1 /*
2 * Copyright (c) 1999, 2016, 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/decoder.hpp"
68 #include "utilities/defaultStream.hpp"
69 #include "utilities/events.hpp"
70 #include "utilities/elfFile.hpp"
71 #include "utilities/growableArray.hpp"
72 #include "utilities/macros.hpp"
73 #include "utilities/vmError.hpp"
74
75 // put OS-includes here
76 # include <sys/types.h>
77 # include <sys/mman.h>
78 # include <sys/stat.h>
79 # include <sys/select.h>
80 # include <pthread.h>
81 # include <signal.h>
82 # include <errno.h>
83 # include <dlfcn.h>
84 # include <stdio.h>
85 # include <unistd.h>
86 # include <sys/resource.h>
87 # include <pthread.h>
88 # include <sys/stat.h>
89 # include <sys/time.h>
90 # include <sys/times.h>
91 # include <sys/utsname.h>
92 # include <sys/socket.h>
93 # include <sys/wait.h>
94 # include <pwd.h>
95 # include <poll.h>
96 # include <semaphore.h>
97 # include <fcntl.h>
98 # include <string.h>
99 # include <syscall.h>
100 # include <sys/sysinfo.h>
101 # include <sys/ipc.h>
102 # include <sys/shm.h>
103 # include <link.h>
104 # include <stdint.h>
105 # include <inttypes.h>
106 # include <sys/ioctl.h>
107
108 #ifndef _GNU_SOURCE
109 #define _GNU_SOURCE
110 #include <sched.h>
111 #undef _GNU_SOURCE
112 #else
113 #include <sched.h>
114 #endif
115
116 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
117 // getrusage() is prepared to handle the associated failure.
118 #ifndef RUSAGE_THREAD
119 #define RUSAGE_THREAD (1) /* only the calling thread */
120 #endif
121
122 #define MAX_PATH (2 * K)
123
124 #define MAX_SECS 100000000
125
126 // for timer info max values which include all bits
127 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
128
129 #define LARGEPAGES_BIT (1 << 6)
130 ////////////////////////////////////////////////////////////////////////////////
131 // global variables
132 julong os::Linux::_physical_memory = 0;
133
134 address os::Linux::_initial_thread_stack_bottom = NULL;
135 uintptr_t os::Linux::_initial_thread_stack_size = 0;
136
137 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
138 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
139 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
140 Mutex* os::Linux::_createThread_lock = NULL;
141 pthread_t os::Linux::_main_thread;
142 int os::Linux::_page_size = -1;
143 bool os::Linux::_supports_fast_thread_cpu_time = false;
144 uint32_t os::Linux::_os_version = 0;
145 const char * os::Linux::_glibc_version = "unknown";
146 const char * os::Linux::_libpthread_version = "unknown";
147 pthread_condattr_t os::Linux::_condattr[1];
148
149 static jlong initial_time_count=0;
150
151 static int clock_tics_per_sec = 100;
152
153 // For diagnostics to print a message once. see run_periodic_checks
154 static sigset_t check_signal_done;
155 static bool check_signals = true;
156
157 // Signal number used to suspend/resume a thread
158
159 // do not use any signal number less than SIGSEGV, see 4355769
160 static int SR_signum = SIGUSR2;
161 sigset_t SR_sigset;
162
163 // Declarations
164 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
165
166 // utility functions
167
168 static int SR_initialize();
169
170 julong os::available_memory() {
171 return Linux::available_memory();
172 }
173
174 julong os::Linux::available_memory() {
175 // values in struct sysinfo are "unsigned long"
176 struct sysinfo si;
177 sysinfo(&si);
178
179 return (julong)si.freeram * si.mem_unit;
180 }
181
182 julong os::physical_memory() {
183 return Linux::physical_memory();
184 }
185
186 // Return true if user is running as root.
187
188 bool os::have_special_privileges() {
189 static bool init = false;
190 static bool privileges = false;
191 if (!init) {
192 privileges = (getuid() != geteuid()) || (getgid() != getegid());
193 init = true;
194 }
195 return privileges;
196 }
197
198
199 #ifndef SYS_gettid
200 // i386: 224, ia64: 1105, amd64: 186, sparc 143
201 #ifdef __ia64__
202 #define SYS_gettid 1105
203 #else
204 #ifdef __i386__
205 #define SYS_gettid 224
206 #else
207 #ifdef __amd64__
208 #define SYS_gettid 186
209 #else
210 #ifdef __sparc__
211 #define SYS_gettid 143
212 #else
213 #error define gettid for the arch
214 #endif
215 #endif
216 #endif
217 #endif
218 #endif
219
220
221 // pid_t gettid()
222 //
223 // Returns the kernel thread id of the currently running thread. Kernel
224 // thread id is used to access /proc.
225 pid_t os::Linux::gettid() {
226 int rslt = syscall(SYS_gettid);
227 assert(rslt != -1, "must be."); // old linuxthreads implementation?
228 return (pid_t)rslt;
229 }
230
231 // Most versions of linux have a bug where the number of processors are
232 // determined by looking at the /proc file system. In a chroot environment,
233 // the system call returns 1. This causes the VM to act as if it is
234 // a single processor and elide locking (see is_MP() call).
235 static bool unsafe_chroot_detected = false;
236 static const char *unstable_chroot_error = "/proc file system not found.\n"
237 "Java may be unstable running multithreaded in a chroot "
238 "environment on Linux when /proc filesystem is not mounted.";
239
240 void os::Linux::initialize_system_info() {
241 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
242 if (processor_count() == 1) {
243 pid_t pid = os::Linux::gettid();
244 char fname[32];
245 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
246 FILE *fp = fopen(fname, "r");
247 if (fp == NULL) {
248 unsafe_chroot_detected = true;
249 } else {
250 fclose(fp);
251 }
252 }
253 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
254 assert(processor_count() > 0, "linux error");
255 }
256
257 void os::init_system_properties_values() {
258 // The next steps are taken in the product version:
259 //
260 // Obtain the JAVA_HOME value from the location of libjvm.so.
