1 /*
2 * Copyright (c) 1999, 2017, 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 "memory/allocation.inline.hpp"
37 #include "memory/filemap.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "os_linux.inline.hpp"
40 #include "os_share_linux.hpp"
41 #include "osContainer_linux.hpp"
42 #include "prims/jniFastGetField.hpp"
43 #include "prims/jvm_misc.hpp"
44 #include "runtime/arguments.hpp"
45 #include "runtime/atomic.hpp"
46 #include "runtime/extendedPC.hpp"
47 #include "runtime/globals.hpp"
48 #include "runtime/interfaceSupport.hpp"
49 #include "runtime/init.hpp"
50 #include "runtime/java.hpp"
51 #include "runtime/javaCalls.hpp"
52 #include "runtime/mutexLocker.hpp"
53 #include "runtime/objectMonitor.hpp"
54 #include "runtime/orderAccess.inline.hpp"
55 #include "runtime/osThread.hpp"
56 #include "runtime/perfMemory.hpp"
57 #include "runtime/sharedRuntime.hpp"
58 #include "runtime/statSampler.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "runtime/thread.inline.hpp"
61 #include "runtime/threadCritical.hpp"
62 #include "runtime/timer.hpp"
63 #include "semaphore_posix.hpp"
64 #include "services/attachListener.hpp"
65 #include "services/memTracker.hpp"
66 #include "services/runtimeService.hpp"
67 #include "utilities/align.hpp"
68 #include "utilities/decoder.hpp"
69 #include "utilities/defaultStream.hpp"
70 #include "utilities/events.hpp"
71 #include "utilities/elfFile.hpp"
72 #include "utilities/growableArray.hpp"
73 #include "utilities/macros.hpp"
74 #include "utilities/vmError.hpp"
75
76 // put OS-includes here
77 # include <sys/types.h>
78 # include <sys/mman.h>
79 # include <sys/stat.h>
80 # include <sys/select.h>
81 # include <pthread.h>
82 # include <signal.h>
83 # include <errno.h>
84 # include <dlfcn.h>
85 # include <stdio.h>
86 # include <unistd.h>
87 # include <sys/resource.h>
88 # include <pthread.h>
89 # include <sys/stat.h>
90 # include <sys/time.h>
91 # include <sys/times.h>
92 # include <sys/utsname.h>
93 # include <sys/socket.h>
94 # include <sys/wait.h>
95 # include <pwd.h>
96 # include <poll.h>
97 # include <semaphore.h>
98 # include <fcntl.h>
99 # include <string.h>
100 # include <syscall.h>
101 # include <sys/sysinfo.h>
102 # include <gnu/libc-version.h>
103 # include <sys/ipc.h>
104 # include <sys/shm.h>
105 # include <link.h>
106 # include <stdint.h>
107 # include <inttypes.h>
108 # include <sys/ioctl.h>
109
110 #ifndef _GNU_SOURCE
111 #define _GNU_SOURCE
112 #include <sched.h>
113 #undef _GNU_SOURCE
114 #else
115 #include <sched.h>
116 #endif
117
118 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
119 // getrusage() is prepared to handle the associated failure.
120 #ifndef RUSAGE_THREAD
121 #define RUSAGE_THREAD (1) /* only the calling thread */
122 #endif
123
124 #define MAX_PATH (2 * K)
125
126 #define MAX_SECS 100000000
127
128 // for timer info max values which include all bits
129 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
130
131 #define LARGEPAGES_BIT (1 << 6)
132 ////////////////////////////////////////////////////////////////////////////////
133 // global variables
134 julong os::Linux::_physical_memory = 0;
135
136 address os::Linux::_initial_thread_stack_bottom = NULL;
137 uintptr_t os::Linux::_initial_thread_stack_size = 0;
138
139 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
140 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
141 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
142 Mutex* os::Linux::_createThread_lock = NULL;
143 pthread_t os::Linux::_main_thread;
144 int os::Linux::_page_size = -1;
145 bool os::Linux::_supports_fast_thread_cpu_time = false;
146 uint32_t os::Linux::_os_version = 0;
147 const char * os::Linux::_glibc_version = NULL;
148 const char * os::Linux::_libpthread_version = NULL;
149
150 static jlong initial_time_count=0;
151
152 static int clock_tics_per_sec = 100;
153
154 // For diagnostics to print a message once. see run_periodic_checks
155 static sigset_t check_signal_done;
156 static bool check_signals = true;
157
158 // Signal number used to suspend/resume a thread
159
160 // do not use any signal number less than SIGSEGV, see 4355769
161 static int SR_signum = SIGUSR2;
162 sigset_t SR_sigset;
163
164 // utility functions
165
166 static int SR_initialize();
167
168 julong os::available_memory() {
169 return Linux::available_memory();
170 }
171
172 julong os::Linux::available_memory() {
173 // values in struct sysinfo are "unsigned long"
174 struct sysinfo si;
175 julong avail_mem;
176
177 if (OSContainer::is_containerized()) {
178 jlong mem_limit, mem_usage;
179 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
180 if ((mem_usage = OSContainer::memory_usage_in_bytes()) > 0) {
181 if (mem_limit > mem_usage) {
182 avail_mem = (julong)mem_limit - (julong)mem_usage;
183 } else {
184 avail_mem = 0;
185 }
186 log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
187 return avail_mem;
188 } else {
189 log_debug(os,container)("container memory usage call failed: " JLONG_FORMAT, mem_usage);
190 }
191 } else {
192 log_debug(os,container)("container memory unlimited or failed: " JLONG_FORMAT, mem_limit);
193 }
194 }
195
196 sysinfo(&si);
197 avail_mem = (julong)si.freeram * si.mem_unit;
198 log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
199 return avail_mem;
200 }
201
202 julong os::physical_memory() {
203 if (OSContainer::is_containerized()) {
204 jlong mem_limit;
205 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
206 log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
207 return (julong)mem_limit;
208 } else {
209 if (mem_limit == OSCONTAINER_ERROR) {
210 log_debug(os,container)("container memory limit call failed");
211 }
212 if (mem_limit == -1) {
213 log_debug(os,container)("container memory unlimited, using host value");
214 }
215 }
216 }
217
218 jlong phys_mem = Linux::physical_memory();
219 log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
220 return phys_mem;
221 }
222
223 // Return true if user is running as root.
224
225 bool os::have_special_privileges() {
226 static bool init = false;
227 static bool privileges = false;
228 if (!init) {
229 privileges = (getuid() != geteuid()) || (getgid() != getegid());
230 init = true;
231 }
232 return privileges;
233 }
234
235
236 #ifndef SYS_gettid
237 // i386: 224, ia64: 1105, amd64: 186, sparc 143
238 #ifdef __ia64__
239 #define SYS_gettid 1105
240 #else
241 #ifdef __i386__
242 #define SYS_gettid 224
243 #else
244 #ifdef __amd64__
245 #define SYS_gettid 186
246 #else
247 #ifdef __sparc__
248 #define SYS_gettid 143
249 #else
250 #error define gettid for the arch
251 #endif
252 #endif
253 #endif
254 #endif
255 #endif
256
257
258 // pid_t gettid()
259 //
260 // Returns the kernel thread id of the currently running thread. Kernel
261 // thread id is used to access /proc.
262 pid_t os::Linux::gettid() {
263 int rslt = syscall(SYS_gettid);
264 assert(rslt != -1, "must be."); // old linuxthreads implementation?
265 return (pid_t)rslt;
266 }
267
268 // Most versions of linux have a bug where the number of processors are
269 // determined by looking at the /proc file system. In a chroot environment,
270 // the system call returns 1. This causes the VM to act as if it is
271 // a single processor and elide locking (see is_MP() call).
272 static bool unsafe_chroot_detected = false;
273 static const char *unstable_chroot_error = "/proc file system not found.\n"
274 "Java may be unstable running multithreaded in a chroot "
275 "environment on Linux when /proc filesystem is not mounted.";
276
277 void os::Linux::initialize_system_info() {
278 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
279 if (processor_count() == 1) {
280 pid_t pid = os::Linux::gettid();
281 char fname[32];
282 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
283 FILE *fp = fopen(fname, "r");
284 if (fp == NULL) {
285 unsafe_chroot_detected = true;
286 } else {
287 fclose(fp);
288 }
289 }
290 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
291 assert(processor_count() > 0, "linux error");
292 }
293
294 void os::init_system_properties_values() {
295 // The next steps are taken in the product version:
296 //
297 // Obtain the JAVA_HOME value from the location of libjvm.so.
298 // This library should be located at:
299 // <JAVA_HOME>/lib/{client|server}/libjvm.so.
300 //
301 // If "/jre/lib/" appears at the right place in the path, then we
302 // assume libjvm.so is installed in a JDK and we use this path.
303 //
304 // Otherwise exit with message: "Could not create the Java virtual machine."
305 //
306 // The following extra steps are taken in the debugging version:
307 //
308 // If "/jre/lib/" does NOT appear at the right place in the path
309 // instead of exit check for $JAVA_HOME environment variable.
310 //
311 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
312 // then we append a fake suffix "hotspot/libjvm.so" to this path so
313 // it looks like libjvm.so is installed there
314 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
315 //
316 // Otherwise exit.
317 //
318 // Important note: if the location of libjvm.so changes this
319 // code needs to be changed accordingly.
320
321 // See ld(1):
322 // The linker uses the following search paths to locate required
323 // shared libraries:
324 // 1: ...
325 // ...
326 // 7: The default directories, normally /lib and /usr/lib.
327 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
328 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
329 #else
330 #define DEFAULT_LIBPATH "/lib:/usr/lib"
331 #endif
332
333 // Base path of extensions installed on the system.
334 #define SYS_EXT_DIR "/usr/java/packages"
335 #define EXTENSIONS_DIR "/lib/ext"
336
337 // Buffer that fits several sprintfs.
338 // Note that the space for the colon and the trailing null are provided
339 // by the nulls included by the sizeof operator.
340 const size_t bufsize =
341 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
342 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
343 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
344
345 // sysclasspath, java_home, dll_dir
346 {
347 char *pslash;
348 os::jvm_path(buf, bufsize);
349
350 // Found the full path to libjvm.so.
351 // Now cut the path to <java_home>/jre if we can.
352 pslash = strrchr(buf, '/');
353 if (pslash != NULL) {
354 *pslash = '\0'; // Get rid of /libjvm.so.
355 }
356 pslash = strrchr(buf, '/');
357 if (pslash != NULL) {
358 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
359 }
360 Arguments::set_dll_dir(buf);
361
362 if (pslash != NULL) {
363 pslash = strrchr(buf, '/');
364 if (pslash != NULL) {
365 *pslash = '\0'; // Get rid of /lib.
366 }
367 }
368 Arguments::set_java_home(buf);
369 set_boot_path('/', ':');
370 }
371
372 // Where to look for native libraries.
373 //
374 // Note: Due to a legacy implementation, most of the library path
375 // is set in the launcher. This was to accomodate linking restrictions
376 // on legacy Linux implementations (which are no longer supported).
377 // Eventually, all the library path setting will be done here.
378 //
379 // However, to prevent the proliferation of improperly built native
380 // libraries, the new path component /usr/java/packages is added here.
381 // Eventually, all the library path setting will be done here.
382 {
383 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
384 // should always exist (until the legacy problem cited above is
385 // addressed).
386 const char *v = ::getenv("LD_LIBRARY_PATH");
387 const char *v_colon = ":";
388 if (v == NULL) { v = ""; v_colon = ""; }
389 // That's +1 for the colon and +1 for the trailing '\0'.
390 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
391 strlen(v) + 1 +
392 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
393 mtInternal);
394 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
395 Arguments::set_library_path(ld_library_path);
396 FREE_C_HEAP_ARRAY(char, ld_library_path);
397 }
398
399 // Extensions directories.
400 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
401 Arguments::set_ext_dirs(buf);
402
403 FREE_C_HEAP_ARRAY(char, buf);
404
405 #undef DEFAULT_LIBPATH
406 #undef SYS_EXT_DIR
407 #undef EXTENSIONS_DIR
408 }
409
410 ////////////////////////////////////////////////////////////////////////////////
411 // breakpoint support
412
413 void os::breakpoint() {
414 BREAKPOINT;
415 }
416
417 extern "C" void breakpoint() {
418 // use debugger to set breakpoint here
419 }
420
421 ////////////////////////////////////////////////////////////////////////////////
422 // signal support
423
424 debug_only(static bool signal_sets_initialized = false);
425 static sigset_t unblocked_sigs, vm_sigs;
426
427 bool os::Linux::is_sig_ignored(int sig) {
428 struct sigaction oact;
429 sigaction(sig, (struct sigaction*)NULL, &oact);
430 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
431 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
432 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
433 return true;
434 } else {
435 return false;
436 }
437 }
438
439 void os::Linux::signal_sets_init() {
440 // Should also have an assertion stating we are still single-threaded.
441 assert(!signal_sets_initialized, "Already initialized");
442 // Fill in signals that are necessarily unblocked for all threads in
443 // the VM. Currently, we unblock the following signals:
444 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
445 // by -Xrs (=ReduceSignalUsage));
446 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
447 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
448 // the dispositions or masks wrt these signals.
449 // Programs embedding the VM that want to use the above signals for their
450 // own purposes must, at this time, use the "-Xrs" option to prevent
451 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
452 // (See bug 4345157, and other related bugs).
453 // In reality, though, unblocking these signals is really a nop, since
454 // these signals are not blocked by default.
455 sigemptyset(&unblocked_sigs);
456 sigaddset(&unblocked_sigs, SIGILL);
457 sigaddset(&unblocked_sigs, SIGSEGV);
458 sigaddset(&unblocked_sigs, SIGBUS);
459 sigaddset(&unblocked_sigs, SIGFPE);
460 #if defined(PPC64)
461 sigaddset(&unblocked_sigs, SIGTRAP);
462 #endif
463 sigaddset(&unblocked_sigs, SR_signum);
464
465 if (!ReduceSignalUsage) {
466 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
467 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
468 }
469 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
470 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
471 }
472 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
473 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
474 }
475 }
476 // Fill in signals that are blocked by all but the VM thread.
477 sigemptyset(&vm_sigs);
478 if (!ReduceSignalUsage) {
479 sigaddset(&vm_sigs, BREAK_SIGNAL);
480 }
481 debug_only(signal_sets_initialized = true);
482
483 }
484
485 // These are signals that are unblocked while a thread is running Java.
486 // (For some reason, they get blocked by default.)
487 sigset_t* os::Linux::unblocked_signals() {
488 assert(signal_sets_initialized, "Not initialized");
489 return &unblocked_sigs;
490 }
491
492 // These are the signals that are blocked while a (non-VM) thread is
493 // running Java. Only the VM thread handles these signals.
494 sigset_t* os::Linux::vm_signals() {
495 assert(signal_sets_initialized, "Not initialized");
496 return &vm_sigs;
497 }
498
499 void os::Linux::hotspot_sigmask(Thread* thread) {
500
501 //Save caller's signal mask before setting VM signal mask
502 sigset_t caller_sigmask;
503 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
504
505 OSThread* osthread = thread->osthread();
506 osthread->set_caller_sigmask(caller_sigmask);
507
508 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
509
510 if (!ReduceSignalUsage) {
511 if (thread->is_VM_thread()) {
512 // Only the VM thread handles BREAK_SIGNAL ...
513 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
514 } else {
515 // ... all other threads block BREAK_SIGNAL
516 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
517 }
518 }
519 }
520
521 //////////////////////////////////////////////////////////////////////////////
522 // detecting pthread library
523
524 void os::Linux::libpthread_init() {
525 // Save glibc and pthread version strings.
526 #if !defined(_CS_GNU_LIBC_VERSION) || \
527 !defined(_CS_GNU_LIBPTHREAD_VERSION)
528 #error "glibc too old (< 2.3.2)"
529 #endif
530
531 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
532 assert(n > 0, "cannot retrieve glibc version");
533 char *str = (char *)malloc(n, mtInternal);
534 confstr(_CS_GNU_LIBC_VERSION, str, n);
535 os::Linux::set_glibc_version(str);
536
537 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
538 assert(n > 0, "cannot retrieve pthread version");
539 str = (char *)malloc(n, mtInternal);
540 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
541 os::Linux::set_libpthread_version(str);
542 }
543
544 /////////////////////////////////////////////////////////////////////////////
545 // thread stack expansion
546
547 // os::Linux::manually_expand_stack() takes care of expanding the thread
548 // stack. Note that this is normally not needed: pthread stacks allocate
549 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
550 // committed. Therefore it is not necessary to expand the stack manually.
551 //
552 // Manually expanding the stack was historically needed on LinuxThreads
553 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
554 // it is kept to deal with very rare corner cases:
555 //
556 // For one, user may run the VM on an own implementation of threads
557 // whose stacks are - like the old LinuxThreads - implemented using
558 // mmap(MAP_GROWSDOWN).
559 //
560 // Also, this coding may be needed if the VM is running on the primordial
561 // thread. Normally we avoid running on the primordial thread; however,
562 // user may still invoke the VM on the primordial thread.
563 //
564 // The following historical comment describes the details about running
565 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
566
567
568 // Force Linux kernel to expand current thread stack. If "bottom" is close
569 // to the stack guard, caller should block all signals.
570 //
571 // MAP_GROWSDOWN:
572 // A special mmap() flag that is used to implement thread stacks. It tells
573 // kernel that the memory region should extend downwards when needed. This
574 // allows early versions of LinuxThreads to only mmap the first few pages
575 // when creating a new thread. Linux kernel will automatically expand thread
576 // stack as needed (on page faults).
577 //
578 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
579 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
580 // region, it's hard to tell if the fault is due to a legitimate stack
581 // access or because of reading/writing non-exist memory (e.g. buffer
582 // overrun). As a rule, if the fault happens below current stack pointer,
583 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
584 // application (see Linux kernel fault.c).
585 //
586 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
587 // stack overflow detection.
588 //
589 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
590 // not use MAP_GROWSDOWN.
591 //
592 // To get around the problem and allow stack banging on Linux, we need to
593 // manually expand thread stack after receiving the SIGSEGV.
594 //
595 // There are two ways to expand thread stack to address "bottom", we used
596 // both of them in JVM before 1.5:
597 // 1. adjust stack pointer first so that it is below "bottom", and then
598 // touch "bottom"
599 // 2. mmap() the page in question
600 //
601 // Now alternate signal stack is gone, it's harder to use 2. For instance,
602 // if current sp is already near the lower end of page 101, and we need to
603 // call mmap() to map page 100, it is possible that part of the mmap() frame
604 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
605 // That will destroy the mmap() frame and cause VM to crash.
606 //
607 // The following code works by adjusting sp first, then accessing the "bottom"
608 // page to force a page fault. Linux kernel will then automatically expand the
609 // stack mapping.
610 //
611 // _expand_stack_to() assumes its frame size is less than page size, which
612 // should always be true if the function is not inlined.
613
614 static void NOINLINE _expand_stack_to(address bottom) {
615 address sp;
616 size_t size;
617 volatile char *p;
618
619 // Adjust bottom to point to the largest address within the same page, it
620 // gives us a one-page buffer if alloca() allocates slightly more memory.
621 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
622 bottom += os::Linux::page_size() - 1;
623
624 // sp might be slightly above current stack pointer; if that's the case, we
625 // will alloca() a little more space than necessary, which is OK. Don't use
626 // os::current_stack_pointer(), as its result can be slightly below current
627 // stack pointer, causing us to not alloca enough to reach "bottom".
628 sp = (address)&sp;
629
630 if (sp > bottom) {
631 size = sp - bottom;
632 p = (volatile char *)alloca(size);
633 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
634 p[0] = '\0';
635 }
636 }
637
638 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
639 assert(t!=NULL, "just checking");
640 assert(t->osthread()->expanding_stack(), "expand should be set");
641 assert(t->stack_base() != NULL, "stack_base was not initialized");
642
643 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
644 sigset_t mask_all, old_sigset;
645 sigfillset(&mask_all);
646 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
647 _expand_stack_to(addr);
648 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
649 return true;
650 }
651 return false;
652 }
653
654 //////////////////////////////////////////////////////////////////////////////
655 // create new thread
656
657 // Thread start routine for all newly created threads
658 static void *thread_native_entry(Thread *thread) {
659 // Try to randomize the cache line index of hot stack frames.
660 // This helps when threads of the same stack traces evict each other's
661 // cache lines. The threads can be either from the same JVM instance, or
662 // from different JVM instances. The benefit is especially true for
663 // processors with hyperthreading technology.
664 static int counter = 0;
665 int pid = os::current_process_id();
666 alloca(((pid ^ counter++) & 7) * 128);
667
668 thread->initialize_thread_current();
669
670 OSThread* osthread = thread->osthread();
671 Monitor* sync = osthread->startThread_lock();
672
673 osthread->set_thread_id(os::current_thread_id());
674
675 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
676 os::current_thread_id(), (uintx) pthread_self());
677
678 if (UseNUMA) {
679 int lgrp_id = os::numa_get_group_id();
680 if (lgrp_id != -1) {
681 thread->set_lgrp_id(lgrp_id);
682 }
683 }
684 // initialize signal mask for this thread
685 os::Linux::hotspot_sigmask(thread);
686
687 // initialize floating point control register
688 os::Linux::init_thread_fpu_state();
689
690 // handshaking with parent thread
691 {
692 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
693
694 // notify parent thread
695 osthread->set_state(INITIALIZED);
696 sync->notify_all();
697
698 // wait until os::start_thread()
699 while (osthread->get_state() == INITIALIZED) {
700 sync->wait(Mutex::_no_safepoint_check_flag);
701 }
702 }
703
704 // call one more level start routine
705 thread->run();
706
707 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
708 os::current_thread_id(), (uintx) pthread_self());
709
710 // If a thread has not deleted itself ("delete this") as part of its
711 // termination sequence, we have to ensure thread-local-storage is
712 // cleared before we actually terminate. No threads should ever be
713 // deleted asynchronously with respect to their termination.
714 if (Thread::current_or_null_safe() != NULL) {
715 assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
716 thread->clear_thread_current();
717 }
718
719 return 0;
720 }
721
722 bool os::create_thread(Thread* thread, ThreadType thr_type,
723 size_t req_stack_size) {
724 assert(thread->osthread() == NULL, "caller responsible");
725
726 // Allocate the OSThread object
727 OSThread* osthread = new OSThread(NULL, NULL);
728 if (osthread == NULL) {
729 return false;
730 }
731
732 // set the correct thread state
733 osthread->set_thread_type(thr_type);
734
735 // Initial state is ALLOCATED but not INITIALIZED
736 osthread->set_state(ALLOCATED);
737
738 thread->set_osthread(osthread);
739
740 // init thread attributes
741 pthread_attr_t attr;
742 pthread_attr_init(&attr);
743 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
744
745 // Calculate stack size if it's not specified by caller.
