847 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5848 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5849 &user_time, &sys_time);
5850 if (count != 13) return -1;
5851 if (user_sys_cpu_time) {
5852 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5853 } else {
5854 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5855 }
5856 }
5857
5858 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5859 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5860 info_ptr->may_skip_backward = false; // elapsed time not wall time
5861 info_ptr->may_skip_forward = false; // elapsed time not wall time
5862 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5863 }
5864
5865 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5866 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5867 info_ptr->may_skip_backward = false; // elapsed time not wall time
5868 info_ptr->may_skip_forward = false; // elapsed time not wall time
5869 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5870 }
5871
5872 bool os::is_thread_cpu_time_supported() {
5873 return true;
5874 }
5875
5876 // System loadavg support. Returns -1 if load average cannot be obtained.
5877 // Linux doesn't yet have a (official) notion of processor sets,
5878 // so just return the system wide load average.
5879 int os::loadavg(double loadavg[], int nelem) {
5880 return ::getloadavg(loadavg, nelem);
5881 }
5882
5883 void os::pause() {
5884 char filename[MAX_PATH];
5885 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5886 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5887 } else {
5888 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5889 }
5890
5891 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5892 if (fd != -1) {
5893 struct stat buf;
5894 ::close(fd);
5895 while (::stat(filename, &buf) == 0) {
5896 (void)::poll(NULL, 0, 100);
5897 }
5898 } else {
5899 jio_fprintf(stderr,
5900 "Could not open pause file '%s', continuing immediately.\n", filename);
5901 }
5902 }
5903
5904 extern char** environ;
5905
5906 // Run the specified command in a separate process. Return its exit value,
5907 // or -1 on failure (e.g. can't fork a new process).
5908 // Unlike system(), this function can be called from signal handler. It
5909 // doesn't block SIGINT et al.
5910 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) {
5911 const char * argv[4] = {"sh", "-c", cmd, NULL};
5912
5913 pid_t pid ;
5914
5915 if (use_vfork_if_available) {
5916 pid = vfork();
5917 } else {
5918 pid = fork();
5919 }
5920
5921 if (pid < 0) {
5922 // fork failed
5923 return -1;
5924
5925 } else if (pid == 0) {
5926 // child process
5927
5928 execve("/bin/sh", (char* const*)argv, environ);
5929
5930 // execve failed
5931 _exit(-1);
5932
5933 } else {
5934 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5935 // care about the actual exit code, for now.
5936
5937 int status;
5938
5939 // Wait for the child process to exit. This returns immediately if
5940 // the child has already exited. */
5941 while (waitpid(pid, &status, 0) < 0) {
5942 switch (errno) {
5943 case ECHILD: return 0;
5944 case EINTR: break;
5945 default: return -1;
5946 }
5947 }
5948
5949 if (WIFEXITED(status)) {
5950 // The child exited normally; get its exit code.
5951 return WEXITSTATUS(status);
5952 } else if (WIFSIGNALED(status)) {
5953 // The child exited because of a signal
5954 // The best value to return is 0x80 + signal number,
5955 // because that is what all Unix shells do, and because
5956 // it allows callers to distinguish between process exit and
5957 // process death by signal.
5958 return 0x80 + WTERMSIG(status);
5959 } else {
5960 // Unknown exit code; pass it through
5961 return status;
5962 }
5963 }
5964 }
5965
5966 // Get the default path to the core file
5967 // Returns the length of the string
5968 int os::get_core_path(char* buffer, size_t bufferSize) {
5969 /*
5970 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5971 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5972 */
5973 const int core_pattern_len = 129;
5974 char core_pattern[core_pattern_len] = {0};
5975
5976 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5977 if (core_pattern_file == -1) {
5978 return -1;
5979 }
5980
5981 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5982 ::close(core_pattern_file);
5983 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5984 return -1;
5985 }
5986 if (core_pattern[ret-1] == '\n') {
5987 core_pattern[ret-1] = '\0';
5988 } else {
5989 core_pattern[ret] = '\0';
5990 }
5991
5992 // Replace the %p in the core pattern with the process id. NOTE: we do this
5993 // only if the pattern doesn't start with "|", and we support only one %p in
5994 // the pattern.