261 // This library should be located at:
262 // <JAVA_HOME>/lib/{client|server}/libjvm.so.
263 //
264 // If "/jre/lib/" appears at the right place in the path, then we
265 // assume libjvm.so is installed in a JDK and we use this path.
266 //
267 // Otherwise exit with message: "Could not create the Java virtual machine."
268 //
269 // The following extra steps are taken in the debugging version:
270 //
271 // If "/jre/lib/" does NOT appear at the right place in the path
272 // instead of exit check for $JAVA_HOME environment variable.
273 //
274 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
275 // then we append a fake suffix "hotspot/libjvm.so" to this path so
276 // it looks like libjvm.so is installed there
277 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
278 //
279 // Otherwise exit.
280 //
281 // Important note: if the location of libjvm.so changes this
282 // code needs to be changed accordingly.
283
284 // See ld(1):
285 // The linker uses the following search paths to locate required
286 // shared libraries:
287 // 1: ...
288 // ...
289 // 7: The default directories, normally /lib and /usr/lib.
290 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
291 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
292 #else
293 #define DEFAULT_LIBPATH "/lib:/usr/lib"
294 #endif
295
296 // Base path of extensions installed on the system.
297 #define SYS_EXT_DIR "/usr/java/packages"
298 #define EXTENSIONS_DIR "/lib/ext"
299
300 // Buffer that fits several sprintfs.
301 // Note that the space for the colon and the trailing null are provided
302 // by the nulls included by the sizeof operator.
303 const size_t bufsize =
304 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
305 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
306 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
307
308 // sysclasspath, java_home, dll_dir
309 {
310 char *pslash;
311 os::jvm_path(buf, bufsize);
312
313 // Found the full path to libjvm.so.
314 // Now cut the path to <java_home>/jre if we can.
315 pslash = strrchr(buf, '/');
316 if (pslash != NULL) {
317 *pslash = '\0'; // Get rid of /libjvm.so.
318 }
319 pslash = strrchr(buf, '/');
320 if (pslash != NULL) {
321 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
322 }
323 Arguments::set_dll_dir(buf);
324
325 if (pslash != NULL) {
326 pslash = strrchr(buf, '/');
327 if (pslash != NULL) {
328 *pslash = '\0'; // Get rid of /lib.
329 }
330 }
331 Arguments::set_java_home(buf);
332 set_boot_path('/', ':');
333 }
334
335 // Where to look for native libraries.
336 //
337 // Note: Due to a legacy implementation, most of the library path
338 // is set in the launcher. This was to accomodate linking restrictions
339 // on legacy Linux implementations (which are no longer supported).
340 // Eventually, all the library path setting will be done here.
341 //
342 // However, to prevent the proliferation of improperly built native
343 // libraries, the new path component /usr/java/packages is added here.
344 // Eventually, all the library path setting will be done here.
345 {
346 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
347 // should always exist (until the legacy problem cited above is
348 // addressed).
349 const char *v = ::getenv("LD_LIBRARY_PATH");
350 const char *v_colon = ":";
351 if (v == NULL) { v = ""; v_colon = ""; }
352 // That's +1 for the colon and +1 for the trailing '\0'.
353 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
354 strlen(v) + 1 +
355 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
356 mtInternal);
357 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
358 Arguments::set_library_path(ld_library_path);
359 FREE_C_HEAP_ARRAY(char, ld_library_path);
360 }
361
362 // Extensions directories.
363 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
364 Arguments::set_ext_dirs(buf);
365
366 FREE_C_HEAP_ARRAY(char, buf);
367
368 #undef DEFAULT_LIBPATH
369 #undef SYS_EXT_DIR
370 #undef EXTENSIONS_DIR
371 }
372
373 ////////////////////////////////////////////////////////////////////////////////
374 // breakpoint support
375
376 void os::breakpoint() {
377 BREAKPOINT;
378 }
379
380 extern "C" void breakpoint() {
381 // use debugger to set breakpoint here
382 }
383
384 ////////////////////////////////////////////////////////////////////////////////
385 // signal support
386
387 debug_only(static bool signal_sets_initialized = false);
388 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
389
390 bool os::Linux::is_sig_ignored(int sig) {
391 struct sigaction oact;
392 sigaction(sig, (struct sigaction*)NULL, &oact);
393 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
394 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
395 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
396 return true;
397 } else {
398 return false;
399 }
400 }
401
402 void os::Linux::signal_sets_init() {
403 // Should also have an assertion stating we are still single-threaded.
404 assert(!signal_sets_initialized, "Already initialized");
405 // Fill in signals that are necessarily unblocked for all threads in
406 // the VM. Currently, we unblock the following signals:
407 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
408 // by -Xrs (=ReduceSignalUsage));
409 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
410 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
411 // the dispositions or masks wrt these signals.
412 // Programs embedding the VM that want to use the above signals for their
413 // own purposes must, at this time, use the "-Xrs" option to prevent
414 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
415 // (See bug 4345157, and other related bugs).
416 // In reality, though, unblocking these signals is really a nop, since
417 // these signals are not blocked by default.