746 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
747 // In the Linux NPTL pthread implementation the guard size mechanism
748 // is not implemented properly. The posix standard requires adding
749 // the size of the guard pages to the stack size, instead Linux
750 // takes the space out of 'stacksize'. Thus we adapt the requested
751 // stack_size by the size of the guard pages to mimick proper
752 // behaviour. However, be careful not to end up with a size
753 // of zero due to overflow. Don't add the guard page in that case.
754 size_t guard_size = os::Linux::default_guard_size(thr_type);
755 if (stack_size <= SIZE_MAX - guard_size) {
756 stack_size += guard_size;
757 }
758 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
759
760 int status = pthread_attr_setstacksize(&attr, stack_size);
761 assert_status(status == 0, status, "pthread_attr_setstacksize");
762
763 // Configure glibc guard page.
764 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
765
766 ThreadState state;
767
768 {
769 pthread_t tid;
770 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
771
772 char buf[64];
773 if (ret == 0) {
774 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
775 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
776 } else {
777 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
778 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
779 }
780
781 pthread_attr_destroy(&attr);
782
783 if (ret != 0) {
784 // Need to clean up stuff we've allocated so far
785 thread->set_osthread(NULL);
786 delete osthread;
787 return false;
788 }
789
790 // Store pthread info into the OSThread
791 osthread->set_pthread_id(tid);
792
793 // Wait until child thread is either initialized or aborted
794 {
795 Monitor* sync_with_child = osthread->startThread_lock();
796 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
797 while ((state = osthread->get_state()) == ALLOCATED) {
798 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
799 }
800 }
801 }
802
803 // Aborted due to thread limit being reached
804 if (state == ZOMBIE) {
805 thread->set_osthread(NULL);
806 delete osthread;
807 return false;
808 }
809
810 // The thread is returned suspended (in state INITIALIZED),
811 // and is started higher up in the call chain
812 assert(state == INITIALIZED, "race condition");
813 return true;
814 }
815
816 /////////////////////////////////////////////////////////////////////////////
817 // attach existing thread
818
819 // bootstrap the main thread
820 bool os::create_main_thread(JavaThread* thread) {
821 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
822 return create_attached_thread(thread);
823 }
824
825 bool os::create_attached_thread(JavaThread* thread) {
826 #ifdef ASSERT
827 thread->verify_not_published();
828 #endif
829
830 // Allocate the OSThread object
831 OSThread* osthread = new OSThread(NULL, NULL);
832
833 if (osthread == NULL) {
834 return false;
835 }
836
837 // Store pthread info into the OSThread
838 osthread->set_thread_id(os::Linux::gettid());
839 osthread->set_pthread_id(::pthread_self());
840
841 // initialize floating point control register
842 os::Linux::init_thread_fpu_state();
843
844 // Initial thread state is RUNNABLE
845 osthread->set_state(RUNNABLE);
846
847 thread->set_osthread(osthread);
848
849 if (UseNUMA) {
850 int lgrp_id = os::numa_get_group_id();
851 if (lgrp_id != -1) {
852 thread->set_lgrp_id(lgrp_id);
853 }
854 }
855
856 if (os::is_primordial_thread()) {
857 // If current thread is primordial thread, its stack is mapped on demand,
858 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
859 // the entire stack region to avoid SEGV in stack banging.
860 // It is also useful to get around the heap-stack-gap problem on SuSE
861 // kernel (see 4821821 for details). We first expand stack to the top
862 // of yellow zone, then enable stack yellow zone (order is significant,
863 // enabling yellow zone first will crash JVM on SuSE Linux), so there
864 // is no gap between the last two virtual memory regions.
865
866 JavaThread *jt = (JavaThread *)thread;
867 address addr = jt->stack_reserved_zone_base();
868 assert(addr != NULL, "initialization problem?");
869 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
870
871 osthread->set_expanding_stack();
872 os::Linux::manually_expand_stack(jt, addr);
873 osthread->clear_expanding_stack();
874 }
875
876 // initialize signal mask for this thread
877 // and save the caller's signal mask
878 os::Linux::hotspot_sigmask(thread);
879
880 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
881 os::current_thread_id(), (uintx) pthread_self());
882
883 return true;
884 }
885
886 void os::pd_start_thread(Thread* thread) {
887 OSThread * osthread = thread->osthread();
888 assert(osthread->get_state() != INITIALIZED, "just checking");
889 Monitor* sync_with_child = osthread->startThread_lock();
890 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
891 sync_with_child->notify();
892 }
893
894 // Free Linux resources related to the OSThread
895 void os::free_thread(OSThread* osthread) {
896 assert(osthread != NULL, "osthread not set");
897
898 // We are told to free resources of the argument thread,
899 // but we can only really operate on the current thread.
900 assert(Thread::current()->osthread() == osthread,
901 "os::free_thread but not current thread");
902
903 #ifdef ASSERT
904 sigset_t current;
905 sigemptyset(¤t);
906 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
907 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
908 #endif
909
910 // Restore caller's signal mask
911 sigset_t sigmask = osthread->caller_sigmask();
912 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
913
914 delete osthread;
915 }
916
917 //////////////////////////////////////////////////////////////////////////////
918 // primordial thread
919
920 // Check if current thread is the primordial thread, similar to Solaris thr_main.
921 bool os::is_primordial_thread(void) {
922 char dummy;
923 // If called before init complete, thread stack bottom will be null.
924 // Can be called if fatal error occurs before initialization.
925 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
926 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
927 os::Linux::initial_thread_stack_size() != 0,
928 "os::init did not locate primordial thread's stack region");
929 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
930 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
931 os::Linux::initial_thread_stack_size()) {
932 return true;
933 } else {
934 return false;
935 }
936 }
937
938 // Find the virtual memory area that contains addr
939 static bool find_vma(address addr, address* vma_low, address* vma_high) {
940 FILE *fp = fopen("/proc/self/maps", "r");
941 if (fp) {
942 address low, high;
943 while (!feof(fp)) {
944 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
945 if (low <= addr && addr < high) {
946 if (vma_low) *vma_low = low;
947 if (vma_high) *vma_high = high;
948 fclose(fp);
949 return true;
950 }
951 }
952 for (;;) {
953 int ch = fgetc(fp);
954 if (ch == EOF || ch == (int)'\n') break;
955 }
956 }
957 fclose(fp);
958 }
959 return false;
960 }
961
962 // Locate primordial thread stack. This special handling of primordial thread stack
963 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
964 // bogus value for the primordial process thread. While the launcher has created
965 // the VM in a new thread since JDK 6, we still have to allow for the use of the
966 // JNI invocation API from a primordial thread.
967 void os::Linux::capture_initial_stack(size_t max_size) {
968
969 // max_size is either 0 (which means accept OS default for thread stacks) or
970 // a user-specified value known to be at least the minimum needed. If we
971 // are actually on the primordial thread we can make it appear that we have a
972 // smaller max_size stack by inserting the guard pages at that location. But we
973 // cannot do anything to emulate a larger stack than what has been provided by
974 // the OS or threading library. In fact if we try to use a stack greater than
975 // what is set by rlimit then we will crash the hosting process.
976
977 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
978 // If this is "unlimited" then it will be a huge value.
979 struct rlimit rlim;
980 getrlimit(RLIMIT_STACK, &rlim);
981 size_t stack_size = rlim.rlim_cur;
982
983 // 6308388: a bug in ld.so will relocate its own .data section to the
984 // lower end of primordial stack; reduce ulimit -s value a little bit
985 // so we won't install guard page on ld.so's data section.
986 // But ensure we don't underflow the stack size - allow 1 page spare
987 if (stack_size >= (size_t)(3 * page_size())) {
988 stack_size -= 2 * page_size();
989 }
990
991 // Try to figure out where the stack base (top) is. This is harder.
992 //
993 // When an application is started, glibc saves the initial stack pointer in
994 // a global variable "__libc_stack_end", which is then used by system
995 // libraries. __libc_stack_end should be pretty close to stack top. The
996 // variable is available since the very early days. However, because it is
997 // a private interface, it could disappear in the future.
998 //
999 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1000 // to __libc_stack_end, it is very close to stack top, but isn't the real
1001 // stack top. Note that /proc may not exist if VM is running as a chroot
1002 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1003 // /proc/<pid>/stat could change in the future (though unlikely).
1004 //
1005 // We try __libc_stack_end first. If that doesn't work, look for
1006 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1007 // as a hint, which should work well in most cases.
1008
1009 uintptr_t stack_start;
1010
1011 // try __libc_stack_end first
1012 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1013 if (p && *p) {
1014 stack_start = *p;
1015 } else {
1016 // see if we can get the start_stack field from /proc/self/stat
1017 FILE *fp;
1018 int pid;
1019 char state;
1020 int ppid;
1021 int pgrp;
1022 int session;
1023 int nr;
1024 int tpgrp;
1025 unsigned long flags;
1026 unsigned long minflt;
1027 unsigned long cminflt;
1028 unsigned long majflt;
1029 unsigned long cmajflt;
1030 unsigned long utime;
1031 unsigned long stime;
1032 long cutime;
1033 long cstime;
1034 long prio;
1035 long nice;
1036 long junk;
1037 long it_real;
1038 uintptr_t start;
1039 uintptr_t vsize;
1040 intptr_t rss;
1041 uintptr_t rsslim;
1042 uintptr_t scodes;
1043 uintptr_t ecode;
1044 int i;
1045
1046 // Figure what the primordial thread stack base is. Code is inspired
1047 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1048 // followed by command name surrounded by parentheses, state, etc.
1049 char stat[2048];
1050 int statlen;
1051
1052 fp = fopen("/proc/self/stat", "r");
1053 if (fp) {
1054 statlen = fread(stat, 1, 2047, fp);
1055 stat[statlen] = '\0';
1056 fclose(fp);
1057
1058 // Skip pid and the command string. Note that we could be dealing with
1059 // weird command names, e.g. user could decide to rename java launcher
1060 // to "java 1.4.2 :)", then the stat file would look like
1061 // 1234 (java 1.4.2 :)) R ... ...
1062 // We don't really need to know the command string, just find the last
1063 // occurrence of ")" and then start parsing from there. See bug 4726580.
1064 char * s = strrchr(stat, ')');
1065
1066 i = 0;
1067 if (s) {
1068 // Skip blank chars
1069 do { s++; } while (s && isspace(*s));
1070
1071 #define _UFM UINTX_FORMAT
1072 #define _DFM INTX_FORMAT
1073
1074 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1075 // 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
1076 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,
1077 &state, // 3 %c
1078 &ppid, // 4 %d
1079 &pgrp, // 5 %d
1080 &session, // 6 %d
1081 &nr, // 7 %d
1082 &tpgrp, // 8 %d
1083 &flags, // 9 %lu
1084 &minflt, // 10 %lu
1085 &cminflt, // 11 %lu
1086 &majflt, // 12 %lu
1087 &cmajflt, // 13 %lu
1088 &utime, // 14 %lu
1089 &stime, // 15 %lu
1090 &cutime, // 16 %ld
1091 &cstime, // 17 %ld
1092 &prio, // 18 %ld
1093 &nice, // 19 %ld
1094 &junk, // 20 %ld
1095 &it_real, // 21 %ld
1096 &start, // 22 UINTX_FORMAT
1097 &vsize, // 23 UINTX_FORMAT
1098 &rss, // 24 INTX_FORMAT
1099 &rsslim, // 25 UINTX_FORMAT
1100 &scodes, // 26 UINTX_FORMAT
1101 &ecode, // 27 UINTX_FORMAT
1102 &stack_start); // 28 UINTX_FORMAT
1103 }
1104
1105 #undef _UFM
1106 #undef _DFM
1107
1108 if (i != 28 - 2) {
1109 assert(false, "Bad conversion from /proc/self/stat");
1110 // product mode - assume we are the primordial thread, good luck in the
1111 // embedded case.
1112 warning("Can't detect primordial thread stack location - bad conversion");
1113 stack_start = (uintptr_t) &rlim;
1114 }
1115 } else {
1116 // For some reason we can't open /proc/self/stat (for example, running on
1117 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1118 // most cases, so don't abort:
1119 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1120 stack_start = (uintptr_t) &rlim;
1121 }
1122 }
1123
1124 // Now we have a pointer (stack_start) very close to the stack top, the
1125 // next thing to do is to figure out the exact location of stack top. We
1126 // can find out the virtual memory area that contains stack_start by
1127 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1128 // and its upper limit is the real stack top. (again, this would fail if
1129 // running inside chroot, because /proc may not exist.)
1130
1131 uintptr_t stack_top;
1132 address low, high;
1133 if (find_vma((address)stack_start, &low, &high)) {
1134 // success, "high" is the true stack top. (ignore "low", because initial
1135 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1136 stack_top = (uintptr_t)high;
1137 } else {
1138 // failed, likely because /proc/self/maps does not exist
1139 warning("Can't detect primordial thread stack location - find_vma failed");
1140 // best effort: stack_start is normally within a few pages below the real
1141 // stack top, use it as stack top, and reduce stack size so we won't put
1142 // guard page outside stack.
1143 stack_top = stack_start;
1144 stack_size -= 16 * page_size();
1145 }
1146
1147 // stack_top could be partially down the page so align it
1148 stack_top = align_up(stack_top, page_size());
1149
1150 // Allowed stack value is minimum of max_size and what we derived from rlimit
1151 if (max_size > 0) {
1152 _initial_thread_stack_size = MIN2(max_size, stack_size);
1153 } else {
1154 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1155 // clamp it at 8MB as we do on Solaris
1156 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1157 }
1158 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1159 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1160
1161 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1162
1163 if (log_is_enabled(Info, os, thread)) {
1164 // See if we seem to be on primordial process thread
1165 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1166 uintptr_t(&rlim) < stack_top;
1167
1168 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1169 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1170 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1171 stack_top, intptr_t(_initial_thread_stack_bottom));
1172 }
1173 }
1174
1175 ////////////////////////////////////////////////////////////////////////////////
1176 // time support
1177
1178 // Time since start-up in seconds to a fine granularity.
1179 // Used by VMSelfDestructTimer and the MemProfiler.
1180 double os::elapsedTime() {
1181
1182 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1183 }
1184
1185 jlong os::elapsed_counter() {
1186 return javaTimeNanos() - initial_time_count;
1187 }
1188
1189 jlong os::elapsed_frequency() {
1190 return NANOSECS_PER_SEC; // nanosecond resolution
1191 }
1192
1193 bool os::supports_vtime() { return true; }
1194 bool os::enable_vtime() { return false; }
1195 bool os::vtime_enabled() { return false; }
1196
1197 double os::elapsedVTime() {
1198 struct rusage usage;
1199 int retval = getrusage(RUSAGE_THREAD, &usage);
1200 if (retval == 0) {
1201 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);
1202 } else {
1203 // better than nothing, but not much
1204 return elapsedTime();
1205 }
1206 }
1207
1208 jlong os::javaTimeMillis() {
1209 timeval time;
1210 int status = gettimeofday(&time, NULL);
1211 assert(status != -1, "linux error");
1212 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1213 }
1214
1215 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1216 timeval time;
1217 int status = gettimeofday(&time, NULL);
1218 assert(status != -1, "linux error");
1219 seconds = jlong(time.tv_sec);
1220 nanos = jlong(time.tv_usec) * 1000;
1221 }
1222
1223
1224 #ifndef CLOCK_MONOTONIC
1225 #define CLOCK_MONOTONIC (1)
1226 #endif
1227
1228 void os::Linux::clock_init() {
1229 // we do dlopen's in this particular order due to bug in linux
1230 // dynamical loader (see 6348968) leading to crash on exit
1231 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1232 if (handle == NULL) {
1233 handle = dlopen("librt.so", RTLD_LAZY);
1234 }
1235
1236 if (handle) {
1237 int (*clock_getres_func)(clockid_t, struct timespec*) =
1238 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1239 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1240 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1241 if (clock_getres_func && clock_gettime_func) {
1242 // See if monotonic clock is supported by the kernel. Note that some
1243 // early implementations simply return kernel jiffies (updated every
1244 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1245 // for nano time (though the monotonic property is still nice to have).
1246 // It's fixed in newer kernels, however clock_getres() still returns
1247 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1248 // resolution for now. Hopefully as people move to new kernels, this
1249 // won't be a problem.
1250 struct timespec res;
1251 struct timespec tp;
1252 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1253 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1254 // yes, monotonic clock is supported
1255 _clock_gettime = clock_gettime_func;
1256 return;
1257 } else {
1258 // close librt if there is no monotonic clock
1259 dlclose(handle);
1260 }
1261 }
1262 }
1263 warning("No monotonic clock was available - timed services may " \
1264 "be adversely affected if the time-of-day clock changes");
1265 }
1266
1267 #ifndef SYS_clock_getres
1268 #if defined(X86) || defined(PPC64) || defined(S390)
1269 #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261)
1270 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1271 #else
1272 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1273 #define sys_clock_getres(x,y) -1
1274 #endif
1275 #else
1276 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1277 #endif
1278
1279 void os::Linux::fast_thread_clock_init() {
1280 if (!UseLinuxPosixThreadCPUClocks) {
1281 return;
1282 }
1283 clockid_t clockid;
1284 struct timespec tp;
1285 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1286 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1287
1288 // Switch to using fast clocks for thread cpu time if
1289 // the sys_clock_getres() returns 0 error code.
1290 // Note, that some kernels may support the current thread
1291 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1292 // returned by the pthread_getcpuclockid().
1293 // If the fast Posix clocks are supported then the sys_clock_getres()
1294 // must return at least tp.tv_sec == 0 which means a resolution
1295 // better than 1 sec. This is extra check for reliability.
1296
1297 if (pthread_getcpuclockid_func &&
1298 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1299 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1300 _supports_fast_thread_cpu_time = true;
1301 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1302 }
1303 }
1304
1305 jlong os::javaTimeNanos() {
1306 if (os::supports_monotonic_clock()) {
1307 struct timespec tp;
1308 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1309 assert(status == 0, "gettime error");
1310 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1311 return result;
1312 } else {
1313 timeval time;
1314 int status = gettimeofday(&time, NULL);
1315 assert(status != -1, "linux error");
1316 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1317 return 1000 * usecs;
1318 }
1319 }
1320
1321 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1322 if (os::supports_monotonic_clock()) {
1323 info_ptr->max_value = ALL_64_BITS;
1324
1325 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1326 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1327 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1328 } else {
1329 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1330 info_ptr->max_value = ALL_64_BITS;
1331
1332 // gettimeofday is a real time clock so it skips
1333 info_ptr->may_skip_backward = true;
1334 info_ptr->may_skip_forward = true;
1335 }
1336
1337 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1338 }
1339
1340 // Return the real, user, and system times in seconds from an
1341 // arbitrary fixed point in the past.
1342 bool os::getTimesSecs(double* process_real_time,
1343 double* process_user_time,
1344 double* process_system_time) {
1345 struct tms ticks;
1346 clock_t real_ticks = times(&ticks);
1347
1348 if (real_ticks == (clock_t) (-1)) {
1349 return false;
1350 } else {
1351 double ticks_per_second = (double) clock_tics_per_sec;
1352 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1353 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1354 *process_real_time = ((double) real_ticks) / ticks_per_second;
1355
1356 return true;
1357 }
1358 }
1359
1360
1361 char * os::local_time_string(char *buf, size_t buflen) {
1362 struct tm t;
1363 time_t long_time;
1364 time(&long_time);
1365 localtime_r(&long_time, &t);
1366 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1367 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1368 t.tm_hour, t.tm_min, t.tm_sec);
1369 return buf;
1370 }
1371
1372 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1373 return localtime_r(clock, res);
1374 }
1375
1376 ////////////////////////////////////////////////////////////////////////////////
1377 // runtime exit support
1378
1379 // Note: os::shutdown() might be called very early during initialization, or
1380 // called from signal handler. Before adding something to os::shutdown(), make
1381 // sure it is async-safe and can handle partially initialized VM.
1382 void os::shutdown() {
1383
1384 // allow PerfMemory to attempt cleanup of any persistent resources
1385 perfMemory_exit();
1386
1387 // needs to remove object in file system
1388 AttachListener::abort();
1389
1390 // flush buffered output, finish log files
1391 ostream_abort();
1392
1393 // Check for abort hook
1394 abort_hook_t abort_hook = Arguments::abort_hook();
1395 if (abort_hook != NULL) {
1396 abort_hook();
1397 }
1398
1399 }
1400
1401 // Note: os::abort() might be called very early during initialization, or
1402 // called from signal handler. Before adding something to os::abort(), make
1403 // sure it is async-safe and can handle partially initialized VM.
1404 void os::abort(bool dump_core, void* siginfo, const void* context) {
1405 os::shutdown();
1406 if (dump_core) {
1407 #ifndef PRODUCT
1408 fdStream out(defaultStream::output_fd());
1409 out.print_raw("Current thread is ");
1410 char buf[16];
1411 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1412 out.print_raw_cr(buf);
1413 out.print_raw_cr("Dumping core ...");
1414 #endif
1415 ::abort(); // dump core
1416 }
1417
1418 ::exit(1);
1419 }
1420
1421 // Die immediately, no exit hook, no abort hook, no cleanup.
1422 void os::die() {
1423 ::abort();
1424 }
1425
1426
1427 // This method is a copy of JDK's sysGetLastErrorString
1428 // from src/solaris/hpi/src/system_md.c
1429
1430 size_t os::lasterror(char *buf, size_t len) {
1431 if (errno == 0) return 0;
1432
1433 const char *s = os::strerror(errno);
1434 size_t n = ::strlen(s);
1435 if (n >= len) {
1436 n = len - 1;
1437 }
1438 ::strncpy(buf, s, n);
1439 buf[n] = '\0';
1440 return n;
1441 }
1442
1443 // thread_id is kernel thread id (similar to Solaris LWP id)
1444 intx os::current_thread_id() { return os::Linux::gettid(); }
1445 int os::current_process_id() {
1446 return ::getpid();
1447 }
1448
1449 // DLL functions
1450
1451 const char* os::dll_file_extension() { return ".so"; }
1452
1453 // This must be hard coded because it's the system's temporary
1454 // directory not the java application's temp directory, ala java.io.tmpdir.
