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