1 /*
2 * Copyright (c) 1999, 2019, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 // no precompiled headers
26 #include "jvm.h"
27 #include "classfile/classLoader.hpp"
28 #include "classfile/systemDictionary.hpp"
29 #include "classfile/vmSymbols.hpp"
30 #include "code/icBuffer.hpp"
31 #include "code/vtableStubs.hpp"
32 #include "compiler/compileBroker.hpp"
33 #include "compiler/disassembler.hpp"
34 #include "interpreter/interpreter.hpp"
35 #include "logging/log.hpp"
36 #include "logging/logStream.hpp"
37 #include "memory/allocation.inline.hpp"
38 #include "memory/filemap.hpp"
39 #include "oops/oop.inline.hpp"
40 #include "os_linux.inline.hpp"
41 #include "os_posix.inline.hpp"
42 #include "os_share_linux.hpp"
43 #include "osContainer_linux.hpp"
44 #include "prims/jniFastGetField.hpp"
45 #include "prims/jvm_misc.hpp"
46 #include "runtime/arguments.hpp"
47 #include "runtime/atomic.hpp"
48 #include "runtime/extendedPC.hpp"
49 #include "runtime/globals.hpp"
50 #include "runtime/interfaceSupport.inline.hpp"
51 #include "runtime/init.hpp"
52 #include "runtime/java.hpp"
53 #include "runtime/javaCalls.hpp"
54 #include "runtime/mutexLocker.hpp"
55 #include "runtime/objectMonitor.hpp"
56 #include "runtime/orderAccess.hpp"
57 #include "runtime/osThread.hpp"
58 #include "runtime/perfMemory.hpp"
59 #include "runtime/sharedRuntime.hpp"
60 #include "runtime/statSampler.hpp"
61 #include "runtime/stubRoutines.hpp"
62 #include "runtime/thread.inline.hpp"
63 #include "runtime/threadCritical.hpp"
64 #include "runtime/threadSMR.hpp"
65 #include "runtime/timer.hpp"
66 #include "runtime/vm_version.hpp"
67 #include "semaphore_posix.hpp"
68 #include "services/attachListener.hpp"
69 #include "services/memTracker.hpp"
70 #include "services/runtimeService.hpp"
71 #include "utilities/align.hpp"
72 #include "utilities/decoder.hpp"
73 #include "utilities/defaultStream.hpp"
74 #include "utilities/events.hpp"
75 #include "utilities/elfFile.hpp"
76 #include "utilities/growableArray.hpp"
77 #include "utilities/macros.hpp"
78 #include "utilities/vmError.hpp"
79
80 // put OS-includes here
81 # include <sys/types.h>
82 # include <sys/mman.h>
83 # include <sys/stat.h>
84 # include <sys/select.h>
85 # include <pthread.h>
86 # include <signal.h>
87 # include <errno.h>
88 # include <dlfcn.h>
89 # include <stdio.h>
90 # include <unistd.h>
91 # include <sys/resource.h>
92 # include <pthread.h>
93 # include <sys/stat.h>
94 # include <sys/time.h>
95 # include <sys/times.h>
96 # include <sys/utsname.h>
97 # include <sys/socket.h>
98 # include <sys/wait.h>
99 # include <pwd.h>
100 # include <poll.h>
101 # include <fcntl.h>
102 # include <string.h>
103 # include <syscall.h>
104 # include <sys/sysinfo.h>
105 # include <gnu/libc-version.h>
106 # include <sys/ipc.h>
107 # include <sys/shm.h>
108 # include <link.h>
109 # include <stdint.h>
110 # include <inttypes.h>
111 # include <sys/ioctl.h>
112
113 #ifndef _GNU_SOURCE
114 #define _GNU_SOURCE
115 #include <sched.h>
116 #undef _GNU_SOURCE
117 #else
118 #include <sched.h>
119 #endif
120
121 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
122 // getrusage() is prepared to handle the associated failure.
123 #ifndef RUSAGE_THREAD
124 #define RUSAGE_THREAD (1) /* only the calling thread */
125 #endif
126
127 #define MAX_PATH (2 * K)
128
129 #define MAX_SECS 100000000
130
131 // for timer info max values which include all bits
132 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
133
134 enum CoredumpFilterBit {
135 FILE_BACKED_PVT_BIT = 1 << 2,
136 FILE_BACKED_SHARED_BIT = 1 << 3,
137 LARGEPAGES_BIT = 1 << 6,
138 DAX_SHARED_BIT = 1 << 8
139 };
140
141 ////////////////////////////////////////////////////////////////////////////////
142 // global variables
143 julong os::Linux::_physical_memory = 0;
144
145 address os::Linux::_initial_thread_stack_bottom = NULL;
146 uintptr_t os::Linux::_initial_thread_stack_size = 0;
147
148 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
149 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
150 Mutex* os::Linux::_createThread_lock = NULL;
151 pthread_t os::Linux::_main_thread;
152 int os::Linux::_page_size = -1;
153 bool os::Linux::_supports_fast_thread_cpu_time = false;
154 uint32_t os::Linux::_os_version = 0;
155 const char * os::Linux::_glibc_version = NULL;
156 const char * os::Linux::_libpthread_version = NULL;
157
158 static jlong initial_time_count=0;
159
160 static int clock_tics_per_sec = 100;
161
162 // If the VM might have been created on the primordial thread, we need to resolve the
163 // primordial thread stack bounds and check if the current thread might be the
164 // primordial thread in places. If we know that the primordial thread is never used,
165 // such as when the VM was created by one of the standard java launchers, we can
166 // avoid this
167 static bool suppress_primordial_thread_resolution = false;
168
169 // For diagnostics to print a message once. see run_periodic_checks
170 static sigset_t check_signal_done;
171 static bool check_signals = true;
172
173 // Signal number used to suspend/resume a thread
174
175 // do not use any signal number less than SIGSEGV, see 4355769
176 static int SR_signum = SIGUSR2;
177 sigset_t SR_sigset;
178
179 // utility functions
180
181 static int SR_initialize();
182
183 julong os::available_memory() {
184 return Linux::available_memory();
185 }
186
187 julong os::Linux::available_memory() {
188 // values in struct sysinfo are "unsigned long"
189 struct sysinfo si;
190 julong avail_mem;
191
192 if (OSContainer::is_containerized()) {
193 jlong mem_limit, mem_usage;
194 if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
195 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
196 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
197 }
198 if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
199 log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
200 }
201 if (mem_limit > 0 && mem_usage > 0 ) {
202 avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
203 log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
204 return avail_mem;
205 }
206 }
207
208 sysinfo(&si);
209 avail_mem = (julong)si.freeram * si.mem_unit;
210 log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
211 return avail_mem;
212 }
213
214 julong os::physical_memory() {
215 jlong phys_mem = 0;
216 if (OSContainer::is_containerized()) {
217 jlong mem_limit;
218 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
219 log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
220 return mem_limit;
221 }
222 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
223 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
224 }
225
226 phys_mem = Linux::physical_memory();
227 log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
228 return phys_mem;
229 }
230
231 static uint64_t initial_total_ticks = 0;
232 static uint64_t initial_steal_ticks = 0;
233 static bool has_initial_tick_info = false;
234
235 static void next_line(FILE *f) {
236 int c;
237 do {
238 c = fgetc(f);
239 } while (c != '\n' && c != EOF);
240 }
241
242 bool os::Linux::get_tick_information(CPUPerfTicks* pticks, int which_logical_cpu) {
243 FILE* fh;
244 uint64_t userTicks, niceTicks, systemTicks, idleTicks;
245 // since at least kernel 2.6 : iowait: time waiting for I/O to complete
246 // irq: time servicing interrupts; softirq: time servicing softirqs
247 uint64_t iowTicks = 0, irqTicks = 0, sirqTicks= 0;
248 // steal (since kernel 2.6.11): time spent in other OS when running in a virtualized environment
249 uint64_t stealTicks = 0;
250 // guest (since kernel 2.6.24): time spent running a virtual CPU for guest OS under the
251 // control of the Linux kernel
252 uint64_t guestNiceTicks = 0;
253 int logical_cpu = -1;
254 const int required_tickinfo_count = (which_logical_cpu == -1) ? 4 : 5;
255 int n;
256
257 memset(pticks, 0, sizeof(CPUPerfTicks));
258
259 if ((fh = fopen("/proc/stat", "r")) == NULL) {
260 return false;
261 }
262
263 if (which_logical_cpu == -1) {
264 n = fscanf(fh, "cpu " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
265 UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
266 UINT64_FORMAT " " UINT64_FORMAT " ",
267 &userTicks, &niceTicks, &systemTicks, &idleTicks,
268 &iowTicks, &irqTicks, &sirqTicks,
269 &stealTicks, &guestNiceTicks);
270 } else {
271 // Move to next line
272 next_line(fh);
273
274 // find the line for requested cpu faster to just iterate linefeeds?
275 for (int i = 0; i < which_logical_cpu; i++) {
276 next_line(fh);
277 }
278
279 n = fscanf(fh, "cpu%u " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
280 UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
281 UINT64_FORMAT " " UINT64_FORMAT " ",
282 &logical_cpu, &userTicks, &niceTicks,
283 &systemTicks, &idleTicks, &iowTicks, &irqTicks, &sirqTicks,
284 &stealTicks, &guestNiceTicks);
285 }
286
287 fclose(fh);
288 if (n < required_tickinfo_count || logical_cpu != which_logical_cpu) {
289 return false;
290 }
291 pticks->used = userTicks + niceTicks;
292 pticks->usedKernel = systemTicks + irqTicks + sirqTicks;
293 pticks->total = userTicks + niceTicks + systemTicks + idleTicks +
294 iowTicks + irqTicks + sirqTicks + stealTicks + guestNiceTicks;
295
296 if (n > required_tickinfo_count + 3) {
297 pticks->steal = stealTicks;
298 pticks->has_steal_ticks = true;
299 } else {
300 pticks->steal = 0;
301 pticks->has_steal_ticks = false;
302 }
303
304 return true;
305 }
306
307 // Return true if user is running as root.
308
309 bool os::have_special_privileges() {
310 static bool init = false;
311 static bool privileges = false;
312 if (!init) {
313 privileges = (getuid() != geteuid()) || (getgid() != getegid());
314 init = true;
315 }
316 return privileges;
317 }
318
319
320 #ifndef SYS_gettid
321 // i386: 224, ia64: 1105, amd64: 186, sparc 143
322 #ifdef __ia64__
323 #define SYS_gettid 1105
324 #else
325 #ifdef __i386__
326 #define SYS_gettid 224
327 #else
328 #ifdef __amd64__
329 #define SYS_gettid 186
330 #else
331 #ifdef __sparc__
332 #define SYS_gettid 143
333 #else
334 #error define gettid for the arch
335 #endif
336 #endif
337 #endif
338 #endif
339 #endif
340
341
342 // pid_t gettid()
343 //
344 // Returns the kernel thread id of the currently running thread. Kernel
345 // thread id is used to access /proc.
346 pid_t os::Linux::gettid() {
347 int rslt = syscall(SYS_gettid);
348 assert(rslt != -1, "must be."); // old linuxthreads implementation?
349 return (pid_t)rslt;
350 }
351
352 // Most versions of linux have a bug where the number of processors are
353 // determined by looking at the /proc file system. In a chroot environment,
354 // the system call returns 1.
355 static bool unsafe_chroot_detected = false;
356 static const char *unstable_chroot_error = "/proc file system not found.\n"
357 "Java may be unstable running multithreaded in a chroot "
358 "environment on Linux when /proc filesystem is not mounted.";
359
360 void os::Linux::initialize_system_info() {
361 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
362 if (processor_count() == 1) {
363 pid_t pid = os::Linux::gettid();
364 char fname[32];
365 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
366 FILE *fp = fopen(fname, "r");
367 if (fp == NULL) {
368 unsafe_chroot_detected = true;
369 } else {
370 fclose(fp);
371 }
372 }
373 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
374 assert(processor_count() > 0, "linux error");
375 }
376
377 void os::init_system_properties_values() {
378 // The next steps are taken in the product version:
379 //
380 // Obtain the JAVA_HOME value from the location of libjvm.so.
381 // This library should be located at:
382 // <JAVA_HOME>/lib/{client|server}/libjvm.so.
383 //
384 // If "/jre/lib/" appears at the right place in the path, then we
385 // assume libjvm.so is installed in a JDK and we use this path.
386 //
387 // Otherwise exit with message: "Could not create the Java virtual machine."
388 //
389 // The following extra steps are taken in the debugging version:
390 //
391 // If "/jre/lib/" does NOT appear at the right place in the path
392 // instead of exit check for $JAVA_HOME environment variable.
393 //
394 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
395 // then we append a fake suffix "hotspot/libjvm.so" to this path so
396 // it looks like libjvm.so is installed there
397 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
398 //
399 // Otherwise exit.
400 //
401 // Important note: if the location of libjvm.so changes this
402 // code needs to be changed accordingly.
403
404 // See ld(1):
405 // The linker uses the following search paths to locate required
406 // shared libraries:
407 // 1: ...
408 // ...
409 // 7: The default directories, normally /lib and /usr/lib.
410 #ifndef OVERRIDE_LIBPATH
411 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
412 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
413 #else
414 #define DEFAULT_LIBPATH "/lib:/usr/lib"
415 #endif
416 #else
417 #define DEFAULT_LIBPATH OVERRIDE_LIBPATH
418 #endif
419
420 // Base path of extensions installed on the system.
421 #define SYS_EXT_DIR "/usr/java/packages"
422 #define EXTENSIONS_DIR "/lib/ext"
423
424 // Buffer that fits several sprintfs.
425 // Note that the space for the colon and the trailing null are provided
426 // by the nulls included by the sizeof operator.
427 const size_t bufsize =
428 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
429 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
430 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
431
432 // sysclasspath, java_home, dll_dir
433 {
434 char *pslash;
435 os::jvm_path(buf, bufsize);
436
437 // Found the full path to libjvm.so.
438 // Now cut the path to <java_home>/jre if we can.
439 pslash = strrchr(buf, '/');
440 if (pslash != NULL) {
441 *pslash = '\0'; // Get rid of /libjvm.so.
442 }
443 pslash = strrchr(buf, '/');
444 if (pslash != NULL) {
445 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
446 }
447 Arguments::set_dll_dir(buf);
448
449 if (pslash != NULL) {
450 pslash = strrchr(buf, '/');
451 if (pslash != NULL) {
452 *pslash = '\0'; // Get rid of /lib.
453 }
454 }
455 Arguments::set_java_home(buf);
456 if (!set_boot_path('/', ':')) {
457 vm_exit_during_initialization("Failed setting boot class path.", NULL);
458 }
459 }
460
461 // Where to look for native libraries.
462 //
463 // Note: Due to a legacy implementation, most of the library path
464 // is set in the launcher. This was to accomodate linking restrictions
465 // on legacy Linux implementations (which are no longer supported).
466 // Eventually, all the library path setting will be done here.
467 //
468 // However, to prevent the proliferation of improperly built native
469 // libraries, the new path component /usr/java/packages is added here.
470 // Eventually, all the library path setting will be done here.
471 {
472 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
473 // should always exist (until the legacy problem cited above is
474 // addressed).
475 const char *v = ::getenv("LD_LIBRARY_PATH");
476 const char *v_colon = ":";
477 if (v == NULL) { v = ""; v_colon = ""; }
478 // That's +1 for the colon and +1 for the trailing '\0'.
479 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
480 strlen(v) + 1 +
481 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
482 mtInternal);
483 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
484 Arguments::set_library_path(ld_library_path);
485 FREE_C_HEAP_ARRAY(char, ld_library_path);
486 }
487
488 // Extensions directories.
489 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
490 Arguments::set_ext_dirs(buf);
491
492 FREE_C_HEAP_ARRAY(char, buf);
493
494 #undef DEFAULT_LIBPATH
495 #undef SYS_EXT_DIR
496 #undef EXTENSIONS_DIR
497 }
498
499 ////////////////////////////////////////////////////////////////////////////////
500 // breakpoint support
501
502 void os::breakpoint() {
503 BREAKPOINT;
504 }
505
506 extern "C" void breakpoint() {
507 // use debugger to set breakpoint here
508 }
509
510 ////////////////////////////////////////////////////////////////////////////////
511 // signal support
512
513 debug_only(static bool signal_sets_initialized = false);
514 static sigset_t unblocked_sigs, vm_sigs;
515
516 void os::Linux::signal_sets_init() {
517 // Should also have an assertion stating we are still single-threaded.
518 assert(!signal_sets_initialized, "Already initialized");
519 // Fill in signals that are necessarily unblocked for all threads in
520 // the VM. Currently, we unblock the following signals:
521 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
522 // by -Xrs (=ReduceSignalUsage));
523 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
524 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
525 // the dispositions or masks wrt these signals.
526 // Programs embedding the VM that want to use the above signals for their
527 // own purposes must, at this time, use the "-Xrs" option to prevent
528 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
529 // (See bug 4345157, and other related bugs).
530 // In reality, though, unblocking these signals is really a nop, since
531 // these signals are not blocked by default.
532 sigemptyset(&unblocked_sigs);
533 sigaddset(&unblocked_sigs, SIGILL);
534 sigaddset(&unblocked_sigs, SIGSEGV);
535 sigaddset(&unblocked_sigs, SIGBUS);
536 sigaddset(&unblocked_sigs, SIGFPE);
537 #if defined(PPC64)
538 sigaddset(&unblocked_sigs, SIGTRAP);
539 #endif
540 sigaddset(&unblocked_sigs, SR_signum);
541
542 if (!ReduceSignalUsage) {
543 if (!os::Posix::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
544 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
545 }
546 if (!os::Posix::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
547 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
548 }
549 if (!os::Posix::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
550 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
551 }
552 }
553 // Fill in signals that are blocked by all but the VM thread.
554 sigemptyset(&vm_sigs);
555 if (!ReduceSignalUsage) {
556 sigaddset(&vm_sigs, BREAK_SIGNAL);
557 }
558 debug_only(signal_sets_initialized = true);
559
560 }
561
562 // These are signals that are unblocked while a thread is running Java.
563 // (For some reason, they get blocked by default.)
564 sigset_t* os::Linux::unblocked_signals() {
565 assert(signal_sets_initialized, "Not initialized");
566 return &unblocked_sigs;
567 }
568
569 // These are the signals that are blocked while a (non-VM) thread is
570 // running Java. Only the VM thread handles these signals.
571 sigset_t* os::Linux::vm_signals() {
572 assert(signal_sets_initialized, "Not initialized");
573 return &vm_sigs;
574 }
575
576 void os::Linux::hotspot_sigmask(Thread* thread) {
577
578 //Save caller's signal mask before setting VM signal mask
579 sigset_t caller_sigmask;
580 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
581
582 OSThread* osthread = thread->osthread();
583 osthread->set_caller_sigmask(caller_sigmask);
584
585 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
586
587 if (!ReduceSignalUsage) {
588 if (thread->is_VM_thread()) {
589 // Only the VM thread handles BREAK_SIGNAL ...
590 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
591 } else {
592 // ... all other threads block BREAK_SIGNAL
593 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
594 }
595 }
596 }
597
598 //////////////////////////////////////////////////////////////////////////////
599 // detecting pthread library
600
601 void os::Linux::libpthread_init() {
602 // Save glibc and pthread version strings.
603 #if !defined(_CS_GNU_LIBC_VERSION) || \
604 !defined(_CS_GNU_LIBPTHREAD_VERSION)
605 #error "glibc too old (< 2.3.2)"
606 #endif
607
608 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
609 assert(n > 0, "cannot retrieve glibc version");
610 char *str = (char *)malloc(n, mtInternal);
611 confstr(_CS_GNU_LIBC_VERSION, str, n);
612 os::Linux::set_glibc_version(str);
613
614 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
615 assert(n > 0, "cannot retrieve pthread version");
616 str = (char *)malloc(n, mtInternal);
617 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
618 os::Linux::set_libpthread_version(str);
619 }
620
621 /////////////////////////////////////////////////////////////////////////////
622 // thread stack expansion
623
624 // os::Linux::manually_expand_stack() takes care of expanding the thread
625 // stack. Note that this is normally not needed: pthread stacks allocate
626 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
627 // committed. Therefore it is not necessary to expand the stack manually.
628 //
629 // Manually expanding the stack was historically needed on LinuxThreads
630 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
631 // it is kept to deal with very rare corner cases:
632 //
633 // For one, user may run the VM on an own implementation of threads
634 // whose stacks are - like the old LinuxThreads - implemented using
635 // mmap(MAP_GROWSDOWN).
636 //
637 // Also, this coding may be needed if the VM is running on the primordial
638 // thread. Normally we avoid running on the primordial thread; however,
639 // user may still invoke the VM on the primordial thread.
640 //
641 // The following historical comment describes the details about running
642 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
643
644
645 // Force Linux kernel to expand current thread stack. If "bottom" is close
646 // to the stack guard, caller should block all signals.
647 //
648 // MAP_GROWSDOWN:
649 // A special mmap() flag that is used to implement thread stacks. It tells
650 // kernel that the memory region should extend downwards when needed. This
651 // allows early versions of LinuxThreads to only mmap the first few pages
652 // when creating a new thread. Linux kernel will automatically expand thread
653 // stack as needed (on page faults).
654 //
655 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
656 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
657 // region, it's hard to tell if the fault is due to a legitimate stack
658 // access or because of reading/writing non-exist memory (e.g. buffer
659 // overrun). As a rule, if the fault happens below current stack pointer,
660 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
661 // application (see Linux kernel fault.c).
662 //
663 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
664 // stack overflow detection.
665 //
666 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
667 // not use MAP_GROWSDOWN.
668 //
669 // To get around the problem and allow stack banging on Linux, we need to
670 // manually expand thread stack after receiving the SIGSEGV.
671 //
672 // There are two ways to expand thread stack to address "bottom", we used
673 // both of them in JVM before 1.5:
674 // 1. adjust stack pointer first so that it is below "bottom", and then
675 // touch "bottom"
676 // 2. mmap() the page in question
677 //
678 // Now alternate signal stack is gone, it's harder to use 2. For instance,
679 // if current sp is already near the lower end of page 101, and we need to
680 // call mmap() to map page 100, it is possible that part of the mmap() frame
681 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
682 // That will destroy the mmap() frame and cause VM to crash.
683 //
684 // The following code works by adjusting sp first, then accessing the "bottom"
685 // page to force a page fault. Linux kernel will then automatically expand the
686 // stack mapping.
687 //
688 // _expand_stack_to() assumes its frame size is less than page size, which
689 // should always be true if the function is not inlined.
690
691 static void NOINLINE _expand_stack_to(address bottom) {
692 address sp;
693 size_t size;
694 volatile char *p;
695
696 // Adjust bottom to point to the largest address within the same page, it
697 // gives us a one-page buffer if alloca() allocates slightly more memory.
698 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
699 bottom += os::Linux::page_size() - 1;
700
701 // sp might be slightly above current stack pointer; if that's the case, we
702 // will alloca() a little more space than necessary, which is OK. Don't use
703 // os::current_stack_pointer(), as its result can be slightly below current
704 // stack pointer, causing us to not alloca enough to reach "bottom".
705 sp = (address)&sp;
706
707 if (sp > bottom) {
708 size = sp - bottom;
709 p = (volatile char *)alloca(size);
710 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
711 p[0] = '\0';
712 }
713 }
714
715 void os::Linux::expand_stack_to(address bottom) {
716 _expand_stack_to(bottom);
717 }
718
719 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
720 assert(t!=NULL, "just checking");
721 assert(t->osthread()->expanding_stack(), "expand should be set");
722 assert(t->stack_base() != NULL, "stack_base was not initialized");
723
724 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
725 sigset_t mask_all, old_sigset;
726 sigfillset(&mask_all);
727 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
728 _expand_stack_to(addr);
729 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
730 return true;
731 }
732 return false;
733 }
734
735 //////////////////////////////////////////////////////////////////////////////
736 // create new thread
737
738 // Thread start routine for all newly created threads
739 static void *thread_native_entry(Thread *thread) {
740
741 thread->record_stack_base_and_size();
742
743 // Try to randomize the cache line index of hot stack frames.
744 // This helps when threads of the same stack traces evict each other's
745 // cache lines. The threads can be either from the same JVM instance, or
746 // from different JVM instances. The benefit is especially true for
747 // processors with hyperthreading technology.