1455 const char* os::get_temp_directory() { return "/tmp"; }
1456
1457 static bool file_exists(const char* filename) {
1458 struct stat statbuf;
1459 if (filename == NULL || strlen(filename) == 0) {
1460 return false;
1461 }
1462 return os::stat(filename, &statbuf) == 0;
1463 }
1464
1465 // check if addr is inside libjvm.so
1466 bool os::address_is_in_vm(address addr) {
1467 static address libjvm_base_addr;
1468 Dl_info dlinfo;
1469
1470 if (libjvm_base_addr == NULL) {
1471 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1472 libjvm_base_addr = (address)dlinfo.dli_fbase;
1473 }
1474 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1475 }
1476
1477 if (dladdr((void *)addr, &dlinfo) != 0) {
1478 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1479 }
1480
1481 return false;
1482 }
1483
1484 bool os::dll_address_to_function_name(address addr, char *buf,
1485 int buflen, int *offset,
1486 bool demangle) {
1487 // buf is not optional, but offset is optional
1488 assert(buf != NULL, "sanity check");
1489
1490 Dl_info dlinfo;
1491
1492 if (dladdr((void*)addr, &dlinfo) != 0) {
1493 // see if we have a matching symbol
1494 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1495 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1496 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1497 }
1498 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1499 return true;
1500 }
1501 // no matching symbol so try for just file info
1502 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1503 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1504 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1505 return true;
1506 }
1507 }
1508 }
1509
1510 buf[0] = '\0';
1511 if (offset != NULL) *offset = -1;
1512 return false;
1513 }
1514
1515 struct _address_to_library_name {
1516 address addr; // input : memory address
1517 size_t buflen; // size of fname
1518 char* fname; // output: library name
1519 address base; // library base addr
1520 };
1521
1522 static int address_to_library_name_callback(struct dl_phdr_info *info,
1523 size_t size, void *data) {
1524 int i;
1525 bool found = false;
1526 address libbase = NULL;
1527 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1528
1529 // iterate through all loadable segments
1530 for (i = 0; i < info->dlpi_phnum; i++) {
1531 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1532 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1533 // base address of a library is the lowest address of its loaded
1534 // segments.
1535 if (libbase == NULL || libbase > segbase) {
1536 libbase = segbase;
1537 }
1538 // see if 'addr' is within current segment
1539 if (segbase <= d->addr &&
1540 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1541 found = true;
1542 }
1543 }
1544 }
1545
1546 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1547 // so dll_address_to_library_name() can fall through to use dladdr() which
1548 // can figure out executable name from argv[0].
1549 if (found && info->dlpi_name && info->dlpi_name[0]) {
1550 d->base = libbase;
1551 if (d->fname) {
1552 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1553 }
1554 return 1;
1555 }
1556 return 0;
1557 }
1558
1559 bool os::dll_address_to_library_name(address addr, char* buf,
1560 int buflen, int* offset) {
1561 // buf is not optional, but offset is optional
1562 assert(buf != NULL, "sanity check");
1563
1564 Dl_info dlinfo;
1565 struct _address_to_library_name data;
1566
1567 // There is a bug in old glibc dladdr() implementation that it could resolve
1568 // to wrong library name if the .so file has a base address != NULL. Here
1569 // we iterate through the program headers of all loaded libraries to find
1570 // out which library 'addr' really belongs to. This workaround can be
1571 // removed once the minimum requirement for glibc is moved to 2.3.x.
1572 data.addr = addr;
1573 data.fname = buf;
1574 data.buflen = buflen;
1575 data.base = NULL;
1576 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1577
1578 if (rslt) {
1579 // buf already contains library name
1580 if (offset) *offset = addr - data.base;
1581 return true;
1582 }
1583 if (dladdr((void*)addr, &dlinfo) != 0) {
1584 if (dlinfo.dli_fname != NULL) {
1585 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1586 }
1587 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1588 *offset = addr - (address)dlinfo.dli_fbase;
1589 }
1590 return true;
1591 }
1592
1593 buf[0] = '\0';
1594 if (offset) *offset = -1;
1595 return false;
1596 }
1597
1598 // Loads .dll/.so and
1599 // in case of error it checks if .dll/.so was built for the
1600 // same architecture as Hotspot is running on
1601
1602
1603 // Remember the stack's state. The Linux dynamic linker will change
1604 // the stack to 'executable' at most once, so we must safepoint only once.
1605 bool os::Linux::_stack_is_executable = false;
1606
1607 // VM operation that loads a library. This is necessary if stack protection
1608 // of the Java stacks can be lost during loading the library. If we
1609 // do not stop the Java threads, they can stack overflow before the stacks
1610 // are protected again.
1611 class VM_LinuxDllLoad: public VM_Operation {
1612 private:
1613 const char *_filename;
1614 char *_ebuf;
1615 int _ebuflen;
1616 void *_lib;
1617 public:
1618 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1619 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1620 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1621 void doit() {
1622 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1623 os::Linux::_stack_is_executable = true;
1624 }
1625 void* loaded_library() { return _lib; }
1626 };
1627
1628 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1629 void * result = NULL;
1630 bool load_attempted = false;
1631
1632 // Check whether the library to load might change execution rights
1633 // of the stack. If they are changed, the protection of the stack
1634 // guard pages will be lost. We need a safepoint to fix this.
1635 //
1636 // See Linux man page execstack(8) for more info.
1637 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1638 if (!ElfFile::specifies_noexecstack(filename)) {
1639 if (!is_init_completed()) {
1640 os::Linux::_stack_is_executable = true;
1641 // This is OK - No Java threads have been created yet, and hence no
1642 // stack guard pages to fix.
1643 //
1644 // This should happen only when you are building JDK7 using a very
1645 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1646 //
1647 // Dynamic loader will make all stacks executable after
1648 // this function returns, and will not do that again.
1649 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1650 } else {
1651 warning("You have loaded library %s which might have disabled stack guard. "
1652 "The VM will try to fix the stack guard now.\n"
1653 "It's highly recommended that you fix the library with "
1654 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1655 filename);
1656
1657 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1658 JavaThread *jt = JavaThread::current();
1659 if (jt->thread_state() != _thread_in_native) {
1660 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1661 // that requires ExecStack. Cannot enter safe point. Let's give up.
1662 warning("Unable to fix stack guard. Giving up.");
1663 } else {
1664 if (!LoadExecStackDllInVMThread) {
1665 // This is for the case where the DLL has an static
1666 // constructor function that executes JNI code. We cannot
1667 // load such DLLs in the VMThread.
1668 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1669 }
1670
1671 ThreadInVMfromNative tiv(jt);
1672 debug_only(VMNativeEntryWrapper vew;)
1673
1674 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1675 VMThread::execute(&op);
1676 if (LoadExecStackDllInVMThread) {
1677 result = op.loaded_library();
1678 }
1679 load_attempted = true;
1680 }
1681 }
1682 }
1683 }
1684
1685 if (!load_attempted) {
1686 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1687 }
1688
1689 if (result != NULL) {
1690 // Successful loading
1691 return result;
1692 }
1693
1694 Elf32_Ehdr elf_head;
1695 int diag_msg_max_length=ebuflen-strlen(ebuf);
1696 char* diag_msg_buf=ebuf+strlen(ebuf);
1697
1698 if (diag_msg_max_length==0) {
1699 // No more space in ebuf for additional diagnostics message
1700 return NULL;
1701 }
1702
1703
1704 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1705
1706 if (file_descriptor < 0) {
1707 // Can't open library, report dlerror() message
1708 return NULL;
1709 }
1710
1711 bool failed_to_read_elf_head=
1712 (sizeof(elf_head)!=
1713 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1714
1715 ::close(file_descriptor);
1716 if (failed_to_read_elf_head) {
1717 // file i/o error - report dlerror() msg
1718 return NULL;
1719 }
1720
1721 typedef struct {
1722 Elf32_Half code; // Actual value as defined in elf.h
1723 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1724 unsigned char elf_class; // 32 or 64 bit
1725 unsigned char endianess; // MSB or LSB
1726 char* name; // String representation
1727 } arch_t;
1728
1729 #ifndef EM_486
1730 #define EM_486 6 /* Intel 80486 */
1731 #endif
1732 #ifndef EM_AARCH64
1733 #define EM_AARCH64 183 /* ARM AARCH64 */
1734 #endif
1735
1736 static const arch_t arch_array[]={
1737 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1738 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1739 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1740 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1741 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1742 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1743 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1744 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1745 #if defined(VM_LITTLE_ENDIAN)
1746 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1747 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1748 #else
1749 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1750 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1751 #endif
1752 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1753 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1754 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1755 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1756 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1757 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1758 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1759 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1760 };
1761
1762 #if (defined IA32)
1763 static Elf32_Half running_arch_code=EM_386;
1764 #elif (defined AMD64)
1765 static Elf32_Half running_arch_code=EM_X86_64;
1766 #elif (defined IA64)
1767 static Elf32_Half running_arch_code=EM_IA_64;
1768 #elif (defined __sparc) && (defined _LP64)
1769 static Elf32_Half running_arch_code=EM_SPARCV9;
1770 #elif (defined __sparc) && (!defined _LP64)
1771 static Elf32_Half running_arch_code=EM_SPARC;
1772 #elif (defined __powerpc64__)
1773 static Elf32_Half running_arch_code=EM_PPC64;
1774 #elif (defined __powerpc__)
1775 static Elf32_Half running_arch_code=EM_PPC;
1776 #elif (defined AARCH64)
1777 static Elf32_Half running_arch_code=EM_AARCH64;
1778 #elif (defined ARM)
1779 static Elf32_Half running_arch_code=EM_ARM;
1780 #elif (defined S390)
1781 static Elf32_Half running_arch_code=EM_S390;
1782 #elif (defined ALPHA)
1783 static Elf32_Half running_arch_code=EM_ALPHA;
1784 #elif (defined MIPSEL)
1785 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1786 #elif (defined PARISC)
1787 static Elf32_Half running_arch_code=EM_PARISC;
1788 #elif (defined MIPS)
1789 static Elf32_Half running_arch_code=EM_MIPS;
1790 #elif (defined M68K)
1791 static Elf32_Half running_arch_code=EM_68K;
1792 #elif (defined SH)
1793 static Elf32_Half running_arch_code=EM_SH;
1794 #else
1795 #error Method os::dll_load requires that one of following is defined:\
1796 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1797 #endif
1798
1799 // Identify compatability class for VM's architecture and library's architecture
1800 // Obtain string descriptions for architectures
1801
1802 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1803 int running_arch_index=-1;
1804
1805 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1806 if (running_arch_code == arch_array[i].code) {
1807 running_arch_index = i;
1808 }
1809 if (lib_arch.code == arch_array[i].code) {
1810 lib_arch.compat_class = arch_array[i].compat_class;
1811 lib_arch.name = arch_array[i].name;
1812 }
1813 }
1814
1815 assert(running_arch_index != -1,
1816 "Didn't find running architecture code (running_arch_code) in arch_array");
1817 if (running_arch_index == -1) {
1818 // Even though running architecture detection failed
1819 // we may still continue with reporting dlerror() message
1820 return NULL;
1821 }
1822
1823 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1824 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1825 return NULL;
1826 }
1827
1828 #ifndef S390
1829 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1830 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1831 return NULL;
1832 }
1833 #endif // !S390
1834
1835 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1836 if (lib_arch.name!=NULL) {
1837 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1838 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1839 lib_arch.name, arch_array[running_arch_index].name);
1840 } else {
1841 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1842 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1843 lib_arch.code,
1844 arch_array[running_arch_index].name);
1845 }
1846 }
1847
1848 return NULL;
1849 }
1850
1851 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1852 int ebuflen) {
1853 void * result = ::dlopen(filename, RTLD_LAZY);
1854 if (result == NULL) {
1855 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1856 ebuf[ebuflen-1] = '\0';
1857 }
1858 return result;
1859 }
1860
1861 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1862 int ebuflen) {
1863 void * result = NULL;
1864 if (LoadExecStackDllInVMThread) {
1865 result = dlopen_helper(filename, ebuf, ebuflen);
1866 }
1867
1868 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1869 // library that requires an executable stack, or which does not have this
1870 // stack attribute set, dlopen changes the stack attribute to executable. The
1871 // read protection of the guard pages gets lost.
1872 //
1873 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1874 // may have been queued at the same time.
1875
1876 if (!_stack_is_executable) {
1877 JavaThread *jt = Threads::first();
1878
1879 while (jt) {
1880 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1881 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1882 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1883 warning("Attempt to reguard stack yellow zone failed.");
1884 }
1885 }
1886 jt = jt->next();
1887 }
1888 }
1889
1890 return result;
1891 }
1892
1893 void* os::dll_lookup(void* handle, const char* name) {
1894 void* res = dlsym(handle, name);
1895 return res;
1896 }
1897
1898 void* os::get_default_process_handle() {
1899 return (void*)::dlopen(NULL, RTLD_LAZY);
1900 }
1901
1902 static bool _print_ascii_file(const char* filename, outputStream* st) {
1903 int fd = ::open(filename, O_RDONLY);
1904 if (fd == -1) {
1905 return false;
1906 }
1907
1908 char buf[33];
1909 int bytes;
1910 buf[32] = '\0';
1911 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1912 st->print_raw(buf, bytes);
1913 }
1914
1915 ::close(fd);
1916
1917 return true;
1918 }
1919
1920 void os::print_dll_info(outputStream *st) {
1921 st->print_cr("Dynamic libraries:");
1922
1923 char fname[32];
1924 pid_t pid = os::Linux::gettid();
1925
1926 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1927
1928 if (!_print_ascii_file(fname, st)) {
1929 st->print("Can not get library information for pid = %d\n", pid);
1930 }
1931 }
1932
1933 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1934 FILE *procmapsFile = NULL;
1935
1936 // Open the procfs maps file for the current process
1937 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1938 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1939 char line[PATH_MAX + 100];
1940
1941 // Read line by line from 'file'
1942 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1943 u8 base, top, offset, inode;
1944 char permissions[5];
1945 char device[6];
1946 char name[PATH_MAX + 1];
1947
1948 // Parse fields from line
1949 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1950 &base, &top, permissions, &offset, device, &inode, name);
1951
1952 // Filter by device id '00:00' so that we only get file system mapped files.
1953 if (strcmp(device, "00:00") != 0) {
1954
1955 // Call callback with the fields of interest
1956 if(callback(name, (address)base, (address)top, param)) {
1957 // Oops abort, callback aborted
1958 fclose(procmapsFile);
1959 return 1;
1960 }
1961 }
1962 }
1963 fclose(procmapsFile);
1964 }
1965 return 0;
1966 }
1967
1968 void os::print_os_info_brief(outputStream* st) {
1969 os::Linux::print_distro_info(st);
1970
1971 os::Posix::print_uname_info(st);
1972
1973 os::Linux::print_libversion_info(st);
1974
1975 }
1976
1977 void os::print_os_info(outputStream* st) {
1978 st->print("OS:");
1979
1980 os::Linux::print_distro_info(st);
1981
1982 os::Posix::print_uname_info(st);
1983
1984 // Print warning if unsafe chroot environment detected
1985 if (unsafe_chroot_detected) {
1986 st->print("WARNING!! ");
1987 st->print_cr("%s", unstable_chroot_error);
1988 }
1989
1990 os::Linux::print_libversion_info(st);
1991
1992 os::Posix::print_rlimit_info(st);
1993
1994 os::Posix::print_load_average(st);
1995
1996 os::Linux::print_full_memory_info(st);
1997
1998 os::Linux::print_container_info(st);
1999 }
2000
2001 // Try to identify popular distros.
2002 // Most Linux distributions have a /etc/XXX-release file, which contains
2003 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2004 // file that also contains the OS version string. Some have more than one
2005 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2006 // /etc/redhat-release.), so the order is important.
2007 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2008 // their own specific XXX-release file as well as a redhat-release file.
2009 // Because of this the XXX-release file needs to be searched for before the
2010 // redhat-release file.
2011 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2012 // search for redhat-release / SuSE-release needs to be before lsb-release.
2013 // Since the lsb-release file is the new standard it needs to be searched
2014 // before the older style release files.
2015 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2016 // next to last resort. The os-release file is a new standard that contains
2017 // distribution information and the system-release file seems to be an old
2018 // standard that has been replaced by the lsb-release and os-release files.
2019 // Searching for the debian_version file is the last resort. It contains
2020 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2021 // "Debian " is printed before the contents of the debian_version file.
2022
2023 const char* distro_files[] = {
2024 "/etc/oracle-release",
2025 "/etc/mandriva-release",
2026 "/etc/mandrake-release",
2027 "/etc/sun-release",
2028 "/etc/redhat-release",
2029 "/etc/SuSE-release",
2030 "/etc/lsb-release",
2031 "/etc/turbolinux-release",
2032 "/etc/gentoo-release",
2033 "/etc/ltib-release",
2034 "/etc/angstrom-version",
2035 "/etc/system-release",
2036 "/etc/os-release",
2037 NULL };
2038
2039 void os::Linux::print_distro_info(outputStream* st) {
2040 for (int i = 0;; i++) {
2041 const char* file = distro_files[i];
2042 if (file == NULL) {
2043 break; // done
2044 }
2045 // If file prints, we found it.
2046 if (_print_ascii_file(file, st)) {
2047 return;
2048 }
2049 }
2050
2051 if (file_exists("/etc/debian_version")) {
2052 st->print("Debian ");
2053 _print_ascii_file("/etc/debian_version", st);
2054 } else {
2055 st->print("Linux");
2056 }
2057 st->cr();
2058 }
2059
2060 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2061 char buf[256];
2062 while (fgets(buf, sizeof(buf), fp)) {
2063 // Edit out extra stuff in expected format
2064 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2065 char* ptr = strstr(buf, "\""); // the name is in quotes
2066 if (ptr != NULL) {
2067 ptr++; // go beyond first quote
2068 char* nl = strchr(ptr, '\"');
2069 if (nl != NULL) *nl = '\0';
2070 strncpy(distro, ptr, length);
2071 } else {
2072 ptr = strstr(buf, "=");
2073 ptr++; // go beyond equals then
2074 char* nl = strchr(ptr, '\n');
2075 if (nl != NULL) *nl = '\0';
2076 strncpy(distro, ptr, length);
2077 }
2078 return;
2079 } else if (get_first_line) {
2080 char* nl = strchr(buf, '\n');
2081 if (nl != NULL) *nl = '\0';
2082 strncpy(distro, buf, length);
2083 return;
2084 }
2085 }
2086 // print last line and close
2087 char* nl = strchr(buf, '\n');
2088 if (nl != NULL) *nl = '\0';
2089 strncpy(distro, buf, length);
2090 }
2091
2092 static void parse_os_info(char* distro, size_t length, const char* file) {
2093 FILE* fp = fopen(file, "r");
2094 if (fp != NULL) {
2095 // if suse format, print out first line
2096 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2097 parse_os_info_helper(fp, distro, length, get_first_line);
2098 fclose(fp);
2099 }
2100 }
2101
2102 void os::get_summary_os_info(char* buf, size_t buflen) {
2103 for (int i = 0;; i++) {
2104 const char* file = distro_files[i];
2105 if (file == NULL) {
2106 break; // ran out of distro_files
2107 }
2108 if (file_exists(file)) {
2109 parse_os_info(buf, buflen, file);
2110 return;
2111 }
2112 }
2113 // special case for debian
2114 if (file_exists("/etc/debian_version")) {
2115 strncpy(buf, "Debian ", buflen);
2116 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2117 } else {
2118 strncpy(buf, "Linux", buflen);
2119 }
2120 }
2121
2122 void os::Linux::print_libversion_info(outputStream* st) {
2123 // libc, pthread
2124 st->print("libc:");
2125 st->print("%s ", os::Linux::glibc_version());
2126 st->print("%s ", os::Linux::libpthread_version());
2127 st->cr();
2128 }
2129
2130 void os::Linux::print_full_memory_info(outputStream* st) {
2131 st->print("\n/proc/meminfo:\n");
2132 _print_ascii_file("/proc/meminfo", st);
2133 st->cr();
2134 }
2135
2136 void os::Linux::print_container_info(outputStream* st) {
2137 if (OSContainer::is_containerized()) {
2138 st->print("container (cgroup) information:\n");
2139
2140 char *p = OSContainer::container_type();
2141 if (p == NULL)
2142 st->print("container_type() failed\n");
2143 else {
2144 st->print("container_type: %s\n", p);
2145 }
2146
2147 p = OSContainer::cpu_cpuset_cpus();
2148 if (p == NULL)
2149 st->print("cpu_cpuset_cpus() failed\n");
2150 else {
2151 st->print("cpu_cpuset_cpus: %s\n", p);
2152 free(p);
2153 }
2154
2155 p = OSContainer::cpu_cpuset_memory_nodes();
2156 if (p < 0)
2157 st->print("cpu_memory_nodes() failed\n");
2158 else {
2159 st->print("cpu_memory_nodes: %s\n", p);
2160 free(p);
2161 }
2162
2163 int i = OSContainer::active_processor_count();
2164 if (i < 0)
2165 st->print("active_processor_count() failed\n");
2166 else
2167 st->print("active_processor_count: %d\n", i);
2168
2169 i = OSContainer::cpu_quota();
2170 st->print("cpu_quota: %d\n", i);
2171
2172 i = OSContainer::cpu_period();
2173 st->print("cpu_period: %d\n", i);
2174
2175 i = OSContainer::cpu_shares();
2176 st->print("cpu_shares: %d\n", i);
2177
2178 jlong j = OSContainer::memory_limit_in_bytes();
2179 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2180
2181 j = OSContainer::memory_and_swap_limit_in_bytes();
2182 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2183
2184 j = OSContainer::memory_soft_limit_in_bytes();
2185 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2186
2187 j = OSContainer::OSContainer::memory_usage_in_bytes();
2188 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2189
2190 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2191 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2192 st->cr();
2193 }
2194 }
2195
2196 void os::print_memory_info(outputStream* st) {
2197
2198 st->print("Memory:");
2199 st->print(" %dk page", os::vm_page_size()>>10);
2200
2201 // values in struct sysinfo are "unsigned long"
2202 struct sysinfo si;
2203 sysinfo(&si);
2204
2205 st->print(", physical " UINT64_FORMAT "k",
2206 os::physical_memory() >> 10);
2207 st->print("(" UINT64_FORMAT "k free)",
2208 os::available_memory() >> 10);
2209 st->print(", swap " UINT64_FORMAT "k",
2210 ((jlong)si.totalswap * si.mem_unit) >> 10);
2211 st->print("(" UINT64_FORMAT "k free)",
2212 ((jlong)si.freeswap * si.mem_unit) >> 10);
2213 st->cr();
2214 }
2215
2216 // Print the first "model name" line and the first "flags" line
2217 // that we find and nothing more. We assume "model name" comes
2218 // before "flags" so if we find a second "model name", then the
2219 // "flags" field is considered missing.
2220 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2221 #if defined(IA32) || defined(AMD64)
2222 // Other platforms have less repetitive cpuinfo files
2223 FILE *fp = fopen("/proc/cpuinfo", "r");
2224 if (fp) {
2225 while (!feof(fp)) {
2226 if (fgets(buf, buflen, fp)) {
2227 // Assume model name comes before flags
2228 bool model_name_printed = false;
2229 if (strstr(buf, "model name") != NULL) {
2230 if (!model_name_printed) {
2231 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2232 st->print_raw(buf);
2233 model_name_printed = true;
2234 } else {
2235 // model name printed but not flags? Odd, just return
2236 fclose(fp);
2237 return true;
2238 }
2239 }
2240 // print the flags line too
2241 if (strstr(buf, "flags") != NULL) {
2242 st->print_raw(buf);
2243 fclose(fp);
2244 return true;
2245 }
2246 }
2247 }
2248 fclose(fp);
2249 }
2250 #endif // x86 platforms
2251 return false;
2252 }
2253
2254 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2255 // Only print the model name if the platform provides this as a summary
2256 if (!print_model_name_and_flags(st, buf, buflen)) {
2257 st->print("\n/proc/cpuinfo:\n");
2258 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2259 st->print_cr(" <Not Available>");
2260 }
2261 }
2262 }
2263
2264 #if defined(AMD64) || defined(IA32) || defined(X32)
2265 const char* search_string = "model name";
2266 #elif defined(M68K)
2267 const char* search_string = "CPU";
2268 #elif defined(PPC64)
2269 const char* search_string = "cpu";
2270 #elif defined(S390)
2271 const char* search_string = "processor";
2272 #elif defined(SPARC)
2273 const char* search_string = "cpu";
2274 #else
2275 const char* search_string = "Processor";
2276 #endif
2277
2278 // Parses the cpuinfo file for string representing the model name.