418 sigemptyset(&unblocked_sigs);
419 sigemptyset(&allowdebug_blocked_sigs);
420 sigaddset(&unblocked_sigs, SIGILL);
421 sigaddset(&unblocked_sigs, SIGSEGV);
422 sigaddset(&unblocked_sigs, SIGBUS);
423 sigaddset(&unblocked_sigs, SIGFPE);
424 #if defined(PPC64)
425 sigaddset(&unblocked_sigs, SIGTRAP);
426 #endif
427 sigaddset(&unblocked_sigs, SR_signum);
428
429 if (!ReduceSignalUsage) {
430 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
431 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
432 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
433 }
434 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
435 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
436 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
437 }
438 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
439 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
440 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
441 }
442 }
443 // Fill in signals that are blocked by all but the VM thread.
444 sigemptyset(&vm_sigs);
445 if (!ReduceSignalUsage) {
446 sigaddset(&vm_sigs, BREAK_SIGNAL);
447 }
448 debug_only(signal_sets_initialized = true);
449
450 }
451
452 // These are signals that are unblocked while a thread is running Java.
453 // (For some reason, they get blocked by default.)
454 sigset_t* os::Linux::unblocked_signals() {
455 assert(signal_sets_initialized, "Not initialized");
456 return &unblocked_sigs;
457 }
458
459 // These are the signals that are blocked while a (non-VM) thread is
460 // running Java. Only the VM thread handles these signals.
461 sigset_t* os::Linux::vm_signals() {
462 assert(signal_sets_initialized, "Not initialized");
463 return &vm_sigs;
464 }
465
466 // These are signals that are blocked during cond_wait to allow debugger in
467 sigset_t* os::Linux::allowdebug_blocked_signals() {
468 assert(signal_sets_initialized, "Not initialized");
469 return &allowdebug_blocked_sigs;
470 }
471
472 void os::Linux::hotspot_sigmask(Thread* thread) {
473
474 //Save caller's signal mask before setting VM signal mask
475 sigset_t caller_sigmask;
476 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
477
478 OSThread* osthread = thread->osthread();
479 osthread->set_caller_sigmask(caller_sigmask);
480
481 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
482
483 if (!ReduceSignalUsage) {
484 if (thread->is_VM_thread()) {
485 // Only the VM thread handles BREAK_SIGNAL ...
486 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
487 } else {
488 // ... all other threads block BREAK_SIGNAL
489 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
490 }
491 }
492 }
493
494 //////////////////////////////////////////////////////////////////////////////
495 // detecting pthread library
496
497 void os::Linux::libpthread_init() {
498 // Save glibc and pthread version strings.
499 #if !defined(_CS_GNU_LIBC_VERSION) || \
500 !defined(_CS_GNU_LIBPTHREAD_VERSION)
501 #error "glibc too old (< 2.3.2)"
502 #endif
503
504 size_t n;
505
506 n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
507 if (n > 0) {
508 char* str = (char *)malloc(n, mtInternal);
509 confstr(_CS_GNU_LIBC_VERSION, str, n);
510 os::Linux::set_glibc_version(str);
511 }
512
513 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
514 if (n > 0) {
515 char* str = (char *)malloc(n, mtInternal);
516 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
517 os::Linux::set_libpthread_version(str);
518 }
519 }
520
521 /////////////////////////////////////////////////////////////////////////////
522 // thread stack expansion
523
524 // os::Linux::manually_expand_stack() takes care of expanding the thread
525 // stack. Note that this is normally not needed: pthread stacks allocate
526 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
527 // committed. Therefore it is not necessary to expand the stack manually.
528 //
529 // Manually expanding the stack was historically needed on LinuxThreads
530 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
531 // it is kept to deal with very rare corner cases:
532 //
533 // For one, user may run the VM on an own implementation of threads
534 // whose stacks are - like the old LinuxThreads - implemented using
535 // mmap(MAP_GROWSDOWN).
536 //
537 // Also, this coding may be needed if the VM is running on the primordial
538 // thread. Normally we avoid running on the primordial thread; however,
539 // user may still invoke the VM on the primordial thread.
540 //
541 // The following historical comment describes the details about running
542 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
543
544
545 // Force Linux kernel to expand current thread stack. If "bottom" is close
546 // to the stack guard, caller should block all signals.
547 //
548 // MAP_GROWSDOWN:
549 // A special mmap() flag that is used to implement thread stacks. It tells
550 // kernel that the memory region should extend downwards when needed. This
551 // allows early versions of LinuxThreads to only mmap the first few pages
552 // when creating a new thread. Linux kernel will automatically expand thread
553 // stack as needed (on page faults).
554 //
555 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
556 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
557 // region, it's hard to tell if the fault is due to a legitimate stack
558 // access or because of reading/writing non-exist memory (e.g. buffer
559 // overrun). As a rule, if the fault happens below current stack pointer,
560 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
561 // application (see Linux kernel fault.c).
562 //
563 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
564 // stack overflow detection.
565 //
566 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
567 // not use MAP_GROWSDOWN.
568 //
569 // To get around the problem and allow stack banging on Linux, we need to
570 // manually expand thread stack after receiving the SIGSEGV.
571 //
572 // There are two ways to expand thread stack to address "bottom", we used
573 // both of them in JVM before 1.5:
574 // 1. adjust stack pointer first so that it is below "bottom", and then
575 // touch "bottom"
576 // 2. mmap() the page in question
577 //
578 // Now alternate signal stack is gone, it's harder to use 2. For instance,
579 // if current sp is already near the lower end of page 101, and we need to
580 // call mmap() to map page 100, it is possible that part of the mmap() frame
581 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
582 // That will destroy the mmap() frame and cause VM to crash.
583 //
584 // The following code works by adjusting sp first, then accessing the "bottom"
585 // page to force a page fault. Linux kernel will then automatically expand the
586 // stack mapping.
587 //
588 // _expand_stack_to() assumes its frame size is less than page size, which
589 // should always be true if the function is not inlined.