748 static int counter = 0;
749 int pid = os::current_process_id();
750 alloca(((pid ^ counter++) & 7) * 128);
751
752 thread->initialize_thread_current();
753
754 OSThread* osthread = thread->osthread();
755 Monitor* sync = osthread->startThread_lock();
756
757 osthread->set_thread_id(os::current_thread_id());
758
759 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
760 os::current_thread_id(), (uintx) pthread_self());
761
762 if (UseNUMA) {
763 int lgrp_id = os::numa_get_group_id();
764 if (lgrp_id != -1) {
765 thread->set_lgrp_id(lgrp_id);
766 }
767 }
768 // initialize signal mask for this thread
769 os::Linux::hotspot_sigmask(thread);
770
771 // initialize floating point control register
772 os::Linux::init_thread_fpu_state();
773
774 // handshaking with parent thread
775 {
776 MutexLocker ml(sync, Mutex::_no_safepoint_check_flag);
777
778 // notify parent thread
779 osthread->set_state(INITIALIZED);
780 sync->notify_all();
781
782 // wait until os::start_thread()
783 while (osthread->get_state() == INITIALIZED) {
784 sync->wait_without_safepoint_check();
785 }
786 }
787
788 assert(osthread->pthread_id() != 0, "pthread_id was not set as expected");
789
790 // call one more level start routine
791 thread->call_run();
792
793 // Note: at this point the thread object may already have deleted itself.
794 // Prevent dereferencing it from here on out.
795 thread = NULL;
796
797 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
798 os::current_thread_id(), (uintx) pthread_self());
799
800 return 0;
801 }
802
803 // On Linux, glibc places static TLS blocks (for __thread variables) on
804 // the thread stack. This decreases the stack size actually available
805 // to threads.
806 //
807 // For large static TLS sizes, this may cause threads to malfunction due
808 // to insufficient stack space. This is a well-known issue in glibc:
809 // http://sourceware.org/bugzilla/show_bug.cgi?id=11787.
810 //
811 // As a workaround, we call a private but assumed-stable glibc function,
812 // __pthread_get_minstack() to obtain the minstack size and derive the
813 // static TLS size from it. We then increase the user requested stack
814 // size by this TLS size.
815 //
816 // Due to compatibility concerns, this size adjustment is opt-in and
817 // controlled via AdjustStackSizeForTLS.
818 typedef size_t (*GetMinStack)(const pthread_attr_t *attr);
819
820 GetMinStack _get_minstack_func = NULL;
821
822 static void get_minstack_init() {
823 _get_minstack_func =
824 (GetMinStack)dlsym(RTLD_DEFAULT, "__pthread_get_minstack");
825 log_info(os, thread)("Lookup of __pthread_get_minstack %s",
826 _get_minstack_func == NULL ? "failed" : "succeeded");
827 }
828
829 // Returns the size of the static TLS area glibc puts on thread stacks.
830 static size_t get_static_tls_area_size(const pthread_attr_t *attr) {
831 size_t tls_size = 0;
832 if (_get_minstack_func != NULL) {
833 // Obtain the pthread minstack size by calling __pthread_get_minstack.
834 size_t minstack_size = _get_minstack_func(attr);
835
836 // Remove non-TLS area size included in minstack size returned
837 // by __pthread_get_minstack() to get the static TLS size.
838 // In glibc before 2.27, minstack size includes guard_size.
839 // In glibc 2.27 and later, guard_size is automatically added
840 // to the stack size by pthread_create and is no longer included
841 // in minstack size. In both cases, the guard_size is taken into
842 // account, so there is no need to adjust the result for that.
843 //
844 // Although __pthread_get_minstack() is a private glibc function,
845 // it is expected to have a stable behavior across future glibc
846 // versions while glibc still allocats the static TLS blocks off
847 // the stack. Following is glibc 2.28 __pthread_get_minstack():
848 //
849 // size_t
850 // __pthread_get_minstack (const pthread_attr_t *attr)
851 // {
852 // return GLRO(dl_pagesize) + __static_tls_size + PTHREAD_STACK_MIN;
853 // }
854 //
855 //
856 // The following 'minstack_size > os::vm_page_size() + PTHREAD_STACK_MIN'
857 // if check is done for precaution.
858 if (minstack_size > (size_t)os::vm_page_size() + PTHREAD_STACK_MIN) {
859 tls_size = minstack_size - os::vm_page_size() - PTHREAD_STACK_MIN;
860 }
861 }
862
863 log_info(os, thread)("Stack size adjustment for TLS is " SIZE_FORMAT,
864 tls_size);
865 return tls_size;
866 }
867
868 bool os::create_thread(Thread* thread, ThreadType thr_type,
869 size_t req_stack_size) {
870 assert(thread->osthread() == NULL, "caller responsible");
871
872 // Allocate the OSThread object
873 OSThread* osthread = new OSThread(NULL, NULL);
874 if (osthread == NULL) {
875 return false;
876 }
877
878 // set the correct thread state
879 osthread->set_thread_type(thr_type);
880
881 // Initial state is ALLOCATED but not INITIALIZED
882 osthread->set_state(ALLOCATED);
883
884 thread->set_osthread(osthread);
885
886 // init thread attributes
887 pthread_attr_t attr;
888 pthread_attr_init(&attr);
889 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
890
891 // Calculate stack size if it's not specified by caller.
892 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
893 // In glibc versions prior to 2.7 the guard size mechanism
894 // is not implemented properly. The posix standard requires adding
895 // the size of the guard pages to the stack size, instead Linux
896 // takes the space out of 'stacksize'. Thus we adapt the requested
897 // stack_size by the size of the guard pages to mimick proper
898 // behaviour. However, be careful not to end up with a size
899 // of zero due to overflow. Don't add the guard page in that case.
900 size_t guard_size = os::Linux::default_guard_size(thr_type);
901 // Configure glibc guard page. Must happen before calling
902 // get_static_tls_area_size(), which uses the guard_size.
903 pthread_attr_setguardsize(&attr, guard_size);
904
905 size_t stack_adjust_size = 0;
906 if (AdjustStackSizeForTLS) {
907 // Adjust the stack_size for on-stack TLS - see get_static_tls_area_size().
908 stack_adjust_size += get_static_tls_area_size(&attr);
909 } else {
910 stack_adjust_size += guard_size;
911 }
912
913 stack_adjust_size = align_up(stack_adjust_size, os::vm_page_size());
914 if (stack_size <= SIZE_MAX - stack_adjust_size) {
915 stack_size += stack_adjust_size;
916 }
917 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
918
919 int status = pthread_attr_setstacksize(&attr, stack_size);
920 assert_status(status == 0, status, "pthread_attr_setstacksize");
921
922 ThreadState state;
923
924 {
925 pthread_t tid;
926 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
927
928 char buf[64];
929 if (ret == 0) {
930 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
931 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
932 } else {
933 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
934 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
935 // Log some OS information which might explain why creating the thread failed.
936 log_info(os, thread)("Number of threads approx. running in the VM: %d", Threads::number_of_threads());
937 LogStream st(Log(os, thread)::info());
938 os::Posix::print_rlimit_info(&st);
939 os::print_memory_info(&st);
940 os::Linux::print_proc_sys_info(&st);
941 os::Linux::print_container_info(&st);
942 }
943
944 pthread_attr_destroy(&attr);
945
946 if (ret != 0) {
947 // Need to clean up stuff we've allocated so far
948 thread->set_osthread(NULL);
949 delete osthread;
950 return false;
951 }
952
953 // Store pthread info into the OSThread
954 osthread->set_pthread_id(tid);
955
956 // Wait until child thread is either initialized or aborted
957 {
958 Monitor* sync_with_child = osthread->startThread_lock();
959 MutexLocker ml(sync_with_child, Mutex::_no_safepoint_check_flag);
960 while ((state = osthread->get_state()) == ALLOCATED) {
961 sync_with_child->wait_without_safepoint_check();
962 }
963 }
964 }
965
966 // Aborted due to thread limit being reached
967 if (state == ZOMBIE) {
968 thread->set_osthread(NULL);
969 delete osthread;
970 return false;
971 }
972
973 // The thread is returned suspended (in state INITIALIZED),
974 // and is started higher up in the call chain
975 assert(state == INITIALIZED, "race condition");
976 return true;
977 }
978
979 /////////////////////////////////////////////////////////////////////////////
980 // attach existing thread
981
982 // bootstrap the main thread
983 bool os::create_main_thread(JavaThread* thread) {
984 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
985 return create_attached_thread(thread);
986 }
987
988 bool os::create_attached_thread(JavaThread* thread) {
989 #ifdef ASSERT
990 thread->verify_not_published();
991 #endif
992
993 // Allocate the OSThread object
994 OSThread* osthread = new OSThread(NULL, NULL);
995
996 if (osthread == NULL) {
997 return false;
998 }
999
1000 // Store pthread info into the OSThread
1001 osthread->set_thread_id(os::Linux::gettid());
1002 osthread->set_pthread_id(::pthread_self());
1003
1004 // initialize floating point control register
1005 os::Linux::init_thread_fpu_state();
1006
1007 // Initial thread state is RUNNABLE
1008 osthread->set_state(RUNNABLE);
1009
1010 thread->set_osthread(osthread);
1011
1012 if (UseNUMA) {
1013 int lgrp_id = os::numa_get_group_id();
1014 if (lgrp_id != -1) {
1015 thread->set_lgrp_id(lgrp_id);
1016 }
1017 }
1018
1019 if (os::is_primordial_thread()) {
1020 // If current thread is primordial thread, its stack is mapped on demand,
1021 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1022 // the entire stack region to avoid SEGV in stack banging.
1023 // It is also useful to get around the heap-stack-gap problem on SuSE
1024 // kernel (see 4821821 for details). We first expand stack to the top
1025 // of yellow zone, then enable stack yellow zone (order is significant,
1026 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1027 // is no gap between the last two virtual memory regions.
1028
1029 JavaThread *jt = (JavaThread *)thread;
1030 address addr = jt->stack_reserved_zone_base();
1031 assert(addr != NULL, "initialization problem?");
1032 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1033
1034 osthread->set_expanding_stack();
1035 os::Linux::manually_expand_stack(jt, addr);
1036 osthread->clear_expanding_stack();
1037 }
1038
1039 // initialize signal mask for this thread
1040 // and save the caller's signal mask
1041 os::Linux::hotspot_sigmask(thread);
1042
1043 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
1044 os::current_thread_id(), (uintx) pthread_self());
1045
1046 return true;
1047 }
1048
1049 void os::pd_start_thread(Thread* thread) {
1050 OSThread * osthread = thread->osthread();
1051 assert(osthread->get_state() != INITIALIZED, "just checking");
1052 Monitor* sync_with_child = osthread->startThread_lock();
1053 MutexLocker ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1054 sync_with_child->notify();
1055 }
1056
1057 // Free Linux resources related to the OSThread
1058 void os::free_thread(OSThread* osthread) {
1059 assert(osthread != NULL, "osthread not set");
1060
1061 // We are told to free resources of the argument thread,
1062 // but we can only really operate on the current thread.
1063 assert(Thread::current()->osthread() == osthread,
1064 "os::free_thread but not current thread");
1065
1066 #ifdef ASSERT
1067 sigset_t current;
1068 sigemptyset(¤t);
1069 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
1070 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
1071 #endif
1072
1073 // Restore caller's signal mask
1074 sigset_t sigmask = osthread->caller_sigmask();
1075 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1076
1077 delete osthread;
1078 }
1079
1080 //////////////////////////////////////////////////////////////////////////////
1081 // primordial thread
1082
1083 // Check if current thread is the primordial thread, similar to Solaris thr_main.
1084 bool os::is_primordial_thread(void) {
1085 if (suppress_primordial_thread_resolution) {
1086 return false;
1087 }
1088 char dummy;
1089 // If called before init complete, thread stack bottom will be null.
1090 // Can be called if fatal error occurs before initialization.
1091 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
1092 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
1093 os::Linux::initial_thread_stack_size() != 0,
1094 "os::init did not locate primordial thread's stack region");
1095 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
1096 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
1097 os::Linux::initial_thread_stack_size()) {
1098 return true;
1099 } else {
1100 return false;
1101 }
1102 }
1103
1104 // Find the virtual memory area that contains addr
1105 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1106 FILE *fp = fopen("/proc/self/maps", "r");
1107 if (fp) {
1108 address low, high;
1109 while (!feof(fp)) {
1110 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1111 if (low <= addr && addr < high) {
1112 if (vma_low) *vma_low = low;
1113 if (vma_high) *vma_high = high;
1114 fclose(fp);
1115 return true;
1116 }
1117 }
1118 for (;;) {
1119 int ch = fgetc(fp);
1120 if (ch == EOF || ch == (int)'\n') break;
1121 }
1122 }
1123 fclose(fp);
1124 }
1125 return false;
1126 }
1127
1128 // Locate primordial thread stack. This special handling of primordial thread stack
1129 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1130 // bogus value for the primordial process thread. While the launcher has created
1131 // the VM in a new thread since JDK 6, we still have to allow for the use of the
1132 // JNI invocation API from a primordial thread.
1133 void os::Linux::capture_initial_stack(size_t max_size) {
1134
1135 // max_size is either 0 (which means accept OS default for thread stacks) or
1136 // a user-specified value known to be at least the minimum needed. If we
1137 // are actually on the primordial thread we can make it appear that we have a
1138 // smaller max_size stack by inserting the guard pages at that location. But we
1139 // cannot do anything to emulate a larger stack than what has been provided by
1140 // the OS or threading library. In fact if we try to use a stack greater than
1141 // what is set by rlimit then we will crash the hosting process.
1142
1143 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
1144 // If this is "unlimited" then it will be a huge value.
1145 struct rlimit rlim;
1146 getrlimit(RLIMIT_STACK, &rlim);
1147 size_t stack_size = rlim.rlim_cur;
1148
1149 // 6308388: a bug in ld.so will relocate its own .data section to the
1150 // lower end of primordial stack; reduce ulimit -s value a little bit
1151 // so we won't install guard page on ld.so's data section.
1152 // But ensure we don't underflow the stack size - allow 1 page spare
1153 if (stack_size >= (size_t)(3 * page_size())) {
1154 stack_size -= 2 * page_size();
1155 }
1156
1157 // Try to figure out where the stack base (top) is. This is harder.
1158 //
1159 // When an application is started, glibc saves the initial stack pointer in
1160 // a global variable "__libc_stack_end", which is then used by system
1161 // libraries. __libc_stack_end should be pretty close to stack top. The
1162 // variable is available since the very early days. However, because it is
1163 // a private interface, it could disappear in the future.
1164 //
1165 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1166 // to __libc_stack_end, it is very close to stack top, but isn't the real
1167 // stack top. Note that /proc may not exist if VM is running as a chroot
1168 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1169 // /proc/<pid>/stat could change in the future (though unlikely).
1170 //
1171 // We try __libc_stack_end first. If that doesn't work, look for
1172 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1173 // as a hint, which should work well in most cases.
1174
1175 uintptr_t stack_start;
1176
1177 // try __libc_stack_end first
1178 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1179 if (p && *p) {
1180 stack_start = *p;
1181 } else {
1182 // see if we can get the start_stack field from /proc/self/stat
1183 FILE *fp;
1184 int pid;
1185 char state;
1186 int ppid;
1187 int pgrp;
1188 int session;
1189 int nr;
1190 int tpgrp;
1191 unsigned long flags;
1192 unsigned long minflt;
1193 unsigned long cminflt;
1194 unsigned long majflt;
1195 unsigned long cmajflt;
1196 unsigned long utime;
1197 unsigned long stime;
1198 long cutime;
1199 long cstime;
1200 long prio;
1201 long nice;
1202 long junk;
1203 long it_real;
1204 uintptr_t start;
1205 uintptr_t vsize;
1206 intptr_t rss;
1207 uintptr_t rsslim;
1208 uintptr_t scodes;
1209 uintptr_t ecode;
1210 int i;
1211
1212 // Figure what the primordial thread stack base is. Code is inspired
1213 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1214 // followed by command name surrounded by parentheses, state, etc.
1215 char stat[2048];
1216 int statlen;
1217
1218 fp = fopen("/proc/self/stat", "r");
1219 if (fp) {
1220 statlen = fread(stat, 1, 2047, fp);
1221 stat[statlen] = '\0';
1222 fclose(fp);
1223
1224 // Skip pid and the command string. Note that we could be dealing with
1225 // weird command names, e.g. user could decide to rename java launcher
1226 // to "java 1.4.2 :)", then the stat file would look like
1227 // 1234 (java 1.4.2 :)) R ... ...
1228 // We don't really need to know the command string, just find the last
1229 // occurrence of ")" and then start parsing from there. See bug 4726580.
1230 char * s = strrchr(stat, ')');
1231
1232 i = 0;
1233 if (s) {
1234 // Skip blank chars
1235 do { s++; } while (s && isspace(*s));
1236
1237 #define _UFM UINTX_FORMAT
1238 #define _DFM INTX_FORMAT
1239
1240 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1241 // 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
1242 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,
1243 &state, // 3 %c
1244 &ppid, // 4 %d
1245 &pgrp, // 5 %d
1246 &session, // 6 %d
1247 &nr, // 7 %d
1248 &tpgrp, // 8 %d
1249 &flags, // 9 %lu
1250 &minflt, // 10 %lu
1251 &cminflt, // 11 %lu
1252 &majflt, // 12 %lu
1253 &cmajflt, // 13 %lu
1254 &utime, // 14 %lu
1255 &stime, // 15 %lu
1256 &cutime, // 16 %ld
1257 &cstime, // 17 %ld
1258 &prio, // 18 %ld
1259 &nice, // 19 %ld
1260 &junk, // 20 %ld
1261 &it_real, // 21 %ld
1262 &start, // 22 UINTX_FORMAT
1263 &vsize, // 23 UINTX_FORMAT
1264 &rss, // 24 INTX_FORMAT
1265 &rsslim, // 25 UINTX_FORMAT
1266 &scodes, // 26 UINTX_FORMAT
1267 &ecode, // 27 UINTX_FORMAT
1268 &stack_start); // 28 UINTX_FORMAT
1269 }
1270
1271 #undef _UFM
1272 #undef _DFM
1273
1274 if (i != 28 - 2) {
1275 assert(false, "Bad conversion from /proc/self/stat");
1276 // product mode - assume we are the primordial thread, good luck in the
1277 // embedded case.
1278 warning("Can't detect primordial thread stack location - bad conversion");
1279 stack_start = (uintptr_t) &rlim;
1280 }
1281 } else {
1282 // For some reason we can't open /proc/self/stat (for example, running on
1283 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1284 // most cases, so don't abort:
1285 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1286 stack_start = (uintptr_t) &rlim;
1287 }
1288 }
1289
1290 // Now we have a pointer (stack_start) very close to the stack top, the
1291 // next thing to do is to figure out the exact location of stack top. We
1292 // can find out the virtual memory area that contains stack_start by
1293 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1294 // and its upper limit is the real stack top. (again, this would fail if
1295 // running inside chroot, because /proc may not exist.)
1296
1297 uintptr_t stack_top;
1298 address low, high;
1299 if (find_vma((address)stack_start, &low, &high)) {
1300 // success, "high" is the true stack top. (ignore "low", because initial
1301 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1302 stack_top = (uintptr_t)high;
1303 } else {
1304 // failed, likely because /proc/self/maps does not exist
1305 warning("Can't detect primordial thread stack location - find_vma failed");
1306 // best effort: stack_start is normally within a few pages below the real
1307 // stack top, use it as stack top, and reduce stack size so we won't put
1308 // guard page outside stack.
1309 stack_top = stack_start;
1310 stack_size -= 16 * page_size();
1311 }
1312
1313 // stack_top could be partially down the page so align it
1314 stack_top = align_up(stack_top, page_size());
1315
1316 // Allowed stack value is minimum of max_size and what we derived from rlimit
1317 if (max_size > 0) {
1318 _initial_thread_stack_size = MIN2(max_size, stack_size);
1319 } else {
1320 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1321 // clamp it at 8MB as we do on Solaris
1322 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1323 }
1324 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1325 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1326
1327 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1328
1329 if (log_is_enabled(Info, os, thread)) {
1330 // See if we seem to be on primordial process thread
1331 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1332 uintptr_t(&rlim) < stack_top;
1333
1334 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1335 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1336 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1337 stack_top, intptr_t(_initial_thread_stack_bottom));
1338 }
1339 }
1340
1341 ////////////////////////////////////////////////////////////////////////////////
1342 // time support
1343
1344 #ifndef SUPPORTS_CLOCK_MONOTONIC
1345 #error "Build platform doesn't support clock_gettime and related functionality"
1346 #endif
1347
1348 // Time since start-up in seconds to a fine granularity.
1349 // Used by VMSelfDestructTimer and the MemProfiler.
1350 double os::elapsedTime() {
1351
1352 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1353 }
1354
1355 jlong os::elapsed_counter() {
1356 return javaTimeNanos() - initial_time_count;
1357 }
1358
1359 jlong os::elapsed_frequency() {
1360 return NANOSECS_PER_SEC; // nanosecond resolution
1361 }
1362
1363 bool os::supports_vtime() { return true; }
1364 bool os::enable_vtime() { return false; }
1365 bool os::vtime_enabled() { return false; }
1366
1367 double os::elapsedVTime() {
1368 struct rusage usage;
1369 int retval = getrusage(RUSAGE_THREAD, &usage);
1370 if (retval == 0) {
1371 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);
1372 } else {
1373 // better than nothing, but not much
1374 return elapsedTime();
1375 }
1376 }
1377
1378 jlong os::javaTimeMillis() {
1379 timeval time;
1380 int status = gettimeofday(&time, NULL);
1381 assert(status != -1, "linux error");
1382 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1383 }
1384
1385 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1386 timeval time;
1387 int status = gettimeofday(&time, NULL);
1388 assert(status != -1, "linux error");
1389 seconds = jlong(time.tv_sec);
1390 nanos = jlong(time.tv_usec) * 1000;
1391 }
1392
1393 void os::Linux::fast_thread_clock_init() {
1394 if (!UseLinuxPosixThreadCPUClocks) {
1395 return;
1396 }
1397 clockid_t clockid;
1398 struct timespec tp;
1399 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1400 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1401
1402 // Switch to using fast clocks for thread cpu time if
1403 // the clock_getres() returns 0 error code.
1404 // Note, that some kernels may support the current thread
1405 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1406 // returned by the pthread_getcpuclockid().
1407 // If the fast Posix clocks are supported then the clock_getres()
1408 // must return at least tp.tv_sec == 0 which means a resolution
1409 // better than 1 sec. This is extra check for reliability.
1410
1411 if (pthread_getcpuclockid_func &&
1412 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1413 os::Posix::clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1414 _supports_fast_thread_cpu_time = true;
1415 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1416 }
1417 }
1418
1419 jlong os::javaTimeNanos() {
1420 if (os::supports_monotonic_clock()) {
1421 struct timespec tp;
1422 int status = os::Posix::clock_gettime(CLOCK_MONOTONIC, &tp);
1423 assert(status == 0, "gettime error");
1424 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1425 return result;
1426 } else {
1427 timeval time;
1428 int status = gettimeofday(&time, NULL);
1429 assert(status != -1, "linux error");
1430 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1431 return 1000 * usecs;
1432 }
1433 }
1434
1435 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1436 if (os::supports_monotonic_clock()) {
1437 info_ptr->max_value = ALL_64_BITS;
1438
1439 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1440 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1441 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1442 } else {
1443 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1444 info_ptr->max_value = ALL_64_BITS;
1445
1446 // gettimeofday is a real time clock so it skips
1447 info_ptr->may_skip_backward = true;
1448 info_ptr->may_skip_forward = true;
1449 }
1450
1451 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1452 }
1453
1454 // Return the real, user, and system times in seconds from an
1455 // arbitrary fixed point in the past.
1456 bool os::getTimesSecs(double* process_real_time,
1457 double* process_user_time,
1458 double* process_system_time) {
1459 struct tms ticks;
1460 clock_t real_ticks = times(&ticks);
1461
1462 if (real_ticks == (clock_t) (-1)) {
1463 return false;
1464 } else {
1465 double ticks_per_second = (double) clock_tics_per_sec;
1466 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1467 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1468 *process_real_time = ((double) real_ticks) / ticks_per_second;
1469
1470 return true;
1471 }
1472 }
1473
1474
1475 char * os::local_time_string(char *buf, size_t buflen) {
1476 struct tm t;
1477 time_t long_time;
1478 time(&long_time);
1479 localtime_r(&long_time, &t);
1480 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1481 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1482 t.tm_hour, t.tm_min, t.tm_sec);
1483 return buf;
1484 }
1485
1486 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1487 return localtime_r(clock, res);
1488 }
1489
1490 ////////////////////////////////////////////////////////////////////////////////
1491 // runtime exit support
1492
1493 // Note: os::shutdown() might be called very early during initialization, or
1494 // called from signal handler. Before adding something to os::shutdown(), make
1495 // sure it is async-safe and can handle partially initialized VM.