2279 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2280 FILE* fp = fopen("/proc/cpuinfo", "r");
2281 if (fp != NULL) {
2282 while (!feof(fp)) {
2283 char buf[256];
2284 if (fgets(buf, sizeof(buf), fp)) {
2285 char* start = strstr(buf, search_string);
2286 if (start != NULL) {
2287 char *ptr = start + strlen(search_string);
2288 char *end = buf + strlen(buf);
2289 while (ptr != end) {
2290 // skip whitespace and colon for the rest of the name.
2291 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2292 break;
2293 }
2294 ptr++;
2295 }
2296 if (ptr != end) {
2297 // reasonable string, get rid of newline and keep the rest
2298 char* nl = strchr(buf, '\n');
2299 if (nl != NULL) *nl = '\0';
2300 strncpy(cpuinfo, ptr, length);
2301 fclose(fp);
2302 return;
2303 }
2304 }
2305 }
2306 }
2307 fclose(fp);
2308 }
2309 // cpuinfo not found or parsing failed, just print generic string. The entire
2310 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2311 #if defined(AARCH64)
2312 strncpy(cpuinfo, "AArch64", length);
2313 #elif defined(AMD64)
2314 strncpy(cpuinfo, "x86_64", length);
2315 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2316 strncpy(cpuinfo, "ARM", length);
2317 #elif defined(IA32)
2318 strncpy(cpuinfo, "x86_32", length);
2319 #elif defined(IA64)
2320 strncpy(cpuinfo, "IA64", length);
2321 #elif defined(PPC)
2322 strncpy(cpuinfo, "PPC64", length);
2323 #elif defined(S390)
2324 strncpy(cpuinfo, "S390", length);
2325 #elif defined(SPARC)
2326 strncpy(cpuinfo, "sparcv9", length);
2327 #elif defined(ZERO_LIBARCH)
2328 strncpy(cpuinfo, ZERO_LIBARCH, length);
2329 #else
2330 strncpy(cpuinfo, "unknown", length);
2331 #endif
2332 }
2333
2334 static void print_signal_handler(outputStream* st, int sig,
2335 char* buf, size_t buflen);
2336
2337 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2338 st->print_cr("Signal Handlers:");
2339 print_signal_handler(st, SIGSEGV, buf, buflen);
2340 print_signal_handler(st, SIGBUS , buf, buflen);
2341 print_signal_handler(st, SIGFPE , buf, buflen);
2342 print_signal_handler(st, SIGPIPE, buf, buflen);
2343 print_signal_handler(st, SIGXFSZ, buf, buflen);
2344 print_signal_handler(st, SIGILL , buf, buflen);
2345 print_signal_handler(st, SR_signum, buf, buflen);
2346 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2347 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2348 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2349 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2350 #if defined(PPC64)
2351 print_signal_handler(st, SIGTRAP, buf, buflen);
2352 #endif
2353 }
2354
2355 static char saved_jvm_path[MAXPATHLEN] = {0};
2356
2357 // Find the full path to the current module, libjvm.so
2358 void os::jvm_path(char *buf, jint buflen) {
2359 // Error checking.
2360 if (buflen < MAXPATHLEN) {
2361 assert(false, "must use a large-enough buffer");
2362 buf[0] = '\0';
2363 return;
2364 }
2365 // Lazy resolve the path to current module.
2366 if (saved_jvm_path[0] != 0) {
2367 strcpy(buf, saved_jvm_path);
2368 return;
2369 }
2370
2371 char dli_fname[MAXPATHLEN];
2372 bool ret = dll_address_to_library_name(
2373 CAST_FROM_FN_PTR(address, os::jvm_path),
2374 dli_fname, sizeof(dli_fname), NULL);
2375 assert(ret, "cannot locate libjvm");
2376 char *rp = NULL;
2377 if (ret && dli_fname[0] != '\0') {
2378 rp = os::Posix::realpath(dli_fname, buf, buflen);
2379 }
2380 if (rp == NULL) {
2381 return;
2382 }
2383
2384 if (Arguments::sun_java_launcher_is_altjvm()) {
2385 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2386 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2387 // If "/jre/lib/" appears at the right place in the string, then
2388 // assume we are installed in a JDK and we're done. Otherwise, check
2389 // for a JAVA_HOME environment variable and fix up the path so it
2390 // looks like libjvm.so is installed there (append a fake suffix
2391 // hotspot/libjvm.so).
2392 const char *p = buf + strlen(buf) - 1;
2393 for (int count = 0; p > buf && count < 5; ++count) {
2394 for (--p; p > buf && *p != '/'; --p)
2395 /* empty */ ;
2396 }
2397
2398 if (strncmp(p, "/jre/lib/", 9) != 0) {
2399 // Look for JAVA_HOME in the environment.
2400 char* java_home_var = ::getenv("JAVA_HOME");
2401 if (java_home_var != NULL && java_home_var[0] != 0) {
2402 char* jrelib_p;
2403 int len;
2404
2405 // Check the current module name "libjvm.so".
2406 p = strrchr(buf, '/');
2407 if (p == NULL) {
2408 return;
2409 }
2410 assert(strstr(p, "/libjvm") == p, "invalid library name");
2411
2412 rp = os::Posix::realpath(java_home_var, buf, buflen);
2413 if (rp == NULL) {
2414 return;
2415 }
2416
2417 // determine if this is a legacy image or modules image
2418 // modules image doesn't have "jre" subdirectory
2419 len = strlen(buf);
2420 assert(len < buflen, "Ran out of buffer room");
2421 jrelib_p = buf + len;
2422 snprintf(jrelib_p, buflen-len, "/jre/lib");
2423 if (0 != access(buf, F_OK)) {
2424 snprintf(jrelib_p, buflen-len, "/lib");
2425 }
2426
2427 if (0 == access(buf, F_OK)) {
2428 // Use current module name "libjvm.so"
2429 len = strlen(buf);
2430 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2431 } else {
2432 // Go back to path of .so
2433 rp = os::Posix::realpath(dli_fname, buf, buflen);
2434 if (rp == NULL) {
2435 return;
2436 }
2437 }
2438 }
2439 }
2440 }
2441
2442 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2443 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2444 }
2445
2446 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2447 // no prefix required, not even "_"
2448 }
2449
2450 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2451 // no suffix required
2452 }
2453
2454 ////////////////////////////////////////////////////////////////////////////////
2455 // sun.misc.Signal support
2456
2457 static volatile jint sigint_count = 0;
2458
2459 static void UserHandler(int sig, void *siginfo, void *context) {
2460 // 4511530 - sem_post is serialized and handled by the manager thread. When
2461 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2462 // don't want to flood the manager thread with sem_post requests.
2463 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2464 return;
2465 }
2466
2467 // Ctrl-C is pressed during error reporting, likely because the error
2468 // handler fails to abort. Let VM die immediately.
2469 if (sig == SIGINT && VMError::is_error_reported()) {
2470 os::die();
2471 }
2472
2473 os::signal_notify(sig);
2474 }
2475
2476 void* os::user_handler() {
2477 return CAST_FROM_FN_PTR(void*, UserHandler);
2478 }
2479
2480 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2481 struct timespec ts;
2482 // Semaphore's are always associated with CLOCK_REALTIME
2483 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2484 // see unpackTime for discussion on overflow checking
2485 if (sec >= MAX_SECS) {
2486 ts.tv_sec += MAX_SECS;
2487 ts.tv_nsec = 0;
2488 } else {
2489 ts.tv_sec += sec;
2490 ts.tv_nsec += nsec;
2491 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2492 ts.tv_nsec -= NANOSECS_PER_SEC;
2493 ++ts.tv_sec; // note: this must be <= max_secs
2494 }
2495 }
2496
2497 return ts;
2498 }
2499
2500 extern "C" {
2501 typedef void (*sa_handler_t)(int);
2502 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2503 }
2504
2505 void* os::signal(int signal_number, void* handler) {
2506 struct sigaction sigAct, oldSigAct;
2507
2508 sigfillset(&(sigAct.sa_mask));
2509 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2510 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2511
2512 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2513 // -1 means registration failed
2514 return (void *)-1;
2515 }
2516
2517 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2518 }
2519
2520 void os::signal_raise(int signal_number) {
2521 ::raise(signal_number);
2522 }
2523
2524 // The following code is moved from os.cpp for making this
2525 // code platform specific, which it is by its very nature.
2526
2527 // Will be modified when max signal is changed to be dynamic
2528 int os::sigexitnum_pd() {
2529 return NSIG;
2530 }
2531
2532 // a counter for each possible signal value
2533 static volatile jint pending_signals[NSIG+1] = { 0 };
2534
2535 // Linux(POSIX) specific hand shaking semaphore.
2536 static sem_t sig_sem;
2537 static PosixSemaphore sr_semaphore;
2538
2539 void os::signal_init_pd() {
2540 // Initialize signal structures
2541 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2542
2543 // Initialize signal semaphore
2544 ::sem_init(&sig_sem, 0, 0);
2545 }
2546
2547 void os::signal_notify(int sig) {
2548 Atomic::inc(&pending_signals[sig]);
2549 ::sem_post(&sig_sem);
2550 }
2551
2552 static int check_pending_signals(bool wait) {
2553 Atomic::store(0, &sigint_count);
2554 for (;;) {
2555 for (int i = 0; i < NSIG + 1; i++) {
2556 jint n = pending_signals[i];
2557 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2558 return i;
2559 }
2560 }
2561 if (!wait) {
2562 return -1;
2563 }
2564 JavaThread *thread = JavaThread::current();
2565 ThreadBlockInVM tbivm(thread);
2566
2567 bool threadIsSuspended;
2568 do {
2569 thread->set_suspend_equivalent();
2570 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2571 ::sem_wait(&sig_sem);
2572
2573 // were we externally suspended while we were waiting?
2574 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2575 if (threadIsSuspended) {
2576 // The semaphore has been incremented, but while we were waiting
2577 // another thread suspended us. We don't want to continue running
2578 // while suspended because that would surprise the thread that
2579 // suspended us.
2580 ::sem_post(&sig_sem);
2581
2582 thread->java_suspend_self();
2583 }
2584 } while (threadIsSuspended);
2585 }
2586 }
2587
2588 int os::signal_lookup() {
2589 return check_pending_signals(false);
2590 }
2591
2592 int os::signal_wait() {
2593 return check_pending_signals(true);
2594 }
2595
2596 ////////////////////////////////////////////////////////////////////////////////
2597 // Virtual Memory
2598
2599 int os::vm_page_size() {
2600 // Seems redundant as all get out
2601 assert(os::Linux::page_size() != -1, "must call os::init");
2602 return os::Linux::page_size();
2603 }
2604
2605 // Solaris allocates memory by pages.
2606 int os::vm_allocation_granularity() {
2607 assert(os::Linux::page_size() != -1, "must call os::init");
2608 return os::Linux::page_size();
2609 }
2610
2611 // Rationale behind this function:
2612 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2613 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2614 // samples for JITted code. Here we create private executable mapping over the code cache
2615 // and then we can use standard (well, almost, as mapping can change) way to provide
2616 // info for the reporting script by storing timestamp and location of symbol
2617 void linux_wrap_code(char* base, size_t size) {
2618 static volatile jint cnt = 0;
2619
2620 if (!UseOprofile) {
2621 return;
2622 }
2623
2624 char buf[PATH_MAX+1];
2625 int num = Atomic::add(1, &cnt);
2626
2627 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2628 os::get_temp_directory(), os::current_process_id(), num);
2629 unlink(buf);
2630
2631 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2632
2633 if (fd != -1) {
2634 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2635 if (rv != (off_t)-1) {
2636 if (::write(fd, "", 1) == 1) {
2637 mmap(base, size,
2638 PROT_READ|PROT_WRITE|PROT_EXEC,
2639 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2640 }
2641 }
2642 ::close(fd);
2643 unlink(buf);
2644 }
2645 }
2646
2647 static bool recoverable_mmap_error(int err) {
2648 // See if the error is one we can let the caller handle. This
2649 // list of errno values comes from JBS-6843484. I can't find a
2650 // Linux man page that documents this specific set of errno
2651 // values so while this list currently matches Solaris, it may
2652 // change as we gain experience with this failure mode.
2653 switch (err) {
2654 case EBADF:
2655 case EINVAL:
2656 case ENOTSUP:
2657 // let the caller deal with these errors
2658 return true;
2659
2660 default:
2661 // Any remaining errors on this OS can cause our reserved mapping
2662 // to be lost. That can cause confusion where different data
2663 // structures think they have the same memory mapped. The worst
2664 // scenario is if both the VM and a library think they have the
2665 // same memory mapped.
2666 return false;
2667 }
2668 }
2669
2670 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2671 int err) {
2672 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2673 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2674 os::strerror(err), err);
2675 }
2676
2677 static void warn_fail_commit_memory(char* addr, size_t size,
2678 size_t alignment_hint, bool exec,
2679 int err) {
2680 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2681 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2682 alignment_hint, exec, os::strerror(err), err);
2683 }
2684
2685 // NOTE: Linux kernel does not really reserve the pages for us.
2686 // All it does is to check if there are enough free pages
2687 // left at the time of mmap(). This could be a potential
2688 // problem.
2689 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2690 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2691 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2692 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2693 if (res != (uintptr_t) MAP_FAILED) {
2694 if (UseNUMAInterleaving) {
2695 numa_make_global(addr, size);
2696 }
2697 return 0;
2698 }
2699
2700 int err = errno; // save errno from mmap() call above
2701
2702 if (!recoverable_mmap_error(err)) {
2703 warn_fail_commit_memory(addr, size, exec, err);
2704 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2705 }
2706
2707 return err;
2708 }
2709
2710 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2711 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2712 }
2713
2714 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2715 const char* mesg) {
2716 assert(mesg != NULL, "mesg must be specified");
2717 int err = os::Linux::commit_memory_impl(addr, size, exec);
2718 if (err != 0) {
2719 // the caller wants all commit errors to exit with the specified mesg:
2720 warn_fail_commit_memory(addr, size, exec, err);
2721 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2722 }
2723 }
2724
2725 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2726 #ifndef MAP_HUGETLB
2727 #define MAP_HUGETLB 0x40000
2728 #endif
2729
2730 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2731 #ifndef MADV_HUGEPAGE
2732 #define MADV_HUGEPAGE 14
2733 #endif
2734
2735 int os::Linux::commit_memory_impl(char* addr, size_t size,
2736 size_t alignment_hint, bool exec) {
2737 int err = os::Linux::commit_memory_impl(addr, size, exec);
2738 if (err == 0) {
2739 realign_memory(addr, size, alignment_hint);
2740 }
2741 return err;
2742 }
2743
2744 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2745 bool exec) {
2746 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2747 }
2748
2749 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2750 size_t alignment_hint, bool exec,
2751 const char* mesg) {
2752 assert(mesg != NULL, "mesg must be specified");
2753 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2754 if (err != 0) {
2755 // the caller wants all commit errors to exit with the specified mesg:
2756 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2757 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2758 }
2759 }
2760
2761 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2762 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2763 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2764 // be supported or the memory may already be backed by huge pages.
2765 ::madvise(addr, bytes, MADV_HUGEPAGE);
2766 }
2767 }
2768
2769 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2770 // This method works by doing an mmap over an existing mmaping and effectively discarding
2771 // the existing pages. However it won't work for SHM-based large pages that cannot be
2772 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2773 // small pages on top of the SHM segment. This method always works for small pages, so we
2774 // allow that in any case.
2775 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2776 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2777 }
2778 }
2779
2780 void os::numa_make_global(char *addr, size_t bytes) {
2781 Linux::numa_interleave_memory(addr, bytes);
2782 }
2783
2784 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2785 // bind policy to MPOL_PREFERRED for the current thread.
2786 #define USE_MPOL_PREFERRED 0
2787
2788 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2789 // To make NUMA and large pages more robust when both enabled, we need to ease
2790 // the requirements on where the memory should be allocated. MPOL_BIND is the
2791 // default policy and it will force memory to be allocated on the specified
2792 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2793 // the specified node, but will not force it. Using this policy will prevent
2794 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2795 // free large pages.
2796 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2797 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2798 }
2799
2800 bool os::numa_topology_changed() { return false; }
2801
2802 size_t os::numa_get_groups_num() {
2803 // Return just the number of nodes in which it's possible to allocate memory
2804 // (in numa terminology, configured nodes).
2805 return Linux::numa_num_configured_nodes();
2806 }
2807
2808 int os::numa_get_group_id() {
2809 int cpu_id = Linux::sched_getcpu();
2810 if (cpu_id != -1) {
2811 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2812 if (lgrp_id != -1) {
2813 return lgrp_id;
2814 }
2815 }
2816 return 0;
2817 }
2818
2819 int os::Linux::get_existing_num_nodes() {
2820 size_t node;
2821 size_t highest_node_number = Linux::numa_max_node();
2822 int num_nodes = 0;
2823
2824 // Get the total number of nodes in the system including nodes without memory.
2825 for (node = 0; node <= highest_node_number; node++) {
2826 if (isnode_in_existing_nodes(node)) {
2827 num_nodes++;
2828 }
2829 }
2830 return num_nodes;
2831 }
2832
2833 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2834 size_t highest_node_number = Linux::numa_max_node();
2835 size_t i = 0;
2836
2837 // Map all node ids in which is possible to allocate memory. Also nodes are
2838 // not always consecutively available, i.e. available from 0 to the highest
2839 // node number.
2840 for (size_t node = 0; node <= highest_node_number; node++) {
2841 if (Linux::isnode_in_configured_nodes(node)) {
2842 ids[i++] = node;
2843 }
2844 }
2845 return i;
2846 }
2847
2848 bool os::get_page_info(char *start, page_info* info) {
2849 return false;
2850 }
2851
2852 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2853 page_info* page_found) {
2854 return end;
2855 }
2856
2857
2858 int os::Linux::sched_getcpu_syscall(void) {
2859 unsigned int cpu = 0;
2860 int retval = -1;
2861
2862 #if defined(IA32)
2863 #ifndef SYS_getcpu
2864 #define SYS_getcpu 318
2865 #endif
2866 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2867 #elif defined(AMD64)
2868 // Unfortunately we have to bring all these macros here from vsyscall.h
2869 // to be able to compile on old linuxes.
2870 #define __NR_vgetcpu 2
2871 #define VSYSCALL_START (-10UL << 20)
2872 #define VSYSCALL_SIZE 1024
2873 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2874 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2875 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2876 retval = vgetcpu(&cpu, NULL, NULL);
2877 #endif
2878
2879 return (retval == -1) ? retval : cpu;
2880 }
2881
2882 void os::Linux::sched_getcpu_init() {
2883 // sched_getcpu() should be in libc.
2884 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2885 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2886
2887 // If it's not, try a direct syscall.
2888 if (sched_getcpu() == -1) {
2889 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2890 (void*)&sched_getcpu_syscall));
2891 }
2892 }
2893
2894 // Something to do with the numa-aware allocator needs these symbols
2895 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2896 extern "C" JNIEXPORT void numa_error(char *where) { }
2897
2898 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2899 // load symbol from base version instead.
2900 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2901 void *f = dlvsym(handle, name, "libnuma_1.1");
2902 if (f == NULL) {
2903 f = dlsym(handle, name);
2904 }
2905 return f;
2906 }
2907
2908 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2909 // Return NULL if the symbol is not defined in this particular version.
2910 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2911 return dlvsym(handle, name, "libnuma_1.2");
2912 }
2913
2914 bool os::Linux::libnuma_init() {
2915 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
2916 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2917 if (handle != NULL) {
2918 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2919 libnuma_dlsym(handle, "numa_node_to_cpus")));
2920 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2921 libnuma_dlsym(handle, "numa_max_node")));
2922 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2923 libnuma_dlsym(handle, "numa_num_configured_nodes")));
2924 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2925 libnuma_dlsym(handle, "numa_available")));
2926 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2927 libnuma_dlsym(handle, "numa_tonode_memory")));
2928 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2929 libnuma_dlsym(handle, "numa_interleave_memory")));
2930 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2931 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2932 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2933 libnuma_dlsym(handle, "numa_set_bind_policy")));
2934 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2935 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2936 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2937 libnuma_dlsym(handle, "numa_distance")));
2938
2939 if (numa_available() != -1) {
2940 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2941 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2942 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2943 // Create an index -> node mapping, since nodes are not always consecutive
2944 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2945 rebuild_nindex_to_node_map();
2946 // Create a cpu -> node mapping
2947 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2948 rebuild_cpu_to_node_map();
2949 return true;
2950 }
2951 }
2952 }
2953 return false;
2954 }
2955
2956 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2957 // Creating guard page is very expensive. Java thread has HotSpot
2958 // guard pages, only enable glibc guard page for non-Java threads.
2959 // (Remember: compiler thread is a Java thread, too!)
2960 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2961 }
2962
2963 void os::Linux::rebuild_nindex_to_node_map() {
2964 int highest_node_number = Linux::numa_max_node();
2965
2966 nindex_to_node()->clear();
2967 for (int node = 0; node <= highest_node_number; node++) {
2968 if (Linux::isnode_in_existing_nodes(node)) {
2969 nindex_to_node()->append(node);
2970 }
2971 }
2972 }
2973
2974 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2975 // The table is later used in get_node_by_cpu().
2976 void os::Linux::rebuild_cpu_to_node_map() {
2977 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2978 // in libnuma (possible values are starting from 16,
2979 // and continuing up with every other power of 2, but less
2980 // than the maximum number of CPUs supported by kernel), and
2981 // is a subject to change (in libnuma version 2 the requirements
2982 // are more reasonable) we'll just hardcode the number they use
2983 // in the library.
2984 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2985
2986 size_t cpu_num = processor_count();
2987 size_t cpu_map_size = NCPUS / BitsPerCLong;
2988 size_t cpu_map_valid_size =
2989 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2990
2991 cpu_to_node()->clear();
2992 cpu_to_node()->at_grow(cpu_num - 1);
2993
2994 size_t node_num = get_existing_num_nodes();
2995
2996 int distance = 0;
2997 int closest_distance = INT_MAX;
2998 int closest_node = 0;
2999 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
3000 for (size_t i = 0; i < node_num; i++) {
3001 // Check if node is configured (not a memory-less node). If it is not, find
3002 // the closest configured node.