5995 char *pid_pos = strstr(core_pattern, "%p");
5996 const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : ""; // skip over the "%p"
5997 int written;
5998
5999 if (core_pattern[0] == '/') {
6000 if (pid_pos != NULL) {
6001 *pid_pos = '\0';
6002 written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern,
6003 current_process_id(), tail);
6004 } else {
6005 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
6006 }
6007 } else {
6008 char cwd[PATH_MAX];
6009
6010 const char* p = get_current_directory(cwd, PATH_MAX);
6011 if (p == NULL) {
6012 return -1;
6013 }
6014
6015 if (core_pattern[0] == '|') {
6016 written = jio_snprintf(buffer, bufferSize,
6017 "\"%s\" (or dumping to %s/core.%d)",
6018 &core_pattern[1], p, current_process_id());
6019 } else if (pid_pos != NULL) {
6020 *pid_pos = '\0';
6021 written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern,
6022 current_process_id(), tail);
6023 } else {
6024 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6025 }
6026 }
6027
6028 if (written < 0) {
6029 return -1;
6030 }
6031
6032 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6033 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6034
6035 if (core_uses_pid_file != -1) {
6036 char core_uses_pid = 0;
6037 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6038 ::close(core_uses_pid_file);
6039
6040 if (core_uses_pid == '1') {
6041 jio_snprintf(buffer + written, bufferSize - written,
6042 ".%d", current_process_id());
6043 }
6044 }
6045 }
6046
6047 return strlen(buffer);
6048 }
6049
6050 bool os::start_debugging(char *buf, int buflen) {
6051 int len = (int)strlen(buf);
6052 char *p = &buf[len];
6053
6054 jio_snprintf(p, buflen-len,
6055 "\n\n"
6056 "Do you want to debug the problem?\n\n"
6057 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
6058 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
6059 "Otherwise, press RETURN to abort...",
6060 os::current_process_id(), os::current_process_id(),
6061 os::current_thread_id(), os::current_thread_id());
6062
6063 bool yes = os::message_box("Unexpected Error", buf);
6064
6065 if (yes) {
6066 // yes, user asked VM to launch debugger
6067 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
6068 os::current_process_id(), os::current_process_id());
6069
6070 os::fork_and_exec(buf);
6071 yes = false;
6072 }
6073 return yes;
6074 }
6075
6076
6077 // Java/Compiler thread:
6078 //
6079 // Low memory addresses
6080 // P0 +------------------------+
6081 // | |\ Java thread created by VM does not have glibc
6082 // | glibc guard page | - guard page, attached Java thread usually has
6083 // | |/ 1 glibc guard page.
6084 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6085 // | |\
6086 // | HotSpot Guard Pages | - red, yellow and reserved pages
6087 // | |/
6088 // +------------------------+ JavaThread::stack_reserved_zone_base()
6089 // | |\
6090 // | Normal Stack | -
6091 // | |/
6092 // P2 +------------------------+ Thread::stack_base()
6093 //
6094 // Non-Java thread:
6095 //
6096 // Low memory addresses
6097 // P0 +------------------------+
6098 // | |\
6099 // | glibc guard page | - usually 1 page
6100 // | |/
6101 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6102 // | |\
6103 // | Normal Stack | -
6104 // | |/
6105 // P2 +------------------------+ Thread::stack_base()
6106 //
6107 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
6108 // returned from pthread_attr_getstack().
6109 // ** Due to NPTL implementation error, linux takes the glibc guard page out
6110 // of the stack size given in pthread_attr. We work around this for
6111 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
6112 //
6113 #ifndef ZERO
6114 static void current_stack_region(address * bottom, size_t * size) {
6115 if (os::is_primordial_thread()) {
6116 // primordial thread needs special handling because pthread_getattr_np()
6117 // may return bogus value.
6118 *bottom = os::Linux::initial_thread_stack_bottom();
6119 *size = os::Linux::initial_thread_stack_size();
6120 } else {
6121 pthread_attr_t attr;
6122
6123 int rslt = pthread_getattr_np(pthread_self(), &attr);
6124
6125 // JVM needs to know exact stack location, abort if it fails
6126 if (rslt != 0) {
6127 if (rslt == ENOMEM) {
6128 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
6129 } else {
6130 fatal("pthread_getattr_np failed with error = %d", rslt);
6131 }
6132 }
6133
6134 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
6135 fatal("Cannot locate current stack attributes!");
6136 }
6137
6138 // Work around NPTL stack guard error.
6139 size_t guard_size = 0;
6140 rslt = pthread_attr_getguardsize(&attr, &guard_size);
6141 if (rslt != 0) {
6142 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
6143 }
6144 *bottom += guard_size;
6145 *size -= guard_size;
6146
6147 pthread_attr_destroy(&attr);
6148
6149 }
6150 assert(os::current_stack_pointer() >= *bottom &&
6151 os::current_stack_pointer() < *bottom + *size, "just checking");
6152 }
6153
6154 address os::current_stack_base() {
6155 address bottom;
6156 size_t size;
6157 current_stack_region(&bottom, &size);
6158 return (bottom + size);
6159 }
6160
6161 size_t os::current_stack_size() {
6162 // This stack size includes the usable stack and HotSpot guard pages
6163 // (for the threads that have Hotspot guard pages).