1496 void os::shutdown() {
1497
1498 // allow PerfMemory to attempt cleanup of any persistent resources
1499 perfMemory_exit();
1500
1501 // needs to remove object in file system
1502 AttachListener::abort();
1503
1504 // flush buffered output, finish log files
1505 ostream_abort();
1506
1507 // Check for abort hook
1508 abort_hook_t abort_hook = Arguments::abort_hook();
1509 if (abort_hook != NULL) {
1510 abort_hook();
1511 }
1512
1513 }
1514
1515 // Note: os::abort() might be called very early during initialization, or
1516 // called from signal handler. Before adding something to os::abort(), make
1517 // sure it is async-safe and can handle partially initialized VM.
1518 void os::abort(bool dump_core, void* siginfo, const void* context) {
1519 os::shutdown();
1520 if (dump_core) {
1521 if (DumpPrivateMappingsInCore) {
1522 ClassLoader::close_jrt_image();
1523 }
1524 #ifndef PRODUCT
1525 fdStream out(defaultStream::output_fd());
1526 out.print_raw("Current thread is ");
1527 char buf[16];
1528 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1529 out.print_raw_cr(buf);
1530 out.print_raw_cr("Dumping core ...");
1531 #endif
1532 ::abort(); // dump core
1533 }
1534
1535 ::exit(1);
1536 }
1537
1538 // Die immediately, no exit hook, no abort hook, no cleanup.
1539 // Dump a core file, if possible, for debugging.
1540 void os::die() {
1541 if (TestUnresponsiveErrorHandler && !CreateCoredumpOnCrash) {
1542 // For TimeoutInErrorHandlingTest.java, we just kill the VM
1543 // and don't take the time to generate a core file.
1544 os::signal_raise(SIGKILL);
1545 } else {
1546 ::abort();
1547 }
1548 }
1549
1550 // thread_id is kernel thread id (similar to Solaris LWP id)
1551 intx os::current_thread_id() { return os::Linux::gettid(); }
1552 int os::current_process_id() {
1553 return ::getpid();
1554 }
1555
1556 // DLL functions
1557
1558 const char* os::dll_file_extension() { return ".so"; }
1559
1560 // This must be hard coded because it's the system's temporary
1561 // directory not the java application's temp directory, ala java.io.tmpdir.
1562 const char* os::get_temp_directory() { return "/tmp"; }
1563
1564 static bool file_exists(const char* filename) {
1565 struct stat statbuf;
1566 if (filename == NULL || strlen(filename) == 0) {
1567 return false;
1568 }
1569 return os::stat(filename, &statbuf) == 0;
1570 }
1571
1572 // check if addr is inside libjvm.so
1573 bool os::address_is_in_vm(address addr) {
1574 static address libjvm_base_addr;
1575 Dl_info dlinfo;
1576
1577 if (libjvm_base_addr == NULL) {
1578 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1579 libjvm_base_addr = (address)dlinfo.dli_fbase;
1580 }
1581 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1582 }
1583
1584 if (dladdr((void *)addr, &dlinfo) != 0) {
1585 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1586 }
1587
1588 return false;
1589 }
1590
1591 bool os::dll_address_to_function_name(address addr, char *buf,
1592 int buflen, int *offset,
1593 bool demangle) {
1594 // buf is not optional, but offset is optional
1595 assert(buf != NULL, "sanity check");
1596
1597 Dl_info dlinfo;
1598
1599 if (dladdr((void*)addr, &dlinfo) != 0) {
1600 // see if we have a matching symbol
1601 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1602 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1603 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1604 }
1605 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1606 return true;
1607 }
1608 // no matching symbol so try for just file info
1609 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1610 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1611 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1612 return true;
1613 }
1614 }
1615 }
1616
1617 buf[0] = '\0';
1618 if (offset != NULL) *offset = -1;
1619 return false;
1620 }
1621
1622 struct _address_to_library_name {
1623 address addr; // input : memory address
1624 size_t buflen; // size of fname
1625 char* fname; // output: library name
1626 address base; // library base addr
1627 };
1628
1629 static int address_to_library_name_callback(struct dl_phdr_info *info,
1630 size_t size, void *data) {
1631 int i;
1632 bool found = false;
1633 address libbase = NULL;
1634 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1635
1636 // iterate through all loadable segments
1637 for (i = 0; i < info->dlpi_phnum; i++) {
1638 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1639 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1640 // base address of a library is the lowest address of its loaded
1641 // segments.
1642 if (libbase == NULL || libbase > segbase) {
1643 libbase = segbase;
1644 }
1645 // see if 'addr' is within current segment
1646 if (segbase <= d->addr &&
1647 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1648 found = true;
1649 }
1650 }
1651 }
1652
1653 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1654 // so dll_address_to_library_name() can fall through to use dladdr() which
1655 // can figure out executable name from argv[0].
1656 if (found && info->dlpi_name && info->dlpi_name[0]) {
1657 d->base = libbase;
1658 if (d->fname) {
1659 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1660 }
1661 return 1;
1662 }
1663 return 0;
1664 }
1665
1666 bool os::dll_address_to_library_name(address addr, char* buf,
1667 int buflen, int* offset) {
1668 // buf is not optional, but offset is optional
1669 assert(buf != NULL, "sanity check");
1670
1671 Dl_info dlinfo;
1672 struct _address_to_library_name data;
1673
1674 // There is a bug in old glibc dladdr() implementation that it could resolve
1675 // to wrong library name if the .so file has a base address != NULL. Here
1676 // we iterate through the program headers of all loaded libraries to find
1677 // out which library 'addr' really belongs to. This workaround can be
1678 // removed once the minimum requirement for glibc is moved to 2.3.x.
1679 data.addr = addr;
1680 data.fname = buf;
1681 data.buflen = buflen;
1682 data.base = NULL;
1683 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1684
1685 if (rslt) {
1686 // buf already contains library name
1687 if (offset) *offset = addr - data.base;
1688 return true;
1689 }
1690 if (dladdr((void*)addr, &dlinfo) != 0) {
1691 if (dlinfo.dli_fname != NULL) {
1692 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1693 }
1694 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1695 *offset = addr - (address)dlinfo.dli_fbase;
1696 }
1697 return true;
1698 }
1699
1700 buf[0] = '\0';
1701 if (offset) *offset = -1;
1702 return false;
1703 }
1704
1705 // Loads .dll/.so and
1706 // in case of error it checks if .dll/.so was built for the
1707 // same architecture as Hotspot is running on
1708
1709
1710 // Remember the stack's state. The Linux dynamic linker will change
1711 // the stack to 'executable' at most once, so we must safepoint only once.
1712 bool os::Linux::_stack_is_executable = false;
1713
1714 // VM operation that loads a library. This is necessary if stack protection
1715 // of the Java stacks can be lost during loading the library. If we
1716 // do not stop the Java threads, they can stack overflow before the stacks
1717 // are protected again.
1718 class VM_LinuxDllLoad: public VM_Operation {
1719 private:
1720 const char *_filename;
1721 char *_ebuf;
1722 int _ebuflen;
1723 void *_lib;
1724 public:
1725 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1726 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1727 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1728 void doit() {
1729 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1730 os::Linux::_stack_is_executable = true;
1731 }
1732 void* loaded_library() { return _lib; }
1733 };
1734
1735 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1736 void * result = NULL;
1737 bool load_attempted = false;
1738
1739 // Check whether the library to load might change execution rights
1740 // of the stack. If they are changed, the protection of the stack
1741 // guard pages will be lost. We need a safepoint to fix this.
1742 //
1743 // See Linux man page execstack(8) for more info.
1744 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1745 if (!ElfFile::specifies_noexecstack(filename)) {
1746 if (!is_init_completed()) {
1747 os::Linux::_stack_is_executable = true;
1748 // This is OK - No Java threads have been created yet, and hence no
1749 // stack guard pages to fix.
1750 //
1751 // Dynamic loader will make all stacks executable after
1752 // this function returns, and will not do that again.
1753 assert(Threads::number_of_threads() == 0, "no Java threads should exist yet.");
1754 } else {
1755 warning("You have loaded library %s which might have disabled stack guard. "
1756 "The VM will try to fix the stack guard now.\n"
1757 "It's highly recommended that you fix the library with "
1758 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1759 filename);
1760
1761 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1762 JavaThread *jt = JavaThread::current();
1763 if (jt->thread_state() != _thread_in_native) {
1764 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1765 // that requires ExecStack. Cannot enter safe point. Let's give up.
1766 warning("Unable to fix stack guard. Giving up.");
1767 } else {
1768 if (!LoadExecStackDllInVMThread) {
1769 // This is for the case where the DLL has an static
1770 // constructor function that executes JNI code. We cannot
1771 // load such DLLs in the VMThread.
1772 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1773 }
1774
1775 ThreadInVMfromNative tiv(jt);
1776 debug_only(VMNativeEntryWrapper vew;)
1777
1778 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1779 VMThread::execute(&op);
1780 if (LoadExecStackDllInVMThread) {
1781 result = op.loaded_library();
1782 }
1783 load_attempted = true;
1784 }
1785 }
1786 }
1787 }
1788
1789 if (!load_attempted) {
1790 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1791 }
1792
1793 if (result != NULL) {
1794 // Successful loading
1795 return result;
1796 }
1797
1798 Elf32_Ehdr elf_head;
1799 int diag_msg_max_length=ebuflen-strlen(ebuf);
1800 char* diag_msg_buf=ebuf+strlen(ebuf);
1801
1802 if (diag_msg_max_length==0) {
1803 // No more space in ebuf for additional diagnostics message
1804 return NULL;
1805 }
1806
1807
1808 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1809
1810 if (file_descriptor < 0) {
1811 // Can't open library, report dlerror() message
1812 return NULL;
1813 }
1814
1815 bool failed_to_read_elf_head=
1816 (sizeof(elf_head)!=
1817 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1818
1819 ::close(file_descriptor);
1820 if (failed_to_read_elf_head) {
1821 // file i/o error - report dlerror() msg
1822 return NULL;
1823 }
1824
1825 typedef struct {
1826 Elf32_Half code; // Actual value as defined in elf.h
1827 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1828 unsigned char elf_class; // 32 or 64 bit
1829 unsigned char endianess; // MSB or LSB
1830 char* name; // String representation
1831 } arch_t;
1832
1833 #ifndef EM_486
1834 #define EM_486 6 /* Intel 80486 */
1835 #endif
1836 #ifndef EM_AARCH64
1837 #define EM_AARCH64 183 /* ARM AARCH64 */
1838 #endif
1839
1840 static const arch_t arch_array[]={
1841 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1842 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1843 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1844 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1845 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1846 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1847 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1848 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1849 #if defined(VM_LITTLE_ENDIAN)
1850 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1851 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1852 #else
1853 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1854 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1855 #endif
1856 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1857 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1858 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1859 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1860 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1861 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1862 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1863 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1864 };
1865
1866 #if (defined IA32)
1867 static Elf32_Half running_arch_code=EM_386;
1868 #elif (defined AMD64) || (defined X32)
1869 static Elf32_Half running_arch_code=EM_X86_64;
1870 #elif (defined IA64)
1871 static Elf32_Half running_arch_code=EM_IA_64;
1872 #elif (defined __sparc) && (defined _LP64)
1873 static Elf32_Half running_arch_code=EM_SPARCV9;
1874 #elif (defined __sparc) && (!defined _LP64)
1875 static Elf32_Half running_arch_code=EM_SPARC;
1876 #elif (defined __powerpc64__)
1877 static Elf32_Half running_arch_code=EM_PPC64;
1878 #elif (defined __powerpc__)
1879 static Elf32_Half running_arch_code=EM_PPC;
1880 #elif (defined AARCH64)
1881 static Elf32_Half running_arch_code=EM_AARCH64;
1882 #elif (defined ARM)
1883 static Elf32_Half running_arch_code=EM_ARM;
1884 #elif (defined S390)
1885 static Elf32_Half running_arch_code=EM_S390;
1886 #elif (defined ALPHA)
1887 static Elf32_Half running_arch_code=EM_ALPHA;
1888 #elif (defined MIPSEL)
1889 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1890 #elif (defined PARISC)
1891 static Elf32_Half running_arch_code=EM_PARISC;
1892 #elif (defined MIPS)
1893 static Elf32_Half running_arch_code=EM_MIPS;
1894 #elif (defined M68K)
1895 static Elf32_Half running_arch_code=EM_68K;
1896 #elif (defined SH)
1897 static Elf32_Half running_arch_code=EM_SH;
1898 #else
1899 #error Method os::dll_load requires that one of following is defined:\
1900 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1901 #endif
1902
1903 // Identify compatability class for VM's architecture and library's architecture
1904 // Obtain string descriptions for architectures
1905
1906 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1907 int running_arch_index=-1;
1908
1909 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1910 if (running_arch_code == arch_array[i].code) {
1911 running_arch_index = i;
1912 }
1913 if (lib_arch.code == arch_array[i].code) {
1914 lib_arch.compat_class = arch_array[i].compat_class;
1915 lib_arch.name = arch_array[i].name;
1916 }
1917 }
1918
1919 assert(running_arch_index != -1,
1920 "Didn't find running architecture code (running_arch_code) in arch_array");
1921 if (running_arch_index == -1) {
1922 // Even though running architecture detection failed
1923 // we may still continue with reporting dlerror() message
1924 return NULL;
1925 }
1926
1927 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1928 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1929 return NULL;
1930 }
1931
1932 #ifndef S390
1933 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1934 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1935 return NULL;
1936 }
1937 #endif // !S390
1938
1939 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1940 if (lib_arch.name!=NULL) {
1941 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1942 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1943 lib_arch.name, arch_array[running_arch_index].name);
1944 } else {
1945 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1946 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1947 lib_arch.code,
1948 arch_array[running_arch_index].name);
1949 }
1950 }
1951
1952 return NULL;
1953 }
1954
1955 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1956 int ebuflen) {
1957 void * result = ::dlopen(filename, RTLD_LAZY);
1958 if (result == NULL) {
1959 const char* error_report = ::dlerror();
1960 if (error_report == NULL) {
1961 error_report = "dlerror returned no error description";
1962 }
1963 if (ebuf != NULL && ebuflen > 0) {
1964 ::strncpy(ebuf, error_report, ebuflen-1);
1965 ebuf[ebuflen-1]='\0';
1966 }
1967 Events::log(NULL, "Loading shared library %s failed, %s", filename, error_report);
1968 } else {
1969 Events::log(NULL, "Loaded shared library %s", filename);
1970 }
1971 return result;
1972 }
1973
1974 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1975 int ebuflen) {
1976 void * result = NULL;
1977 if (LoadExecStackDllInVMThread) {
1978 result = dlopen_helper(filename, ebuf, ebuflen);
1979 }
1980
1981 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1982 // library that requires an executable stack, or which does not have this
1983 // stack attribute set, dlopen changes the stack attribute to executable. The
1984 // read protection of the guard pages gets lost.
1985 //
1986 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1987 // may have been queued at the same time.
1988
1989 if (!_stack_is_executable) {
1990 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
1991 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1992 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1993 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1994 warning("Attempt to reguard stack yellow zone failed.");
1995 }
1996 }
1997 }
1998 }
1999
2000 return result;
2001 }
2002
2003 void* os::dll_lookup(void* handle, const char* name) {
2004 void* res = dlsym(handle, name);
2005 return res;
2006 }
2007
2008 void* os::get_default_process_handle() {
2009 return (void*)::dlopen(NULL, RTLD_LAZY);
2010 }
2011
2012 static bool _print_ascii_file(const char* filename, outputStream* st, const char* hdr = NULL) {
2013 int fd = ::open(filename, O_RDONLY);
2014 if (fd == -1) {
2015 return false;
2016 }
2017
2018 if (hdr != NULL) {
2019 st->print_cr("%s", hdr);
2020 }
2021
2022 char buf[33];
2023 int bytes;
2024 buf[32] = '\0';
2025 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
2026 st->print_raw(buf, bytes);
2027 }
2028
2029 ::close(fd);
2030
2031 return true;
2032 }
2033
2034 void os::print_dll_info(outputStream *st) {
2035 st->print_cr("Dynamic libraries:");
2036
2037 char fname[32];
2038 pid_t pid = os::Linux::gettid();
2039
2040 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2041
2042 if (!_print_ascii_file(fname, st)) {
2043 st->print("Can not get library information for pid = %d\n", pid);
2044 }
2045 }
2046
2047 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2048 FILE *procmapsFile = NULL;
2049
2050 // Open the procfs maps file for the current process
2051 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2052 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2053 char line[PATH_MAX + 100];
2054
2055 // Read line by line from 'file'
2056 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2057 u8 base, top, offset, inode;
2058 char permissions[5];
2059 char device[6];
2060 char name[PATH_MAX + 1];
2061
2062 // Parse fields from line
2063 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %7s " INT64_FORMAT " %s",
2064 &base, &top, permissions, &offset, device, &inode, name);
2065
2066 // Filter by device id '00:00' so that we only get file system mapped files.
2067 if (strcmp(device, "00:00") != 0) {
2068
2069 // Call callback with the fields of interest
2070 if(callback(name, (address)base, (address)top, param)) {
2071 // Oops abort, callback aborted
2072 fclose(procmapsFile);
2073 return 1;
2074 }
2075 }
2076 }
2077 fclose(procmapsFile);
2078 }
2079 return 0;
2080 }
2081
2082 void os::print_os_info_brief(outputStream* st) {
2083 os::Linux::print_distro_info(st);
2084
2085 os::Posix::print_uname_info(st);
2086
2087 os::Linux::print_libversion_info(st);
2088
2089 }
2090
2091 void os::print_os_info(outputStream* st) {
2092 st->print("OS:");
2093
2094 os::Linux::print_distro_info(st);
2095
2096 os::Posix::print_uname_info(st);
2097
2098 // Print warning if unsafe chroot environment detected
2099 if (unsafe_chroot_detected) {
2100 st->print("WARNING!! ");
2101 st->print_cr("%s", unstable_chroot_error);
2102 }
2103
2104 os::Linux::print_libversion_info(st);
2105
2106 os::Posix::print_rlimit_info(st);
2107
2108 os::Posix::print_load_average(st);
2109
2110 os::Linux::print_full_memory_info(st);
2111
2112 os::Linux::print_proc_sys_info(st);
2113
2114 os::Linux::print_ld_preload_file(st);
2115
2116 os::Linux::print_container_info(st);
2117
2118 VM_Version::print_platform_virtualization_info(st);
2119
2120 os::Linux::print_steal_info(st);
2121 }
2122
2123 // Try to identify popular distros.
2124 // Most Linux distributions have a /etc/XXX-release file, which contains
2125 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2126 // file that also contains the OS version string. Some have more than one
2127 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2128 // /etc/redhat-release.), so the order is important.
2129 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2130 // their own specific XXX-release file as well as a redhat-release file.
2131 // Because of this the XXX-release file needs to be searched for before the
2132 // redhat-release file.
2133 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2134 // search for redhat-release / SuSE-release needs to be before lsb-release.
2135 // Since the lsb-release file is the new standard it needs to be searched
2136 // before the older style release files.
2137 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2138 // next to last resort. The os-release file is a new standard that contains
2139 // distribution information and the system-release file seems to be an old
2140 // standard that has been replaced by the lsb-release and os-release files.
2141 // Searching for the debian_version file is the last resort. It contains
2142 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2143 // "Debian " is printed before the contents of the debian_version file.
2144
2145 const char* distro_files[] = {
2146 "/etc/oracle-release",
2147 "/etc/mandriva-release",
2148 "/etc/mandrake-release",
2149 "/etc/sun-release",
2150 "/etc/redhat-release",
2151 "/etc/SuSE-release",
2152 "/etc/lsb-release",
2153 "/etc/turbolinux-release",
2154 "/etc/gentoo-release",
2155 "/etc/ltib-release",
2156 "/etc/angstrom-version",
2157 "/etc/system-release",
2158 "/etc/os-release",
2159 NULL };
2160
2161 void os::Linux::print_distro_info(outputStream* st) {
2162 for (int i = 0;; i++) {
2163 const char* file = distro_files[i];
2164 if (file == NULL) {
2165 break; // done
2166 }
2167 // If file prints, we found it.
2168 if (_print_ascii_file(file, st)) {
2169 return;
2170 }
2171 }
2172
2173 if (file_exists("/etc/debian_version")) {
2174 st->print("Debian ");
2175 _print_ascii_file("/etc/debian_version", st);
2176 } else {
2177 st->print("Linux");
2178 }
2179 st->cr();
2180 }
2181
2182 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2183 char buf[256];
2184 while (fgets(buf, sizeof(buf), fp)) {
2185 // Edit out extra stuff in expected format
2186 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2187 char* ptr = strstr(buf, "\""); // the name is in quotes
2188 if (ptr != NULL) {
2189 ptr++; // go beyond first quote
2190 char* nl = strchr(ptr, '\"');
2191 if (nl != NULL) *nl = '\0';
2192 strncpy(distro, ptr, length);
2193 } else {
2194 ptr = strstr(buf, "=");
2195 ptr++; // go beyond equals then
2196 char* nl = strchr(ptr, '\n');
2197 if (nl != NULL) *nl = '\0';
2198 strncpy(distro, ptr, length);
2199 }
2200 return;
2201 } else if (get_first_line) {
2202 char* nl = strchr(buf, '\n');
2203 if (nl != NULL) *nl = '\0';
2204 strncpy(distro, buf, length);
2205 return;
2206 }
2207 }
2208 // print last line and close
2209 char* nl = strchr(buf, '\n');
2210 if (nl != NULL) *nl = '\0';
2211 strncpy(distro, buf, length);
2212 }
2213
2214 static void parse_os_info(char* distro, size_t length, const char* file) {
2215 FILE* fp = fopen(file, "r");
2216 if (fp != NULL) {
2217 // if suse format, print out first line
2218 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2219 parse_os_info_helper(fp, distro, length, get_first_line);
2220 fclose(fp);
2221 }
2222 }
2223
2224 void os::get_summary_os_info(char* buf, size_t buflen) {
2225 for (int i = 0;; i++) {
2226 const char* file = distro_files[i];
2227 if (file == NULL) {
2228 break; // ran out of distro_files
2229 }
2230 if (file_exists(file)) {
2231 parse_os_info(buf, buflen, file);
2232 return;
2233 }
2234 }
2235 // special case for debian
2236 if (file_exists("/etc/debian_version")) {
2237 strncpy(buf, "Debian ", buflen);
2238 if (buflen > 7) {
2239 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2240 }
2241 } else {
2242 strncpy(buf, "Linux", buflen);
2243 }
2244 }
2245
2246 void os::Linux::print_libversion_info(outputStream* st) {
2247 // libc, pthread
2248 st->print("libc:");
2249 st->print("%s ", os::Linux::glibc_version());
2250 st->print("%s ", os::Linux::libpthread_version());
2251 st->cr();
2252 }
2253
2254 void os::Linux::print_proc_sys_info(outputStream* st) {
2255 st->cr();
2256 st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):");
2257 _print_ascii_file("/proc/sys/kernel/threads-max", st);
2258 st->cr();
2259 st->cr();
2260
2261 st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):");
2262 _print_ascii_file("/proc/sys/vm/max_map_count", st);
2263 st->cr();
2264 st->cr();
2265
2266 st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):");
2267 _print_ascii_file("/proc/sys/kernel/pid_max", st);
2268 st->cr();
2269 st->cr();
2270 }
2271
2272 void os::Linux::print_full_memory_info(outputStream* st) {
2273 st->print("\n/proc/meminfo:\n");
2274 _print_ascii_file("/proc/meminfo", st);
2275 st->cr();
2276 }
2277
2278 void os::Linux::print_ld_preload_file(outputStream* st) {
2279 _print_ascii_file("/etc/ld.so.preload", st, "\n/etc/ld.so.preload:");
2280 st->cr();
2281 }
2282
2283 void os::Linux::print_container_info(outputStream* st) {
2284 if (!OSContainer::is_containerized()) {
2285 return;
2286 }
2287
2288 st->print("container (cgroup) information:\n");
2289
2290 const char *p_ct = OSContainer::container_type();
2291 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "not supported");
2292
2293 char *p = OSContainer::cpu_cpuset_cpus();
2294 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "not supported");
2295 free(p);
2296
2297 p = OSContainer::cpu_cpuset_memory_nodes();
2298 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "not supported");
2299 free(p);
2300
2301 int i = OSContainer::active_processor_count();
2302 st->print("active_processor_count: ");
2303 if (i > 0) {
2304 st->print("%d\n", i);
2305 } else {
2306 st->print("not supported\n");
2307 }
2308
2309 i = OSContainer::cpu_quota();
2310 st->print("cpu_quota: ");
2311 if (i > 0) {
2312 st->print("%d\n", i);
2313 } else {
2314 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no quota");
2315 }
2316
2317 i = OSContainer::cpu_period();
2318 st->print("cpu_period: ");
2319 if (i > 0) {
2320 st->print("%d\n", i);
2321 } else {
2322 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no period");
2323 }
2324
2325 i = OSContainer::cpu_shares();
2326 st->print("cpu_shares: ");
2327 if (i > 0) {
2328 st->print("%d\n", i);
2329 } else {
2330 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no shares");
2331 }
2332
2333 jlong j = OSContainer::memory_limit_in_bytes();
2334 st->print("memory_limit_in_bytes: ");
2335 if (j > 0) {
2336 st->print(JLONG_FORMAT "\n", j);
2337 } else {
2338 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2339 }
2340
2341 j = OSContainer::memory_and_swap_limit_in_bytes();
2342 st->print("memory_and_swap_limit_in_bytes: ");
2343 if (j > 0) {
2344 st->print(JLONG_FORMAT "\n", j);
2345 } else {
2346 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2347 }
2348
2349 j = OSContainer::memory_soft_limit_in_bytes();
2350 st->print("memory_soft_limit_in_bytes: ");
2351 if (j > 0) {
2352 st->print(JLONG_FORMAT "\n", j);
2353 } else {
2354 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2355 }
2356
2357 j = OSContainer::OSContainer::memory_usage_in_bytes();
2358 st->print("memory_usage_in_bytes: ");
2359 if (j > 0) {
2360 st->print(JLONG_FORMAT "\n", j);
2361 } else {
2362 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2363 }
2364
2365 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2366 st->print("memory_max_usage_in_bytes: ");
2367 if (j > 0) {
2368 st->print(JLONG_FORMAT "\n", j);
2369 } else {
2370 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2371 }
2372 st->cr();
2373 }
2374
2375 void os::Linux::print_steal_info(outputStream* st) {
2376 if (has_initial_tick_info) {
2377 CPUPerfTicks pticks;
2378 bool res = os::Linux::get_tick_information(&pticks, -1);
2379
2380 if (res && pticks.has_steal_ticks) {
2381 uint64_t steal_ticks_difference = pticks.steal - initial_steal_ticks;
2382 uint64_t total_ticks_difference = pticks.total - initial_total_ticks;
2383 double steal_ticks_perc = 0.0;
2384 if (total_ticks_difference != 0) {
2385 steal_ticks_perc = (double) steal_ticks_difference / total_ticks_difference;
2386 }
2387 st->print_cr("Steal ticks since vm start: " UINT64_FORMAT, steal_ticks_difference);
2388 st->print_cr("Steal ticks percentage since vm start:%7.3f", steal_ticks_perc);
2389 }
2390 }
2391 }
2392
2393 void os::print_memory_info(outputStream* st) {
2394
2395 st->print("Memory:");
2396 st->print(" %dk page", os::vm_page_size()>>10);
2397
2398 // values in struct sysinfo are "unsigned long"
2399 struct sysinfo si;
2400 sysinfo(&si);
2401
2402 st->print(", physical " UINT64_FORMAT "k",
2403 os::physical_memory() >> 10);
2404 st->print("(" UINT64_FORMAT "k free)",
2405 os::available_memory() >> 10);
2406 st->print(", swap " UINT64_FORMAT "k",
2407 ((jlong)si.totalswap * si.mem_unit) >> 10);
2408 st->print("(" UINT64_FORMAT "k free)",
2409 ((jlong)si.freeswap * si.mem_unit) >> 10);
2410 st->cr();
2411 }
2412
2413 // Print the first "model name" line and the first "flags" line
2414 // that we find and nothing more. We assume "model name" comes
2415 // before "flags" so if we find a second "model name", then the
2416 // "flags" field is considered missing.