590
591 static void NOINLINE _expand_stack_to(address bottom) {
592 address sp;
593 size_t size;
594 volatile char *p;
595
596 // Adjust bottom to point to the largest address within the same page, it
597 // gives us a one-page buffer if alloca() allocates slightly more memory.
598 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
599 bottom += os::Linux::page_size() - 1;
600
601 // sp might be slightly above current stack pointer; if that's the case, we
602 // will alloca() a little more space than necessary, which is OK. Don't use
603 // os::current_stack_pointer(), as its result can be slightly below current
604 // stack pointer, causing us to not alloca enough to reach "bottom".
605 sp = (address)&sp;
606
607 if (sp > bottom) {
608 size = sp - bottom;
609 p = (volatile char *)alloca(size);
610 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
611 p[0] = '\0';
612 }
613 }
614
615 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
616 assert(t!=NULL, "just checking");
617 assert(t->osthread()->expanding_stack(), "expand should be set");
618 assert(t->stack_base() != NULL, "stack_base was not initialized");
619
620 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
621 sigset_t mask_all, old_sigset;
622 sigfillset(&mask_all);
623 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
624 _expand_stack_to(addr);
625 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
626 return true;
627 }
628 return false;
629 }
630
631 //////////////////////////////////////////////////////////////////////////////
632 // create new thread
633
634 // Thread start routine for all newly created threads
635 static void *thread_native_entry(Thread *thread) {
636 // Try to randomize the cache line index of hot stack frames.
637 // This helps when threads of the same stack traces evict each other's
638 // cache lines. The threads can be either from the same JVM instance, or
639 // from different JVM instances. The benefit is especially true for
640 // processors with hyperthreading technology.
641 static int counter = 0;
642 int pid = os::current_process_id();
643 alloca(((pid ^ counter++) & 7) * 128);
644
645 thread->initialize_thread_current();
646
647 OSThread* osthread = thread->osthread();
648 Monitor* sync = osthread->startThread_lock();
649
650 osthread->set_thread_id(os::current_thread_id());
651
652 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
653 os::current_thread_id(), (uintx) pthread_self());
654
655 if (UseNUMA) {
656 int lgrp_id = os::numa_get_group_id();
657 if (lgrp_id != -1) {
658 thread->set_lgrp_id(lgrp_id);
659 }
660 }
661 // initialize signal mask for this thread
662 os::Linux::hotspot_sigmask(thread);
663
664 // initialize floating point control register
665 os::Linux::init_thread_fpu_state();
666
667 // handshaking with parent thread
668 {
669 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
670
671 // notify parent thread
672 osthread->set_state(INITIALIZED);
673 sync->notify_all();
674
675 // wait until os::start_thread()
676 while (osthread->get_state() == INITIALIZED) {
677 sync->wait(Mutex::_no_safepoint_check_flag);
678 }
679 }
680
681 // call one more level start routine
682 thread->run();
683
684 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
685 os::current_thread_id(), (uintx) pthread_self());
686
687 // If a thread has not deleted itself ("delete this") as part of its
688 // termination sequence, we have to ensure thread-local-storage is
689 // cleared before we actually terminate. No threads should ever be
690 // deleted asynchronously with respect to their termination.
691 if (Thread::current_or_null_safe() != NULL) {
692 assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
693 thread->clear_thread_current();
694 }
695
696 return 0;
697 }
698
699 bool os::create_thread(Thread* thread, ThreadType thr_type,
700 size_t req_stack_size) {
701 assert(thread->osthread() == NULL, "caller responsible");
702
703 // Allocate the OSThread object
704 OSThread* osthread = new OSThread(NULL, NULL);
705 if (osthread == NULL) {
706 return false;
707 }
708
709 // set the correct thread state
710 osthread->set_thread_type(thr_type);
711
712 // Initial state is ALLOCATED but not INITIALIZED
713 osthread->set_state(ALLOCATED);
714
715 thread->set_osthread(osthread);
716
717 // init thread attributes
718 pthread_attr_t attr;
719 pthread_attr_init(&attr);
720 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
721
722 // Calculate stack size if it's not specified by caller.
723 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
724 // In the Linux NPTL pthread implementation the guard size mechanism
725 // is not implemented properly. The posix standard requires adding
726 // the size of the guard pages to the stack size, instead Linux
727 // takes the space out of 'stacksize'. Thus we adapt the requested
728 // stack_size by the size of the guard pages to mimick proper
729 // behaviour.
730 stack_size = align_size_up(stack_size + os::Linux::default_guard_size(thr_type), vm_page_size());
731 pthread_attr_setstacksize(&attr, stack_size);
732
733 // Configure glibc guard page.