3003 if (!isnode_in_configured_nodes(nindex_to_node()->at(i))) {
3004 closest_distance = INT_MAX;
3005 // Check distance from all remaining nodes in the system. Ignore distance
3006 // from itself and from another non-configured node.
3007 for (size_t m = 0; m < node_num; m++) {
3008 if (m != i && isnode_in_configured_nodes(nindex_to_node()->at(m))) {
3009 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3010 // If a closest node is found, update. There is always at least one
3011 // configured node in the system so there is always at least one node
3012 // close.
3013 if (distance != 0 && distance < closest_distance) {
3014 closest_distance = distance;
3015 closest_node = nindex_to_node()->at(m);
3016 }
3017 }
3018 }
3019 } else {
3020 // Current node is already a configured node.
3021 closest_node = nindex_to_node()->at(i);
3022 }
3023
3024 // Get cpus from the original node and map them to the closest node. If node
3025 // is a configured node (not a memory-less node), then original node and
3026 // closest node are the same.
3027 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3028 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3029 if (cpu_map[j] != 0) {
3030 for (size_t k = 0; k < BitsPerCLong; k++) {
3031 if (cpu_map[j] & (1UL << k)) {
3032 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3033 }
3034 }
3035 }
3036 }
3037 }
3038 }
3039 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3040 }
3041
3042 int os::Linux::get_node_by_cpu(int cpu_id) {
3043 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3044 return cpu_to_node()->at(cpu_id);
3045 }
3046 return -1;
3047 }
3048
3049 GrowableArray<int>* os::Linux::_cpu_to_node;
3050 GrowableArray<int>* os::Linux::_nindex_to_node;
3051 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3052 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3053 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3054 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3055 os::Linux::numa_available_func_t os::Linux::_numa_available;
3056 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3057 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3058 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3059 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3060 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3061 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3062 unsigned long* os::Linux::_numa_all_nodes;
3063 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3064 struct bitmask* os::Linux::_numa_nodes_ptr;
3065
3066 bool os::pd_uncommit_memory(char* addr, size_t size) {
3067 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3068 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3069 return res != (uintptr_t) MAP_FAILED;
3070 }
3071
3072 static address get_stack_commited_bottom(address bottom, size_t size) {
3073 address nbot = bottom;
3074 address ntop = bottom + size;
3075
3076 size_t page_sz = os::vm_page_size();
3077 unsigned pages = size / page_sz;
3078
3079 unsigned char vec[1];
3080 unsigned imin = 1, imax = pages + 1, imid;
3081 int mincore_return_value = 0;
3082
3083 assert(imin <= imax, "Unexpected page size");
3084
3085 while (imin < imax) {
3086 imid = (imax + imin) / 2;
3087 nbot = ntop - (imid * page_sz);
3088
3089 // Use a trick with mincore to check whether the page is mapped or not.
3090 // mincore sets vec to 1 if page resides in memory and to 0 if page
3091 // is swapped output but if page we are asking for is unmapped
3092 // it returns -1,ENOMEM
3093 mincore_return_value = mincore(nbot, page_sz, vec);
3094
3095 if (mincore_return_value == -1) {
3096 // Page is not mapped go up
3097 // to find first mapped page
3098 if (errno != EAGAIN) {
3099 assert(errno == ENOMEM, "Unexpected mincore errno");
3100 imax = imid;
3101 }
3102 } else {
3103 // Page is mapped go down
3104 // to find first not mapped page
3105 imin = imid + 1;
3106 }
3107 }
3108
3109 nbot = nbot + page_sz;
3110
3111 // Adjust stack bottom one page up if last checked page is not mapped
3112 if (mincore_return_value == -1) {
3113 nbot = nbot + page_sz;
3114 }
3115
3116 return nbot;
3117 }
3118
3119
3120 // Linux uses a growable mapping for the stack, and if the mapping for
3121 // the stack guard pages is not removed when we detach a thread the
3122 // stack cannot grow beyond the pages where the stack guard was
3123 // mapped. If at some point later in the process the stack expands to
3124 // that point, the Linux kernel cannot expand the stack any further
3125 // because the guard pages are in the way, and a segfault occurs.
3126 //
3127 // However, it's essential not to split the stack region by unmapping
3128 // a region (leaving a hole) that's already part of the stack mapping,
3129 // so if the stack mapping has already grown beyond the guard pages at
3130 // the time we create them, we have to truncate the stack mapping.
3131 // So, we need to know the extent of the stack mapping when
3132 // create_stack_guard_pages() is called.
3133
3134 // We only need this for stacks that are growable: at the time of
3135 // writing thread stacks don't use growable mappings (i.e. those
3136 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3137 // only applies to the main thread.
3138
3139 // If the (growable) stack mapping already extends beyond the point
3140 // where we're going to put our guard pages, truncate the mapping at
3141 // that point by munmap()ping it. This ensures that when we later
3142 // munmap() the guard pages we don't leave a hole in the stack
3143 // mapping. This only affects the main/primordial thread
3144
3145 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3146 if (os::is_primordial_thread()) {
3147 // As we manually grow stack up to bottom inside create_attached_thread(),
3148 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3149 // we don't need to do anything special.
3150 // Check it first, before calling heavy function.
3151 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3152 unsigned char vec[1];
3153
3154 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3155 // Fallback to slow path on all errors, including EAGAIN
3156 stack_extent = (uintptr_t) get_stack_commited_bottom(
3157 os::Linux::initial_thread_stack_bottom(),
3158 (size_t)addr - stack_extent);
3159 }
3160
3161 if (stack_extent < (uintptr_t)addr) {
3162 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3163 }
3164 }
3165
3166 return os::commit_memory(addr, size, !ExecMem);
3167 }
3168
3169 // If this is a growable mapping, remove the guard pages entirely by
3170 // munmap()ping them. If not, just call uncommit_memory(). This only
3171 // affects the main/primordial thread, but guard against future OS changes.
3172 // It's safe to always unmap guard pages for primordial thread because we
3173 // always place it right after end of the mapped region.
3174
3175 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3176 uintptr_t stack_extent, stack_base;
3177
3178 if (os::is_primordial_thread()) {
3179 return ::munmap(addr, size) == 0;
3180 }
3181
3182 return os::uncommit_memory(addr, size);
3183 }
3184
3185 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3186 // at 'requested_addr'. If there are existing memory mappings at the same
3187 // location, however, they will be overwritten. If 'fixed' is false,
3188 // 'requested_addr' is only treated as a hint, the return value may or
3189 // may not start from the requested address. Unlike Linux mmap(), this
3190 // function returns NULL to indicate failure.
3191 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3192 char * addr;
3193 int flags;
3194
3195 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3196 if (fixed) {
3197 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3198 flags |= MAP_FIXED;
3199 }
3200
3201 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3202 // touch an uncommitted page. Otherwise, the read/write might
3203 // succeed if we have enough swap space to back the physical page.
3204 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3205 flags, -1, 0);
3206
3207 return addr == MAP_FAILED ? NULL : addr;
3208 }
3209
3210 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3211 // (req_addr != NULL) or with a given alignment.
3212 // - bytes shall be a multiple of alignment.
3213 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3214 // - alignment sets the alignment at which memory shall be allocated.
3215 // It must be a multiple of allocation granularity.
3216 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3217 // req_addr or NULL.
3218 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3219
3220 size_t extra_size = bytes;
3221 if (req_addr == NULL && alignment > 0) {
3222 extra_size += alignment;
3223 }
3224
3225 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3226 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3227 -1, 0);
3228 if (start == MAP_FAILED) {
3229 start = NULL;
3230 } else {
3231 if (req_addr != NULL) {
3232 if (start != req_addr) {
3233 ::munmap(start, extra_size);
3234 start = NULL;
3235 }
3236 } else {
3237 char* const start_aligned = align_up(start, alignment);
3238 char* const end_aligned = start_aligned + bytes;
3239 char* const end = start + extra_size;
3240 if (start_aligned > start) {
3241 ::munmap(start, start_aligned - start);
3242 }
3243 if (end_aligned < end) {
3244 ::munmap(end_aligned, end - end_aligned);
3245 }
3246 start = start_aligned;
3247 }
3248 }
3249 return start;
3250 }
3251
3252 static int anon_munmap(char * addr, size_t size) {
3253 return ::munmap(addr, size) == 0;
3254 }
3255
3256 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3257 size_t alignment_hint) {
3258 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3259 }
3260
3261 bool os::pd_release_memory(char* addr, size_t size) {
3262 return anon_munmap(addr, size);
3263 }
3264
3265 static bool linux_mprotect(char* addr, size_t size, int prot) {
3266 // Linux wants the mprotect address argument to be page aligned.
3267 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3268
3269 // According to SUSv3, mprotect() should only be used with mappings
3270 // established by mmap(), and mmap() always maps whole pages. Unaligned
3271 // 'addr' likely indicates problem in the VM (e.g. trying to change
3272 // protection of malloc'ed or statically allocated memory). Check the
3273 // caller if you hit this assert.
3274 assert(addr == bottom, "sanity check");
3275
3276 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3277 return ::mprotect(bottom, size, prot) == 0;
3278 }
3279
3280 // Set protections specified
3281 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3282 bool is_committed) {
3283 unsigned int p = 0;
3284 switch (prot) {
3285 case MEM_PROT_NONE: p = PROT_NONE; break;
3286 case MEM_PROT_READ: p = PROT_READ; break;
3287 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3288 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3289 default:
3290 ShouldNotReachHere();
3291 }
3292 // is_committed is unused.
3293 return linux_mprotect(addr, bytes, p);
3294 }
3295
3296 bool os::guard_memory(char* addr, size_t size) {
3297 return linux_mprotect(addr, size, PROT_NONE);
3298 }
3299
3300 bool os::unguard_memory(char* addr, size_t size) {
3301 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3302 }
3303
3304 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3305 size_t page_size) {
3306 bool result = false;
3307 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3308 MAP_ANONYMOUS|MAP_PRIVATE,
3309 -1, 0);
3310 if (p != MAP_FAILED) {
3311 void *aligned_p = align_up(p, page_size);
3312
3313 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3314
3315 munmap(p, page_size * 2);
3316 }
3317
3318 if (warn && !result) {
3319 warning("TransparentHugePages is not supported by the operating system.");
3320 }
3321
3322 return result;
3323 }
3324
3325 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3326 bool result = false;
3327 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3328 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3329 -1, 0);
3330
3331 if (p != MAP_FAILED) {
3332 // We don't know if this really is a huge page or not.
3333 FILE *fp = fopen("/proc/self/maps", "r");
3334 if (fp) {
3335 while (!feof(fp)) {
3336 char chars[257];
3337 long x = 0;
3338 if (fgets(chars, sizeof(chars), fp)) {
3339 if (sscanf(chars, "%lx-%*x", &x) == 1
3340 && x == (long)p) {
3341 if (strstr (chars, "hugepage")) {
3342 result = true;
3343 break;
3344 }
3345 }
3346 }
3347 }
3348 fclose(fp);
3349 }
3350 munmap(p, page_size);
3351 }
3352
3353 if (warn && !result) {
3354 warning("HugeTLBFS is not supported by the operating system.");
3355 }
3356
3357 return result;
3358 }
3359
3360 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3361 //
3362 // From the coredump_filter documentation:
3363 //
3364 // - (bit 0) anonymous private memory
3365 // - (bit 1) anonymous shared memory
3366 // - (bit 2) file-backed private memory
3367 // - (bit 3) file-backed shared memory
3368 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3369 // effective only if the bit 2 is cleared)
3370 // - (bit 5) hugetlb private memory
3371 // - (bit 6) hugetlb shared memory
3372 //
3373 static void set_coredump_filter(void) {
3374 FILE *f;
3375 long cdm;
3376
3377 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3378 return;
3379 }
3380
3381 if (fscanf(f, "%lx", &cdm) != 1) {
3382 fclose(f);
3383 return;
3384 }
3385
3386 rewind(f);
3387
3388 if ((cdm & LARGEPAGES_BIT) == 0) {
3389 cdm |= LARGEPAGES_BIT;
3390 fprintf(f, "%#lx", cdm);
3391 }
3392
3393 fclose(f);
3394 }
3395
3396 // Large page support
3397
3398 static size_t _large_page_size = 0;
3399
3400 size_t os::Linux::find_large_page_size() {
3401 size_t large_page_size = 0;
3402
3403 // large_page_size on Linux is used to round up heap size. x86 uses either
3404 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3405 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3406 // page as large as 256M.
3407 //
3408 // Here we try to figure out page size by parsing /proc/meminfo and looking
3409 // for a line with the following format:
3410 // Hugepagesize: 2048 kB
3411 //
3412 // If we can't determine the value (e.g. /proc is not mounted, or the text
3413 // format has been changed), we'll use the largest page size supported by
3414 // the processor.
3415
3416 #ifndef ZERO
3417 large_page_size =
3418 AARCH64_ONLY(2 * M)
3419 AMD64_ONLY(2 * M)
3420 ARM32_ONLY(2 * M)
3421 IA32_ONLY(4 * M)
3422 IA64_ONLY(256 * M)
3423 PPC_ONLY(4 * M)
3424 S390_ONLY(1 * M)
3425 SPARC_ONLY(4 * M);
3426 #endif // ZERO
3427
3428 FILE *fp = fopen("/proc/meminfo", "r");
3429 if (fp) {
3430 while (!feof(fp)) {
3431 int x = 0;
3432 char buf[16];
3433 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3434 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3435 large_page_size = x * K;
3436 break;
3437 }
3438 } else {
3439 // skip to next line
3440 for (;;) {
3441 int ch = fgetc(fp);
3442 if (ch == EOF || ch == (int)'\n') break;
3443 }
3444 }
3445 }
3446 fclose(fp);
3447 }
3448
3449 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3450 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3451 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3452 proper_unit_for_byte_size(large_page_size));
3453 }
3454
3455 return large_page_size;
3456 }
3457
3458 size_t os::Linux::setup_large_page_size() {
3459 _large_page_size = Linux::find_large_page_size();
3460 const size_t default_page_size = (size_t)Linux::page_size();
3461 if (_large_page_size > default_page_size) {
3462 _page_sizes[0] = _large_page_size;
3463 _page_sizes[1] = default_page_size;
3464 _page_sizes[2] = 0;
3465 }
3466
3467 return _large_page_size;
3468 }
3469
3470 bool os::Linux::setup_large_page_type(size_t page_size) {
3471 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3472 FLAG_IS_DEFAULT(UseSHM) &&
3473 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3474
3475 // The type of large pages has not been specified by the user.
3476
3477 // Try UseHugeTLBFS and then UseSHM.
3478 UseHugeTLBFS = UseSHM = true;
3479
3480 // Don't try UseTransparentHugePages since there are known
3481 // performance issues with it turned on. This might change in the future.
3482 UseTransparentHugePages = false;
3483 }
3484
3485 if (UseTransparentHugePages) {
3486 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3487 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3488 UseHugeTLBFS = false;
3489 UseSHM = false;
3490 return true;
3491 }
3492 UseTransparentHugePages = false;
3493 }
3494
3495 if (UseHugeTLBFS) {
3496 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3497 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3498 UseSHM = false;
3499 return true;
3500 }
3501 UseHugeTLBFS = false;
3502 }
3503
3504 return UseSHM;
3505 }
3506
3507 void os::large_page_init() {
3508 if (!UseLargePages &&
3509 !UseTransparentHugePages &&
3510 !UseHugeTLBFS &&
3511 !UseSHM) {
3512 // Not using large pages.
3513 return;
3514 }
3515
3516 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3517 // The user explicitly turned off large pages.
3518 // Ignore the rest of the large pages flags.
3519 UseTransparentHugePages = false;
3520 UseHugeTLBFS = false;
3521 UseSHM = false;
3522 return;
3523 }
3524
3525 size_t large_page_size = Linux::setup_large_page_size();
3526 UseLargePages = Linux::setup_large_page_type(large_page_size);
3527
3528 set_coredump_filter();
3529 }
3530
3531 #ifndef SHM_HUGETLB
3532 #define SHM_HUGETLB 04000
3533 #endif
3534
3535 #define shm_warning_format(format, ...) \
3536 do { \
3537 if (UseLargePages && \
3538 (!FLAG_IS_DEFAULT(UseLargePages) || \
3539 !FLAG_IS_DEFAULT(UseSHM) || \
3540 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3541 warning(format, __VA_ARGS__); \
3542 } \
3543 } while (0)
3544
3545 #define shm_warning(str) shm_warning_format("%s", str)
3546
3547 #define shm_warning_with_errno(str) \
3548 do { \
3549 int err = errno; \
3550 shm_warning_format(str " (error = %d)", err); \
3551 } while (0)
3552
3553 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3554 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3555
3556 if (!is_aligned(alignment, SHMLBA)) {
3557 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3558 return NULL;
3559 }
3560
3561 // To ensure that we get 'alignment' aligned memory from shmat,
3562 // we pre-reserve aligned virtual memory and then attach to that.
3563
3564 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3565 if (pre_reserved_addr == NULL) {
3566 // Couldn't pre-reserve aligned memory.
3567 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3568 return NULL;
3569 }
3570
3571 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3572 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3573
3574 if ((intptr_t)addr == -1) {
3575 int err = errno;
3576 shm_warning_with_errno("Failed to attach shared memory.");
3577
3578 assert(err != EACCES, "Unexpected error");
3579 assert(err != EIDRM, "Unexpected error");
3580 assert(err != EINVAL, "Unexpected error");
3581
3582 // Since we don't know if the kernel unmapped the pre-reserved memory area
3583 // we can't unmap it, since that would potentially unmap memory that was
3584 // mapped from other threads.
3585 return NULL;
3586 }
3587
3588 return addr;
3589 }
3590
3591 static char* shmat_at_address(int shmid, char* req_addr) {
3592 if (!is_aligned(req_addr, SHMLBA)) {
3593 assert(false, "Requested address needs to be SHMLBA aligned");
3594 return NULL;
3595 }
3596
3597 char* addr = (char*)shmat(shmid, req_addr, 0);
3598
3599 if ((intptr_t)addr == -1) {
3600 shm_warning_with_errno("Failed to attach shared memory.");
3601 return NULL;
3602 }
3603
3604 return addr;
3605 }
3606
3607 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3608 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3609 if (req_addr != NULL) {
3610 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3611 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3612 return shmat_at_address(shmid, req_addr);
3613 }
3614
3615 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3616 // return large page size aligned memory addresses when req_addr == NULL.
3617 // However, if the alignment is larger than the large page size, we have
3618 // to manually ensure that the memory returned is 'alignment' aligned.
3619 if (alignment > os::large_page_size()) {
3620 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3621 return shmat_with_alignment(shmid, bytes, alignment);
3622 } else {
3623 return shmat_at_address(shmid, NULL);
3624 }
3625 }
3626
3627 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3628 char* req_addr, bool exec) {
3629 // "exec" is passed in but not used. Creating the shared image for
3630 // the code cache doesn't have an SHM_X executable permission to check.
3631 assert(UseLargePages && UseSHM, "only for SHM large pages");
3632 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3633 assert(is_aligned(req_addr, alignment), "Unaligned address");
3634
3635 if (!is_aligned(bytes, os::large_page_size())) {
3636 return NULL; // Fallback to small pages.
3637 }
3638
3639 // Create a large shared memory region to attach to based on size.
3640 // Currently, size is the total size of the heap.
3641 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3642 if (shmid == -1) {
3643 // Possible reasons for shmget failure:
3644 // 1. shmmax is too small for Java heap.
3645 // > check shmmax value: cat /proc/sys/kernel/shmmax
3646 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3647 // 2. not enough large page memory.
3648 // > check available large pages: cat /proc/meminfo
3649 // > increase amount of large pages:
3650 // echo new_value > /proc/sys/vm/nr_hugepages
3651 // Note 1: different Linux may use different name for this property,
3652 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3653 // Note 2: it's possible there's enough physical memory available but
3654 // they are so fragmented after a long run that they can't
3655 // coalesce into large pages. Try to reserve large pages when
3656 // the system is still "fresh".
3657 shm_warning_with_errno("Failed to reserve shared memory.");
3658 return NULL;
3659 }
3660
3661 // Attach to the region.
3662 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3663
3664 // Remove shmid. If shmat() is successful, the actual shared memory segment
3665 // will be deleted when it's detached by shmdt() or when the process
3666 // terminates. If shmat() is not successful this will remove the shared
3667 // segment immediately.
3668 shmctl(shmid, IPC_RMID, NULL);
3669
3670 return addr;
3671 }
3672
3673 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3674 int error) {
3675 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3676
3677 bool warn_on_failure = UseLargePages &&
3678 (!FLAG_IS_DEFAULT(UseLargePages) ||
3679 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3680 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3681
3682 if (warn_on_failure) {
3683 char msg[128];
3684 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3685 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3686 warning("%s", msg);
3687 }
3688 }
3689
3690 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3691 char* req_addr,
3692 bool exec) {
3693 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3694 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3695 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3696
3697 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3698 char* addr = (char*)::mmap(req_addr, bytes, prot,
3699 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3700 -1, 0);
3701
3702 if (addr == MAP_FAILED) {
3703 warn_on_large_pages_failure(req_addr, bytes, errno);
3704 return NULL;
3705 }
3706
3707 assert(is_aligned(addr, os::large_page_size()), "Must be");
3708
3709 return addr;
3710 }
3711
3712 // Reserve memory using mmap(MAP_HUGETLB).
3713 // - bytes shall be a multiple of alignment.
3714 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3715 // - alignment sets the alignment at which memory shall be allocated.
3716 // It must be a multiple of allocation granularity.
3717 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3718 // req_addr or NULL.
3719 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3720 size_t alignment,
3721 char* req_addr,
3722 bool exec) {
3723 size_t large_page_size = os::large_page_size();
3724 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3725
3726 assert(is_aligned(req_addr, alignment), "Must be");
3727 assert(is_aligned(bytes, alignment), "Must be");
3728
3729 // First reserve - but not commit - the address range in small pages.
3730 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3731
3732 if (start == NULL) {
3733 return NULL;
3734 }
3735
3736 assert(is_aligned(start, alignment), "Must be");
3737
3738 char* end = start + bytes;
3739
3740 // Find the regions of the allocated chunk that can be promoted to large pages.
3741 char* lp_start = align_up(start, large_page_size);
3742 char* lp_end = align_down(end, large_page_size);
3743
3744 size_t lp_bytes = lp_end - lp_start;
3745
3746 assert(is_aligned(lp_bytes, large_page_size), "Must be");
3747
3748 if (lp_bytes == 0) {
3749 // The mapped region doesn't even span the start and the end of a large page.
3750 // Fall back to allocate a non-special area.
3751 ::munmap(start, end - start);
3752 return NULL;
3753 }
3754
3755 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3756
3757 void* result;
3758
3759 // Commit small-paged leading area.