2417 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2418 #if defined(IA32) || defined(AMD64)
2419 // Other platforms have less repetitive cpuinfo files
2420 FILE *fp = fopen("/proc/cpuinfo", "r");
2421 if (fp) {
2422 while (!feof(fp)) {
2423 if (fgets(buf, buflen, fp)) {
2424 // Assume model name comes before flags
2425 bool model_name_printed = false;
2426 if (strstr(buf, "model name") != NULL) {
2427 if (!model_name_printed) {
2428 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2429 st->print_raw(buf);
2430 model_name_printed = true;
2431 } else {
2432 // model name printed but not flags? Odd, just return
2433 fclose(fp);
2434 return true;
2435 }
2436 }
2437 // print the flags line too
2438 if (strstr(buf, "flags") != NULL) {
2439 st->print_raw(buf);
2440 fclose(fp);
2441 return true;
2442 }
2443 }
2444 }
2445 fclose(fp);
2446 }
2447 #endif // x86 platforms
2448 return false;
2449 }
2450
2451 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2452 // Only print the model name if the platform provides this as a summary
2453 if (!print_model_name_and_flags(st, buf, buflen)) {
2454 st->print("\n/proc/cpuinfo:\n");
2455 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2456 st->print_cr(" <Not Available>");
2457 }
2458 }
2459 }
2460
2461 #if defined(AMD64) || defined(IA32) || defined(X32)
2462 const char* search_string = "model name";
2463 #elif defined(M68K)
2464 const char* search_string = "CPU";
2465 #elif defined(PPC64)
2466 const char* search_string = "cpu";
2467 #elif defined(S390)
2468 const char* search_string = "machine =";
2469 #elif defined(SPARC)
2470 const char* search_string = "cpu";
2471 #else
2472 const char* search_string = "Processor";
2473 #endif
2474
2475 // Parses the cpuinfo file for string representing the model name.
2476 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2477 FILE* fp = fopen("/proc/cpuinfo", "r");
2478 if (fp != NULL) {
2479 while (!feof(fp)) {
2480 char buf[256];
2481 if (fgets(buf, sizeof(buf), fp)) {
2482 char* start = strstr(buf, search_string);
2483 if (start != NULL) {
2484 char *ptr = start + strlen(search_string);
2485 char *end = buf + strlen(buf);
2486 while (ptr != end) {
2487 // skip whitespace and colon for the rest of the name.
2488 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2489 break;
2490 }
2491 ptr++;
2492 }
2493 if (ptr != end) {
2494 // reasonable string, get rid of newline and keep the rest
2495 char* nl = strchr(buf, '\n');
2496 if (nl != NULL) *nl = '\0';
2497 strncpy(cpuinfo, ptr, length);
2498 fclose(fp);
2499 return;
2500 }
2501 }
2502 }
2503 }
2504 fclose(fp);
2505 }
2506 // cpuinfo not found or parsing failed, just print generic string. The entire
2507 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2508 #if defined(AARCH64)
2509 strncpy(cpuinfo, "AArch64", length);
2510 #elif defined(AMD64)
2511 strncpy(cpuinfo, "x86_64", length);
2512 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2513 strncpy(cpuinfo, "ARM", length);
2514 #elif defined(IA32)
2515 strncpy(cpuinfo, "x86_32", length);
2516 #elif defined(IA64)
2517 strncpy(cpuinfo, "IA64", length);
2518 #elif defined(PPC)
2519 strncpy(cpuinfo, "PPC64", length);
2520 #elif defined(S390)
2521 strncpy(cpuinfo, "S390", length);
2522 #elif defined(SPARC)
2523 strncpy(cpuinfo, "sparcv9", length);
2524 #elif defined(ZERO_LIBARCH)
2525 strncpy(cpuinfo, ZERO_LIBARCH, length);
2526 #else
2527 strncpy(cpuinfo, "unknown", length);
2528 #endif
2529 }
2530
2531 static void print_signal_handler(outputStream* st, int sig,
2532 char* buf, size_t buflen);
2533
2534 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2535 st->print_cr("Signal Handlers:");
2536 print_signal_handler(st, SIGSEGV, buf, buflen);
2537 print_signal_handler(st, SIGBUS , buf, buflen);
2538 print_signal_handler(st, SIGFPE , buf, buflen);
2539 print_signal_handler(st, SIGPIPE, buf, buflen);
2540 print_signal_handler(st, SIGXFSZ, buf, buflen);
2541 print_signal_handler(st, SIGILL , buf, buflen);
2542 print_signal_handler(st, SR_signum, buf, buflen);
2543 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2544 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2545 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2546 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2547 #if defined(PPC64)
2548 print_signal_handler(st, SIGTRAP, buf, buflen);
2549 #endif
2550 }
2551
2552 static char saved_jvm_path[MAXPATHLEN] = {0};
2553
2554 // Find the full path to the current module, libjvm.so
2555 void os::jvm_path(char *buf, jint buflen) {
2556 // Error checking.
2557 if (buflen < MAXPATHLEN) {
2558 assert(false, "must use a large-enough buffer");
2559 buf[0] = '\0';
2560 return;
2561 }
2562 // Lazy resolve the path to current module.
2563 if (saved_jvm_path[0] != 0) {
2564 strcpy(buf, saved_jvm_path);
2565 return;
2566 }
2567
2568 char dli_fname[MAXPATHLEN];
2569 bool ret = dll_address_to_library_name(
2570 CAST_FROM_FN_PTR(address, os::jvm_path),
2571 dli_fname, sizeof(dli_fname), NULL);
2572 assert(ret, "cannot locate libjvm");
2573 char *rp = NULL;
2574 if (ret && dli_fname[0] != '\0') {
2575 rp = os::Posix::realpath(dli_fname, buf, buflen);
2576 }
2577 if (rp == NULL) {
2578 return;
2579 }
2580
2581 if (Arguments::sun_java_launcher_is_altjvm()) {
2582 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2583 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2584 // If "/jre/lib/" appears at the right place in the string, then
2585 // assume we are installed in a JDK and we're done. Otherwise, check
2586 // for a JAVA_HOME environment variable and fix up the path so it
2587 // looks like libjvm.so is installed there (append a fake suffix
2588 // hotspot/libjvm.so).
2589 const char *p = buf + strlen(buf) - 1;
2590 for (int count = 0; p > buf && count < 5; ++count) {
2591 for (--p; p > buf && *p != '/'; --p)
2592 /* empty */ ;
2593 }
2594
2595 if (strncmp(p, "/jre/lib/", 9) != 0) {
2596 // Look for JAVA_HOME in the environment.
2597 char* java_home_var = ::getenv("JAVA_HOME");
2598 if (java_home_var != NULL && java_home_var[0] != 0) {
2599 char* jrelib_p;
2600 int len;
2601
2602 // Check the current module name "libjvm.so".
2603 p = strrchr(buf, '/');
2604 if (p == NULL) {
2605 return;
2606 }
2607 assert(strstr(p, "/libjvm") == p, "invalid library name");
2608
2609 rp = os::Posix::realpath(java_home_var, buf, buflen);
2610 if (rp == NULL) {
2611 return;
2612 }
2613
2614 // determine if this is a legacy image or modules image
2615 // modules image doesn't have "jre" subdirectory
2616 len = strlen(buf);
2617 assert(len < buflen, "Ran out of buffer room");
2618 jrelib_p = buf + len;
2619 snprintf(jrelib_p, buflen-len, "/jre/lib");
2620 if (0 != access(buf, F_OK)) {
2621 snprintf(jrelib_p, buflen-len, "/lib");
2622 }
2623
2624 if (0 == access(buf, F_OK)) {
2625 // Use current module name "libjvm.so"
2626 len = strlen(buf);
2627 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2628 } else {
2629 // Go back to path of .so
2630 rp = os::Posix::realpath(dli_fname, buf, buflen);
2631 if (rp == NULL) {
2632 return;
2633 }
2634 }
2635 }
2636 }
2637 }
2638
2639 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2640 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2641 }
2642
2643 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2644 // no prefix required, not even "_"
2645 }
2646
2647 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2648 // no suffix required
2649 }
2650
2651 ////////////////////////////////////////////////////////////////////////////////
2652 // sun.misc.Signal support
2653
2654 static volatile jint sigint_count = 0;
2655
2656 static void UserHandler(int sig, void *siginfo, void *context) {
2657 // 4511530 - sem_post is serialized and handled by the manager thread. When
2658 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2659 // don't want to flood the manager thread with sem_post requests.
2660 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2661 return;
2662 }
2663
2664 // Ctrl-C is pressed during error reporting, likely because the error
2665 // handler fails to abort. Let VM die immediately.
2666 if (sig == SIGINT && VMError::is_error_reported()) {
2667 os::die();
2668 }
2669
2670 os::signal_notify(sig);
2671 }
2672
2673 void* os::user_handler() {
2674 return CAST_FROM_FN_PTR(void*, UserHandler);
2675 }
2676
2677 extern "C" {
2678 typedef void (*sa_handler_t)(int);
2679 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2680 }
2681
2682 void* os::signal(int signal_number, void* handler) {
2683 struct sigaction sigAct, oldSigAct;
2684
2685 sigfillset(&(sigAct.sa_mask));
2686 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2687 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2688
2689 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2690 // -1 means registration failed
2691 return (void *)-1;
2692 }
2693
2694 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2695 }
2696
2697 void os::signal_raise(int signal_number) {
2698 ::raise(signal_number);
2699 }
2700
2701 // The following code is moved from os.cpp for making this
2702 // code platform specific, which it is by its very nature.
2703
2704 // Will be modified when max signal is changed to be dynamic
2705 int os::sigexitnum_pd() {
2706 return NSIG;
2707 }
2708
2709 // a counter for each possible signal value
2710 static volatile jint pending_signals[NSIG+1] = { 0 };
2711
2712 // Linux(POSIX) specific hand shaking semaphore.
2713 static Semaphore* sig_sem = NULL;
2714 static PosixSemaphore sr_semaphore;
2715
2716 static void jdk_misc_signal_init() {
2717 // Initialize signal structures
2718 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2719
2720 // Initialize signal semaphore
2721 sig_sem = new Semaphore();
2722 }
2723
2724 void os::signal_notify(int sig) {
2725 if (sig_sem != NULL) {
2726 Atomic::inc(&pending_signals[sig]);
2727 sig_sem->signal();
2728 } else {
2729 // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2730 // initialization isn't called.
2731 assert(ReduceSignalUsage, "signal semaphore should be created");
2732 }
2733 }
2734
2735 static int check_pending_signals() {
2736 Atomic::store(0, &sigint_count);
2737 for (;;) {
2738 for (int i = 0; i < NSIG + 1; i++) {
2739 jint n = pending_signals[i];
2740 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2741 return i;
2742 }
2743 }
2744 JavaThread *thread = JavaThread::current();
2745 ThreadBlockInVM tbivm(thread);
2746
2747 bool threadIsSuspended;
2748 do {
2749 thread->set_suspend_equivalent();
2750 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2751 sig_sem->wait();
2752
2753 // were we externally suspended while we were waiting?
2754 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2755 if (threadIsSuspended) {
2756 // The semaphore has been incremented, but while we were waiting
2757 // another thread suspended us. We don't want to continue running
2758 // while suspended because that would surprise the thread that
2759 // suspended us.
2760 sig_sem->signal();
2761
2762 thread->java_suspend_self();
2763 }
2764 } while (threadIsSuspended);
2765 }
2766 }
2767
2768 int os::signal_wait() {
2769 return check_pending_signals();
2770 }
2771
2772 ////////////////////////////////////////////////////////////////////////////////
2773 // Virtual Memory
2774
2775 int os::vm_page_size() {
2776 // Seems redundant as all get out
2777 assert(os::Linux::page_size() != -1, "must call os::init");
2778 return os::Linux::page_size();
2779 }
2780
2781 // Solaris allocates memory by pages.
2782 int os::vm_allocation_granularity() {
2783 assert(os::Linux::page_size() != -1, "must call os::init");
2784 return os::Linux::page_size();
2785 }
2786
2787 // Rationale behind this function:
2788 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2789 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2790 // samples for JITted code. Here we create private executable mapping over the code cache
2791 // and then we can use standard (well, almost, as mapping can change) way to provide
2792 // info for the reporting script by storing timestamp and location of symbol
2793 void linux_wrap_code(char* base, size_t size) {
2794 static volatile jint cnt = 0;
2795
2796 if (!UseOprofile) {
2797 return;
2798 }
2799
2800 char buf[PATH_MAX+1];
2801 int num = Atomic::add(1, &cnt);
2802
2803 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2804 os::get_temp_directory(), os::current_process_id(), num);
2805 unlink(buf);
2806
2807 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2808
2809 if (fd != -1) {
2810 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2811 if (rv != (off_t)-1) {
2812 if (::write(fd, "", 1) == 1) {
2813 mmap(base, size,
2814 PROT_READ|PROT_WRITE|PROT_EXEC,
2815 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2816 }
2817 }
2818 ::close(fd);
2819 unlink(buf);
2820 }
2821 }
2822
2823 static bool recoverable_mmap_error(int err) {
2824 // See if the error is one we can let the caller handle. This
2825 // list of errno values comes from JBS-6843484. I can't find a
2826 // Linux man page that documents this specific set of errno
2827 // values so while this list currently matches Solaris, it may
2828 // change as we gain experience with this failure mode.
2829 switch (err) {
2830 case EBADF:
2831 case EINVAL:
2832 case ENOTSUP:
2833 // let the caller deal with these errors
2834 return true;
2835
2836 default:
2837 // Any remaining errors on this OS can cause our reserved mapping
2838 // to be lost. That can cause confusion where different data
2839 // structures think they have the same memory mapped. The worst
2840 // scenario is if both the VM and a library think they have the
2841 // same memory mapped.
2842 return false;
2843 }
2844 }
2845
2846 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2847 int err) {
2848 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2849 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2850 os::strerror(err), err);
2851 }
2852
2853 static void warn_fail_commit_memory(char* addr, size_t size,
2854 size_t alignment_hint, bool exec,
2855 int err) {
2856 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2857 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2858 alignment_hint, exec, os::strerror(err), err);
2859 }
2860
2861 // NOTE: Linux kernel does not really reserve the pages for us.
2862 // All it does is to check if there are enough free pages
2863 // left at the time of mmap(). This could be a potential
2864 // problem.
2865 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2866 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2867 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2868 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2869 if (res != (uintptr_t) MAP_FAILED) {
2870 if (UseNUMAInterleaving) {
2871 numa_make_global(addr, size);
2872 }
2873 return 0;
2874 }
2875
2876 int err = errno; // save errno from mmap() call above
2877
2878 if (!recoverable_mmap_error(err)) {
2879 warn_fail_commit_memory(addr, size, exec, err);
2880 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2881 }
2882
2883 return err;
2884 }
2885
2886 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2887 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2888 }
2889
2890 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2891 const char* mesg) {
2892 assert(mesg != NULL, "mesg must be specified");
2893 int err = os::Linux::commit_memory_impl(addr, size, exec);
2894 if (err != 0) {
2895 // the caller wants all commit errors to exit with the specified mesg:
2896 warn_fail_commit_memory(addr, size, exec, err);
2897 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2898 }
2899 }
2900
2901 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2902 #ifndef MAP_HUGETLB
2903 #define MAP_HUGETLB 0x40000
2904 #endif
2905
2906 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2907 #ifndef MADV_HUGEPAGE
2908 #define MADV_HUGEPAGE 14
2909 #endif
2910
2911 int os::Linux::commit_memory_impl(char* addr, size_t size,
2912 size_t alignment_hint, bool exec) {
2913 int err = os::Linux::commit_memory_impl(addr, size, exec);
2914 if (err == 0) {
2915 realign_memory(addr, size, alignment_hint);
2916 }
2917 return err;
2918 }
2919
2920 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2921 bool exec) {
2922 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2923 }
2924
2925 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2926 size_t alignment_hint, bool exec,
2927 const char* mesg) {
2928 assert(mesg != NULL, "mesg must be specified");
2929 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2930 if (err != 0) {
2931 // the caller wants all commit errors to exit with the specified mesg:
2932 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2933 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2934 }
2935 }
2936
2937 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2938 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2939 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2940 // be supported or the memory may already be backed by huge pages.
2941 ::madvise(addr, bytes, MADV_HUGEPAGE);
2942 }
2943 }
2944
2945 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2946 // This method works by doing an mmap over an existing mmaping and effectively discarding
2947 // the existing pages. However it won't work for SHM-based large pages that cannot be
2948 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2949 // small pages on top of the SHM segment. This method always works for small pages, so we
2950 // allow that in any case.
2951 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2952 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2953 }
2954 }
2955
2956 void os::numa_make_global(char *addr, size_t bytes) {
2957 Linux::numa_interleave_memory(addr, bytes);
2958 }
2959
2960 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2961 // bind policy to MPOL_PREFERRED for the current thread.
2962 #define USE_MPOL_PREFERRED 0
2963
2964 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2965 // To make NUMA and large pages more robust when both enabled, we need to ease
2966 // the requirements on where the memory should be allocated. MPOL_BIND is the
2967 // default policy and it will force memory to be allocated on the specified
2968 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2969 // the specified node, but will not force it. Using this policy will prevent
2970 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2971 // free large pages.
2972 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2973 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2974 }
2975
2976 bool os::numa_topology_changed() { return false; }
2977
2978 size_t os::numa_get_groups_num() {
2979 // Return just the number of nodes in which it's possible to allocate memory
2980 // (in numa terminology, configured nodes).
2981 return Linux::numa_num_configured_nodes();
2982 }
2983
2984 int os::numa_get_group_id() {
2985 int cpu_id = Linux::sched_getcpu();
2986 if (cpu_id != -1) {
2987 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2988 if (lgrp_id != -1) {
2989 return lgrp_id;
2990 }
2991 }
2992 return 0;
2993 }
2994
2995 int os::Linux::get_existing_num_nodes() {
2996 int node;
2997 int highest_node_number = Linux::numa_max_node();
2998 int num_nodes = 0;
2999
3000 // Get the total number of nodes in the system including nodes without memory.
3001 for (node = 0; node <= highest_node_number; node++) {
3002 if (is_node_in_existing_nodes(node)) {
3003 num_nodes++;
3004 }
3005 }
3006 return num_nodes;
3007 }
3008
3009 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
3010 int highest_node_number = Linux::numa_max_node();
3011 size_t i = 0;
3012
3013 // Map all node ids in which it is possible to allocate memory. Also nodes are
3014 // not always consecutively available, i.e. available from 0 to the highest
3015 // node number. If the nodes have been bound explicitly using numactl membind,
3016 // then allocate memory from those nodes only.
3017 for (int node = 0; node <= highest_node_number; node++) {
3018 if (Linux::is_node_in_bound_nodes((unsigned int)node)) {
3019 ids[i++] = node;
3020 }
3021 }
3022 return i;
3023 }
3024
3025 bool os::get_page_info(char *start, page_info* info) {
3026 return false;
3027 }
3028
3029 char *os::scan_pages(char *start, char* end, page_info* page_expected,
3030 page_info* page_found) {
3031 return end;
3032 }
3033
3034
3035 int os::Linux::sched_getcpu_syscall(void) {
3036 unsigned int cpu = 0;
3037 int retval = -1;
3038
3039 #if defined(IA32)
3040 #ifndef SYS_getcpu
3041 #define SYS_getcpu 318
3042 #endif
3043 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
3044 #elif defined(AMD64)
3045 // Unfortunately we have to bring all these macros here from vsyscall.h
3046 // to be able to compile on old linuxes.
3047 #define __NR_vgetcpu 2
3048 #define VSYSCALL_START (-10UL << 20)
3049 #define VSYSCALL_SIZE 1024
3050 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
3051 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
3052 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
3053 retval = vgetcpu(&cpu, NULL, NULL);
3054 #endif
3055
3056 return (retval == -1) ? retval : cpu;
3057 }
3058
3059 void os::Linux::sched_getcpu_init() {
3060 // sched_getcpu() should be in libc.
3061 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
3062 dlsym(RTLD_DEFAULT, "sched_getcpu")));
3063
3064 // If it's not, try a direct syscall.
3065 if (sched_getcpu() == -1) {
3066 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
3067 (void*)&sched_getcpu_syscall));
3068 }
3069
3070 if (sched_getcpu() == -1) {
3071 vm_exit_during_initialization("getcpu(2) system call not supported by kernel");
3072 }
3073 }
3074
3075 // Something to do with the numa-aware allocator needs these symbols
3076 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
3077 extern "C" JNIEXPORT void numa_error(char *where) { }
3078
3079 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
3080 // load symbol from base version instead.
3081 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
3082 void *f = dlvsym(handle, name, "libnuma_1.1");
3083 if (f == NULL) {
3084 f = dlsym(handle, name);
3085 }
3086 return f;
3087 }
3088
3089 // Handle request to load libnuma symbol version 1.2 (API v2) only.
3090 // Return NULL if the symbol is not defined in this particular version.