734 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
735
736 ThreadState state;
737
738 {
739 pthread_t tid;
740 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
741
742 char buf[64];
743 if (ret == 0) {
744 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
745 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
746 } else {
747 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
748 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
749 }
750
751 pthread_attr_destroy(&attr);
752
753 if (ret != 0) {
754 // Need to clean up stuff we've allocated so far
755 thread->set_osthread(NULL);
756 delete osthread;
757 return false;
758 }
759
760 // Store pthread info into the OSThread
761 osthread->set_pthread_id(tid);
762
763 // Wait until child thread is either initialized or aborted
764 {
765 Monitor* sync_with_child = osthread->startThread_lock();
766 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
767 while ((state = osthread->get_state()) == ALLOCATED) {
768 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
769 }
770 }
771 }
772
773 // Aborted due to thread limit being reached
774 if (state == ZOMBIE) {
775 thread->set_osthread(NULL);
776 delete osthread;
777 return false;
778 }
779
780 // The thread is returned suspended (in state INITIALIZED),
781 // and is started higher up in the call chain
782 assert(state == INITIALIZED, "race condition");
783 return true;
784 }
785
786 /////////////////////////////////////////////////////////////////////////////
787 // attach existing thread
788
789 // bootstrap the main thread
790 bool os::create_main_thread(JavaThread* thread) {
791 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
792 return create_attached_thread(thread);
793 }
794
795 bool os::create_attached_thread(JavaThread* thread) {
796 #ifdef ASSERT
797 thread->verify_not_published();
798 #endif
799
800 // Allocate the OSThread object
801 OSThread* osthread = new OSThread(NULL, NULL);
802
803 if (osthread == NULL) {
804 return false;
805 }
806
807 // Store pthread info into the OSThread
808 osthread->set_thread_id(os::Linux::gettid());
809 osthread->set_pthread_id(::pthread_self());
810
811 // initialize floating point control register
812 os::Linux::init_thread_fpu_state();
813
814 // Initial thread state is RUNNABLE
815 osthread->set_state(RUNNABLE);
816
817 thread->set_osthread(osthread);
818
819 if (UseNUMA) {
820 int lgrp_id = os::numa_get_group_id();
821 if (lgrp_id != -1) {
822 thread->set_lgrp_id(lgrp_id);
823 }
824 }
825
826 if (os::Linux::is_initial_thread()) {
827 // If current thread is initial thread, its stack is mapped on demand,
828 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
829 // the entire stack region to avoid SEGV in stack banging.
830 // It is also useful to get around the heap-stack-gap problem on SuSE
831 // kernel (see 4821821 for details). We first expand stack to the top
832 // of yellow zone, then enable stack yellow zone (order is significant,
833 // enabling yellow zone first will crash JVM on SuSE Linux), so there
834 // is no gap between the last two virtual memory regions.
835
836 JavaThread *jt = (JavaThread *)thread;
837 address addr = jt->stack_reserved_zone_base();
838 assert(addr != NULL, "initialization problem?");
839 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
840
841 osthread->set_expanding_stack();
842 os::Linux::manually_expand_stack(jt, addr);
843 osthread->clear_expanding_stack();
844 }
845
846 // initialize signal mask for this thread
847 // and save the caller's signal mask
848 os::Linux::hotspot_sigmask(thread);
849
850 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
851 os::current_thread_id(), (uintx) pthread_self());
852
853 return true;
854 }
855
856 void os::pd_start_thread(Thread* thread) {
857 OSThread * osthread = thread->osthread();
858 assert(osthread->get_state() != INITIALIZED, "just checking");
859 Monitor* sync_with_child = osthread->startThread_lock();
860 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
861 sync_with_child->notify();
862 }
863
864 // Free Linux resources related to the OSThread
865 void os::free_thread(OSThread* osthread) {
866 assert(osthread != NULL, "osthread not set");
867
868 // We are told to free resources of the argument thread,
869 // but we can only really operate on the current thread.
870 assert(Thread::current()->osthread() == osthread,
871 "os::free_thread but not current thread");
872
873 #ifdef ASSERT
874 sigset_t current;
875 sigemptyset(¤t);
876 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
877 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
878 #endif
879
880 // Restore caller's signal mask
881 sigset_t sigmask = osthread->caller_sigmask();
882 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
883
884 delete osthread;
885 }
886
887 //////////////////////////////////////////////////////////////////////////////
888 // initial thread
889
890 // Check if current thread is the initial thread, similar to Solaris thr_main.
891 bool os::Linux::is_initial_thread(void) {
892 char dummy;
893 // If called before init complete, thread stack bottom will be null.
894 // Can be called if fatal error occurs before initialization.
895 if (initial_thread_stack_bottom() == NULL) return false;
896 assert(initial_thread_stack_bottom() != NULL &&
897 initial_thread_stack_size() != 0,
898 "os::init did not locate initial thread's stack region");
899 if ((address)&dummy >= initial_thread_stack_bottom() &&
900 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
901 return true;
902 } else {
903 return false;
904 }
905 }
906
907 // Find the virtual memory area that contains addr
908 static bool find_vma(address addr, address* vma_low, address* vma_high) {
909 FILE *fp = fopen("/proc/self/maps", "r");
910 if (fp) {
911 address low, high;
912 while (!feof(fp)) {
913 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
914 if (low <= addr && addr < high) {
915 if (vma_low) *vma_low = low;
916 if (vma_high) *vma_high = high;
917 fclose(fp);
918 return true;
919 }
920 }
921 for (;;) {
922 int ch = fgetc(fp);
923 if (ch == EOF || ch == (int)'\n') break;
924 }
925 }
926 fclose(fp);
927 }
928 return false;
929 }
930
931 // Locate initial thread stack. This special handling of initial thread stack
932 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
933 // bogus value for the primordial process thread. While the launcher has created
934 // the VM in a new thread since JDK 6, we still have to allow for the use of the
935 // JNI invocation API from a primordial thread.
936 void os::Linux::capture_initial_stack(size_t max_size) {
937
938 // max_size is either 0 (which means accept OS default for thread stacks) or
939 // a user-specified value known to be at least the minimum needed. If we
940 // are actually on the primordial thread we can make it appear that we have a
941 // smaller max_size stack by inserting the guard pages at that location. But we
942 // cannot do anything to emulate a larger stack than what has been provided by
943 // the OS or threading library. In fact if we try to use a stack greater than
944 // what is set by rlimit then we will crash the hosting process.
945
946 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
947 // If this is "unlimited" then it will be a huge value.