3760 if (start != lp_start) {
3761 result = ::mmap(start, lp_start - start, prot,
3762 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3763 -1, 0);
3764 if (result == MAP_FAILED) {
3765 ::munmap(lp_start, end - lp_start);
3766 return NULL;
3767 }
3768 }
3769
3770 // Commit large-paged area.
3771 result = ::mmap(lp_start, lp_bytes, prot,
3772 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3773 -1, 0);
3774 if (result == MAP_FAILED) {
3775 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3776 // If the mmap above fails, the large pages region will be unmapped and we
3777 // have regions before and after with small pages. Release these regions.
3778 //
3779 // | mapped | unmapped | mapped |
3780 // ^ ^ ^ ^
3781 // start lp_start lp_end end
3782 //
3783 ::munmap(start, lp_start - start);
3784 ::munmap(lp_end, end - lp_end);
3785 return NULL;
3786 }
3787
3788 // Commit small-paged trailing area.
3789 if (lp_end != end) {
3790 result = ::mmap(lp_end, end - lp_end, prot,
3791 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3792 -1, 0);
3793 if (result == MAP_FAILED) {
3794 ::munmap(start, lp_end - start);
3795 return NULL;
3796 }
3797 }
3798
3799 return start;
3800 }
3801
3802 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3803 size_t alignment,
3804 char* req_addr,
3805 bool exec) {
3806 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3807 assert(is_aligned(req_addr, alignment), "Must be");
3808 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3809 assert(is_power_of_2(os::large_page_size()), "Must be");
3810 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3811
3812 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3813 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3814 } else {
3815 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3816 }
3817 }
3818
3819 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3820 char* req_addr, bool exec) {
3821 assert(UseLargePages, "only for large pages");
3822
3823 char* addr;
3824 if (UseSHM) {
3825 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3826 } else {
3827 assert(UseHugeTLBFS, "must be");
3828 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3829 }
3830
3831 if (addr != NULL) {
3832 if (UseNUMAInterleaving) {
3833 numa_make_global(addr, bytes);
3834 }
3835
3836 // The memory is committed
3837 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3838 }
3839
3840 return addr;
3841 }
3842
3843 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3844 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3845 return shmdt(base) == 0;
3846 }
3847
3848 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3849 return pd_release_memory(base, bytes);
3850 }
3851
3852 bool os::release_memory_special(char* base, size_t bytes) {
3853 bool res;
3854 if (MemTracker::tracking_level() > NMT_minimal) {
3855 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3856 res = os::Linux::release_memory_special_impl(base, bytes);
3857 if (res) {
3858 tkr.record((address)base, bytes);
3859 }
3860
3861 } else {
3862 res = os::Linux::release_memory_special_impl(base, bytes);
3863 }
3864 return res;
3865 }
3866
3867 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3868 assert(UseLargePages, "only for large pages");
3869 bool res;
3870
3871 if (UseSHM) {
3872 res = os::Linux::release_memory_special_shm(base, bytes);
3873 } else {
3874 assert(UseHugeTLBFS, "must be");
3875 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3876 }
3877 return res;
3878 }
3879
3880 size_t os::large_page_size() {
3881 return _large_page_size;
3882 }
3883
3884 // With SysV SHM the entire memory region must be allocated as shared
3885 // memory.
3886 // HugeTLBFS allows application to commit large page memory on demand.
3887 // However, when committing memory with HugeTLBFS fails, the region
3888 // that was supposed to be committed will lose the old reservation
3889 // and allow other threads to steal that memory region. Because of this
3890 // behavior we can't commit HugeTLBFS memory.
3891 bool os::can_commit_large_page_memory() {
3892 return UseTransparentHugePages;
3893 }
3894
3895 bool os::can_execute_large_page_memory() {
3896 return UseTransparentHugePages || UseHugeTLBFS;
3897 }
3898
3899 // Reserve memory at an arbitrary address, only if that area is
3900 // available (and not reserved for something else).
3901
3902 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3903 const int max_tries = 10;
3904 char* base[max_tries];
3905 size_t size[max_tries];
3906 const size_t gap = 0x000000;
3907
3908 // Assert only that the size is a multiple of the page size, since
3909 // that's all that mmap requires, and since that's all we really know
3910 // about at this low abstraction level. If we need higher alignment,
3911 // we can either pass an alignment to this method or verify alignment
3912 // in one of the methods further up the call chain. See bug 5044738.
3913 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3914
3915 // Repeatedly allocate blocks until the block is allocated at the
3916 // right spot.
3917
3918 // Linux mmap allows caller to pass an address as hint; give it a try first,
3919 // if kernel honors the hint then we can return immediately.
3920 char * addr = anon_mmap(requested_addr, bytes, false);
3921 if (addr == requested_addr) {
3922 return requested_addr;
3923 }
3924
3925 if (addr != NULL) {
3926 // mmap() is successful but it fails to reserve at the requested address
3927 anon_munmap(addr, bytes);
3928 }
3929
3930 int i;
3931 for (i = 0; i < max_tries; ++i) {
3932 base[i] = reserve_memory(bytes);
3933
3934 if (base[i] != NULL) {
3935 // Is this the block we wanted?
3936 if (base[i] == requested_addr) {
3937 size[i] = bytes;
3938 break;
3939 }
3940
3941 // Does this overlap the block we wanted? Give back the overlapped
3942 // parts and try again.
3943
3944 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3945 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3946 unmap_memory(base[i], top_overlap);
3947 base[i] += top_overlap;
3948 size[i] = bytes - top_overlap;
3949 } else {
3950 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3951 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3952 unmap_memory(requested_addr, bottom_overlap);
3953 size[i] = bytes - bottom_overlap;
3954 } else {
3955 size[i] = bytes;
3956 }
3957 }
3958 }
3959 }
3960
3961 // Give back the unused reserved pieces.
3962
3963 for (int j = 0; j < i; ++j) {
3964 if (base[j] != NULL) {
3965 unmap_memory(base[j], size[j]);
3966 }
3967 }
3968
3969 if (i < max_tries) {
3970 return requested_addr;
3971 } else {
3972 return NULL;
3973 }
3974 }
3975
3976 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3977 return ::read(fd, buf, nBytes);
3978 }
3979
3980 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3981 return ::pread(fd, buf, nBytes, offset);
3982 }
3983
3984 // Short sleep, direct OS call.
3985 //
3986 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3987 // sched_yield(2) will actually give up the CPU:
3988 //
3989 // * Alone on this pariticular CPU, keeps running.
3990 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3991 // (pre 2.6.39).
3992 //
3993 // So calling this with 0 is an alternative.
3994 //
3995 void os::naked_short_sleep(jlong ms) {
3996 struct timespec req;
3997
3998 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3999 req.tv_sec = 0;
4000 if (ms > 0) {
4001 req.tv_nsec = (ms % 1000) * 1000000;
4002 } else {
4003 req.tv_nsec = 1;
4004 }
4005
4006 nanosleep(&req, NULL);
4007
4008 return;
4009 }
4010
4011 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4012 void os::infinite_sleep() {
4013 while (true) { // sleep forever ...
4014 ::sleep(100); // ... 100 seconds at a time
4015 }
4016 }
4017
4018 // Used to convert frequent JVM_Yield() to nops
4019 bool os::dont_yield() {
4020 return DontYieldALot;
4021 }
4022
4023 void os::naked_yield() {
4024 sched_yield();
4025 }
4026
4027 ////////////////////////////////////////////////////////////////////////////////
4028 // thread priority support
4029
4030 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4031 // only supports dynamic priority, static priority must be zero. For real-time
4032 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4033 // However, for large multi-threaded applications, SCHED_RR is not only slower
4034 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4035 // of 5 runs - Sep 2005).
4036 //
4037 // The following code actually changes the niceness of kernel-thread/LWP. It
4038 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4039 // not the entire user process, and user level threads are 1:1 mapped to kernel
4040 // threads. It has always been the case, but could change in the future. For
4041 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4042 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4043
4044 int os::java_to_os_priority[CriticalPriority + 1] = {
4045 19, // 0 Entry should never be used
4046
4047 4, // 1 MinPriority
4048 3, // 2
4049 2, // 3
4050
4051 1, // 4
4052 0, // 5 NormPriority
4053 -1, // 6
4054
4055 -2, // 7
4056 -3, // 8
4057 -4, // 9 NearMaxPriority
4058
4059 -5, // 10 MaxPriority
4060
4061 -5 // 11 CriticalPriority
4062 };
4063
4064 static int prio_init() {
4065 if (ThreadPriorityPolicy == 1) {
4066 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4067 // if effective uid is not root. Perhaps, a more elegant way of doing
4068 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4069 if (geteuid() != 0) {
4070 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4071 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4072 }
4073 ThreadPriorityPolicy = 0;
4074 }
4075 }
4076 if (UseCriticalJavaThreadPriority) {
4077 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4078 }
4079 return 0;
4080 }
4081
4082 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4083 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4084
4085 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4086 return (ret == 0) ? OS_OK : OS_ERR;
4087 }
4088
4089 OSReturn os::get_native_priority(const Thread* const thread,
4090 int *priority_ptr) {
4091 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4092 *priority_ptr = java_to_os_priority[NormPriority];
4093 return OS_OK;
4094 }
4095
4096 errno = 0;
4097 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4098 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4099 }
4100
4101 // Hint to the underlying OS that a task switch would not be good.
4102 // Void return because it's a hint and can fail.
4103 void os::hint_no_preempt() {}
4104
4105 ////////////////////////////////////////////////////////////////////////////////
4106 // suspend/resume support
4107
4108 // The low-level signal-based suspend/resume support is a remnant from the
4109 // old VM-suspension that used to be for java-suspension, safepoints etc,
4110 // within hotspot. Currently used by JFR's OSThreadSampler
4111 //
4112 // The remaining code is greatly simplified from the more general suspension
4113 // code that used to be used.
4114 //
4115 // The protocol is quite simple:
4116 // - suspend:
4117 // - sends a signal to the target thread
4118 // - polls the suspend state of the osthread using a yield loop
4119 // - target thread signal handler (SR_handler) sets suspend state
4120 // and blocks in sigsuspend until continued
4121 // - resume:
4122 // - sets target osthread state to continue
4123 // - sends signal to end the sigsuspend loop in the SR_handler
4124 //
4125 // Note that the SR_lock plays no role in this suspend/resume protocol,
4126 // but is checked for NULL in SR_handler as a thread termination indicator.
4127 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4128 //
4129 // Note that resume_clear_context() and suspend_save_context() are needed
4130 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4131 // which in part is used by:
4132 // - Forte Analyzer: AsyncGetCallTrace()
4133 // - StackBanging: get_frame_at_stack_banging_point()
4134
4135 static void resume_clear_context(OSThread *osthread) {
4136 osthread->set_ucontext(NULL);
4137 osthread->set_siginfo(NULL);
4138 }
4139
4140 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4141 ucontext_t* context) {
4142 osthread->set_ucontext(context);
4143 osthread->set_siginfo(siginfo);
4144 }
4145
4146 // Handler function invoked when a thread's execution is suspended or
4147 // resumed. We have to be careful that only async-safe functions are
4148 // called here (Note: most pthread functions are not async safe and
4149 // should be avoided.)
4150 //
4151 // Note: sigwait() is a more natural fit than sigsuspend() from an
4152 // interface point of view, but sigwait() prevents the signal hander
4153 // from being run. libpthread would get very confused by not having
4154 // its signal handlers run and prevents sigwait()'s use with the
4155 // mutex granting granting signal.
4156 //
4157 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4158 //
4159 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4160 // Save and restore errno to avoid confusing native code with EINTR
4161 // after sigsuspend.
4162 int old_errno = errno;
4163
4164 Thread* thread = Thread::current_or_null_safe();
4165 assert(thread != NULL, "Missing current thread in SR_handler");
4166
4167 // On some systems we have seen signal delivery get "stuck" until the signal
4168 // mask is changed as part of thread termination. Check that the current thread
4169 // has not already terminated (via SR_lock()) - else the following assertion
4170 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4171 // destructor has completed.
4172
4173 if (thread->SR_lock() == NULL) {
4174 return;
4175 }
4176
4177 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4178
4179 OSThread* osthread = thread->osthread();
4180
4181 os::SuspendResume::State current = osthread->sr.state();
4182 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4183 suspend_save_context(osthread, siginfo, context);
4184
4185 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4186 os::SuspendResume::State state = osthread->sr.suspended();
4187 if (state == os::SuspendResume::SR_SUSPENDED) {
4188 sigset_t suspend_set; // signals for sigsuspend()
4189 sigemptyset(&suspend_set);
4190 // get current set of blocked signals and unblock resume signal
4191 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4192 sigdelset(&suspend_set, SR_signum);
4193
4194 sr_semaphore.signal();
4195 // wait here until we are resumed
4196 while (1) {
4197 sigsuspend(&suspend_set);
4198
4199 os::SuspendResume::State result = osthread->sr.running();
4200 if (result == os::SuspendResume::SR_RUNNING) {
4201 sr_semaphore.signal();
4202 break;
4203 }
4204 }
4205
4206 } else if (state == os::SuspendResume::SR_RUNNING) {
4207 // request was cancelled, continue
4208 } else {
4209 ShouldNotReachHere();
4210 }
4211
4212 resume_clear_context(osthread);
4213 } else if (current == os::SuspendResume::SR_RUNNING) {
4214 // request was cancelled, continue
4215 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4216 // ignore
4217 } else {
4218 // ignore
4219 }
4220
4221 errno = old_errno;
4222 }
4223
4224 static int SR_initialize() {
4225 struct sigaction act;
4226 char *s;
4227
4228 // Get signal number to use for suspend/resume
4229 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4230 int sig = ::strtol(s, 0, 10);
4231 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4232 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4233 SR_signum = sig;
4234 } else {
4235 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4236 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4237 }
4238 }
4239
4240 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4241 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4242
4243 sigemptyset(&SR_sigset);
4244 sigaddset(&SR_sigset, SR_signum);
4245
4246 // Set up signal handler for suspend/resume
4247 act.sa_flags = SA_RESTART|SA_SIGINFO;
4248 act.sa_handler = (void (*)(int)) SR_handler;
4249
4250 // SR_signum is blocked by default.
4251 // 4528190 - We also need to block pthread restart signal (32 on all
4252 // supported Linux platforms). Note that LinuxThreads need to block
4253 // this signal for all threads to work properly. So we don't have
4254 // to use hard-coded signal number when setting up the mask.
4255 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4256
4257 if (sigaction(SR_signum, &act, 0) == -1) {
4258 return -1;
4259 }
4260
4261 // Save signal flag
4262 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4263 return 0;
4264 }
4265
4266 static int sr_notify(OSThread* osthread) {
4267 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4268 assert_status(status == 0, status, "pthread_kill");
4269 return status;
4270 }
4271
4272 // "Randomly" selected value for how long we want to spin
4273 // before bailing out on suspending a thread, also how often
4274 // we send a signal to a thread we want to resume
4275 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4276 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4277
4278 // returns true on success and false on error - really an error is fatal
4279 // but this seems the normal response to library errors
4280 static bool do_suspend(OSThread* osthread) {
4281 assert(osthread->sr.is_running(), "thread should be running");
4282 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4283
4284 // mark as suspended and send signal
4285 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4286 // failed to switch, state wasn't running?
4287 ShouldNotReachHere();
4288 return false;
4289 }
4290
4291 if (sr_notify(osthread) != 0) {
4292 ShouldNotReachHere();
4293 }
4294
4295 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4296 while (true) {
4297 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4298 break;
4299 } else {
4300 // timeout
4301 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4302 if (cancelled == os::SuspendResume::SR_RUNNING) {
4303 return false;
4304 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4305 // make sure that we consume the signal on the semaphore as well
4306 sr_semaphore.wait();
4307 break;
4308 } else {
4309 ShouldNotReachHere();
4310 return false;
4311 }
4312 }
4313 }
4314
4315 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4316 return true;
4317 }
4318
4319 static void do_resume(OSThread* osthread) {
4320 assert(osthread->sr.is_suspended(), "thread should be suspended");
4321 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4322
4323 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4324 // failed to switch to WAKEUP_REQUEST
4325 ShouldNotReachHere();
4326 return;
4327 }
4328
4329 while (true) {
4330 if (sr_notify(osthread) == 0) {
4331 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4332 if (osthread->sr.is_running()) {
4333 return;
4334 }
4335 }
4336 } else {
4337 ShouldNotReachHere();
4338 }
4339 }
4340
4341 guarantee(osthread->sr.is_running(), "Must be running!");
4342 }
4343
4344 ///////////////////////////////////////////////////////////////////////////////////
4345 // signal handling (except suspend/resume)
4346
4347 // This routine may be used by user applications as a "hook" to catch signals.
4348 // The user-defined signal handler must pass unrecognized signals to this
4349 // routine, and if it returns true (non-zero), then the signal handler must
4350 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4351 // routine will never retun false (zero), but instead will execute a VM panic
4352 // routine kill the process.
4353 //
4354 // If this routine returns false, it is OK to call it again. This allows
4355 // the user-defined signal handler to perform checks either before or after
4356 // the VM performs its own checks. Naturally, the user code would be making
4357 // a serious error if it tried to handle an exception (such as a null check
4358 // or breakpoint) that the VM was generating for its own correct operation.
4359 //
4360 // This routine may recognize any of the following kinds of signals:
4361 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4362 // It should be consulted by handlers for any of those signals.
4363 //
4364 // The caller of this routine must pass in the three arguments supplied
4365 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4366 // field of the structure passed to sigaction(). This routine assumes that
4367 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4368 //
4369 // Note that the VM will print warnings if it detects conflicting signal
4370 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4371 //
4372 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4373 siginfo_t* siginfo,
4374 void* ucontext,
4375 int abort_if_unrecognized);
4376
4377 void signalHandler(int sig, siginfo_t* info, void* uc) {
4378 assert(info != NULL && uc != NULL, "it must be old kernel");
4379 int orig_errno = errno; // Preserve errno value over signal handler.
4380 JVM_handle_linux_signal(sig, info, uc, true);
4381 errno = orig_errno;
4382 }
4383
4384
4385 // This boolean allows users to forward their own non-matching signals
4386 // to JVM_handle_linux_signal, harmlessly.
4387 bool os::Linux::signal_handlers_are_installed = false;
4388
4389 // For signal-chaining
4390 struct sigaction sigact[NSIG];
4391 uint64_t sigs = 0;
4392 #if (64 < NSIG-1)
4393 #error "Not all signals can be encoded in sigs. Adapt its type!"
4394 #endif
4395 bool os::Linux::libjsig_is_loaded = false;
4396 typedef struct sigaction *(*get_signal_t)(int);
4397 get_signal_t os::Linux::get_signal_action = NULL;
4398
4399 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4400 struct sigaction *actp = NULL;
4401
4402 if (libjsig_is_loaded) {
4403 // Retrieve the old signal handler from libjsig
4404 actp = (*get_signal_action)(sig);
4405 }
4406 if (actp == NULL) {
4407 // Retrieve the preinstalled signal handler from jvm
4408 actp = get_preinstalled_handler(sig);
4409 }
4410
4411 return actp;
4412 }
4413
4414 static bool call_chained_handler(struct sigaction *actp, int sig,
4415 siginfo_t *siginfo, void *context) {
4416 // Call the old signal handler
4417 if (actp->sa_handler == SIG_DFL) {
4418 // It's more reasonable to let jvm treat it as an unexpected exception
4419 // instead of taking the default action.
4420 return false;
4421 } else if (actp->sa_handler != SIG_IGN) {
4422 if ((actp->sa_flags & SA_NODEFER) == 0) {
4423 // automaticlly block the signal
4424 sigaddset(&(actp->sa_mask), sig);
4425 }
4426
4427 sa_handler_t hand = NULL;
4428 sa_sigaction_t sa = NULL;
4429 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4430 // retrieve the chained handler
4431 if (siginfo_flag_set) {
4432 sa = actp->sa_sigaction;
4433 } else {
4434 hand = actp->sa_handler;
4435 }
4436
4437 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4438 actp->sa_handler = SIG_DFL;
4439 }
4440
4441 // try to honor the signal mask
4442 sigset_t oset;
4443 sigemptyset(&oset);
4444 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4445
4446 // call into the chained handler
4447 if (siginfo_flag_set) {
4448 (*sa)(sig, siginfo, context);
4449 } else {
4450 (*hand)(sig);
4451 }
4452
4453 // restore the signal mask
4454 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4455 }
4456 // Tell jvm's signal handler the signal is taken care of.
4457 return true;
4458 }
4459
4460 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4461 bool chained = false;
4462 // signal-chaining
4463 if (UseSignalChaining) {
4464 struct sigaction *actp = get_chained_signal_action(sig);
4465 if (actp != NULL) {
4466 chained = call_chained_handler(actp, sig, siginfo, context);
4467 }
4468 }
4469 return chained;
4470 }
4471
4472 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4473 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4474 return &sigact[sig];
4475 }
4476 return NULL;
4477 }
4478
4479 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4480 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4481 sigact[sig] = oldAct;
4482 sigs |= (uint64_t)1 << (sig-1);
4483 }
4484
4485 // for diagnostic
4486 int sigflags[NSIG];
4487
4488 int os::Linux::get_our_sigflags(int sig) {
4489 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4490 return sigflags[sig];
4491 }
4492
4493 void os::Linux::set_our_sigflags(int sig, int flags) {
4494 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4495 if (sig > 0 && sig < NSIG) {
4496 sigflags[sig] = flags;
4497 }
4498 }
4499
4500 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4501 // Check for overwrite.
4502 struct sigaction oldAct;
4503 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4504
4505 void* oldhand = oldAct.sa_sigaction
4506 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4507 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4508 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4509 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4510 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4511 if (AllowUserSignalHandlers || !set_installed) {
4512 // Do not overwrite; user takes responsibility to forward to us.
4513 return;
4514 } else if (UseSignalChaining) {
4515 // save the old handler in jvm
4516 save_preinstalled_handler(sig, oldAct);
4517 // libjsig also interposes the sigaction() call below and saves the
4518 // old sigaction on it own.
4519 } else {
4520 fatal("Encountered unexpected pre-existing sigaction handler "
4521 "%#lx for signal %d.", (long)oldhand, sig);
4522 }
4523 }
4524
4525 struct sigaction sigAct;
4526 sigfillset(&(sigAct.sa_mask));
4527 sigAct.sa_handler = SIG_DFL;
4528 if (!set_installed) {
4529 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4530 } else {
4531 sigAct.sa_sigaction = signalHandler;
4532 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4533 }
4534 // Save flags, which are set by ours
4535 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4536 sigflags[sig] = sigAct.sa_flags;
4537
4538 int ret = sigaction(sig, &sigAct, &oldAct);
4539 assert(ret == 0, "check");
4540
4541 void* oldhand2 = oldAct.sa_sigaction
4542 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4543 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4544 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4545 }
4546
4547 // install signal handlers for signals that HotSpot needs to
4548 // handle in order to support Java-level exception handling.