3091 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
3092 return dlvsym(handle, name, "libnuma_1.2");
3093 }
3094
3095 bool os::Linux::libnuma_init() {
3096 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
3097 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
3098 if (handle != NULL) {
3099 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
3100 libnuma_dlsym(handle, "numa_node_to_cpus")));
3101 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
3102 libnuma_dlsym(handle, "numa_max_node")));
3103 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
3104 libnuma_dlsym(handle, "numa_num_configured_nodes")));
3105 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
3106 libnuma_dlsym(handle, "numa_available")));
3107 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
3108 libnuma_dlsym(handle, "numa_tonode_memory")));
3109 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
3110 libnuma_dlsym(handle, "numa_interleave_memory")));
3111 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
3112 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
3113 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
3114 libnuma_dlsym(handle, "numa_set_bind_policy")));
3115 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
3116 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
3117 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
3118 libnuma_dlsym(handle, "numa_distance")));
3119 set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t,
3120 libnuma_v2_dlsym(handle, "numa_get_membind")));
3121 set_numa_get_interleave_mask(CAST_TO_FN_PTR(numa_get_interleave_mask_func_t,
3122 libnuma_v2_dlsym(handle, "numa_get_interleave_mask")));
3123
3124 if (numa_available() != -1) {
3125 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
3126 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
3127 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
3128 set_numa_interleave_bitmask(_numa_get_interleave_mask());
3129 set_numa_membind_bitmask(_numa_get_membind());
3130 // Create an index -> node mapping, since nodes are not always consecutive
3131 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3132 rebuild_nindex_to_node_map();
3133 // Create a cpu -> node mapping
3134 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3135 rebuild_cpu_to_node_map();
3136 return true;
3137 }
3138 }
3139 }
3140 return false;
3141 }
3142
3143 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
3144 // Creating guard page is very expensive. Java thread has HotSpot
3145 // guard pages, only enable glibc guard page for non-Java threads.
3146 // (Remember: compiler thread is a Java thread, too!)
3147 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
3148 }
3149
3150 void os::Linux::rebuild_nindex_to_node_map() {
3151 int highest_node_number = Linux::numa_max_node();
3152
3153 nindex_to_node()->clear();
3154 for (int node = 0; node <= highest_node_number; node++) {
3155 if (Linux::is_node_in_existing_nodes(node)) {
3156 nindex_to_node()->append(node);
3157 }
3158 }
3159 }
3160
3161 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
3162 // The table is later used in get_node_by_cpu().
3163 void os::Linux::rebuild_cpu_to_node_map() {
3164 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
3165 // in libnuma (possible values are starting from 16,
3166 // and continuing up with every other power of 2, but less
3167 // than the maximum number of CPUs supported by kernel), and
3168 // is a subject to change (in libnuma version 2 the requirements
3169 // are more reasonable) we'll just hardcode the number they use
3170 // in the library.
3171 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
3172
3173 size_t cpu_num = processor_count();
3174 size_t cpu_map_size = NCPUS / BitsPerCLong;
3175 size_t cpu_map_valid_size =
3176 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
3177
3178 cpu_to_node()->clear();
3179 cpu_to_node()->at_grow(cpu_num - 1);
3180
3181 size_t node_num = get_existing_num_nodes();
3182
3183 int distance = 0;
3184 int closest_distance = INT_MAX;
3185 int closest_node = 0;
3186 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
3187 for (size_t i = 0; i < node_num; i++) {
3188 // Check if node is configured (not a memory-less node). If it is not, find
3189 // the closest configured node. Check also if node is bound, i.e. it's allowed
3190 // to allocate memory from the node. If it's not allowed, map cpus in that node
3191 // to the closest node from which memory allocation is allowed.
3192 if (!is_node_in_configured_nodes(nindex_to_node()->at(i)) ||
3193 !is_node_in_bound_nodes(nindex_to_node()->at(i))) {
3194 closest_distance = INT_MAX;
3195 // Check distance from all remaining nodes in the system. Ignore distance
3196 // from itself, from another non-configured node, and from another non-bound
3197 // node.
3198 for (size_t m = 0; m < node_num; m++) {
3199 if (m != i &&
3200 is_node_in_configured_nodes(nindex_to_node()->at(m)) &&
3201 is_node_in_bound_nodes(nindex_to_node()->at(m))) {
3202 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3203 // If a closest node is found, update. There is always at least one
3204 // configured and bound node in the system so there is always at least
3205 // one node close.
3206 if (distance != 0 && distance < closest_distance) {
3207 closest_distance = distance;
3208 closest_node = nindex_to_node()->at(m);
3209 }
3210 }
3211 }
3212 } else {
3213 // Current node is already a configured node.
3214 closest_node = nindex_to_node()->at(i);
3215 }
3216
3217 // Get cpus from the original node and map them to the closest node. If node
3218 // is a configured node (not a memory-less node), then original node and
3219 // closest node are the same.
3220 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3221 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3222 if (cpu_map[j] != 0) {
3223 for (size_t k = 0; k < BitsPerCLong; k++) {
3224 if (cpu_map[j] & (1UL << k)) {
3225 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3226 }
3227 }
3228 }
3229 }
3230 }
3231 }
3232 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3233 }
3234
3235 int os::Linux::get_node_by_cpu(int cpu_id) {
3236 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3237 return cpu_to_node()->at(cpu_id);
3238 }
3239 return -1;
3240 }
3241
3242 GrowableArray<int>* os::Linux::_cpu_to_node;
3243 GrowableArray<int>* os::Linux::_nindex_to_node;
3244 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3245 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3246 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3247 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3248 os::Linux::numa_available_func_t os::Linux::_numa_available;
3249 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3250 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3251 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3252 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3253 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3254 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3255 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind;
3256 os::Linux::numa_get_interleave_mask_func_t os::Linux::_numa_get_interleave_mask;
3257 os::Linux::NumaAllocationPolicy os::Linux::_current_numa_policy;
3258 unsigned long* os::Linux::_numa_all_nodes;
3259 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3260 struct bitmask* os::Linux::_numa_nodes_ptr;
3261 struct bitmask* os::Linux::_numa_interleave_bitmask;
3262 struct bitmask* os::Linux::_numa_membind_bitmask;
3263
3264 bool os::pd_uncommit_memory(char* addr, size_t size) {
3265 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3266 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3267 return res != (uintptr_t) MAP_FAILED;
3268 }
3269
3270 static address get_stack_commited_bottom(address bottom, size_t size) {
3271 address nbot = bottom;
3272 address ntop = bottom + size;
3273
3274 size_t page_sz = os::vm_page_size();
3275 unsigned pages = size / page_sz;
3276
3277 unsigned char vec[1];
3278 unsigned imin = 1, imax = pages + 1, imid;
3279 int mincore_return_value = 0;
3280
3281 assert(imin <= imax, "Unexpected page size");
3282
3283 while (imin < imax) {
3284 imid = (imax + imin) / 2;
3285 nbot = ntop - (imid * page_sz);
3286
3287 // Use a trick with mincore to check whether the page is mapped or not.
3288 // mincore sets vec to 1 if page resides in memory and to 0 if page
3289 // is swapped output but if page we are asking for is unmapped
3290 // it returns -1,ENOMEM
3291 mincore_return_value = mincore(nbot, page_sz, vec);
3292
3293 if (mincore_return_value == -1) {
3294 // Page is not mapped go up
3295 // to find first mapped page
3296 if (errno != EAGAIN) {
3297 assert(errno == ENOMEM, "Unexpected mincore errno");
3298 imax = imid;
3299 }
3300 } else {
3301 // Page is mapped go down
3302 // to find first not mapped page
3303 imin = imid + 1;
3304 }
3305 }
3306
3307 nbot = nbot + page_sz;
3308
3309 // Adjust stack bottom one page up if last checked page is not mapped
3310 if (mincore_return_value == -1) {
3311 nbot = nbot + page_sz;
3312 }
3313
3314 return nbot;
3315 }
3316
3317 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3318 int mincore_return_value;
3319 const size_t stripe = 1024; // query this many pages each time
3320 unsigned char vec[stripe + 1];
3321 // set a guard
3322 vec[stripe] = 'X';
3323
3324 const size_t page_sz = os::vm_page_size();
3325 size_t pages = size / page_sz;
3326
3327 assert(is_aligned(start, page_sz), "Start address must be page aligned");
3328 assert(is_aligned(size, page_sz), "Size must be page aligned");
3329
3330 committed_start = NULL;
3331
3332 int loops = (pages + stripe - 1) / stripe;
3333 int committed_pages = 0;
3334 address loop_base = start;
3335 bool found_range = false;
3336
3337 for (int index = 0; index < loops && !found_range; index ++) {
3338 assert(pages > 0, "Nothing to do");
3339 int pages_to_query = (pages >= stripe) ? stripe : pages;
3340 pages -= pages_to_query;
3341
3342 // Get stable read
3343 while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3344
3345 // During shutdown, some memory goes away without properly notifying NMT,
3346 // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3347 // Bailout and return as not committed for now.
3348 if (mincore_return_value == -1 && errno == ENOMEM) {
3349 return false;
3350 }
3351
3352 assert(vec[stripe] == 'X', "overflow guard");
3353 assert(mincore_return_value == 0, "Range must be valid");
3354 // Process this stripe
3355 for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3356 if ((vec[vecIdx] & 0x01) == 0) { // not committed
3357 // End of current contiguous region
3358 if (committed_start != NULL) {
3359 found_range = true;
3360 break;
3361 }
3362 } else { // committed
3363 // Start of region
3364 if (committed_start == NULL) {
3365 committed_start = loop_base + page_sz * vecIdx;
3366 }
3367 committed_pages ++;
3368 }
3369 }
3370
3371 loop_base += pages_to_query * page_sz;
3372 }
3373
3374 if (committed_start != NULL) {
3375 assert(committed_pages > 0, "Must have committed region");
3376 assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3377 assert(committed_start >= start && committed_start < start + size, "Out of range");
3378 committed_size = page_sz * committed_pages;
3379 return true;
3380 } else {
3381 assert(committed_pages == 0, "Should not have committed region");
3382 return false;
3383 }
3384 }
3385
3386
3387 // Linux uses a growable mapping for the stack, and if the mapping for
3388 // the stack guard pages is not removed when we detach a thread the
3389 // stack cannot grow beyond the pages where the stack guard was
3390 // mapped. If at some point later in the process the stack expands to
3391 // that point, the Linux kernel cannot expand the stack any further
3392 // because the guard pages are in the way, and a segfault occurs.
3393 //
3394 // However, it's essential not to split the stack region by unmapping
3395 // a region (leaving a hole) that's already part of the stack mapping,
3396 // so if the stack mapping has already grown beyond the guard pages at
3397 // the time we create them, we have to truncate the stack mapping.
3398 // So, we need to know the extent of the stack mapping when
3399 // create_stack_guard_pages() is called.
3400
3401 // We only need this for stacks that are growable: at the time of
3402 // writing thread stacks don't use growable mappings (i.e. those
3403 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3404 // only applies to the main thread.
3405
3406 // If the (growable) stack mapping already extends beyond the point
3407 // where we're going to put our guard pages, truncate the mapping at
3408 // that point by munmap()ping it. This ensures that when we later
3409 // munmap() the guard pages we don't leave a hole in the stack
3410 // mapping. This only affects the main/primordial thread
3411
3412 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3413 if (os::is_primordial_thread()) {
3414 // As we manually grow stack up to bottom inside create_attached_thread(),
3415 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3416 // we don't need to do anything special.
3417 // Check it first, before calling heavy function.
3418 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3419 unsigned char vec[1];
3420
3421 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3422 // Fallback to slow path on all errors, including EAGAIN
3423 stack_extent = (uintptr_t) get_stack_commited_bottom(
3424 os::Linux::initial_thread_stack_bottom(),
3425 (size_t)addr - stack_extent);
3426 }
3427
3428 if (stack_extent < (uintptr_t)addr) {
3429 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3430 }
3431 }
3432
3433 return os::commit_memory(addr, size, !ExecMem);
3434 }
3435
3436 // If this is a growable mapping, remove the guard pages entirely by
3437 // munmap()ping them. If not, just call uncommit_memory(). This only
3438 // affects the main/primordial thread, but guard against future OS changes.
3439 // It's safe to always unmap guard pages for primordial thread because we
3440 // always place it right after end of the mapped region.
3441
3442 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3443 uintptr_t stack_extent, stack_base;
3444
3445 if (os::is_primordial_thread()) {
3446 return ::munmap(addr, size) == 0;
3447 }
3448
3449 return os::uncommit_memory(addr, size);
3450 }
3451
3452 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3453 // at 'requested_addr'. If there are existing memory mappings at the same
3454 // location, however, they will be overwritten. If 'fixed' is false,
3455 // 'requested_addr' is only treated as a hint, the return value may or
3456 // may not start from the requested address. Unlike Linux mmap(), this
3457 // function returns NULL to indicate failure.
3458 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3459 char * addr;
3460 int flags;
3461
3462 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3463 if (fixed) {
3464 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3465 flags |= MAP_FIXED;
3466 }
3467
3468 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3469 // touch an uncommitted page. Otherwise, the read/write might
3470 // succeed if we have enough swap space to back the physical page.
3471 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3472 flags, -1, 0);
3473
3474 return addr == MAP_FAILED ? NULL : addr;
3475 }
3476
3477 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3478 // (req_addr != NULL) or with a given alignment.
3479 // - bytes shall be a multiple of alignment.
3480 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3481 // - alignment sets the alignment at which memory shall be allocated.
3482 // It must be a multiple of allocation granularity.
3483 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3484 // req_addr or NULL.
3485 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3486
3487 size_t extra_size = bytes;
3488 if (req_addr == NULL && alignment > 0) {
3489 extra_size += alignment;
3490 }
3491
3492 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3493 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3494 -1, 0);
3495 if (start == MAP_FAILED) {
3496 start = NULL;
3497 } else {
3498 if (req_addr != NULL) {
3499 if (start != req_addr) {
3500 ::munmap(start, extra_size);
3501 start = NULL;
3502 }
3503 } else {
3504 char* const start_aligned = align_up(start, alignment);
3505 char* const end_aligned = start_aligned + bytes;
3506 char* const end = start + extra_size;
3507 if (start_aligned > start) {
3508 ::munmap(start, start_aligned - start);
3509 }
3510 if (end_aligned < end) {
3511 ::munmap(end_aligned, end - end_aligned);
3512 }
3513 start = start_aligned;
3514 }
3515 }
3516 return start;
3517 }
3518
3519 static int anon_munmap(char * addr, size_t size) {
3520 return ::munmap(addr, size) == 0;
3521 }
3522
3523 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3524 size_t alignment_hint) {
3525 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3526 }
3527
3528 bool os::pd_release_memory(char* addr, size_t size) {
3529 return anon_munmap(addr, size);
3530 }
3531
3532 static bool linux_mprotect(char* addr, size_t size, int prot) {
3533 // Linux wants the mprotect address argument to be page aligned.
3534 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3535
3536 // According to SUSv3, mprotect() should only be used with mappings
3537 // established by mmap(), and mmap() always maps whole pages. Unaligned
3538 // 'addr' likely indicates problem in the VM (e.g. trying to change
3539 // protection of malloc'ed or statically allocated memory). Check the
3540 // caller if you hit this assert.
3541 assert(addr == bottom, "sanity check");
3542
3543 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3544 Events::log(NULL, "Protecting memory [" INTPTR_FORMAT "," INTPTR_FORMAT "] with protection modes %x", p2i(bottom), p2i(bottom+size), prot);
3545 return ::mprotect(bottom, size, prot) == 0;
3546 }
3547
3548 // Set protections specified
3549 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3550 bool is_committed) {
3551 unsigned int p = 0;
3552 switch (prot) {
3553 case MEM_PROT_NONE: p = PROT_NONE; break;
3554 case MEM_PROT_READ: p = PROT_READ; break;
3555 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3556 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3557 default:
3558 ShouldNotReachHere();
3559 }
3560 // is_committed is unused.
3561 return linux_mprotect(addr, bytes, p);
3562 }
3563
3564 bool os::guard_memory(char* addr, size_t size) {
3565 return linux_mprotect(addr, size, PROT_NONE);
3566 }
3567
3568 bool os::unguard_memory(char* addr, size_t size) {
3569 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3570 }
3571
3572 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3573 size_t page_size) {
3574 bool result = false;
3575 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3576 MAP_ANONYMOUS|MAP_PRIVATE,
3577 -1, 0);
3578 if (p != MAP_FAILED) {
3579 void *aligned_p = align_up(p, page_size);
3580
3581 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3582
3583 munmap(p, page_size * 2);
3584 }
3585
3586 if (warn && !result) {
3587 warning("TransparentHugePages is not supported by the operating system.");
3588 }
3589
3590 return result;
3591 }
3592
3593 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3594 bool result = false;
3595 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3596 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3597 -1, 0);
3598
3599 if (p != MAP_FAILED) {
3600 // We don't know if this really is a huge page or not.
3601 FILE *fp = fopen("/proc/self/maps", "r");
3602 if (fp) {
3603 while (!feof(fp)) {
3604 char chars[257];
3605 long x = 0;
3606 if (fgets(chars, sizeof(chars), fp)) {
3607 if (sscanf(chars, "%lx-%*x", &x) == 1
3608 && x == (long)p) {
3609 if (strstr (chars, "hugepage")) {
3610 result = true;
3611 break;
3612 }
3613 }
3614 }
3615 }
3616 fclose(fp);
3617 }
3618 munmap(p, page_size);
3619 }
3620
3621 if (warn && !result) {
3622 warning("HugeTLBFS is not supported by the operating system.");
3623 }
3624
3625 return result;
3626 }
3627
3628 // From the coredump_filter documentation:
3629 //
3630 // - (bit 0) anonymous private memory
3631 // - (bit 1) anonymous shared memory
3632 // - (bit 2) file-backed private memory
3633 // - (bit 3) file-backed shared memory
3634 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3635 // effective only if the bit 2 is cleared)
3636 // - (bit 5) hugetlb private memory
3637 // - (bit 6) hugetlb shared memory
3638 // - (bit 7) dax private memory
3639 // - (bit 8) dax shared memory
3640 //
3641 static void set_coredump_filter(CoredumpFilterBit bit) {
3642 FILE *f;
3643 long cdm;
3644
3645 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3646 return;
3647 }
3648
3649 if (fscanf(f, "%lx", &cdm) != 1) {
3650 fclose(f);
3651 return;
3652 }
3653
3654 long saved_cdm = cdm;
3655 rewind(f);
3656 cdm |= bit;
3657
3658 if (cdm != saved_cdm) {
3659 fprintf(f, "%#lx", cdm);
3660 }
3661
3662 fclose(f);
3663 }
3664
3665 // Large page support
3666
3667 static size_t _large_page_size = 0;
3668
3669 size_t os::Linux::find_large_page_size() {
3670 size_t large_page_size = 0;
3671
3672 // large_page_size on Linux is used to round up heap size. x86 uses either
3673 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3674 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3675 // page as large as 256M.
3676 //
3677 // Here we try to figure out page size by parsing /proc/meminfo and looking
3678 // for a line with the following format:
3679 // Hugepagesize: 2048 kB
3680 //
3681 // If we can't determine the value (e.g. /proc is not mounted, or the text
3682 // format has been changed), we'll use the largest page size supported by
3683 // the processor.
3684
3685 #ifndef ZERO
3686 large_page_size =
3687 AARCH64_ONLY(2 * M)
3688 AMD64_ONLY(2 * M)
3689 ARM32_ONLY(2 * M)
3690 IA32_ONLY(4 * M)
3691 IA64_ONLY(256 * M)
3692 PPC_ONLY(4 * M)
3693 S390_ONLY(1 * M)
3694 SPARC_ONLY(4 * M);
3695 #endif // ZERO
3696
3697 FILE *fp = fopen("/proc/meminfo", "r");
3698 if (fp) {
3699 while (!feof(fp)) {
3700 int x = 0;
3701 char buf[16];
3702 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3703 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3704 large_page_size = x * K;
3705 break;
3706 }
3707 } else {
3708 // skip to next line
3709 for (;;) {
3710 int ch = fgetc(fp);
3711 if (ch == EOF || ch == (int)'\n') break;
3712 }
3713 }
3714 }
3715 fclose(fp);
3716 }
3717
3718 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3719 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3720 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3721 proper_unit_for_byte_size(large_page_size));
3722 }
3723
3724 return large_page_size;
3725 }
3726
3727 size_t os::Linux::setup_large_page_size() {
3728 _large_page_size = Linux::find_large_page_size();
3729 const size_t default_page_size = (size_t)Linux::page_size();
3730 if (_large_page_size > default_page_size) {
3731 _page_sizes[0] = _large_page_size;
3732 _page_sizes[1] = default_page_size;
3733 _page_sizes[2] = 0;
3734 }
3735
3736 return _large_page_size;
3737 }
3738
3739 bool os::Linux::setup_large_page_type(size_t page_size) {
3740 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3741 FLAG_IS_DEFAULT(UseSHM) &&
3742 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3743
3744 // The type of large pages has not been specified by the user.
3745
3746 // Try UseHugeTLBFS and then UseSHM.
3747 UseHugeTLBFS = UseSHM = true;
3748
3749 // Don't try UseTransparentHugePages since there are known
3750 // performance issues with it turned on. This might change in the future.
3751 UseTransparentHugePages = false;
3752 }
3753
3754 if (UseTransparentHugePages) {
3755 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3756 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3757 UseHugeTLBFS = false;
3758 UseSHM = false;
3759 return true;
3760 }
3761 UseTransparentHugePages = false;
3762 }
3763
3764 if (UseHugeTLBFS) {
3765 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3766 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3767 UseSHM = false;
3768 return true;
3769 }
3770 UseHugeTLBFS = false;
3771 }
3772
3773 return UseSHM;
3774 }
3775
3776 void os::large_page_init() {
3777 if (!UseLargePages &&
3778 !UseTransparentHugePages &&
3779 !UseHugeTLBFS &&
3780 !UseSHM) {
3781 // Not using large pages.
3782 return;
3783 }
3784
3785 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3786 // The user explicitly turned off large pages.
3787 // Ignore the rest of the large pages flags.
3788 UseTransparentHugePages = false;
3789 UseHugeTLBFS = false;
3790 UseSHM = false;
3791 return;
3792 }
3793
3794 size_t large_page_size = Linux::setup_large_page_size();
3795 UseLargePages = Linux::setup_large_page_type(large_page_size);
3796
3797 set_coredump_filter(LARGEPAGES_BIT);
3798 }
3799
3800 #ifndef SHM_HUGETLB
3801 #define SHM_HUGETLB 04000
3802 #endif
3803
3804 #define shm_warning_format(format, ...) \
3805 do { \
3806 if (UseLargePages && \
3807 (!FLAG_IS_DEFAULT(UseLargePages) || \
3808 !FLAG_IS_DEFAULT(UseSHM) || \
3809 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3810 warning(format, __VA_ARGS__); \
3811 } \
3812 } while (0)
3813
3814 #define shm_warning(str) shm_warning_format("%s", str)
3815
3816 #define shm_warning_with_errno(str) \
3817 do { \
3818 int err = errno; \
3819 shm_warning_format(str " (error = %d)", err); \
3820 } while (0)
3821
3822 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3823 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3824
3825 if (!is_aligned(alignment, SHMLBA)) {
3826 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3827 return NULL;
3828 }
3829
3830 // To ensure that we get 'alignment' aligned memory from shmat,
3831 // we pre-reserve aligned virtual memory and then attach to that.
3832
3833 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3834 if (pre_reserved_addr == NULL) {
3835 // Couldn't pre-reserve aligned memory.
3836 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3837 return NULL;
3838 }
3839
3840 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3841 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3842
3843 if ((intptr_t)addr == -1) {
3844 int err = errno;
3845 shm_warning_with_errno("Failed to attach shared memory.");
3846
3847 assert(err != EACCES, "Unexpected error");
3848 assert(err != EIDRM, "Unexpected error");
3849 assert(err != EINVAL, "Unexpected error");
3850
3851 // Since we don't know if the kernel unmapped the pre-reserved memory area
3852 // we can't unmap it, since that would potentially unmap memory that was
3853 // mapped from other threads.
3854 return NULL;
3855 }
3856
3857 return addr;
3858 }
3859
3860 static char* shmat_at_address(int shmid, char* req_addr) {
3861 if (!is_aligned(req_addr, SHMLBA)) {
3862 assert(false, "Requested address needs to be SHMLBA aligned");
3863 return NULL;
3864 }
3865
3866 char* addr = (char*)shmat(shmid, req_addr, 0);
3867
3868 if ((intptr_t)addr == -1) {
3869 shm_warning_with_errno("Failed to attach shared memory.");
3870 return NULL;
3871 }
3872
3873 return addr;
3874 }
3875
3876 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3877 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3878 if (req_addr != NULL) {
3879 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3880 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3881 return shmat_at_address(shmid, req_addr);
3882 }
3883
3884 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3885 // return large page size aligned memory addresses when req_addr == NULL.