6164 address bottom;
6165 size_t size;
6166 current_stack_region(&bottom, &size);
6167 return size;
6168 }
6169 #endif
6170
6171 static inline struct timespec get_mtime(const char* filename) {
6172 struct stat st;
6173 int ret = os::stat(filename, &st);
6174 assert(ret == 0, "failed to stat() file '%s': %s", filename, os::strerror(errno));
6175 return st.st_mtim;
6176 }
6177
6178 int os::compare_file_modified_times(const char* file1, const char* file2) {
6179 struct timespec filetime1 = get_mtime(file1);
6180 struct timespec filetime2 = get_mtime(file2);
6181 int diff = filetime1.tv_sec - filetime2.tv_sec;
6182 if (diff == 0) {
6183 return filetime1.tv_nsec - filetime2.tv_nsec;
6184 }
6185 return diff;
6186 }
6187
6188 bool os::supports_map_sync() {
6189 return true;
6190 }
6191
6192 /////////////// Unit tests ///////////////
6193
6194 #ifndef PRODUCT
6195
6196 class TestReserveMemorySpecial : AllStatic {
6197 public:
6198 static void small_page_write(void* addr, size_t size) {
6199 size_t page_size = os::vm_page_size();
6200
6201 char* end = (char*)addr + size;
6202 for (char* p = (char*)addr; p < end; p += page_size) {
6203 *p = 1;
6204 }
6205 }
6206
6207 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6208 if (!UseHugeTLBFS) {
6209 return;
6210 }
6211
6212 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6213
6214 if (addr != NULL) {
6215 small_page_write(addr, size);
6216
6217 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6218 }
6219 }
6220
6221 static void test_reserve_memory_special_huge_tlbfs_only() {
6222 if (!UseHugeTLBFS) {
6223 return;
6224 }
6225
6226 size_t lp = os::large_page_size();
6227
6228 for (size_t size = lp; size <= lp * 10; size += lp) {
6229 test_reserve_memory_special_huge_tlbfs_only(size);
6230 }
6231 }
6232
6233 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6234 size_t lp = os::large_page_size();
6235 size_t ag = os::vm_allocation_granularity();
6236
6237 // sizes to test
6238 const size_t sizes[] = {
6239 lp, lp + ag, lp + lp / 2, lp * 2,
6240 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6241 lp * 10, lp * 10 + lp / 2
6242 };
6243 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6244
6245 // For each size/alignment combination, we test three scenarios:
6246 // 1) with req_addr == NULL
6247 // 2) with a non-null req_addr at which we expect to successfully allocate
6248 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6249 // expect the allocation to either fail or to ignore req_addr
6250
6251 // Pre-allocate two areas; they shall be as large as the largest allocation
6252 // and aligned to the largest alignment we will be testing.
6253 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6254 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6255 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6256 -1, 0);
6257 assert(mapping1 != MAP_FAILED, "should work");
6258
6259 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6260 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6261 -1, 0);
6262 assert(mapping2 != MAP_FAILED, "should work");
6263
6264 // Unmap the first mapping, but leave the second mapping intact: the first
6265 // mapping will serve as a value for a "good" req_addr (case 2). The second
6266 // mapping, still intact, as "bad" req_addr (case 3).
6267 ::munmap(mapping1, mapping_size);
6268
6269 // Case 1
6270 for (int i = 0; i < num_sizes; i++) {
6271 const size_t size = sizes[i];
6272 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6273 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6274 if (p != NULL) {
6275 assert(is_aligned(p, alignment), "must be");
6276 small_page_write(p, size);
6277 os::Linux::release_memory_special_huge_tlbfs(p, size);
6278 }
6279 }
6280 }
6281
6282 // Case 2
6283 for (int i = 0; i < num_sizes; i++) {
6284 const size_t size = sizes[i];
6285 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6286 char* const req_addr = align_up(mapping1, alignment);
6287 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6288 if (p != NULL) {
6289 assert(p == req_addr, "must be");
6290 small_page_write(p, size);
6291 os::Linux::release_memory_special_huge_tlbfs(p, size);
6292 }
6293 }
6294 }
6295
6296 // Case 3
6297 for (int i = 0; i < num_sizes; i++) {
6298 const size_t size = sizes[i];
6299 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6300 char* const req_addr = align_up(mapping2, alignment);
6301 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6302 // as the area around req_addr contains already existing mappings, the API should always
6303 // return NULL (as per contract, it cannot return another address)
6304 assert(p == NULL, "must be");
6305 }
6306 }
6307
6308 ::munmap(mapping2, mapping_size);
6309
6310 }
6311
6312 static void test_reserve_memory_special_huge_tlbfs() {
6313 if (!UseHugeTLBFS) {
6314 return;
6315 }
6316
6317 test_reserve_memory_special_huge_tlbfs_only();
6318 test_reserve_memory_special_huge_tlbfs_mixed();
6319 }
6320
6321 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6322 if (!UseSHM) {
6323 return;
6324 }
6325
6326 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6327
6328 if (addr != NULL) {
6329 assert(is_aligned(addr, alignment), "Check");
6330 assert(is_aligned(addr, os::large_page_size()), "Check");
6331
6332 small_page_write(addr, size);
6333
6334 os::Linux::release_memory_special_shm(addr, size);
6335 }
6336 }
6337
6338 static void test_reserve_memory_special_shm() {
6339 size_t lp = os::large_page_size();
6340 size_t ag = os::vm_allocation_granularity();
6341
6342 for (size_t size = ag; size < lp * 3; size += ag) {
6343 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6344 test_reserve_memory_special_shm(size, alignment);
6345 }
6346 }
6347 }
6348
6349 static void test() {
6350 test_reserve_memory_special_huge_tlbfs();
6351 test_reserve_memory_special_shm();
6352 }
6353 };
6354
6355 void TestReserveMemorySpecial_test() {
6356 TestReserveMemorySpecial::test();
6357 }
6358
6359 #endif