948 struct rlimit rlim;
949 getrlimit(RLIMIT_STACK, &rlim);
950 size_t stack_size = rlim.rlim_cur;
951
952 // 6308388: a bug in ld.so will relocate its own .data section to the
953 // lower end of primordial stack; reduce ulimit -s value a little bit
954 // so we won't install guard page on ld.so's data section.
955 stack_size -= 2 * page_size();
956
957 // Try to figure out where the stack base (top) is. This is harder.
958 //
959 // When an application is started, glibc saves the initial stack pointer in
960 // a global variable "__libc_stack_end", which is then used by system
961 // libraries. __libc_stack_end should be pretty close to stack top. The
962 // variable is available since the very early days. However, because it is
963 // a private interface, it could disappear in the future.
964 //
965 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
966 // to __libc_stack_end, it is very close to stack top, but isn't the real
967 // stack top. Note that /proc may not exist if VM is running as a chroot
968 // program, so reading /proc/<pid>/stat could fail. Also the contents of
969 // /proc/<pid>/stat could change in the future (though unlikely).
970 //
971 // We try __libc_stack_end first. If that doesn't work, look for
972 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
973 // as a hint, which should work well in most cases.
974
975 uintptr_t stack_start;
976
977 // try __libc_stack_end first
978 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
979 if (p && *p) {
980 stack_start = *p;
981 } else {
982 // see if we can get the start_stack field from /proc/self/stat
983 FILE *fp;
984 int pid;
985 char state;
986 int ppid;
987 int pgrp;
988 int session;
989 int nr;
990 int tpgrp;
991 unsigned long flags;
992 unsigned long minflt;
993 unsigned long cminflt;
994 unsigned long majflt;
995 unsigned long cmajflt;
996 unsigned long utime;
997 unsigned long stime;
998 long cutime;
999 long cstime;
1000 long prio;
1001 long nice;
1002 long junk;
1003 long it_real;
1004 uintptr_t start;
1005 uintptr_t vsize;
1006 intptr_t rss;
1007 uintptr_t rsslim;
1008 uintptr_t scodes;
1009 uintptr_t ecode;
1010 int i;
1011
1012 // Figure what the primordial thread stack base is. Code is inspired
1013 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1014 // followed by command name surrounded by parentheses, state, etc.
1015 char stat[2048];
1016 int statlen;
1017
1018 fp = fopen("/proc/self/stat", "r");
1019 if (fp) {
1020 statlen = fread(stat, 1, 2047, fp);
1021 stat[statlen] = '\0';
1022 fclose(fp);
1023
1024 // Skip pid and the command string. Note that we could be dealing with
1025 // weird command names, e.g. user could decide to rename java launcher
1026 // to "java 1.4.2 :)", then the stat file would look like
1027 // 1234 (java 1.4.2 :)) R ... ...
1028 // We don't really need to know the command string, just find the last
1029 // occurrence of ")" and then start parsing from there. See bug 4726580.
1030 char * s = strrchr(stat, ')');
1031
1032 i = 0;
1033 if (s) {
1034 // Skip blank chars
1035 do { s++; } while (s && isspace(*s));
1036
1037 #define _UFM UINTX_FORMAT
1038 #define _DFM INTX_FORMAT
1039
1040 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1041 // 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
1042 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,
1043 &state, // 3 %c
1044 &ppid, // 4 %d
1045 &pgrp, // 5 %d
1046 &session, // 6 %d
1047 &nr, // 7 %d
1048 &tpgrp, // 8 %d
1049 &flags, // 9 %lu
1050 &minflt, // 10 %lu
1051 &cminflt, // 11 %lu
1052 &majflt, // 12 %lu
1053 &cmajflt, // 13 %lu
1054 &utime, // 14 %lu
1055 &stime, // 15 %lu
1056 &cutime, // 16 %ld
1057 &cstime, // 17 %ld
1058 &prio, // 18 %ld
1059 &nice, // 19 %ld
1060 &junk, // 20 %ld
1061 &it_real, // 21 %ld
1062 &start, // 22 UINTX_FORMAT
1063 &vsize, // 23 UINTX_FORMAT
1064 &rss, // 24 INTX_FORMAT
1065 &rsslim, // 25 UINTX_FORMAT
1066 &scodes, // 26 UINTX_FORMAT
1067 &ecode, // 27 UINTX_FORMAT
1068 &stack_start); // 28 UINTX_FORMAT
1069 }
1070
1071 #undef _UFM
1072 #undef _DFM
1073
1074 if (i != 28 - 2) {
1075 assert(false, "Bad conversion from /proc/self/stat");
1076 // product mode - assume we are the initial thread, good luck in the
1077 // embedded case.
1078 warning("Can't detect initial thread stack location - bad conversion");
1079 stack_start = (uintptr_t) &rlim;
1080 }
1081 } else {
1082 // For some reason we can't open /proc/self/stat (for example, running on
1083 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1084 // most cases, so don't abort:
1085 warning("Can't detect initial thread stack location - no /proc/self/stat");
1086 stack_start = (uintptr_t) &rlim;
1087 }
1088 }
1089
1090 // Now we have a pointer (stack_start) very close to the stack top, the
1091 // next thing to do is to figure out the exact location of stack top. We
1092 // can find out the virtual memory area that contains stack_start by
1093 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1094 // and its upper limit is the real stack top. (again, this would fail if
1095 // running inside chroot, because /proc may not exist.)
1096
1097 uintptr_t stack_top;
1098 address low, high;
1099 if (find_vma((address)stack_start, &low, &high)) {
1100 // success, "high" is the true stack top. (ignore "low", because initial
1101 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1102 stack_top = (uintptr_t)high;
1103 } else {
1104 // failed, likely because /proc/self/maps does not exist
1105 warning("Can't detect initial thread stack location - find_vma failed");
1106 // best effort: stack_start is normally within a few pages below the real
1107 // stack top, use it as stack top, and reduce stack size so we won't put
1108 // guard page outside stack.