3886 // However, if the alignment is larger than the large page size, we have
3887 // to manually ensure that the memory returned is 'alignment' aligned.
3888 if (alignment > os::large_page_size()) {
3889 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3890 return shmat_with_alignment(shmid, bytes, alignment);
3891 } else {
3892 return shmat_at_address(shmid, NULL);
3893 }
3894 }
3895
3896 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3897 char* req_addr, bool exec) {
3898 // "exec" is passed in but not used. Creating the shared image for
3899 // the code cache doesn't have an SHM_X executable permission to check.
3900 assert(UseLargePages && UseSHM, "only for SHM large pages");
3901 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3902 assert(is_aligned(req_addr, alignment), "Unaligned address");
3903
3904 if (!is_aligned(bytes, os::large_page_size())) {
3905 return NULL; // Fallback to small pages.
3906 }
3907
3908 // Create a large shared memory region to attach to based on size.
3909 // Currently, size is the total size of the heap.
3910 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3911 if (shmid == -1) {
3912 // Possible reasons for shmget failure:
3913 // 1. shmmax is too small for Java heap.
3914 // > check shmmax value: cat /proc/sys/kernel/shmmax
3915 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3916 // 2. not enough large page memory.
3917 // > check available large pages: cat /proc/meminfo
3918 // > increase amount of large pages:
3919 // echo new_value > /proc/sys/vm/nr_hugepages
3920 // Note 1: different Linux may use different name for this property,
3921 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3922 // Note 2: it's possible there's enough physical memory available but
3923 // they are so fragmented after a long run that they can't
3924 // coalesce into large pages. Try to reserve large pages when
3925 // the system is still "fresh".
3926 shm_warning_with_errno("Failed to reserve shared memory.");
3927 return NULL;
3928 }
3929
3930 // Attach to the region.
3931 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3932
3933 // Remove shmid. If shmat() is successful, the actual shared memory segment
3934 // will be deleted when it's detached by shmdt() or when the process
3935 // terminates. If shmat() is not successful this will remove the shared
3936 // segment immediately.
3937 shmctl(shmid, IPC_RMID, NULL);
3938
3939 return addr;
3940 }
3941
3942 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3943 int error) {
3944 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3945
3946 bool warn_on_failure = UseLargePages &&
3947 (!FLAG_IS_DEFAULT(UseLargePages) ||
3948 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3949 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3950
3951 if (warn_on_failure) {
3952 char msg[128];
3953 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3954 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3955 warning("%s", msg);
3956 }
3957 }
3958
3959 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3960 char* req_addr,
3961 bool exec) {
3962 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3963 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3964 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3965
3966 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3967 char* addr = (char*)::mmap(req_addr, bytes, prot,
3968 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3969 -1, 0);
3970
3971 if (addr == MAP_FAILED) {
3972 warn_on_large_pages_failure(req_addr, bytes, errno);
3973 return NULL;
3974 }
3975
3976 assert(is_aligned(addr, os::large_page_size()), "Must be");
3977
3978 return addr;
3979 }
3980
3981 // Reserve memory using mmap(MAP_HUGETLB).
3982 // - bytes shall be a multiple of alignment.
3983 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3984 // - alignment sets the alignment at which memory shall be allocated.
3985 // It must be a multiple of allocation granularity.
3986 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3987 // req_addr or NULL.
3988 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3989 size_t alignment,
3990 char* req_addr,
3991 bool exec) {
3992 size_t large_page_size = os::large_page_size();
3993 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3994
3995 assert(is_aligned(req_addr, alignment), "Must be");
3996 assert(is_aligned(bytes, alignment), "Must be");
3997
3998 // First reserve - but not commit - the address range in small pages.
3999 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
4000
4001 if (start == NULL) {
4002 return NULL;
4003 }
4004
4005 assert(is_aligned(start, alignment), "Must be");
4006
4007 char* end = start + bytes;
4008
4009 // Find the regions of the allocated chunk that can be promoted to large pages.
4010 char* lp_start = align_up(start, large_page_size);
4011 char* lp_end = align_down(end, large_page_size);
4012
4013 size_t lp_bytes = lp_end - lp_start;
4014
4015 assert(is_aligned(lp_bytes, large_page_size), "Must be");
4016
4017 if (lp_bytes == 0) {
4018 // The mapped region doesn't even span the start and the end of a large page.
4019 // Fall back to allocate a non-special area.
4020 ::munmap(start, end - start);
4021 return NULL;
4022 }
4023
4024 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
4025
4026 void* result;
4027
4028 // Commit small-paged leading area.
4029 if (start != lp_start) {
4030 result = ::mmap(start, lp_start - start, prot,
4031 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
4032 -1, 0);
4033 if (result == MAP_FAILED) {
4034 ::munmap(lp_start, end - lp_start);
4035 return NULL;
4036 }
4037 }
4038
4039 // Commit large-paged area.
4040 result = ::mmap(lp_start, lp_bytes, prot,
4041 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
4042 -1, 0);
4043 if (result == MAP_FAILED) {
4044 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
4045 // If the mmap above fails, the large pages region will be unmapped and we
4046 // have regions before and after with small pages. Release these regions.
4047 //
4048 // | mapped | unmapped | mapped |
4049 // ^ ^ ^ ^
4050 // start lp_start lp_end end
4051 //
4052 ::munmap(start, lp_start - start);
4053 ::munmap(lp_end, end - lp_end);
4054 return NULL;
4055 }
4056
4057 // Commit small-paged trailing area.
4058 if (lp_end != end) {
4059 result = ::mmap(lp_end, end - lp_end, prot,
4060 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
4061 -1, 0);
4062 if (result == MAP_FAILED) {
4063 ::munmap(start, lp_end - start);
4064 return NULL;
4065 }
4066 }
4067
4068 return start;
4069 }
4070
4071 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
4072 size_t alignment,
4073 char* req_addr,
4074 bool exec) {
4075 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
4076 assert(is_aligned(req_addr, alignment), "Must be");
4077 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
4078 assert(is_power_of_2(os::large_page_size()), "Must be");
4079 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
4080
4081 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
4082 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
4083 } else {
4084 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
4085 }
4086 }
4087
4088 char* os::reserve_memory_special(size_t bytes, size_t alignment,
4089 char* req_addr, bool exec) {
4090 assert(UseLargePages, "only for large pages");
4091
4092 char* addr;
4093 if (UseSHM) {
4094 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
4095 } else {
4096 assert(UseHugeTLBFS, "must be");
4097 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
4098 }
4099
4100 if (addr != NULL) {
4101 if (UseNUMAInterleaving) {
4102 numa_make_global(addr, bytes);
4103 }
4104
4105 // The memory is committed
4106 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
4107 }
4108
4109 return addr;
4110 }
4111
4112 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
4113 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
4114 return shmdt(base) == 0;
4115 }
4116
4117 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
4118 return pd_release_memory(base, bytes);
4119 }
4120
4121 bool os::release_memory_special(char* base, size_t bytes) {
4122 bool res;
4123 if (MemTracker::tracking_level() > NMT_minimal) {
4124 Tracker tkr(Tracker::release);
4125 res = os::Linux::release_memory_special_impl(base, bytes);
4126 if (res) {
4127 tkr.record((address)base, bytes);
4128 }
4129
4130 } else {
4131 res = os::Linux::release_memory_special_impl(base, bytes);
4132 }
4133 return res;
4134 }
4135
4136 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
4137 assert(UseLargePages, "only for large pages");
4138 bool res;
4139
4140 if (UseSHM) {
4141 res = os::Linux::release_memory_special_shm(base, bytes);
4142 } else {
4143 assert(UseHugeTLBFS, "must be");
4144 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
4145 }
4146 return res;
4147 }
4148
4149 size_t os::large_page_size() {
4150 return _large_page_size;
4151 }
4152
4153 // With SysV SHM the entire memory region must be allocated as shared
4154 // memory.
4155 // HugeTLBFS allows application to commit large page memory on demand.
4156 // However, when committing memory with HugeTLBFS fails, the region
4157 // that was supposed to be committed will lose the old reservation
4158 // and allow other threads to steal that memory region. Because of this
4159 // behavior we can't commit HugeTLBFS memory.
4160 bool os::can_commit_large_page_memory() {
4161 return UseTransparentHugePages;
4162 }
4163
4164 bool os::can_execute_large_page_memory() {
4165 return UseTransparentHugePages || UseHugeTLBFS;
4166 }
4167
4168 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
4169 assert(file_desc >= 0, "file_desc is not valid");
4170 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
4171 if (result != NULL) {
4172 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
4173 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
4174 }
4175 }
4176 return result;
4177 }
4178
4179 // Reserve memory at an arbitrary address, only if that area is
4180 // available (and not reserved for something else).
4181
4182 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
4183 // Assert only that the size is a multiple of the page size, since
4184 // that's all that mmap requires, and since that's all we really know
4185 // about at this low abstraction level. If we need higher alignment,
4186 // we can either pass an alignment to this method or verify alignment
4187 // in one of the methods further up the call chain. See bug 5044738.
4188 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
4189
4190 // Repeatedly allocate blocks until the block is allocated at the
4191 // right spot.
4192
4193 // Linux mmap allows caller to pass an address as hint; give it a try first,
4194 // if kernel honors the hint then we can return immediately.
4195 char * addr = anon_mmap(requested_addr, bytes, false);
4196 if (addr == requested_addr) {
4197 return requested_addr;
4198 }
4199
4200 if (addr != NULL) {
4201 // mmap() is successful but it fails to reserve at the requested address
4202 anon_munmap(addr, bytes);
4203 }
4204
4205 return NULL;
4206 }
4207
4208 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4209 void os::infinite_sleep() {
4210 while (true) { // sleep forever ...
4211 ::sleep(100); // ... 100 seconds at a time
4212 }
4213 }
4214
4215 // Used to convert frequent JVM_Yield() to nops
4216 bool os::dont_yield() {
4217 return DontYieldALot;
4218 }
4219
4220 // Linux CFS scheduler (since 2.6.23) does not guarantee sched_yield(2) will
4221 // actually give up the CPU. Since skip buddy (v2.6.28):
4222 //
4223 // * Sets the yielding task as skip buddy for current CPU's run queue.
4224 // * Picks next from run queue, if empty, picks a skip buddy (can be the yielding task).
4225 // * Clears skip buddies for this run queue (yielding task no longer a skip buddy).
4226 //
4227 // An alternative is calling os::naked_short_nanosleep with a small number to avoid
4228 // getting re-scheduled immediately.
4229 //
4230 void os::naked_yield() {
4231 sched_yield();
4232 }
4233
4234 ////////////////////////////////////////////////////////////////////////////////
4235 // thread priority support
4236
4237 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4238 // only supports dynamic priority, static priority must be zero. For real-time
4239 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4240 // However, for large multi-threaded applications, SCHED_RR is not only slower
4241 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4242 // of 5 runs - Sep 2005).
4243 //
4244 // The following code actually changes the niceness of kernel-thread/LWP. It
4245 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4246 // not the entire user process, and user level threads are 1:1 mapped to kernel
4247 // threads. It has always been the case, but could change in the future. For
4248 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4249 // It is only used when ThreadPriorityPolicy=1 and may require system level permission
4250 // (e.g., root privilege or CAP_SYS_NICE capability).
4251
4252 int os::java_to_os_priority[CriticalPriority + 1] = {
4253 19, // 0 Entry should never be used
4254
4255 4, // 1 MinPriority
4256 3, // 2
4257 2, // 3
4258
4259 1, // 4
4260 0, // 5 NormPriority
4261 -1, // 6
4262
4263 -2, // 7
4264 -3, // 8
4265 -4, // 9 NearMaxPriority
4266
4267 -5, // 10 MaxPriority
4268
4269 -5 // 11 CriticalPriority
4270 };
4271
4272 static int prio_init() {
4273 if (ThreadPriorityPolicy == 1) {
4274 if (geteuid() != 0) {
4275 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4276 warning("-XX:ThreadPriorityPolicy=1 may require system level permission, " \
4277 "e.g., being the root user. If the necessary permission is not " \
4278 "possessed, changes to priority will be silently ignored.");
4279 }
4280 }
4281 }
4282 if (UseCriticalJavaThreadPriority) {
4283 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4284 }
4285 return 0;
4286 }
4287
4288 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4289 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4290
4291 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4292 return (ret == 0) ? OS_OK : OS_ERR;
4293 }
4294
4295 OSReturn os::get_native_priority(const Thread* const thread,
4296 int *priority_ptr) {
4297 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4298 *priority_ptr = java_to_os_priority[NormPriority];
4299 return OS_OK;
4300 }
4301
4302 errno = 0;
4303 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4304 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4305 }
4306
4307 ////////////////////////////////////////////////////////////////////////////////
4308 // suspend/resume support
4309
4310 // The low-level signal-based suspend/resume support is a remnant from the
4311 // old VM-suspension that used to be for java-suspension, safepoints etc,
4312 // within hotspot. Currently used by JFR's OSThreadSampler
4313 //
4314 // The remaining code is greatly simplified from the more general suspension
4315 // code that used to be used.
4316 //
4317 // The protocol is quite simple:
4318 // - suspend:
4319 // - sends a signal to the target thread
4320 // - polls the suspend state of the osthread using a yield loop
4321 // - target thread signal handler (SR_handler) sets suspend state
4322 // and blocks in sigsuspend until continued
4323 // - resume:
4324 // - sets target osthread state to continue
4325 // - sends signal to end the sigsuspend loop in the SR_handler
4326 //
4327 // Note that the SR_lock plays no role in this suspend/resume protocol,
4328 // but is checked for NULL in SR_handler as a thread termination indicator.
4329 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4330 //
4331 // Note that resume_clear_context() and suspend_save_context() are needed
4332 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4333 // which in part is used by:
4334 // - Forte Analyzer: AsyncGetCallTrace()
4335 // - StackBanging: get_frame_at_stack_banging_point()
4336
4337 static void resume_clear_context(OSThread *osthread) {
4338 osthread->set_ucontext(NULL);
4339 osthread->set_siginfo(NULL);
4340 }
4341
4342 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4343 ucontext_t* context) {
4344 osthread->set_ucontext(context);
4345 osthread->set_siginfo(siginfo);
4346 }
4347
4348 // Handler function invoked when a thread's execution is suspended or
4349 // resumed. We have to be careful that only async-safe functions are
4350 // called here (Note: most pthread functions are not async safe and
4351 // should be avoided.)
4352 //
4353 // Note: sigwait() is a more natural fit than sigsuspend() from an
4354 // interface point of view, but sigwait() prevents the signal hander
4355 // from being run. libpthread would get very confused by not having
4356 // its signal handlers run and prevents sigwait()'s use with the
4357 // mutex granting granting signal.
4358 //
4359 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4360 //
4361 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4362 // Save and restore errno to avoid confusing native code with EINTR
4363 // after sigsuspend.
4364 int old_errno = errno;
4365
4366 Thread* thread = Thread::current_or_null_safe();
4367 assert(thread != NULL, "Missing current thread in SR_handler");
4368
4369 // On some systems we have seen signal delivery get "stuck" until the signal
4370 // mask is changed as part of thread termination. Check that the current thread
4371 // has not already terminated (via SR_lock()) - else the following assertion
4372 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4373 // destructor has completed.
4374
4375 if (thread->SR_lock() == NULL) {
4376 return;
4377 }
4378
4379 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4380
4381 OSThread* osthread = thread->osthread();
4382
4383 os::SuspendResume::State current = osthread->sr.state();
4384 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4385 suspend_save_context(osthread, siginfo, context);
4386
4387 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4388 os::SuspendResume::State state = osthread->sr.suspended();
4389 if (state == os::SuspendResume::SR_SUSPENDED) {
4390 sigset_t suspend_set; // signals for sigsuspend()
4391 sigemptyset(&suspend_set);
4392 // get current set of blocked signals and unblock resume signal
4393 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4394 sigdelset(&suspend_set, SR_signum);
4395
4396 sr_semaphore.signal();
4397 // wait here until we are resumed
4398 while (1) {
4399 sigsuspend(&suspend_set);
4400
4401 os::SuspendResume::State result = osthread->sr.running();
4402 if (result == os::SuspendResume::SR_RUNNING) {
4403 sr_semaphore.signal();
4404 break;
4405 }
4406 }
4407
4408 } else if (state == os::SuspendResume::SR_RUNNING) {
4409 // request was cancelled, continue
4410 } else {
4411 ShouldNotReachHere();
4412 }
4413
4414 resume_clear_context(osthread);
4415 } else if (current == os::SuspendResume::SR_RUNNING) {
4416 // request was cancelled, continue
4417 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4418 // ignore
4419 } else {
4420 // ignore
4421 }
4422
4423 errno = old_errno;
4424 }
4425
4426 static int SR_initialize() {
4427 struct sigaction act;
4428 char *s;
4429
4430 // Get signal number to use for suspend/resume
4431 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4432 int sig = ::strtol(s, 0, 10);
4433 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4434 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4435 SR_signum = sig;
4436 } else {
4437 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4438 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4439 }
4440 }
4441
4442 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4443 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4444
4445 sigemptyset(&SR_sigset);
4446 sigaddset(&SR_sigset, SR_signum);
4447
4448 // Set up signal handler for suspend/resume
4449 act.sa_flags = SA_RESTART|SA_SIGINFO;
4450 act.sa_handler = (void (*)(int)) SR_handler;
4451
4452 // SR_signum is blocked by default.
4453 // 4528190 - We also need to block pthread restart signal (32 on all
4454 // supported Linux platforms). Note that LinuxThreads need to block
4455 // this signal for all threads to work properly. So we don't have
4456 // to use hard-coded signal number when setting up the mask.
4457 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4458
4459 if (sigaction(SR_signum, &act, 0) == -1) {
4460 return -1;
4461 }
4462
4463 // Save signal flag
4464 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4465 return 0;
4466 }
4467
4468 static int sr_notify(OSThread* osthread) {
4469 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4470 assert_status(status == 0, status, "pthread_kill");
4471 return status;
4472 }
4473
4474 // "Randomly" selected value for how long we want to spin
4475 // before bailing out on suspending a thread, also how often
4476 // we send a signal to a thread we want to resume
4477 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4478 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4479
4480 // returns true on success and false on error - really an error is fatal
4481 // but this seems the normal response to library errors
4482 static bool do_suspend(OSThread* osthread) {
4483 assert(osthread->sr.is_running(), "thread should be running");
4484 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4485
4486 // mark as suspended and send signal
4487 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4488 // failed to switch, state wasn't running?
4489 ShouldNotReachHere();
4490 return false;
4491 }
4492
4493 if (sr_notify(osthread) != 0) {
4494 ShouldNotReachHere();
4495 }
4496
4497 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4498 while (true) {
4499 if (sr_semaphore.timedwait(2)) {
4500 break;
4501 } else {
4502 // timeout
4503 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4504 if (cancelled == os::SuspendResume::SR_RUNNING) {
4505 return false;
4506 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4507 // make sure that we consume the signal on the semaphore as well
4508 sr_semaphore.wait();
4509 break;
4510 } else {
4511 ShouldNotReachHere();
4512 return false;
4513 }
4514 }
4515 }
4516
4517 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4518 return true;
4519 }
4520
4521 static void do_resume(OSThread* osthread) {
4522 assert(osthread->sr.is_suspended(), "thread should be suspended");
4523 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4524
4525 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4526 // failed to switch to WAKEUP_REQUEST
4527 ShouldNotReachHere();
4528 return;
4529 }
4530
4531 while (true) {
4532 if (sr_notify(osthread) == 0) {
4533 if (sr_semaphore.timedwait(2)) {
4534 if (osthread->sr.is_running()) {
4535 return;
4536 }
4537 }
4538 } else {
4539 ShouldNotReachHere();
4540 }
4541 }
4542
4543 guarantee(osthread->sr.is_running(), "Must be running!");
4544 }
4545
4546 ///////////////////////////////////////////////////////////////////////////////////
4547 // signal handling (except suspend/resume)
4548
4549 // This routine may be used by user applications as a "hook" to catch signals.
4550 // The user-defined signal handler must pass unrecognized signals to this
4551 // routine, and if it returns true (non-zero), then the signal handler must
4552 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4553 // routine will never retun false (zero), but instead will execute a VM panic
4554 // routine kill the process.
4555 //
4556 // If this routine returns false, it is OK to call it again. This allows
4557 // the user-defined signal handler to perform checks either before or after
4558 // the VM performs its own checks. Naturally, the user code would be making
4559 // a serious error if it tried to handle an exception (such as a null check
4560 // or breakpoint) that the VM was generating for its own correct operation.
4561 //
4562 // This routine may recognize any of the following kinds of signals:
4563 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4564 // It should be consulted by handlers for any of those signals.
4565 //
4566 // The caller of this routine must pass in the three arguments supplied
4567 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4568 // field of the structure passed to sigaction(). This routine assumes that
4569 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4570 //
4571 // Note that the VM will print warnings if it detects conflicting signal
4572 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4573 //
4574 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4575 siginfo_t* siginfo,
4576 void* ucontext,
4577 int abort_if_unrecognized);
4578
4579 static void signalHandler(int sig, siginfo_t* info, void* uc) {
4580 assert(info != NULL && uc != NULL, "it must be old kernel");
4581 int orig_errno = errno; // Preserve errno value over signal handler.
4582 JVM_handle_linux_signal(sig, info, uc, true);
4583 errno = orig_errno;
4584 }
4585
4586
4587 // This boolean allows users to forward their own non-matching signals
4588 // to JVM_handle_linux_signal, harmlessly.
4589 bool os::Linux::signal_handlers_are_installed = false;
4590
4591 // For signal-chaining
4592 bool os::Linux::libjsig_is_loaded = false;
4593 typedef struct sigaction *(*get_signal_t)(int);
4594 get_signal_t os::Linux::get_signal_action = NULL;
4595
4596 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4597 struct sigaction *actp = NULL;
4598
4599 if (libjsig_is_loaded) {
4600 // Retrieve the old signal handler from libjsig
4601 actp = (*get_signal_action)(sig);
4602 }
4603 if (actp == NULL) {
4604 // Retrieve the preinstalled signal handler from jvm
4605 actp = os::Posix::get_preinstalled_handler(sig);
4606 }
4607
4608 return actp;
4609 }
4610
4611 static bool call_chained_handler(struct sigaction *actp, int sig,
4612 siginfo_t *siginfo, void *context) {
4613 // Call the old signal handler
4614 if (actp->sa_handler == SIG_DFL) {
4615 // It's more reasonable to let jvm treat it as an unexpected exception
4616 // instead of taking the default action.
4617 return false;
4618 } else if (actp->sa_handler != SIG_IGN) {
4619 if ((actp->sa_flags & SA_NODEFER) == 0) {
4620 // automaticlly block the signal
4621 sigaddset(&(actp->sa_mask), sig);
4622 }
4623
4624 sa_handler_t hand = NULL;
4625 sa_sigaction_t sa = NULL;
4626 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4627 // retrieve the chained handler
4628 if (siginfo_flag_set) {
4629 sa = actp->sa_sigaction;
4630 } else {
4631 hand = actp->sa_handler;
4632 }
4633
4634 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4635 actp->sa_handler = SIG_DFL;
4636 }
4637
4638 // try to honor the signal mask
4639 sigset_t oset;
4640 sigemptyset(&oset);
4641 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4642
4643 // call into the chained handler
4644 if (siginfo_flag_set) {
4645 (*sa)(sig, siginfo, context);
4646 } else {
4647 (*hand)(sig);
4648 }
4649
4650 // restore the signal mask
4651 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4652 }
4653 // Tell jvm's signal handler the signal is taken care of.
4654 return true;
4655 }
4656
4657 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4658 bool chained = false;
4659 // signal-chaining
4660 if (UseSignalChaining) {
4661 struct sigaction *actp = get_chained_signal_action(sig);
4662 if (actp != NULL) {
4663 chained = call_chained_handler(actp, sig, siginfo, context);
4664 }
4665 }
4666 return chained;
4667 }
4668
4669 // for diagnostic
4670 int sigflags[NSIG];
4671
4672 int os::Linux::get_our_sigflags(int sig) {
4673 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4674 return sigflags[sig];
4675 }
4676
4677 void os::Linux::set_our_sigflags(int sig, int flags) {
4678 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4679 if (sig > 0 && sig < NSIG) {
4680 sigflags[sig] = flags;
4681 }
4682 }
4683
4684 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4685 // Check for overwrite.