4549
4550 void os::Linux::install_signal_handlers() {
4551 if (!signal_handlers_are_installed) {
4552 signal_handlers_are_installed = true;
4553
4554 // signal-chaining
4555 typedef void (*signal_setting_t)();
4556 signal_setting_t begin_signal_setting = NULL;
4557 signal_setting_t end_signal_setting = NULL;
4558 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4559 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4560 if (begin_signal_setting != NULL) {
4561 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4562 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4563 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4564 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4565 libjsig_is_loaded = true;
4566 assert(UseSignalChaining, "should enable signal-chaining");
4567 }
4568 if (libjsig_is_loaded) {
4569 // Tell libjsig jvm is setting signal handlers
4570 (*begin_signal_setting)();
4571 }
4572
4573 set_signal_handler(SIGSEGV, true);
4574 set_signal_handler(SIGPIPE, true);
4575 set_signal_handler(SIGBUS, true);
4576 set_signal_handler(SIGILL, true);
4577 set_signal_handler(SIGFPE, true);
4578 #if defined(PPC64)
4579 set_signal_handler(SIGTRAP, true);
4580 #endif
4581 set_signal_handler(SIGXFSZ, true);
4582
4583 if (libjsig_is_loaded) {
4584 // Tell libjsig jvm finishes setting signal handlers
4585 (*end_signal_setting)();
4586 }
4587
4588 // We don't activate signal checker if libjsig is in place, we trust ourselves
4589 // and if UserSignalHandler is installed all bets are off.
4590 // Log that signal checking is off only if -verbose:jni is specified.
4591 if (CheckJNICalls) {
4592 if (libjsig_is_loaded) {
4593 if (PrintJNIResolving) {
4594 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4595 }
4596 check_signals = false;
4597 }
4598 if (AllowUserSignalHandlers) {
4599 if (PrintJNIResolving) {
4600 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4601 }
4602 check_signals = false;
4603 }
4604 }
4605 }
4606 }
4607
4608 // This is the fastest way to get thread cpu time on Linux.
4609 // Returns cpu time (user+sys) for any thread, not only for current.
4610 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4611 // It might work on 2.6.10+ with a special kernel/glibc patch.
4612 // For reference, please, see IEEE Std 1003.1-2004:
4613 // http://www.unix.org/single_unix_specification
4614
4615 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4616 struct timespec tp;
4617 int rc = os::Linux::clock_gettime(clockid, &tp);
4618 assert(rc == 0, "clock_gettime is expected to return 0 code");
4619
4620 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4621 }
4622
4623 void os::Linux::initialize_os_info() {
4624 assert(_os_version == 0, "OS info already initialized");
4625
4626 struct utsname _uname;
4627
4628 uint32_t major;
4629 uint32_t minor;
4630 uint32_t fix;
4631
4632 int rc;
4633
4634 // Kernel version is unknown if
4635 // verification below fails.
4636 _os_version = 0x01000000;
4637
4638 rc = uname(&_uname);
4639 if (rc != -1) {
4640
4641 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4642 if (rc == 3) {
4643
4644 if (major < 256 && minor < 256 && fix < 256) {
4645 // Kernel version format is as expected,
4646 // set it overriding unknown state.
4647 _os_version = (major << 16) |
4648 (minor << 8 ) |
4649 (fix << 0 ) ;
4650 }
4651 }
4652 }
4653 }
4654
4655 uint32_t os::Linux::os_version() {
4656 assert(_os_version != 0, "not initialized");
4657 return _os_version & 0x00FFFFFF;
4658 }
4659
4660 bool os::Linux::os_version_is_known() {
4661 assert(_os_version != 0, "not initialized");
4662 return _os_version & 0x01000000 ? false : true;
4663 }
4664
4665 /////
4666 // glibc on Linux platform uses non-documented flag
4667 // to indicate, that some special sort of signal
4668 // trampoline is used.
4669 // We will never set this flag, and we should
4670 // ignore this flag in our diagnostic
4671 #ifdef SIGNIFICANT_SIGNAL_MASK
4672 #undef SIGNIFICANT_SIGNAL_MASK
4673 #endif
4674 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4675
4676 static const char* get_signal_handler_name(address handler,
4677 char* buf, int buflen) {
4678 int offset = 0;
4679 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4680 if (found) {
4681 // skip directory names
4682 const char *p1, *p2;
4683 p1 = buf;
4684 size_t len = strlen(os::file_separator());
4685 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4686 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4687 } else {
4688 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4689 }
4690 return buf;
4691 }
4692
4693 static void print_signal_handler(outputStream* st, int sig,
4694 char* buf, size_t buflen) {
4695 struct sigaction sa;
4696
4697 sigaction(sig, NULL, &sa);
4698
4699 // See comment for SIGNIFICANT_SIGNAL_MASK define
4700 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4701
4702 st->print("%s: ", os::exception_name(sig, buf, buflen));
4703
4704 address handler = (sa.sa_flags & SA_SIGINFO)
4705 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4706 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4707
4708 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4709 st->print("SIG_DFL");
4710 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4711 st->print("SIG_IGN");
4712 } else {
4713 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4714 }
4715
4716 st->print(", sa_mask[0]=");
4717 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4718
4719 address rh = VMError::get_resetted_sighandler(sig);
4720 // May be, handler was resetted by VMError?
4721 if (rh != NULL) {
4722 handler = rh;
4723 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4724 }
4725
4726 st->print(", sa_flags=");
4727 os::Posix::print_sa_flags(st, sa.sa_flags);
4728
4729 // Check: is it our handler?
4730 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4731 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4732 // It is our signal handler
4733 // check for flags, reset system-used one!
4734 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4735 st->print(
4736 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4737 os::Linux::get_our_sigflags(sig));
4738 }
4739 }
4740 st->cr();
4741 }
4742
4743
4744 #define DO_SIGNAL_CHECK(sig) \
4745 do { \
4746 if (!sigismember(&check_signal_done, sig)) { \
4747 os::Linux::check_signal_handler(sig); \
4748 } \
4749 } while (0)
4750
4751 // This method is a periodic task to check for misbehaving JNI applications
4752 // under CheckJNI, we can add any periodic checks here
4753
4754 void os::run_periodic_checks() {
4755 if (check_signals == false) return;
4756
4757 // SEGV and BUS if overridden could potentially prevent
4758 // generation of hs*.log in the event of a crash, debugging
4759 // such a case can be very challenging, so we absolutely
4760 // check the following for a good measure:
4761 DO_SIGNAL_CHECK(SIGSEGV);
4762 DO_SIGNAL_CHECK(SIGILL);
4763 DO_SIGNAL_CHECK(SIGFPE);
4764 DO_SIGNAL_CHECK(SIGBUS);
4765 DO_SIGNAL_CHECK(SIGPIPE);
4766 DO_SIGNAL_CHECK(SIGXFSZ);
4767 #if defined(PPC64)
4768 DO_SIGNAL_CHECK(SIGTRAP);
4769 #endif
4770
4771 // ReduceSignalUsage allows the user to override these handlers
4772 // see comments at the very top and jvm_md.h
4773 if (!ReduceSignalUsage) {
4774 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4775 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4776 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4777 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4778 }
4779
4780 DO_SIGNAL_CHECK(SR_signum);
4781 }
4782
4783 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4784
4785 static os_sigaction_t os_sigaction = NULL;
4786
4787 void os::Linux::check_signal_handler(int sig) {
4788 char buf[O_BUFLEN];
4789 address jvmHandler = NULL;
4790
4791
4792 struct sigaction act;
4793 if (os_sigaction == NULL) {
4794 // only trust the default sigaction, in case it has been interposed
4795 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4796 if (os_sigaction == NULL) return;
4797 }
4798
4799 os_sigaction(sig, (struct sigaction*)NULL, &act);
4800
4801
4802 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4803
4804 address thisHandler = (act.sa_flags & SA_SIGINFO)
4805 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4806 : CAST_FROM_FN_PTR(address, act.sa_handler);
4807
4808
4809 switch (sig) {
4810 case SIGSEGV:
4811 case SIGBUS:
4812 case SIGFPE:
4813 case SIGPIPE:
4814 case SIGILL:
4815 case SIGXFSZ:
4816 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4817 break;
4818
4819 case SHUTDOWN1_SIGNAL:
4820 case SHUTDOWN2_SIGNAL:
4821 case SHUTDOWN3_SIGNAL:
4822 case BREAK_SIGNAL:
4823 jvmHandler = (address)user_handler();
4824 break;
4825
4826 default:
4827 if (sig == SR_signum) {
4828 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4829 } else {
4830 return;
4831 }
4832 break;
4833 }
4834
4835 if (thisHandler != jvmHandler) {
4836 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4837 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4838 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4839 // No need to check this sig any longer
4840 sigaddset(&check_signal_done, sig);
4841 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4842 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4843 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4844 exception_name(sig, buf, O_BUFLEN));
4845 }
4846 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4847 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4848 tty->print("expected:");
4849 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4850 tty->cr();
4851 tty->print(" found:");
4852 os::Posix::print_sa_flags(tty, act.sa_flags);
4853 tty->cr();
4854 // No need to check this sig any longer
4855 sigaddset(&check_signal_done, sig);
4856 }
4857
4858 // Dump all the signal
4859 if (sigismember(&check_signal_done, sig)) {
4860 print_signal_handlers(tty, buf, O_BUFLEN);
4861 }
4862 }
4863
4864 extern void report_error(char* file_name, int line_no, char* title,
4865 char* format, ...);
4866
4867 // this is called _before_ most of the global arguments have been parsed
4868 void os::init(void) {
4869 char dummy; // used to get a guess on initial stack address
4870
4871 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4872
4873 init_random(1234567);
4874
4875 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4876 if (Linux::page_size() == -1) {
4877 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4878 os::strerror(errno));
4879 }
4880 init_page_sizes((size_t) Linux::page_size());
4881
4882 Linux::initialize_system_info();
4883
4884 Linux::initialize_os_info();
4885
4886 // _main_thread points to the thread that created/loaded the JVM.
4887 Linux::_main_thread = pthread_self();
4888
4889 Linux::clock_init();
4890 initial_time_count = javaTimeNanos();
4891
4892 // retrieve entry point for pthread_setname_np
4893 Linux::_pthread_setname_np =
4894 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4895
4896 os::Posix::init();
4897 }
4898
4899 // To install functions for atexit system call
4900 extern "C" {
4901 static void perfMemory_exit_helper() {
4902 perfMemory_exit();
4903 }
4904 }
4905
4906 void os::pd_init_container_support() {
4907 OSContainer::init();
4908 }
4909
4910 // this is called _after_ the global arguments have been parsed
4911 jint os::init_2(void) {
4912
4913 os::Posix::init_2();
4914
4915 Linux::fast_thread_clock_init();
4916
4917 // initialize suspend/resume support - must do this before signal_sets_init()
4918 if (SR_initialize() != 0) {
4919 perror("SR_initialize failed");
4920 return JNI_ERR;
4921 }
4922
4923 Linux::signal_sets_init();
4924 Linux::install_signal_handlers();
4925
4926 // Check and sets minimum stack sizes against command line options
4927 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4928 return JNI_ERR;
4929 }
4930 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4931
4932 #if defined(IA32)
4933 workaround_expand_exec_shield_cs_limit();
4934 #endif
4935
4936 Linux::libpthread_init();
4937 Linux::sched_getcpu_init();
4938 log_info(os)("HotSpot is running with %s, %s",
4939 Linux::glibc_version(), Linux::libpthread_version());
4940
4941 if (UseNUMA) {
4942 if (!Linux::libnuma_init()) {
4943 UseNUMA = false;
4944 } else {
4945 if ((Linux::numa_max_node() < 1)) {
4946 // There's only one node(they start from 0), disable NUMA.
4947 UseNUMA = false;
4948 }
4949 }
4950
4951 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4952 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4953 // we can make the adaptive lgrp chunk resizing work. If the user specified both
4954 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
4955 // and disable adaptive resizing.
4956 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4957 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
4958 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4959 UseAdaptiveSizePolicy = false;
4960 UseAdaptiveNUMAChunkSizing = false;
4961 }
4962 }
4963
4964 if (!UseNUMA && ForceNUMA) {
4965 UseNUMA = true;
4966 }
4967 }
4968
4969 if (MaxFDLimit) {
4970 // set the number of file descriptors to max. print out error
4971 // if getrlimit/setrlimit fails but continue regardless.
4972 struct rlimit nbr_files;
4973 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4974 if (status != 0) {
4975 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4976 } else {
4977 nbr_files.rlim_cur = nbr_files.rlim_max;
4978 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4979 if (status != 0) {
4980 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4981 }
4982 }
4983 }
4984
4985 // Initialize lock used to serialize thread creation (see os::create_thread)
4986 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4987
4988 // at-exit methods are called in the reverse order of their registration.
4989 // atexit functions are called on return from main or as a result of a
4990 // call to exit(3C). There can be only 32 of these functions registered
4991 // and atexit() does not set errno.
4992
4993 if (PerfAllowAtExitRegistration) {
4994 // only register atexit functions if PerfAllowAtExitRegistration is set.
4995 // atexit functions can be delayed until process exit time, which
4996 // can be problematic for embedded VM situations. Embedded VMs should
4997 // call DestroyJavaVM() to assure that VM resources are released.
4998
4999 // note: perfMemory_exit_helper atexit function may be removed in
5000 // the future if the appropriate cleanup code can be added to the
5001 // VM_Exit VMOperation's doit method.
5002 if (atexit(perfMemory_exit_helper) != 0) {
5003 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5004 }
5005 }
5006
5007 // initialize thread priority policy
5008 prio_init();
5009
5010 return JNI_OK;
5011 }
5012
5013 // Mark the polling page as unreadable
5014 void os::make_polling_page_unreadable(void) {
5015 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5016 fatal("Could not disable polling page");
5017 }
5018 }
5019
5020 // Mark the polling page as readable
5021 void os::make_polling_page_readable(void) {
5022 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5023 fatal("Could not enable polling page");
5024 }
5025 }
5026
5027 // older glibc versions don't have this macro (which expands to
5028 // an optimized bit-counting function) so we have to roll our own
5029 #ifndef CPU_COUNT
5030
5031 static int _cpu_count(const cpu_set_t* cpus) {
5032 int count = 0;
5033 // only look up to the number of configured processors
5034 for (int i = 0; i < os::processor_count(); i++) {
5035 if (CPU_ISSET(i, cpus)) {
5036 count++;
5037 }
5038 }
5039 return count;
5040 }
5041
5042 #define CPU_COUNT(cpus) _cpu_count(cpus)
5043
5044 #endif // CPU_COUNT
5045
5046 // Get the current number of available processors for this process.
5047 // This value can change at any time during a process's lifetime.
5048 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5049 // If it appears there may be more than 1024 processors then we do a
5050 // dynamic check - see 6515172 for details.
5051 // If anything goes wrong we fallback to returning the number of online
5052 // processors - which can be greater than the number available to the process.
5053 int os::Linux::active_processor_count() {
5054 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5055 cpu_set_t* cpus_p = &cpus;
5056 int cpus_size = sizeof(cpu_set_t);
5057
5058 int configured_cpus = os::processor_count(); // upper bound on available cpus
5059 int cpu_count = 0;
5060
5061 // old build platforms may not support dynamic cpu sets
5062 #ifdef CPU_ALLOC
5063
5064 // To enable easy testing of the dynamic path on different platforms we
5065 // introduce a diagnostic flag: UseCpuAllocPath
5066 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5067 // kernel may use a mask bigger than cpu_set_t
5068 log_trace(os)("active_processor_count: using dynamic path %s"
5069 "- configured processors: %d",
5070 UseCpuAllocPath ? "(forced) " : "",
5071 configured_cpus);
5072 cpus_p = CPU_ALLOC(configured_cpus);
5073 if (cpus_p != NULL) {
5074 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5075 // zero it just to be safe
5076 CPU_ZERO_S(cpus_size, cpus_p);
5077 }
5078 else {
5079 // failed to allocate so fallback to online cpus
5080 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5081 log_trace(os)("active_processor_count: "
5082 "CPU_ALLOC failed (%s) - using "
5083 "online processor count: %d",
5084 os::strerror(errno), online_cpus);
5085 return online_cpus;
5086 }
5087 }
5088 else {
5089 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5090 configured_cpus);
5091 }
5092 #else // CPU_ALLOC
5093 // these stubs won't be executed
5094 #define CPU_COUNT_S(size, cpus) -1
5095 #define CPU_FREE(cpus)
5096
5097 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5098 configured_cpus);
5099 #endif // CPU_ALLOC
5100
5101 // pid 0 means the current thread - which we have to assume represents the process
5102 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5103 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5104 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5105 }
5106 else {
5107 cpu_count = CPU_COUNT(cpus_p);
5108 }
5109 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5110 }
5111 else {
5112 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5113 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5114 "which may exceed available processors", os::strerror(errno), cpu_count);
5115 }
5116
5117 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5118 CPU_FREE(cpus_p);
5119 }
5120
5121 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5122 return cpu_count;
5123 }
5124
5125 // Determine the active processor count from one of
5126 // three different sources:
5127 //
5128 // 1. User option -XX:ActiveProcessorCount
5129 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5130 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5131 //
5132 // Option 1, if specified, will always override.
5133 // If the cgroup subsystem is active and configured, we
5134 // will return the min of the cgroup and option 2 results.
5135 // This is required since tools, such as numactl, that
5136 // alter cpu affinity do not update cgroup subsystem
5137 // cpuset configuration files.
5138 int os::active_processor_count() {
5139 // User has overridden the number of active processors
5140 if (ActiveProcessorCount > 0) {
5141 log_trace(os)("active_processor_count: "
5142 "active processor count set by user : %d",
5143 ActiveProcessorCount);
5144 return ActiveProcessorCount;
5145 }
5146
5147 int active_cpus;
5148 if (OSContainer::is_containerized()) {
5149 active_cpus = OSContainer::active_processor_count();
5150 log_trace(os)("active_processor_count: determined by OSContainer: %d",
5151 active_cpus);
5152 } else {
5153 active_cpus = os::Linux::active_processor_count();
5154 }
5155
5156 return active_cpus;
5157 }
5158
5159 void os::set_native_thread_name(const char *name) {
5160 if (Linux::_pthread_setname_np) {
5161 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5162 snprintf(buf, sizeof(buf), "%s", name);
5163 buf[sizeof(buf) - 1] = '\0';
5164 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5165 // ERANGE should not happen; all other errors should just be ignored.
5166 assert(rc != ERANGE, "pthread_setname_np failed");
5167 }
5168 }
5169
5170 bool os::distribute_processes(uint length, uint* distribution) {
5171 // Not yet implemented.
5172 return false;
5173 }
5174
5175 bool os::bind_to_processor(uint processor_id) {
5176 // Not yet implemented.
5177 return false;
5178 }
5179
5180 ///
5181
5182 void os::SuspendedThreadTask::internal_do_task() {
5183 if (do_suspend(_thread->osthread())) {
5184 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5185 do_task(context);
5186 do_resume(_thread->osthread());
5187 }
5188 }
5189
5190 ////////////////////////////////////////////////////////////////////////////////
5191 // debug support
5192
5193 bool os::find(address addr, outputStream* st) {
5194 Dl_info dlinfo;
5195 memset(&dlinfo, 0, sizeof(dlinfo));
5196 if (dladdr(addr, &dlinfo) != 0) {
5197 st->print(PTR_FORMAT ": ", p2i(addr));
5198 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5199 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5200 p2i(addr) - p2i(dlinfo.dli_saddr));
5201 } else if (dlinfo.dli_fbase != NULL) {
5202 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5203 } else {
5204 st->print("<absolute address>");
5205 }
5206 if (dlinfo.dli_fname != NULL) {
5207 st->print(" in %s", dlinfo.dli_fname);
5208 }
5209 if (dlinfo.dli_fbase != NULL) {
5210 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5211 }
5212 st->cr();
5213
5214 if (Verbose) {
5215 // decode some bytes around the PC
5216 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5217 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5218 address lowest = (address) dlinfo.dli_sname;
5219 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5220 if (begin < lowest) begin = lowest;
5221 Dl_info dlinfo2;
5222 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5223 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5224 end = (address) dlinfo2.dli_saddr;
5225 }
5226 Disassembler::decode(begin, end, st);
5227 }
5228 return true;
5229 }
5230 return false;
5231 }
5232
5233 ////////////////////////////////////////////////////////////////////////////////
5234 // misc
5235
5236 // This does not do anything on Linux. This is basically a hook for being
5237 // able to use structured exception handling (thread-local exception filters)
5238 // on, e.g., Win32.
5239 void
5240 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5241 JavaCallArguments* args, Thread* thread) {
5242 f(value, method, args, thread);
5243 }
5244
5245 void os::print_statistics() {
5246 }
5247
5248 bool os::message_box(const char* title, const char* message) {
5249 int i;
5250 fdStream err(defaultStream::error_fd());
5251 for (i = 0; i < 78; i++) err.print_raw("=");
5252 err.cr();
5253 err.print_raw_cr(title);
5254 for (i = 0; i < 78; i++) err.print_raw("-");
5255 err.cr();
5256 err.print_raw_cr(message);
5257 for (i = 0; i < 78; i++) err.print_raw("=");
5258 err.cr();
5259
5260 char buf[16];
5261 // Prevent process from exiting upon "read error" without consuming all CPU
5262 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5263
5264 return buf[0] == 'y' || buf[0] == 'Y';
5265 }
5266
5267 int os::stat(const char *path, struct stat *sbuf) {
5268 char pathbuf[MAX_PATH];
5269 if (strlen(path) > MAX_PATH - 1) {
5270 errno = ENAMETOOLONG;
5271 return -1;
5272 }
5273 os::native_path(strcpy(pathbuf, path));
5274 return ::stat(pathbuf, sbuf);
5275 }
5276
5277 // Is a (classpath) directory empty?
5278 bool os::dir_is_empty(const char* path) {
5279 DIR *dir = NULL;
5280 struct dirent *ptr;
5281
5282 dir = opendir(path);
5283 if (dir == NULL) return true;
5284
5285 // Scan the directory
5286 bool result = true;
5287 char buf[sizeof(struct dirent) + MAX_PATH];
5288 while (result && (ptr = ::readdir(dir)) != NULL) {
5289 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5290 result = false;
5291 }
5292 }
5293 closedir(dir);
5294 return result;
5295 }
5296
5297 // This code originates from JDK's sysOpen and open64_w
5298 // from src/solaris/hpi/src/system_md.c
5299
5300 int os::open(const char *path, int oflag, int mode) {
5301 if (strlen(path) > MAX_PATH - 1) {
5302 errno = ENAMETOOLONG;
5303 return -1;
5304 }
5305
5306 // All file descriptors that are opened in the Java process and not
5307 // specifically destined for a subprocess should have the close-on-exec
5308 // flag set. If we don't set it, then careless 3rd party native code
5309 // might fork and exec without closing all appropriate file descriptors
5310 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5311 // turn might:
5312 //
5313 // - cause end-of-file to fail to be detected on some file
5314 // descriptors, resulting in mysterious hangs, or
5315 //
5316 // - might cause an fopen in the subprocess to fail on a system
5317 // suffering from bug 1085341.