1109 stack_top = stack_start;
1110 stack_size -= 16 * page_size();
1111 }
1112
1113 // stack_top could be partially down the page so align it
1114 stack_top = align_size_up(stack_top, page_size());
1115
1116 // Allowed stack value is minimum of max_size and what we derived from rlimit
1117 if (max_size > 0) {
1118 _initial_thread_stack_size = MIN2(max_size, stack_size);
1119 } else {
1120 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1121 // clamp it at 8MB as we do on Solaris
1122 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1123 }
1124 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1125 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1126
1127 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1128
1129 if (log_is_enabled(Info, os, thread)) {
1130 // See if we seem to be on primordial process thread
1131 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1132 uintptr_t(&rlim) < stack_top;
1133
1134 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1135 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1136 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1137 stack_top, intptr_t(_initial_thread_stack_bottom));
1138 }
1139 }
1140
1141 ////////////////////////////////////////////////////////////////////////////////
1142 // time support
1143
1144 // Time since start-up in seconds to a fine granularity.
1145 // Used by VMSelfDestructTimer and the MemProfiler.
1146 double os::elapsedTime() {
1147
1148 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1149 }
1150
1151 jlong os::elapsed_counter() {
1152 return javaTimeNanos() - initial_time_count;
1153 }
1154
1155 jlong os::elapsed_frequency() {
1156 return NANOSECS_PER_SEC; // nanosecond resolution
1157 }
1158
1159 bool os::supports_vtime() { return true; }
1160 bool os::enable_vtime() { return false; }
1161 bool os::vtime_enabled() { return false; }
1162
1163 double os::elapsedVTime() {
1164 struct rusage usage;
1165 int retval = getrusage(RUSAGE_THREAD, &usage);
1166 if (retval == 0) {
1167 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);
1168 } else {
1169 // better than nothing, but not much
1170 return elapsedTime();
1171 }
1172 }
1173
1174 jlong os::javaTimeMillis() {
1175 timeval time;
1176 int status = gettimeofday(&time, NULL);
1177 assert(status != -1, "linux error");
1178 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1179 }
1180
1181 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1182 timeval time;
1183 int status = gettimeofday(&time, NULL);
1184 assert(status != -1, "linux error");
1185 seconds = jlong(time.tv_sec);
1186 nanos = jlong(time.tv_usec) * 1000;
1187 }
1188
1189
1190 #ifndef CLOCK_MONOTONIC
1191 #define CLOCK_MONOTONIC (1)
1192 #endif
1193
1194 void os::Linux::clock_init() {
1195 // we do dlopen's in this particular order due to bug in linux
1196 // dynamical loader (see 6348968) leading to crash on exit
1197 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1198 if (handle == NULL) {
1199 handle = dlopen("librt.so", RTLD_LAZY);
1200 }
1201
1202 if (handle) {
1203 int (*clock_getres_func)(clockid_t, struct timespec*) =
1204 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1205 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1206 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1207 if (clock_getres_func && clock_gettime_func) {
1208 // See if monotonic clock is supported by the kernel. Note that some
1209 // early implementations simply return kernel jiffies (updated every
1210 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1211 // for nano time (though the monotonic property is still nice to have).
1212 // It's fixed in newer kernels, however clock_getres() still returns
1213 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1214 // resolution for now. Hopefully as people move to new kernels, this
1215 // won't be a problem.
1216 struct timespec res;
1217 struct timespec tp;
1218 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1219 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1220 // yes, monotonic clock is supported
1221 _clock_gettime = clock_gettime_func;
1222 return;
1223 } else {
1224 // close librt if there is no monotonic clock
1225 dlclose(handle);
1226 }
1227 }
1228 }
1229 warning("No monotonic clock was available - timed services may " \
1230 "be adversely affected if the time-of-day clock changes");
1231 }
1232
1233 #ifndef SYS_clock_getres
1234 #if defined(X86) || defined(PPC64) || defined(S390)
1235 #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261)
1236 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1237 #else
1238 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1239 #define sys_clock_getres(x,y) -1
1240 #endif
1241 #else
1242 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1243 #endif
1244
1245 void os::Linux::fast_thread_clock_init() {
1246 if (!UseLinuxPosixThreadCPUClocks) {
1247 return;
1248 }
1249 clockid_t clockid;
1250 struct timespec tp;
1251 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1252 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1253
1254 // Switch to using fast clocks for thread cpu time if
1255 // the sys_clock_getres() returns 0 error code.
1256 // Note, that some kernels may support the current thread
1257 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1258 // returned by the pthread_getcpuclockid().
1259 // If the fast Posix clocks are supported then the sys_clock_getres()
1260 // must return at least tp.tv_sec == 0 which means a resolution
1261 // better than 1 sec. This is extra check for reliability.
1262
1263 if (pthread_getcpuclockid_func &&
1264 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1265 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1266 _supports_fast_thread_cpu_time = true;
1267 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1268 }
1269 }
1270
1271 jlong os::javaTimeNanos() {
1272 if (os::supports_monotonic_clock()) {
1273 struct timespec tp;
1274 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1275 assert(status == 0, "gettime error");
1276 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1277 return result;
1278 } else {
1279 timeval time;
1280 int status = gettimeofday(&time, NULL);
1281 assert(status != -1, "linux error");
1282 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1283 return 1000 * usecs;
1284 }
1285 }
1286
1287 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1288 if (os::supports_monotonic_clock()) {
1289 info_ptr->max_value = ALL_64_BITS;
1290
1291 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1292 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1293 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1294 } else {
1295 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1296 info_ptr->max_value = ALL_64_BITS;
1297
1298 // gettimeofday is a real time clock so it skips
1299 info_ptr->may_skip_backward = true;
1300 info_ptr->may_skip_forward = true;
1301 }
1302
1303 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1304 }
1305
1306 // Return the real, user, and system times in seconds from an
1307 // arbitrary fixed point in the past.