4686 struct sigaction oldAct;
4687 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4688
4689 void* oldhand = oldAct.sa_sigaction
4690 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4691 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4692 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4693 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4694 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4695 if (AllowUserSignalHandlers || !set_installed) {
4696 // Do not overwrite; user takes responsibility to forward to us.
4697 return;
4698 } else if (UseSignalChaining) {
4699 // save the old handler in jvm
4700 os::Posix::save_preinstalled_handler(sig, oldAct);
4701 // libjsig also interposes the sigaction() call below and saves the
4702 // old sigaction on it own.
4703 } else {
4704 fatal("Encountered unexpected pre-existing sigaction handler "
4705 "%#lx for signal %d.", (long)oldhand, sig);
4706 }
4707 }
4708
4709 struct sigaction sigAct;
4710 sigfillset(&(sigAct.sa_mask));
4711 sigAct.sa_handler = SIG_DFL;
4712 if (!set_installed) {
4713 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4714 } else {
4715 sigAct.sa_sigaction = signalHandler;
4716 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4717 }
4718 // Save flags, which are set by ours
4719 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4720 sigflags[sig] = sigAct.sa_flags;
4721
4722 int ret = sigaction(sig, &sigAct, &oldAct);
4723 assert(ret == 0, "check");
4724
4725 void* oldhand2 = oldAct.sa_sigaction
4726 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4727 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4728 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4729 }
4730
4731 // install signal handlers for signals that HotSpot needs to
4732 // handle in order to support Java-level exception handling.
4733
4734 void os::Linux::install_signal_handlers() {
4735 if (!signal_handlers_are_installed) {
4736 signal_handlers_are_installed = true;
4737
4738 // signal-chaining
4739 typedef void (*signal_setting_t)();
4740 signal_setting_t begin_signal_setting = NULL;
4741 signal_setting_t end_signal_setting = NULL;
4742 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4743 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4744 if (begin_signal_setting != NULL) {
4745 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4746 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4747 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4748 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4749 libjsig_is_loaded = true;
4750 assert(UseSignalChaining, "should enable signal-chaining");
4751 }
4752 if (libjsig_is_loaded) {
4753 // Tell libjsig jvm is setting signal handlers
4754 (*begin_signal_setting)();
4755 }
4756
4757 set_signal_handler(SIGSEGV, true);
4758 set_signal_handler(SIGPIPE, true);
4759 set_signal_handler(SIGBUS, true);
4760 set_signal_handler(SIGILL, true);
4761 set_signal_handler(SIGFPE, true);
4762 #if defined(PPC64)
4763 set_signal_handler(SIGTRAP, true);
4764 #endif
4765 set_signal_handler(SIGXFSZ, true);
4766
4767 if (libjsig_is_loaded) {
4768 // Tell libjsig jvm finishes setting signal handlers
4769 (*end_signal_setting)();
4770 }
4771
4772 // We don't activate signal checker if libjsig is in place, we trust ourselves
4773 // and if UserSignalHandler is installed all bets are off.
4774 // Log that signal checking is off only if -verbose:jni is specified.
4775 if (CheckJNICalls) {
4776 if (libjsig_is_loaded) {
4777 if (PrintJNIResolving) {
4778 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4779 }
4780 check_signals = false;
4781 }
4782 if (AllowUserSignalHandlers) {
4783 if (PrintJNIResolving) {
4784 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4785 }
4786 check_signals = false;
4787 }
4788 }
4789 }
4790 }
4791
4792 // This is the fastest way to get thread cpu time on Linux.
4793 // Returns cpu time (user+sys) for any thread, not only for current.
4794 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4795 // It might work on 2.6.10+ with a special kernel/glibc patch.
4796 // For reference, please, see IEEE Std 1003.1-2004:
4797 // http://www.unix.org/single_unix_specification
4798
4799 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4800 struct timespec tp;
4801 int rc = os::Posix::clock_gettime(clockid, &tp);
4802 assert(rc == 0, "clock_gettime is expected to return 0 code");
4803
4804 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4805 }
4806
4807 void os::Linux::initialize_os_info() {
4808 assert(_os_version == 0, "OS info already initialized");
4809
4810 struct utsname _uname;
4811
4812 uint32_t major;
4813 uint32_t minor;
4814 uint32_t fix;
4815
4816 int rc;
4817
4818 // Kernel version is unknown if
4819 // verification below fails.
4820 _os_version = 0x01000000;
4821
4822 rc = uname(&_uname);
4823 if (rc != -1) {
4824
4825 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4826 if (rc == 3) {
4827
4828 if (major < 256 && minor < 256 && fix < 256) {
4829 // Kernel version format is as expected,
4830 // set it overriding unknown state.
4831 _os_version = (major << 16) |
4832 (minor << 8 ) |
4833 (fix << 0 ) ;
4834 }
4835 }
4836 }
4837 }
4838
4839 uint32_t os::Linux::os_version() {
4840 assert(_os_version != 0, "not initialized");
4841 return _os_version & 0x00FFFFFF;
4842 }
4843
4844 bool os::Linux::os_version_is_known() {
4845 assert(_os_version != 0, "not initialized");
4846 return _os_version & 0x01000000 ? false : true;
4847 }
4848
4849 /////
4850 // glibc on Linux platform uses non-documented flag
4851 // to indicate, that some special sort of signal
4852 // trampoline is used.
4853 // We will never set this flag, and we should
4854 // ignore this flag in our diagnostic
4855 #ifdef SIGNIFICANT_SIGNAL_MASK
4856 #undef SIGNIFICANT_SIGNAL_MASK
4857 #endif
4858 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4859
4860 static const char* get_signal_handler_name(address handler,
4861 char* buf, int buflen) {
4862 int offset = 0;
4863 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4864 if (found) {
4865 // skip directory names
4866 const char *p1, *p2;
4867 p1 = buf;
4868 size_t len = strlen(os::file_separator());
4869 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4870 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4871 } else {
4872 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4873 }
4874 return buf;
4875 }
4876
4877 static void print_signal_handler(outputStream* st, int sig,
4878 char* buf, size_t buflen) {
4879 struct sigaction sa;
4880
4881 sigaction(sig, NULL, &sa);
4882
4883 // See comment for SIGNIFICANT_SIGNAL_MASK define
4884 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4885
4886 st->print("%s: ", os::exception_name(sig, buf, buflen));
4887
4888 address handler = (sa.sa_flags & SA_SIGINFO)
4889 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4890 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4891
4892 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4893 st->print("SIG_DFL");
4894 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4895 st->print("SIG_IGN");
4896 } else {
4897 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4898 }
4899
4900 st->print(", sa_mask[0]=");
4901 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4902
4903 address rh = VMError::get_resetted_sighandler(sig);
4904 // May be, handler was resetted by VMError?
4905 if (rh != NULL) {
4906 handler = rh;
4907 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4908 }
4909
4910 st->print(", sa_flags=");
4911 os::Posix::print_sa_flags(st, sa.sa_flags);
4912
4913 // Check: is it our handler?
4914 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4915 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4916 // It is our signal handler
4917 // check for flags, reset system-used one!
4918 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4919 st->print(
4920 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4921 os::Linux::get_our_sigflags(sig));
4922 }
4923 }
4924 st->cr();
4925 }
4926
4927
4928 #define DO_SIGNAL_CHECK(sig) \
4929 do { \
4930 if (!sigismember(&check_signal_done, sig)) { \
4931 os::Linux::check_signal_handler(sig); \
4932 } \
4933 } while (0)
4934
4935 // This method is a periodic task to check for misbehaving JNI applications
4936 // under CheckJNI, we can add any periodic checks here
4937
4938 void os::run_periodic_checks() {
4939 if (check_signals == false) return;
4940
4941 // SEGV and BUS if overridden could potentially prevent
4942 // generation of hs*.log in the event of a crash, debugging
4943 // such a case can be very challenging, so we absolutely
4944 // check the following for a good measure:
4945 DO_SIGNAL_CHECK(SIGSEGV);
4946 DO_SIGNAL_CHECK(SIGILL);
4947 DO_SIGNAL_CHECK(SIGFPE);
4948 DO_SIGNAL_CHECK(SIGBUS);
4949 DO_SIGNAL_CHECK(SIGPIPE);
4950 DO_SIGNAL_CHECK(SIGXFSZ);
4951 #if defined(PPC64)
4952 DO_SIGNAL_CHECK(SIGTRAP);
4953 #endif
4954
4955 // ReduceSignalUsage allows the user to override these handlers
4956 // see comments at the very top and jvm_md.h
4957 if (!ReduceSignalUsage) {
4958 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4959 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4960 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4961 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4962 }
4963
4964 DO_SIGNAL_CHECK(SR_signum);
4965 }
4966
4967 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4968
4969 static os_sigaction_t os_sigaction = NULL;
4970
4971 void os::Linux::check_signal_handler(int sig) {
4972 char buf[O_BUFLEN];
4973 address jvmHandler = NULL;
4974
4975
4976 struct sigaction act;
4977 if (os_sigaction == NULL) {
4978 // only trust the default sigaction, in case it has been interposed
4979 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4980 if (os_sigaction == NULL) return;
4981 }
4982
4983 os_sigaction(sig, (struct sigaction*)NULL, &act);
4984
4985
4986 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4987
4988 address thisHandler = (act.sa_flags & SA_SIGINFO)
4989 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4990 : CAST_FROM_FN_PTR(address, act.sa_handler);
4991
4992
4993 switch (sig) {
4994 case SIGSEGV:
4995 case SIGBUS:
4996 case SIGFPE:
4997 case SIGPIPE:
4998 case SIGILL:
4999 case SIGXFSZ:
5000 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
5001 break;
5002
5003 case SHUTDOWN1_SIGNAL:
5004 case SHUTDOWN2_SIGNAL:
5005 case SHUTDOWN3_SIGNAL:
5006 case BREAK_SIGNAL:
5007 jvmHandler = (address)user_handler();
5008 break;
5009
5010 default:
5011 if (sig == SR_signum) {
5012 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
5013 } else {
5014 return;
5015 }
5016 break;
5017 }
5018
5019 if (thisHandler != jvmHandler) {
5020 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
5021 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
5022 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
5023 // No need to check this sig any longer
5024 sigaddset(&check_signal_done, sig);
5025 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
5026 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
5027 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
5028 exception_name(sig, buf, O_BUFLEN));
5029 }
5030 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
5031 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
5032 tty->print("expected:");
5033 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
5034 tty->cr();
5035 tty->print(" found:");
5036 os::Posix::print_sa_flags(tty, act.sa_flags);
5037 tty->cr();
5038 // No need to check this sig any longer
5039 sigaddset(&check_signal_done, sig);
5040 }
5041
5042 // Dump all the signal
5043 if (sigismember(&check_signal_done, sig)) {
5044 print_signal_handlers(tty, buf, O_BUFLEN);
5045 }
5046 }
5047
5048 extern void report_error(char* file_name, int line_no, char* title,
5049 char* format, ...);
5050
5051 // this is called _before_ most of the global arguments have been parsed
5052 void os::init(void) {
5053 char dummy; // used to get a guess on initial stack address
5054
5055 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
5056
5057 init_random(1234567);
5058
5059 Linux::set_page_size(sysconf(_SC_PAGESIZE));
5060 if (Linux::page_size() == -1) {
5061 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
5062 os::strerror(errno));
5063 }
5064 init_page_sizes((size_t) Linux::page_size());
5065
5066 Linux::initialize_system_info();
5067
5068 Linux::initialize_os_info();
5069
5070 os::Linux::CPUPerfTicks pticks;
5071 bool res = os::Linux::get_tick_information(&pticks, -1);
5072
5073 if (res && pticks.has_steal_ticks) {
5074 has_initial_tick_info = true;
5075 initial_total_ticks = pticks.total;
5076 initial_steal_ticks = pticks.steal;
5077 }
5078
5079 // _main_thread points to the thread that created/loaded the JVM.
5080 Linux::_main_thread = pthread_self();
5081
5082 // retrieve entry point for pthread_setname_np
5083 Linux::_pthread_setname_np =
5084 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
5085
5086 os::Posix::init();
5087
5088 initial_time_count = javaTimeNanos();
5089
5090 // Always warn if no monotonic clock available
5091 if (!os::Posix::supports_monotonic_clock()) {
5092 warning("No monotonic clock was available - timed services may " \
5093 "be adversely affected if the time-of-day clock changes");
5094 }
5095 }
5096
5097 // To install functions for atexit system call
5098 extern "C" {
5099 static void perfMemory_exit_helper() {
5100 perfMemory_exit();
5101 }
5102 }
5103
5104 void os::pd_init_container_support() {
5105 OSContainer::init();
5106 }
5107
5108 void os::Linux::numa_init() {
5109
5110 // Java can be invoked as
5111 // 1. Without numactl and heap will be allocated/configured on all nodes as
5112 // per the system policy.
5113 // 2. With numactl --interleave:
5114 // Use numa_get_interleave_mask(v2) API to get nodes bitmask. The same
5115 // API for membind case bitmask is reset.
5116 // Interleave is only hint and Kernel can fallback to other nodes if
5117 // no memory is available on the target nodes.
5118 // 3. With numactl --membind:
5119 // Use numa_get_membind(v2) API to get nodes bitmask. The same API for
5120 // interleave case returns bitmask of all nodes.
5121 // numa_all_nodes_ptr holds bitmask of all nodes.
5122 // numa_get_interleave_mask(v2) and numa_get_membind(v2) APIs returns correct
5123 // bitmask when externally configured to run on all or fewer nodes.
5124
5125 if (!Linux::libnuma_init()) {
5126 UseNUMA = false;
5127 } else {
5128 if ((Linux::numa_max_node() < 1) || Linux::is_bound_to_single_node()) {
5129 // If there's only one node (they start from 0) or if the process
5130 // is bound explicitly to a single node using membind, disable NUMA.
5131 UseNUMA = false;
5132 } else {
5133
5134 LogTarget(Info,os) log;
5135 LogStream ls(log);
5136
5137 Linux::set_configured_numa_policy(Linux::identify_numa_policy());
5138
5139 struct bitmask* bmp = Linux::_numa_membind_bitmask;
5140 const char* numa_mode = "membind";
5141
5142 if (Linux::is_running_in_interleave_mode()) {
5143 bmp = Linux::_numa_interleave_bitmask;
5144 numa_mode = "interleave";
5145 }
5146
5147 ls.print("UseNUMA is enabled and invoked in '%s' mode."
5148 " Heap will be configured using NUMA memory nodes:", numa_mode);
5149
5150 for (int node = 0; node <= Linux::numa_max_node(); node++) {
5151 if (Linux::_numa_bitmask_isbitset(bmp, node)) {
5152 ls.print(" %d", node);
5153 }
5154 }
5155 }
5156 }
5157
5158 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5159 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5160 // we can make the adaptive lgrp chunk resizing work. If the user specified both
5161 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5162 // and disable adaptive resizing.
5163 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5164 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5165 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5166 UseAdaptiveSizePolicy = false;
5167 UseAdaptiveNUMAChunkSizing = false;
5168 }
5169 }
5170
5171 if (!UseNUMA && ForceNUMA) {
5172 UseNUMA = true;
5173 }
5174 }
5175
5176 // this is called _after_ the global arguments have been parsed
5177 jint os::init_2(void) {
5178
5179 // This could be set after os::Posix::init() but all platforms
5180 // have to set it the same so we have to mirror Solaris.
5181 DEBUG_ONLY(os::set_mutex_init_done();)
5182
5183 os::Posix::init_2();
5184
5185 Linux::fast_thread_clock_init();
5186
5187 // initialize suspend/resume support - must do this before signal_sets_init()
5188 if (SR_initialize() != 0) {
5189 perror("SR_initialize failed");
5190 return JNI_ERR;
5191 }
5192
5193 Linux::signal_sets_init();
5194 Linux::install_signal_handlers();
5195 // Initialize data for jdk.internal.misc.Signal
5196 if (!ReduceSignalUsage) {
5197 jdk_misc_signal_init();
5198 }
5199
5200 if (AdjustStackSizeForTLS) {
5201 get_minstack_init();
5202 }
5203
5204 // Check and sets minimum stack sizes against command line options
5205 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
5206 return JNI_ERR;
5207 }
5208
5209 #if defined(IA32)
5210 // Need to ensure we've determined the process's initial stack to
5211 // perform the workaround
5212 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5213 workaround_expand_exec_shield_cs_limit();
5214 #else
5215 suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
5216 if (!suppress_primordial_thread_resolution) {
5217 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5218 }
5219 #endif
5220
5221 Linux::libpthread_init();
5222 Linux::sched_getcpu_init();
5223 log_info(os)("HotSpot is running with %s, %s",
5224 Linux::glibc_version(), Linux::libpthread_version());
5225
5226 if (UseNUMA) {
5227 Linux::numa_init();
5228 }
5229
5230 if (MaxFDLimit) {
5231 // set the number of file descriptors to max. print out error
5232 // if getrlimit/setrlimit fails but continue regardless.
5233 struct rlimit nbr_files;
5234 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5235 if (status != 0) {
5236 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5237 } else {
5238 nbr_files.rlim_cur = nbr_files.rlim_max;
5239 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5240 if (status != 0) {
5241 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5242 }
5243 }
5244 }
5245
5246 // Initialize lock used to serialize thread creation (see os::create_thread)
5247 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5248
5249 // at-exit methods are called in the reverse order of their registration.
5250 // atexit functions are called on return from main or as a result of a
5251 // call to exit(3C). There can be only 32 of these functions registered
5252 // and atexit() does not set errno.
5253
5254 if (PerfAllowAtExitRegistration) {
5255 // only register atexit functions if PerfAllowAtExitRegistration is set.
5256 // atexit functions can be delayed until process exit time, which
5257 // can be problematic for embedded VM situations. Embedded VMs should
5258 // call DestroyJavaVM() to assure that VM resources are released.
5259
5260 // note: perfMemory_exit_helper atexit function may be removed in
5261 // the future if the appropriate cleanup code can be added to the
5262 // VM_Exit VMOperation's doit method.
5263 if (atexit(perfMemory_exit_helper) != 0) {
5264 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5265 }
5266 }
5267
5268 // initialize thread priority policy
5269 prio_init();
5270
5271 if (!FLAG_IS_DEFAULT(AllocateHeapAt) || !FLAG_IS_DEFAULT(AllocateOldGenAt)) {
5272 set_coredump_filter(DAX_SHARED_BIT);
5273 }
5274
5275 if (DumpPrivateMappingsInCore) {
5276 set_coredump_filter(FILE_BACKED_PVT_BIT);
5277 }
5278
5279 if (DumpSharedMappingsInCore) {
5280 set_coredump_filter(FILE_BACKED_SHARED_BIT);
5281 }
5282
5283 return JNI_OK;
5284 }
5285
5286 // Mark the polling page as unreadable
5287 void os::make_polling_page_unreadable(void) {
5288 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5289 fatal("Could not disable polling page");
5290 }
5291 }
5292
5293 // Mark the polling page as readable
5294 void os::make_polling_page_readable(void) {
5295 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5296 fatal("Could not enable polling page");
5297 }
5298 }
5299
5300 // older glibc versions don't have this macro (which expands to
5301 // an optimized bit-counting function) so we have to roll our own
5302 #ifndef CPU_COUNT
5303
5304 static int _cpu_count(const cpu_set_t* cpus) {
5305 int count = 0;
5306 // only look up to the number of configured processors
5307 for (int i = 0; i < os::processor_count(); i++) {
5308 if (CPU_ISSET(i, cpus)) {
5309 count++;
5310 }
5311 }
5312 return count;
5313 }
5314
5315 #define CPU_COUNT(cpus) _cpu_count(cpus)
5316
5317 #endif // CPU_COUNT
5318
5319 // Get the current number of available processors for this process.
5320 // This value can change at any time during a process's lifetime.
5321 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5322 // If it appears there may be more than 1024 processors then we do a
5323 // dynamic check - see 6515172 for details.
5324 // If anything goes wrong we fallback to returning the number of online
5325 // processors - which can be greater than the number available to the process.
5326 int os::Linux::active_processor_count() {
5327 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5328 cpu_set_t* cpus_p = &cpus;
5329 int cpus_size = sizeof(cpu_set_t);
5330
5331 int configured_cpus = os::processor_count(); // upper bound on available cpus
5332 int cpu_count = 0;
5333
5334 // old build platforms may not support dynamic cpu sets
5335 #ifdef CPU_ALLOC
5336
5337 // To enable easy testing of the dynamic path on different platforms we
5338 // introduce a diagnostic flag: UseCpuAllocPath
5339 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5340 // kernel may use a mask bigger than cpu_set_t
5341 log_trace(os)("active_processor_count: using dynamic path %s"
5342 "- configured processors: %d",
5343 UseCpuAllocPath ? "(forced) " : "",
5344 configured_cpus);
5345 cpus_p = CPU_ALLOC(configured_cpus);
5346 if (cpus_p != NULL) {
5347 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5348 // zero it just to be safe
5349 CPU_ZERO_S(cpus_size, cpus_p);
5350 }
5351 else {
5352 // failed to allocate so fallback to online cpus
5353 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5354 log_trace(os)("active_processor_count: "
5355 "CPU_ALLOC failed (%s) - using "
5356 "online processor count: %d",
5357 os::strerror(errno), online_cpus);
5358 return online_cpus;
5359 }
5360 }
5361 else {
5362 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5363 configured_cpus);
5364 }
5365 #else // CPU_ALLOC
5366 // these stubs won't be executed
5367 #define CPU_COUNT_S(size, cpus) -1
5368 #define CPU_FREE(cpus)
5369
5370 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5371 configured_cpus);
5372 #endif // CPU_ALLOC
5373
5374 // pid 0 means the current thread - which we have to assume represents the process
5375 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5376 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5377 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5378 }
5379 else {
5380 cpu_count = CPU_COUNT(cpus_p);
5381 }
5382 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5383 }
5384 else {
5385 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5386 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5387 "which may exceed available processors", os::strerror(errno), cpu_count);
5388 }
5389
5390 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5391 CPU_FREE(cpus_p);
5392 }
5393
5394 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5395 return cpu_count;
5396 }
5397
5398 // Determine the active processor count from one of
5399 // three different sources:
5400 //
5401 // 1. User option -XX:ActiveProcessorCount
5402 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5403 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5404 //
5405 // Option 1, if specified, will always override.
5406 // If the cgroup subsystem is active and configured, we
5407 // will return the min of the cgroup and option 2 results.
5408 // This is required since tools, such as numactl, that
5409 // alter cpu affinity do not update cgroup subsystem
5410 // cpuset configuration files.
5411 int os::active_processor_count() {
5412 // User has overridden the number of active processors
5413 if (ActiveProcessorCount > 0) {
5414 log_trace(os)("active_processor_count: "
5415 "active processor count set by user : %d",
5416 ActiveProcessorCount);
5417 return ActiveProcessorCount;
5418 }
5419
5420 int active_cpus;
5421 if (OSContainer::is_containerized()) {
5422 active_cpus = OSContainer::active_processor_count();
5423 log_trace(os)("active_processor_count: determined by OSContainer: %d",
5424 active_cpus);
5425 } else {
5426 active_cpus = os::Linux::active_processor_count();
5427 }
5428
5429 return active_cpus;
5430 }
5431
5432 uint os::processor_id() {
5433 const int id = Linux::sched_getcpu();
5434 assert(id >= 0 && id < _processor_count, "Invalid processor id");
5435 return (uint)id;
5436 }
5437
5438 void os::set_native_thread_name(const char *name) {
5439 if (Linux::_pthread_setname_np) {
5440 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5441 snprintf(buf, sizeof(buf), "%s", name);
5442 buf[sizeof(buf) - 1] = '\0';
5443 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5444 // ERANGE should not happen; all other errors should just be ignored.
5445 assert(rc != ERANGE, "pthread_setname_np failed");
5446 }
5447 }
5448
5449 bool os::distribute_processes(uint length, uint* distribution) {
5450 // Not yet implemented.
5451 return false;
5452 }
5453
5454 bool os::bind_to_processor(uint processor_id) {
5455 // Not yet implemented.