5318 //
5319 // (Yes, the default setting of the close-on-exec flag is a Unix
5320 // design flaw)
5321 //
5322 // See:
5323 // 1085341: 32-bit stdio routines should support file descriptors >255
5324 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5325 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5326 //
5327 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5328 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5329 // because it saves a system call and removes a small window where the flag
5330 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5331 // and we fall back to using FD_CLOEXEC (see below).
5332 #ifdef O_CLOEXEC
5333 oflag |= O_CLOEXEC;
5334 #endif
5335
5336 int fd = ::open64(path, oflag, mode);
5337 if (fd == -1) return -1;
5338
5339 //If the open succeeded, the file might still be a directory
5340 {
5341 struct stat64 buf64;
5342 int ret = ::fstat64(fd, &buf64);
5343 int st_mode = buf64.st_mode;
5344
5345 if (ret != -1) {
5346 if ((st_mode & S_IFMT) == S_IFDIR) {
5347 errno = EISDIR;
5348 ::close(fd);
5349 return -1;
5350 }
5351 } else {
5352 ::close(fd);
5353 return -1;
5354 }
5355 }
5356
5357 #ifdef FD_CLOEXEC
5358 // Validate that the use of the O_CLOEXEC flag on open above worked.
5359 // With recent kernels, we will perform this check exactly once.
5360 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5361 if (!O_CLOEXEC_is_known_to_work) {
5362 int flags = ::fcntl(fd, F_GETFD);
5363 if (flags != -1) {
5364 if ((flags & FD_CLOEXEC) != 0)
5365 O_CLOEXEC_is_known_to_work = 1;
5366 else
5367 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5368 }
5369 }
5370 #endif
5371
5372 return fd;
5373 }
5374
5375
5376 // create binary file, rewriting existing file if required
5377 int os::create_binary_file(const char* path, bool rewrite_existing) {
5378 int oflags = O_WRONLY | O_CREAT;
5379 if (!rewrite_existing) {
5380 oflags |= O_EXCL;
5381 }
5382 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5383 }
5384
5385 // return current position of file pointer
5386 jlong os::current_file_offset(int fd) {
5387 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5388 }
5389
5390 // move file pointer to the specified offset
5391 jlong os::seek_to_file_offset(int fd, jlong offset) {
5392 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5393 }
5394
5395 // This code originates from JDK's sysAvailable
5396 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5397
5398 int os::available(int fd, jlong *bytes) {
5399 jlong cur, end;
5400 int mode;
5401 struct stat64 buf64;
5402
5403 if (::fstat64(fd, &buf64) >= 0) {
5404 mode = buf64.st_mode;
5405 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5406 int n;
5407 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5408 *bytes = n;
5409 return 1;
5410 }
5411 }
5412 }
5413 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5414 return 0;
5415 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5416 return 0;
5417 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5418 return 0;
5419 }
5420 *bytes = end - cur;
5421 return 1;
5422 }
5423
5424 // Map a block of memory.
5425 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5426 char *addr, size_t bytes, bool read_only,
5427 bool allow_exec) {
5428 int prot;
5429 int flags = MAP_PRIVATE;
5430
5431 if (read_only) {
5432 prot = PROT_READ;
5433 } else {
5434 prot = PROT_READ | PROT_WRITE;
5435 }
5436
5437 if (allow_exec) {
5438 prot |= PROT_EXEC;
5439 }
5440
5441 if (addr != NULL) {
5442 flags |= MAP_FIXED;
5443 }
5444
5445 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5446 fd, file_offset);
5447 if (mapped_address == MAP_FAILED) {
5448 return NULL;
5449 }
5450 return mapped_address;
5451 }
5452
5453
5454 // Remap a block of memory.
5455 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5456 char *addr, size_t bytes, bool read_only,
5457 bool allow_exec) {
5458 // same as map_memory() on this OS
5459 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5460 allow_exec);
5461 }
5462
5463
5464 // Unmap a block of memory.
5465 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5466 return munmap(addr, bytes) == 0;
5467 }
5468
5469 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5470
5471 static clockid_t thread_cpu_clockid(Thread* thread) {
5472 pthread_t tid = thread->osthread()->pthread_id();
5473 clockid_t clockid;
5474
5475 // Get thread clockid
5476 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5477 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5478 return clockid;
5479 }
5480
5481 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5482 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5483 // of a thread.
5484 //
5485 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5486 // the fast estimate available on the platform.
5487
5488 jlong os::current_thread_cpu_time() {
5489 if (os::Linux::supports_fast_thread_cpu_time()) {
5490 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5491 } else {
5492 // return user + sys since the cost is the same
5493 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5494 }
5495 }
5496
5497 jlong os::thread_cpu_time(Thread* thread) {
5498 // consistent with what current_thread_cpu_time() returns
5499 if (os::Linux::supports_fast_thread_cpu_time()) {
5500 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5501 } else {
5502 return slow_thread_cpu_time(thread, true /* user + sys */);
5503 }
5504 }
5505
5506 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5507 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5508 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5509 } else {
5510 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5511 }
5512 }
5513
5514 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5515 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5516 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5517 } else {
5518 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5519 }
5520 }
5521
5522 // -1 on error.
5523 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5524 pid_t tid = thread->osthread()->thread_id();
5525 char *s;
5526 char stat[2048];
5527 int statlen;
5528 char proc_name[64];
5529 int count;
5530 long sys_time, user_time;
5531 char cdummy;
5532 int idummy;
5533 long ldummy;
5534 FILE *fp;
5535
5536 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5537 fp = fopen(proc_name, "r");
5538 if (fp == NULL) return -1;
5539 statlen = fread(stat, 1, 2047, fp);
5540 stat[statlen] = '\0';
5541 fclose(fp);
5542
5543 // Skip pid and the command string. Note that we could be dealing with
5544 // weird command names, e.g. user could decide to rename java launcher
5545 // to "java 1.4.2 :)", then the stat file would look like
5546 // 1234 (java 1.4.2 :)) R ... ...
5547 // We don't really need to know the command string, just find the last
5548 // occurrence of ")" and then start parsing from there. See bug 4726580.
5549 s = strrchr(stat, ')');
5550 if (s == NULL) return -1;
5551
5552 // Skip blank chars
5553 do { s++; } while (s && isspace(*s));
5554
5555 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5556 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5557 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5558 &user_time, &sys_time);
5559 if (count != 13) return -1;
5560 if (user_sys_cpu_time) {
5561 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5562 } else {
5563 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5564 }
5565 }
5566
5567 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5568 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5569 info_ptr->may_skip_backward = false; // elapsed time not wall time
5570 info_ptr->may_skip_forward = false; // elapsed time not wall time
5571 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5572 }
5573
5574 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5575 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5576 info_ptr->may_skip_backward = false; // elapsed time not wall time
5577 info_ptr->may_skip_forward = false; // elapsed time not wall time
5578 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5579 }
5580
5581 bool os::is_thread_cpu_time_supported() {
5582 return true;
5583 }
5584
5585 // System loadavg support. Returns -1 if load average cannot be obtained.
5586 // Linux doesn't yet have a (official) notion of processor sets,
5587 // so just return the system wide load average.
5588 int os::loadavg(double loadavg[], int nelem) {
5589 return ::getloadavg(loadavg, nelem);
5590 }
5591
5592 void os::pause() {
5593 char filename[MAX_PATH];
5594 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5595 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5596 } else {
5597 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5598 }
5599
5600 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5601 if (fd != -1) {
5602 struct stat buf;
5603 ::close(fd);
5604 while (::stat(filename, &buf) == 0) {
5605 (void)::poll(NULL, 0, 100);
5606 }
5607 } else {
5608 jio_fprintf(stderr,
5609 "Could not open pause file '%s', continuing immediately.\n", filename);
5610 }
5611 }
5612
5613 extern char** environ;
5614
5615 // Run the specified command in a separate process. Return its exit value,
5616 // or -1 on failure (e.g. can't fork a new process).
5617 // Unlike system(), this function can be called from signal handler. It
5618 // doesn't block SIGINT et al.
5619 int os::fork_and_exec(char* cmd) {
5620 const char * argv[4] = {"sh", "-c", cmd, NULL};
5621
5622 pid_t pid = fork();
5623
5624 if (pid < 0) {
5625 // fork failed
5626 return -1;
5627
5628 } else if (pid == 0) {
5629 // child process
5630
5631 execve("/bin/sh", (char* const*)argv, environ);
5632
5633 // execve failed
5634 _exit(-1);
5635
5636 } else {
5637 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5638 // care about the actual exit code, for now.
5639
5640 int status;
5641
5642 // Wait for the child process to exit. This returns immediately if
5643 // the child has already exited. */
5644 while (waitpid(pid, &status, 0) < 0) {
5645 switch (errno) {
5646 case ECHILD: return 0;
5647 case EINTR: break;
5648 default: return -1;
5649 }
5650 }
5651
5652 if (WIFEXITED(status)) {
5653 // The child exited normally; get its exit code.
5654 return WEXITSTATUS(status);
5655 } else if (WIFSIGNALED(status)) {
5656 // The child exited because of a signal
5657 // The best value to return is 0x80 + signal number,
5658 // because that is what all Unix shells do, and because
5659 // it allows callers to distinguish between process exit and
5660 // process death by signal.
5661 return 0x80 + WTERMSIG(status);
5662 } else {
5663 // Unknown exit code; pass it through
5664 return status;
5665 }
5666 }
5667 }
5668
5669 // is_headless_jre()
5670 //
5671 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5672 // in order to report if we are running in a headless jre
5673 //
5674 // Since JDK8 xawt/libmawt.so was moved into the same directory
5675 // as libawt.so, and renamed libawt_xawt.so
5676 //
5677 bool os::is_headless_jre() {
5678 struct stat statbuf;
5679 char buf[MAXPATHLEN];
5680 char libmawtpath[MAXPATHLEN];
5681 const char *xawtstr = "/xawt/libmawt.so";
5682 const char *new_xawtstr = "/libawt_xawt.so";
5683 char *p;
5684
5685 // Get path to libjvm.so
5686 os::jvm_path(buf, sizeof(buf));
5687
5688 // Get rid of libjvm.so
5689 p = strrchr(buf, '/');
5690 if (p == NULL) {
5691 return false;
5692 } else {
5693 *p = '\0';
5694 }
5695
5696 // Get rid of client or server
5697 p = strrchr(buf, '/');
5698 if (p == NULL) {
5699 return false;
5700 } else {
5701 *p = '\0';
5702 }
5703
5704 // check xawt/libmawt.so
5705 strcpy(libmawtpath, buf);
5706 strcat(libmawtpath, xawtstr);
5707 if (::stat(libmawtpath, &statbuf) == 0) return false;
5708
5709 // check libawt_xawt.so
5710 strcpy(libmawtpath, buf);
5711 strcat(libmawtpath, new_xawtstr);
5712 if (::stat(libmawtpath, &statbuf) == 0) return false;
5713
5714 return true;
5715 }
5716
5717 // Get the default path to the core file
5718 // Returns the length of the string
5719 int os::get_core_path(char* buffer, size_t bufferSize) {
5720 /*
5721 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5722 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5723 */
5724 const int core_pattern_len = 129;
5725 char core_pattern[core_pattern_len] = {0};
5726
5727 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5728 if (core_pattern_file == -1) {
5729 return -1;
5730 }
5731
5732 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5733 ::close(core_pattern_file);
5734 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5735 return -1;
5736 }
5737 if (core_pattern[ret-1] == '\n') {
5738 core_pattern[ret-1] = '\0';
5739 } else {
5740 core_pattern[ret] = '\0';
5741 }
5742
5743 char *pid_pos = strstr(core_pattern, "%p");
5744 int written;
5745
5746 if (core_pattern[0] == '/') {
5747 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5748 } else {
5749 char cwd[PATH_MAX];
5750
5751 const char* p = get_current_directory(cwd, PATH_MAX);
5752 if (p == NULL) {
5753 return -1;
5754 }
5755
5756 if (core_pattern[0] == '|') {
5757 written = jio_snprintf(buffer, bufferSize,
5758 "\"%s\" (or dumping to %s/core.%d)",
5759 &core_pattern[1], p, current_process_id());
5760 } else {
5761 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5762 }
5763 }
5764
5765 if (written < 0) {
5766 return -1;
5767 }
5768
5769 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5770 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5771
5772 if (core_uses_pid_file != -1) {
5773 char core_uses_pid = 0;
5774 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5775 ::close(core_uses_pid_file);
5776
5777 if (core_uses_pid == '1') {
5778 jio_snprintf(buffer + written, bufferSize - written,
5779 ".%d", current_process_id());
5780 }
5781 }
5782 }
5783
5784 return strlen(buffer);
5785 }
5786
5787 bool os::start_debugging(char *buf, int buflen) {
5788 int len = (int)strlen(buf);
5789 char *p = &buf[len];
5790
5791 jio_snprintf(p, buflen-len,
5792 "\n\n"
5793 "Do you want to debug the problem?\n\n"
5794 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5795 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5796 "Otherwise, press RETURN to abort...",
5797 os::current_process_id(), os::current_process_id(),
5798 os::current_thread_id(), os::current_thread_id());
5799
5800 bool yes = os::message_box("Unexpected Error", buf);
5801
5802 if (yes) {
5803 // yes, user asked VM to launch debugger
5804 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5805 os::current_process_id(), os::current_process_id());
5806
5807 os::fork_and_exec(buf);
5808 yes = false;
5809 }
5810 return yes;
5811 }
5812
5813
5814 // Java/Compiler thread:
5815 //
5816 // Low memory addresses
5817 // P0 +------------------------+
5818 // | |\ Java thread created by VM does not have glibc
5819 // | glibc guard page | - guard page, attached Java thread usually has
5820 // | |/ 1 glibc guard page.
5821 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5822 // | |\
5823 // | HotSpot Guard Pages | - red, yellow and reserved pages
5824 // | |/
5825 // +------------------------+ JavaThread::stack_reserved_zone_base()
5826 // | |\
5827 // | Normal Stack | -
5828 // | |/
5829 // P2 +------------------------+ Thread::stack_base()
5830 //
5831 // Non-Java thread:
5832 //
5833 // Low memory addresses
5834 // P0 +------------------------+
5835 // | |\
5836 // | glibc guard page | - usually 1 page
5837 // | |/
5838 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5839 // | |\
5840 // | Normal Stack | -
5841 // | |/
5842 // P2 +------------------------+ Thread::stack_base()
5843 //
5844 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
5845 // returned from pthread_attr_getstack().
5846 // ** Due to NPTL implementation error, linux takes the glibc guard page out
5847 // of the stack size given in pthread_attr. We work around this for
5848 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
5849 //
5850 #ifndef ZERO
5851 static void current_stack_region(address * bottom, size_t * size) {
5852 if (os::is_primordial_thread()) {
5853 // primordial thread needs special handling because pthread_getattr_np()
5854 // may return bogus value.
5855 *bottom = os::Linux::initial_thread_stack_bottom();
5856 *size = os::Linux::initial_thread_stack_size();
5857 } else {
5858 pthread_attr_t attr;
5859
5860 int rslt = pthread_getattr_np(pthread_self(), &attr);
5861
5862 // JVM needs to know exact stack location, abort if it fails
5863 if (rslt != 0) {
5864 if (rslt == ENOMEM) {
5865 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
5866 } else {
5867 fatal("pthread_getattr_np failed with error = %d", rslt);
5868 }
5869 }
5870
5871 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
5872 fatal("Cannot locate current stack attributes!");
5873 }
5874
5875 // Work around NPTL stack guard error.
5876 size_t guard_size = 0;
5877 rslt = pthread_attr_getguardsize(&attr, &guard_size);
5878 if (rslt != 0) {
5879 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
5880 }
5881 *bottom += guard_size;
5882 *size -= guard_size;
5883
5884 pthread_attr_destroy(&attr);
5885
5886 }
5887 assert(os::current_stack_pointer() >= *bottom &&
5888 os::current_stack_pointer() < *bottom + *size, "just checking");
5889 }
5890
5891 address os::current_stack_base() {
5892 address bottom;
5893 size_t size;
5894 current_stack_region(&bottom, &size);
5895 return (bottom + size);
5896 }
5897
5898 size_t os::current_stack_size() {
5899 // This stack size includes the usable stack and HotSpot guard pages
5900 // (for the threads that have Hotspot guard pages).
5901 address bottom;
5902 size_t size;
5903 current_stack_region(&bottom, &size);
5904 return size;
5905 }
5906 #endif
5907
5908 static inline struct timespec get_mtime(const char* filename) {
5909 struct stat st;
5910 int ret = os::stat(filename, &st);
5911 assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
5912 return st.st_mtim;
5913 }
5914
5915 int os::compare_file_modified_times(const char* file1, const char* file2) {
5916 struct timespec filetime1 = get_mtime(file1);
5917 struct timespec filetime2 = get_mtime(file2);
5918 int diff = filetime1.tv_sec - filetime2.tv_sec;
5919 if (diff == 0) {
5920 return filetime1.tv_nsec - filetime2.tv_nsec;
5921 }
5922 return diff;
5923 }
5924
5925 /////////////// Unit tests ///////////////
5926
5927 #ifndef PRODUCT
5928
5929 #define test_log(...) \
5930 do { \
5931 if (VerboseInternalVMTests) { \
5932 tty->print_cr(__VA_ARGS__); \
5933 tty->flush(); \
5934 } \
5935 } while (false)
5936
5937 class TestReserveMemorySpecial : AllStatic {
5938 public:
5939 static void small_page_write(void* addr, size_t size) {
5940 size_t page_size = os::vm_page_size();
5941
5942 char* end = (char*)addr + size;
5943 for (char* p = (char*)addr; p < end; p += page_size) {
5944 *p = 1;
5945 }
5946 }
5947
5948 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
5949 if (!UseHugeTLBFS) {
5950 return;
5951 }
5952
5953 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
5954
5955 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
5956
5957 if (addr != NULL) {
5958 small_page_write(addr, size);
5959
5960 os::Linux::release_memory_special_huge_tlbfs(addr, size);
5961 }
5962 }
5963
5964 static void test_reserve_memory_special_huge_tlbfs_only() {
5965 if (!UseHugeTLBFS) {
5966 return;
5967 }
5968
5969 size_t lp = os::large_page_size();
5970
5971 for (size_t size = lp; size <= lp * 10; size += lp) {
5972 test_reserve_memory_special_huge_tlbfs_only(size);
5973 }
5974 }
5975
5976 static void test_reserve_memory_special_huge_tlbfs_mixed() {
5977 size_t lp = os::large_page_size();
5978 size_t ag = os::vm_allocation_granularity();
5979
5980 // sizes to test
5981 const size_t sizes[] = {
5982 lp, lp + ag, lp + lp / 2, lp * 2,
5983 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
5984 lp * 10, lp * 10 + lp / 2
5985 };
5986 const int num_sizes = sizeof(sizes) / sizeof(size_t);
5987
5988 // For each size/alignment combination, we test three scenarios:
5989 // 1) with req_addr == NULL
5990 // 2) with a non-null req_addr at which we expect to successfully allocate
5991 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
5992 // expect the allocation to either fail or to ignore req_addr
5993
5994 // Pre-allocate two areas; they shall be as large as the largest allocation
5995 // and aligned to the largest alignment we will be testing.
5996 const size_t mapping_size = sizes[num_sizes - 1] * 2;
5997 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
5998 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
5999 -1, 0);
6000 assert(mapping1 != MAP_FAILED, "should work");
6001
6002 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6003 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6004 -1, 0);
6005 assert(mapping2 != MAP_FAILED, "should work");
6006
6007 // Unmap the first mapping, but leave the second mapping intact: the first
6008 // mapping will serve as a value for a "good" req_addr (case 2). The second
6009 // mapping, still intact, as "bad" req_addr (case 3).
6010 ::munmap(mapping1, mapping_size);
6011
6012 // Case 1
6013 test_log("%s, req_addr NULL:", __FUNCTION__);
6014 test_log("size align result");
6015
6016 for (int i = 0; i < num_sizes; i++) {
6017 const size_t size = sizes[i];
6018 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6019 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6020 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6021 size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6022 if (p != NULL) {
6023 assert(is_aligned(p, alignment), "must be");
6024 small_page_write(p, size);
6025 os::Linux::release_memory_special_huge_tlbfs(p, size);
6026 }
6027 }
6028 }
6029
6030 // Case 2
6031 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6032 test_log("size align req_addr result");
6033
6034 for (int i = 0; i < num_sizes; i++) {
6035 const size_t size = sizes[i];
6036 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6037 char* const req_addr = align_up(mapping1, alignment);
6038 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6039 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6040 size, alignment, p2i(req_addr), p2i(p),
6041 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6042 if (p != NULL) {
6043 assert(p == req_addr, "must be");
6044 small_page_write(p, size);
6045 os::Linux::release_memory_special_huge_tlbfs(p, size);
6046 }
6047 }
6048 }
6049
6050 // Case 3
6051 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6052 test_log("size align req_addr result");
6053
6054 for (int i = 0; i < num_sizes; i++) {
6055 const size_t size = sizes[i];
6056 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6057 char* const req_addr = align_up(mapping2, alignment);
6058 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6059 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6060 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6061 // as the area around req_addr contains already existing mappings, the API should always
6062 // return NULL (as per contract, it cannot return another address)
6063 assert(p == NULL, "must be");
6064 }
6065 }
6066
6067 ::munmap(mapping2, mapping_size);
6068
6069 }
6070
6071 static void test_reserve_memory_special_huge_tlbfs() {
6072 if (!UseHugeTLBFS) {
6073 return;
6074 }
6075
6076 test_reserve_memory_special_huge_tlbfs_only();
6077 test_reserve_memory_special_huge_tlbfs_mixed();
6078 }
6079
6080 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6081 if (!UseSHM) {
6082 return;
6083 }
6084
6085 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6086
6087 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6088
6089 if (addr != NULL) {
6090 assert(is_aligned(addr, alignment), "Check");
6091 assert(is_aligned(addr, os::large_page_size()), "Check");
6092
6093 small_page_write(addr, size);
6094
6095 os::Linux::release_memory_special_shm(addr, size);
6096 }
6097 }
6098
6099 static void test_reserve_memory_special_shm() {
6100 size_t lp = os::large_page_size();
6101 size_t ag = os::vm_allocation_granularity();
6102
6103 for (size_t size = ag; size < lp * 3; size += ag) {
6104 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6105 test_reserve_memory_special_shm(size, alignment);
6106 }
6107 }
6108 }
6109
6110 static void test() {
6111 test_reserve_memory_special_huge_tlbfs();
6112 test_reserve_memory_special_shm();
6113 }
6114 };
6115
6116 void TestReserveMemorySpecial_test() {
6117 TestReserveMemorySpecial::test();
6118 }
6119
6120 #endif