5456 return false;
5457 }
5458
5459 ///
5460
5461 void os::SuspendedThreadTask::internal_do_task() {
5462 if (do_suspend(_thread->osthread())) {
5463 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5464 do_task(context);
5465 do_resume(_thread->osthread());
5466 }
5467 }
5468
5469 ////////////////////////////////////////////////////////////////////////////////
5470 // debug support
5471
5472 bool os::find(address addr, outputStream* st) {
5473 Dl_info dlinfo;
5474 memset(&dlinfo, 0, sizeof(dlinfo));
5475 if (dladdr(addr, &dlinfo) != 0) {
5476 st->print(PTR_FORMAT ": ", p2i(addr));
5477 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5478 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5479 p2i(addr) - p2i(dlinfo.dli_saddr));
5480 } else if (dlinfo.dli_fbase != NULL) {
5481 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5482 } else {
5483 st->print("<absolute address>");
5484 }
5485 if (dlinfo.dli_fname != NULL) {
5486 st->print(" in %s", dlinfo.dli_fname);
5487 }
5488 if (dlinfo.dli_fbase != NULL) {
5489 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5490 }
5491 st->cr();
5492
5493 if (Verbose) {
5494 // decode some bytes around the PC
5495 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5496 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5497 address lowest = (address) dlinfo.dli_sname;
5498 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5499 if (begin < lowest) begin = lowest;
5500 Dl_info dlinfo2;
5501 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5502 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5503 end = (address) dlinfo2.dli_saddr;
5504 }
5505 Disassembler::decode(begin, end, st);
5506 }
5507 return true;
5508 }
5509 return false;
5510 }
5511
5512 ////////////////////////////////////////////////////////////////////////////////
5513 // misc
5514
5515 // This does not do anything on Linux. This is basically a hook for being
5516 // able to use structured exception handling (thread-local exception filters)
5517 // on, e.g., Win32.
5518 void
5519 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5520 JavaCallArguments* args, Thread* thread) {
5521 f(value, method, args, thread);
5522 }
5523
5524 void os::print_statistics() {
5525 }
5526
5527 bool os::message_box(const char* title, const char* message) {
5528 int i;
5529 fdStream err(defaultStream::error_fd());
5530 for (i = 0; i < 78; i++) err.print_raw("=");
5531 err.cr();
5532 err.print_raw_cr(title);
5533 for (i = 0; i < 78; i++) err.print_raw("-");
5534 err.cr();
5535 err.print_raw_cr(message);
5536 for (i = 0; i < 78; i++) err.print_raw("=");
5537 err.cr();
5538
5539 char buf[16];
5540 // Prevent process from exiting upon "read error" without consuming all CPU
5541 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5542
5543 return buf[0] == 'y' || buf[0] == 'Y';
5544 }
5545
5546 // Is a (classpath) directory empty?
5547 bool os::dir_is_empty(const char* path) {
5548 DIR *dir = NULL;
5549 struct dirent *ptr;
5550
5551 dir = opendir(path);
5552 if (dir == NULL) return true;
5553
5554 // Scan the directory
5555 bool result = true;
5556 while (result && (ptr = readdir(dir)) != NULL) {
5557 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5558 result = false;
5559 }
5560 }
5561 closedir(dir);
5562 return result;
5563 }
5564
5565 // This code originates from JDK's sysOpen and open64_w
5566 // from src/solaris/hpi/src/system_md.c
5567
5568 int os::open(const char *path, int oflag, int mode) {
5569 if (strlen(path) > MAX_PATH - 1) {
5570 errno = ENAMETOOLONG;
5571 return -1;
5572 }
5573
5574 // All file descriptors that are opened in the Java process and not
5575 // specifically destined for a subprocess should have the close-on-exec
5576 // flag set. If we don't set it, then careless 3rd party native code
5577 // might fork and exec without closing all appropriate file descriptors
5578 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5579 // turn might:
5580 //
5581 // - cause end-of-file to fail to be detected on some file
5582 // descriptors, resulting in mysterious hangs, or
5583 //
5584 // - might cause an fopen in the subprocess to fail on a system
5585 // suffering from bug 1085341.
5586 //
5587 // (Yes, the default setting of the close-on-exec flag is a Unix
5588 // design flaw)
5589 //
5590 // See:
5591 // 1085341: 32-bit stdio routines should support file descriptors >255
5592 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5593 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5594 //
5595 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5596 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5597 // because it saves a system call and removes a small window where the flag
5598 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5599 // and we fall back to using FD_CLOEXEC (see below).
5600 #ifdef O_CLOEXEC
5601 oflag |= O_CLOEXEC;
5602 #endif
5603
5604 int fd = ::open64(path, oflag, mode);
5605 if (fd == -1) return -1;
5606
5607 //If the open succeeded, the file might still be a directory
5608 {
5609 struct stat64 buf64;
5610 int ret = ::fstat64(fd, &buf64);
5611 int st_mode = buf64.st_mode;
5612
5613 if (ret != -1) {
5614 if ((st_mode & S_IFMT) == S_IFDIR) {
5615 errno = EISDIR;
5616 ::close(fd);
5617 return -1;
5618 }
5619 } else {
5620 ::close(fd);
5621 return -1;
5622 }
5623 }
5624
5625 #ifdef FD_CLOEXEC
5626 // Validate that the use of the O_CLOEXEC flag on open above worked.
5627 // With recent kernels, we will perform this check exactly once.
5628 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5629 if (!O_CLOEXEC_is_known_to_work) {
5630 int flags = ::fcntl(fd, F_GETFD);
5631 if (flags != -1) {
5632 if ((flags & FD_CLOEXEC) != 0)
5633 O_CLOEXEC_is_known_to_work = 1;
5634 else
5635 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5636 }
5637 }
5638 #endif
5639
5640 return fd;
5641 }
5642
5643
5644 // create binary file, rewriting existing file if required
5645 int os::create_binary_file(const char* path, bool rewrite_existing) {
5646 int oflags = O_WRONLY | O_CREAT;
5647 if (!rewrite_existing) {
5648 oflags |= O_EXCL;
5649 }
5650 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5651 }
5652
5653 // return current position of file pointer
5654 jlong os::current_file_offset(int fd) {
5655 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5656 }
5657
5658 // move file pointer to the specified offset
5659 jlong os::seek_to_file_offset(int fd, jlong offset) {
5660 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5661 }
5662
5663 // This code originates from JDK's sysAvailable
5664 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5665
5666 int os::available(int fd, jlong *bytes) {
5667 jlong cur, end;
5668 int mode;
5669 struct stat64 buf64;
5670
5671 if (::fstat64(fd, &buf64) >= 0) {
5672 mode = buf64.st_mode;
5673 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5674 int n;
5675 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5676 *bytes = n;
5677 return 1;
5678 }
5679 }
5680 }
5681 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5682 return 0;
5683 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5684 return 0;
5685 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5686 return 0;
5687 }
5688 *bytes = end - cur;
5689 return 1;
5690 }
5691
5692 // Map a block of memory.
5693 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5694 char *addr, size_t bytes, bool read_only,
5695 bool allow_exec) {
5696 int prot;
5697 int flags = MAP_PRIVATE;
5698
5699 if (read_only) {
5700 prot = PROT_READ;
5701 } else {
5702 prot = PROT_READ | PROT_WRITE;
5703 }
5704
5705 if (allow_exec) {
5706 prot |= PROT_EXEC;
5707 }
5708
5709 if (addr != NULL) {
5710 flags |= MAP_FIXED;
5711 }
5712
5713 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5714 fd, file_offset);
5715 if (mapped_address == MAP_FAILED) {
5716 return NULL;
5717 }
5718 return mapped_address;
5719 }
5720
5721
5722 // Remap a block of memory.
5723 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5724 char *addr, size_t bytes, bool read_only,
5725 bool allow_exec) {
5726 // same as map_memory() on this OS
5727 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5728 allow_exec);
5729 }
5730
5731
5732 // Unmap a block of memory.
5733 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5734 return munmap(addr, bytes) == 0;
5735 }
5736
5737 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5738
5739 static jlong fast_cpu_time(Thread *thread) {
5740 clockid_t clockid;
5741 int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(),
5742 &clockid);
5743 if (rc == 0) {
5744 return os::Linux::fast_thread_cpu_time(clockid);
5745 } else {
5746 // It's possible to encounter a terminated native thread that failed
5747 // to detach itself from the VM - which should result in ESRCH.
5748 assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed");
5749 return -1;
5750 }
5751 }
5752
5753 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5754 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5755 // of a thread.
5756 //
5757 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5758 // the fast estimate available on the platform.
5759
5760 jlong os::current_thread_cpu_time() {
5761 if (os::Linux::supports_fast_thread_cpu_time()) {
5762 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5763 } else {
5764 // return user + sys since the cost is the same
5765 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5766 }
5767 }
5768
5769 jlong os::thread_cpu_time(Thread* thread) {
5770 // consistent with what current_thread_cpu_time() returns
5771 if (os::Linux::supports_fast_thread_cpu_time()) {
5772 return fast_cpu_time(thread);
5773 } else {
5774 return slow_thread_cpu_time(thread, true /* user + sys */);
5775 }
5776 }
5777
5778 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5779 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5780 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5781 } else {
5782 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5783 }
5784 }
5785
5786 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5787 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5788 return fast_cpu_time(thread);
5789 } else {
5790 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5791 }
5792 }
5793
5794 // -1 on error.
5795 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5796 pid_t tid = thread->osthread()->thread_id();
5797 char *s;
5798 char stat[2048];
5799 int statlen;
5800 char proc_name[64];
5801 int count;
5802 long sys_time, user_time;
5803 char cdummy;
5804 int idummy;
5805 long ldummy;
5806 FILE *fp;
5807
5808 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5809 fp = fopen(proc_name, "r");
5810 if (fp == NULL) return -1;
5811 statlen = fread(stat, 1, 2047, fp);
5812 stat[statlen] = '\0';
5813 fclose(fp);
5814
5815 // Skip pid and the command string. Note that we could be dealing with
5816 // weird command names, e.g. user could decide to rename java launcher
5817 // to "java 1.4.2 :)", then the stat file would look like
5818 // 1234 (java 1.4.2 :)) R ... ...
5819 // We don't really need to know the command string, just find the last
5820 // occurrence of ")" and then start parsing from there. See bug 4726580.
5821 s = strrchr(stat, ')');
5822 if (s == NULL) return -1;
5823
5824 // Skip blank chars
5825 do { s++; } while (s && isspace(*s));
5826
5827 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5828 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5829 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5830 &user_time, &sys_time);
5831 if (count != 13) return -1;
5832 if (user_sys_cpu_time) {
5833 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5834 } else {
5835 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5836 }
5837 }
5838
5839 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5840 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5841 info_ptr->may_skip_backward = false; // elapsed time not wall time
5842 info_ptr->may_skip_forward = false; // elapsed time not wall time
5843 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5844 }
5845
5846 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5847 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5848 info_ptr->may_skip_backward = false; // elapsed time not wall time
5849 info_ptr->may_skip_forward = false; // elapsed time not wall time
5850 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5851 }
5852
5853 bool os::is_thread_cpu_time_supported() {
5854 return true;
5855 }
5856
5857 // System loadavg support. Returns -1 if load average cannot be obtained.
5858 // Linux doesn't yet have a (official) notion of processor sets,
5859 // so just return the system wide load average.
5860 int os::loadavg(double loadavg[], int nelem) {
5861 return ::getloadavg(loadavg, nelem);
5862 }
5863
5864 void os::pause() {
5865 char filename[MAX_PATH];
5866 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5867 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5868 } else {
5869 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5870 }
5871
5872 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5873 if (fd != -1) {
5874 struct stat buf;
5875 ::close(fd);
5876 while (::stat(filename, &buf) == 0) {
5877 (void)::poll(NULL, 0, 100);
5878 }
5879 } else {
5880 jio_fprintf(stderr,
5881 "Could not open pause file '%s', continuing immediately.\n", filename);
5882 }
5883 }
5884
5885 extern char** environ;
5886
5887 // Run the specified command in a separate process. Return its exit value,
5888 // or -1 on failure (e.g. can't fork a new process).
5889 // Unlike system(), this function can be called from signal handler. It
5890 // doesn't block SIGINT et al.
5891 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) {
5892 const char * argv[4] = {"sh", "-c", cmd, NULL};
5893
5894 pid_t pid ;
5895
5896 if (use_vfork_if_available) {
5897 pid = vfork();
5898 } else {
5899 pid = fork();
5900 }
5901
5902 if (pid < 0) {
5903 // fork failed
5904 return -1;
5905
5906 } else if (pid == 0) {
5907 // child process
5908
5909 execve("/bin/sh", (char* const*)argv, environ);
5910
5911 // execve failed
5912 _exit(-1);
5913
5914 } else {
5915 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5916 // care about the actual exit code, for now.
5917
5918 int status;
5919
5920 // Wait for the child process to exit. This returns immediately if
5921 // the child has already exited. */
5922 while (waitpid(pid, &status, 0) < 0) {
5923 switch (errno) {
5924 case ECHILD: return 0;
5925 case EINTR: break;
5926 default: return -1;
5927 }
5928 }
5929
5930 if (WIFEXITED(status)) {
5931 // The child exited normally; get its exit code.
5932 return WEXITSTATUS(status);
5933 } else if (WIFSIGNALED(status)) {
5934 // The child exited because of a signal
5935 // The best value to return is 0x80 + signal number,
5936 // because that is what all Unix shells do, and because
5937 // it allows callers to distinguish between process exit and
5938 // process death by signal.
5939 return 0x80 + WTERMSIG(status);
5940 } else {
5941 // Unknown exit code; pass it through
5942 return status;
5943 }
5944 }
5945 }
5946
5947 // Get the default path to the core file
5948 // Returns the length of the string
5949 int os::get_core_path(char* buffer, size_t bufferSize) {
5950 /*
5951 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5952 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5953 */
5954 const int core_pattern_len = 129;
5955 char core_pattern[core_pattern_len] = {0};
5956
5957 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5958 if (core_pattern_file == -1) {
5959 return -1;
5960 }
5961
5962 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5963 ::close(core_pattern_file);
5964 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5965 return -1;
5966 }
5967 if (core_pattern[ret-1] == '\n') {
5968 core_pattern[ret-1] = '\0';
5969 } else {
5970 core_pattern[ret] = '\0';
5971 }
5972
5973 // Replace the %p in the core pattern with the process id. NOTE: we do this
5974 // only if the pattern doesn't start with "|", and we support only one %p in
5975 // the pattern.
5976 char *pid_pos = strstr(core_pattern, "%p");
5977 const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : ""; // skip over the "%p"
5978 int written;
5979
5980 if (core_pattern[0] == '/') {
5981 if (pid_pos != NULL) {
5982 *pid_pos = '\0';
5983 written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern,
5984 current_process_id(), tail);
5985 } else {
5986 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5987 }
5988 } else {
5989 char cwd[PATH_MAX];
5990
5991 const char* p = get_current_directory(cwd, PATH_MAX);
5992 if (p == NULL) {
5993 return -1;
5994 }
5995
5996 if (core_pattern[0] == '|') {
5997 written = jio_snprintf(buffer, bufferSize,
5998 "\"%s\" (or dumping to %s/core.%d)",
5999 &core_pattern[1], p, current_process_id());
6000 } else if (pid_pos != NULL) {
6001 *pid_pos = '\0';
6002 written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern,
6003 current_process_id(), tail);
6004 } else {
6005 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6006 }
6007 }
6008
6009 if (written < 0) {
6010 return -1;
6011 }
6012
6013 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6014 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6015
6016 if (core_uses_pid_file != -1) {
6017 char core_uses_pid = 0;
6018 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6019 ::close(core_uses_pid_file);
6020
6021 if (core_uses_pid == '1') {
6022 jio_snprintf(buffer + written, bufferSize - written,
6023 ".%d", current_process_id());
6024 }
6025 }
6026 }
6027
6028 return strlen(buffer);
6029 }
6030
6031 bool os::start_debugging(char *buf, int buflen) {
6032 int len = (int)strlen(buf);
6033 char *p = &buf[len];
6034
6035 jio_snprintf(p, buflen-len,
6036 "\n\n"
6037 "Do you want to debug the problem?\n\n"
6038 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
6039 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
6040 "Otherwise, press RETURN to abort...",
6041 os::current_process_id(), os::current_process_id(),
6042 os::current_thread_id(), os::current_thread_id());
6043
6044 bool yes = os::message_box("Unexpected Error", buf);
6045
6046 if (yes) {
6047 // yes, user asked VM to launch debugger
6048 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
6049 os::current_process_id(), os::current_process_id());
6050
6051 os::fork_and_exec(buf);
6052 yes = false;
6053 }
6054 return yes;
6055 }
6056
6057
6058 // Java/Compiler thread:
6059 //
6060 // Low memory addresses
6061 // P0 +------------------------+
6062 // | |\ Java thread created by VM does not have glibc
6063 // | glibc guard page | - guard page, attached Java thread usually has
6064 // | |/ 1 glibc guard page.
6065 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6066 // | |\
6067 // | HotSpot Guard Pages | - red, yellow and reserved pages
6068 // | |/
6069 // +------------------------+ JavaThread::stack_reserved_zone_base()
6070 // | |\
6071 // | Normal Stack | -
6072 // | |/
6073 // P2 +------------------------+ Thread::stack_base()
6074 //
6075 // Non-Java thread:
6076 //
6077 // Low memory addresses
6078 // P0 +------------------------+
6079 // | |\
6080 // | glibc guard page | - usually 1 page
6081 // | |/
6082 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6083 // | |\
6084 // | Normal Stack | -
6085 // | |/
6086 // P2 +------------------------+ Thread::stack_base()
6087 //
6088 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
6089 // returned from pthread_attr_getstack().
6090 // ** Due to NPTL implementation error, linux takes the glibc guard page out
6091 // of the stack size given in pthread_attr. We work around this for
6092 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
6093 //
6094 #ifndef ZERO
6095 static void current_stack_region(address * bottom, size_t * size) {
6096 if (os::is_primordial_thread()) {
6097 // primordial thread needs special handling because pthread_getattr_np()
6098 // may return bogus value.
6099 *bottom = os::Linux::initial_thread_stack_bottom();
6100 *size = os::Linux::initial_thread_stack_size();
6101 } else {
6102 pthread_attr_t attr;
6103
6104 int rslt = pthread_getattr_np(pthread_self(), &attr);
6105
6106 // JVM needs to know exact stack location, abort if it fails
6107 if (rslt != 0) {
6108 if (rslt == ENOMEM) {
6109 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
6110 } else {
6111 fatal("pthread_getattr_np failed with error = %d", rslt);
6112 }
6113 }
6114
6115 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
6116 fatal("Cannot locate current stack attributes!");
6117 }
6118
6119 // Work around NPTL stack guard error.
6120 size_t guard_size = 0;
6121 rslt = pthread_attr_getguardsize(&attr, &guard_size);
6122 if (rslt != 0) {
6123 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
6124 }
6125 *bottom += guard_size;
6126 *size -= guard_size;
6127
6128 pthread_attr_destroy(&attr);
6129
6130 }
6131 assert(os::current_stack_pointer() >= *bottom &&
6132 os::current_stack_pointer() < *bottom + *size, "just checking");
6133 }
6134
6135 address os::current_stack_base() {
6136 address bottom;
6137 size_t size;
6138 current_stack_region(&bottom, &size);
6139 return (bottom + size);
6140 }
6141
6142 size_t os::current_stack_size() {
6143 // This stack size includes the usable stack and HotSpot guard pages
6144 // (for the threads that have Hotspot guard pages).
6145 address bottom;
6146 size_t size;
6147 current_stack_region(&bottom, &size);
6148 return size;
6149 }
6150 #endif
6151
6152 static inline struct timespec get_mtime(const char* filename) {
6153 struct stat st;
6154 int ret = os::stat(filename, &st);
6155 assert(ret == 0, "failed to stat() file '%s': %s", filename, os::strerror(errno));
6156 return st.st_mtim;
6157 }
6158
6159 int os::compare_file_modified_times(const char* file1, const char* file2) {
6160 struct timespec filetime1 = get_mtime(file1);
6161 struct timespec filetime2 = get_mtime(file2);
6162 int diff = filetime1.tv_sec - filetime2.tv_sec;
6163 if (diff == 0) {
6164 return filetime1.tv_nsec - filetime2.tv_nsec;
6165 }
6166 return diff;
6167 }
6168
6169 /////////////// Unit tests ///////////////
6170
6171 #ifndef PRODUCT
6172
6173 class TestReserveMemorySpecial : AllStatic {
6174 public:
6175 static void small_page_write(void* addr, size_t size) {
6176 size_t page_size = os::vm_page_size();
6177
6178 char* end = (char*)addr + size;
6179 for (char* p = (char*)addr; p < end; p += page_size) {
6180 *p = 1;
6181 }
6182 }
6183
6184 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6185 if (!UseHugeTLBFS) {
6186 return;
6187 }
6188
6189 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6190
6191 if (addr != NULL) {
6192 small_page_write(addr, size);
6193
6194 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6195 }
6196 }
6197
6198 static void test_reserve_memory_special_huge_tlbfs_only() {
6199 if (!UseHugeTLBFS) {
6200 return;
6201 }
6202
6203 size_t lp = os::large_page_size();
6204
6205 for (size_t size = lp; size <= lp * 10; size += lp) {
6206 test_reserve_memory_special_huge_tlbfs_only(size);
6207 }
6208 }
6209
6210 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6211 size_t lp = os::large_page_size();
6212 size_t ag = os::vm_allocation_granularity();
6213
6214 // sizes to test
6215 const size_t sizes[] = {
6216 lp, lp + ag, lp + lp / 2, lp * 2,
6217 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6218 lp * 10, lp * 10 + lp / 2
6219 };
6220 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6221
6222 // For each size/alignment combination, we test three scenarios:
6223 // 1) with req_addr == NULL
6224 // 2) with a non-null req_addr at which we expect to successfully allocate
6225 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6226 // expect the allocation to either fail or to ignore req_addr
6227
6228 // Pre-allocate two areas; they shall be as large as the largest allocation
6229 // and aligned to the largest alignment we will be testing.
6230 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6231 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6232 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6233 -1, 0);
6234 assert(mapping1 != MAP_FAILED, "should work");
6235
6236 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6237 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6238 -1, 0);
6239 assert(mapping2 != MAP_FAILED, "should work");
6240
6241 // Unmap the first mapping, but leave the second mapping intact: the first
6242 // mapping will serve as a value for a "good" req_addr (case 2). The second
6243 // mapping, still intact, as "bad" req_addr (case 3).
6244 ::munmap(mapping1, mapping_size);
6245
6246 // Case 1
6247 for (int i = 0; i < num_sizes; i++) {
6248 const size_t size = sizes[i];
6249 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6250 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6251 if (p != NULL) {
6252 assert(is_aligned(p, alignment), "must be");
6253 small_page_write(p, size);
6254 os::Linux::release_memory_special_huge_tlbfs(p, size);
6255 }
6256 }
6257 }
6258
6259 // Case 2
6260 for (int i = 0; i < num_sizes; i++) {
6261 const size_t size = sizes[i];
6262 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6263 char* const req_addr = align_up(mapping1, alignment);
6264 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6265 if (p != NULL) {
6266 assert(p == req_addr, "must be");
6267 small_page_write(p, size);
6268 os::Linux::release_memory_special_huge_tlbfs(p, size);
6269 }
6270 }
6271 }
6272
6273 // Case 3
6274 for (int i = 0; i < num_sizes; i++) {
6275 const size_t size = sizes[i];
6276 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6277 char* const req_addr = align_up(mapping2, alignment);
6278 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6279 // as the area around req_addr contains already existing mappings, the API should always
6280 // return NULL (as per contract, it cannot return another address)
6281 assert(p == NULL, "must be");
6282 }
6283 }
6284
6285 ::munmap(mapping2, mapping_size);
6286
6287 }
6288
6289 static void test_reserve_memory_special_huge_tlbfs() {
6290 if (!UseHugeTLBFS) {
6291 return;
6292 }
6293
6294 test_reserve_memory_special_huge_tlbfs_only();
6295 test_reserve_memory_special_huge_tlbfs_mixed();
6296 }
6297
6298 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6299 if (!UseSHM) {
6300 return;
6301 }
6302
6303 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6304
6305 if (addr != NULL) {
6306 assert(is_aligned(addr, alignment), "Check");
6307 assert(is_aligned(addr, os::large_page_size()), "Check");
6308
6309 small_page_write(addr, size);
6310
6311 os::Linux::release_memory_special_shm(addr, size);
6312 }
6313 }
6314
6315 static void test_reserve_memory_special_shm() {
6316 size_t lp = os::large_page_size();
6317 size_t ag = os::vm_allocation_granularity();
6318
6319 for (size_t size = ag; size < lp * 3; size += ag) {
6320 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6321 test_reserve_memory_special_shm(size, alignment);
6322 }
6323 }
6324 }
6325
6326 static void test() {
6327 test_reserve_memory_special_huge_tlbfs();
6328 test_reserve_memory_special_shm();
6329 }
6330 };
6331
6332 void TestReserveMemorySpecial_test() {
6333 TestReserveMemorySpecial::test();
6334 }
6335
6336 #endif