1308 bool os::getTimesSecs(double* process_real_time,
1309 double* process_user_time,
1310 double* process_system_time) {
1311 struct tms ticks;
1312 clock_t real_ticks = times(&ticks);
1313
1314 if (real_ticks == (clock_t) (-1)) {
1315 return false;
1316 } else {
1317 double ticks_per_second = (double) clock_tics_per_sec;
1318 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1319 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1320 *process_real_time = ((double) real_ticks) / ticks_per_second;
1321
1322 return true;
1323 }
1324 }
1325
1326
1327 char * os::local_time_string(char *buf, size_t buflen) {
1328 struct tm t;
1329 time_t long_time;
1330 time(&long_time);
1331 localtime_r(&long_time, &t);
1332 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1333 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1334 t.tm_hour, t.tm_min, t.tm_sec);
1335 return buf;
1336 }
1337
1338 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1339 return localtime_r(clock, res);
1340 }
1341
1342 ////////////////////////////////////////////////////////////////////////////////
1343 // runtime exit support
1344
1345 // Note: os::shutdown() might be called very early during initialization, or
1346 // called from signal handler. Before adding something to os::shutdown(), make
1347 // sure it is async-safe and can handle partially initialized VM.
1348 void os::shutdown() {
1349
1350 // allow PerfMemory to attempt cleanup of any persistent resources
1351 perfMemory_exit();
1352
1353 // needs to remove object in file system
1354 AttachListener::abort();
1355
1356 // flush buffered output, finish log files
1357 ostream_abort();
1358
1359 // Check for abort hook
1360 abort_hook_t abort_hook = Arguments::abort_hook();
1361 if (abort_hook != NULL) {
1362 abort_hook();
1363 }
1364
1365 }
1366
1367 // Note: os::abort() might be called very early during initialization, or
1368 // called from signal handler. Before adding something to os::abort(), make
1369 // sure it is async-safe and can handle partially initialized VM.
1370 void os::abort(bool dump_core, void* siginfo, const void* context) {
1371 os::shutdown();
1372 if (dump_core) {
1373 #ifndef PRODUCT
1374 fdStream out(defaultStream::output_fd());
1375 out.print_raw("Current thread is ");
1376 char buf[16];
1377 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1378 out.print_raw_cr(buf);
1379 out.print_raw_cr("Dumping core ...");
1380 #endif
1381 ::abort(); // dump core
1382 }
1383
1384 ::exit(1);
1385 }
1386
1387 // Die immediately, no exit hook, no abort hook, no cleanup.
1388 void os::die() {
1389 ::abort();
1390 }
1391
1392
1393 // This method is a copy of JDK's sysGetLastErrorString
1394 // from src/solaris/hpi/src/system_md.c
1395
1396 size_t os::lasterror(char *buf, size_t len) {
1397 if (errno == 0) return 0;
1398
1399 const char *s = os::strerror(errno);
1400 size_t n = ::strlen(s);
1401 if (n >= len) {
1402 n = len - 1;
1403 }
1404 ::strncpy(buf, s, n);
1405 buf[n] = '\0';
1406 return n;
1407 }
1408
1409 // thread_id is kernel thread id (similar to Solaris LWP id)
1410 intx os::current_thread_id() { return os::Linux::gettid(); }
1411 int os::current_process_id() {
1412 return ::getpid();
1413 }
1414
1415 // DLL functions
1416
1417 const char* os::dll_file_extension() { return ".so"; }
1418
1419 // This must be hard coded because it's the system's temporary
1420 // directory not the java application's temp directory, ala java.io.tmpdir.
1421 const char* os::get_temp_directory() { return "/tmp"; }
1422
1423 static bool file_exists(const char* filename) {
1424 struct stat statbuf;
1425 if (filename == NULL || strlen(filename) == 0) {
1426 return false;
1427 }
1428 return os::stat(filename, &statbuf) == 0;
1429 }
1430
1431 bool os::dll_build_name(char* buffer, size_t buflen,
1432 const char* pname, const char* fname) {
1433 bool retval = false;
1434 // Copied from libhpi
1435 const size_t pnamelen = pname ? strlen(pname) : 0;
1436
1437 // Return error on buffer overflow.
1438 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1439 return retval;
1440 }
1441
1442 if (pnamelen == 0) {
1443 snprintf(buffer, buflen, "lib%s.so", fname);
1444 retval = true;
1445 } else if (strchr(pname, *os::path_separator()) != NULL) {
1446 int n;
1447 char** pelements = split_path(pname, &n);
1448 if (pelements == NULL) {
1449 return false;
1450 }
1451 for (int i = 0; i < n; i++) {
1452 // Really shouldn't be NULL, but check can't hurt
1453 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1454 continue; // skip the empty path values
1455 }
1456 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1457 if (file_exists(buffer)) {
1458 retval = true;
1459 break;
1460 }
1461 }
1462 // release the storage
1463 for (int i = 0; i < n; i++) {
1464 if (pelements[i] != NULL) {
1465 FREE_C_HEAP_ARRAY(char, pelements[i]);
1466 }
1467 }
1468 if (pelements != NULL) {
1469 FREE_C_HEAP_ARRAY(char*, pelements);
1470 }
1471 } else {
1472 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1473 retval = true;
1474 }
1475 return retval;
1476 }
1477
1478 // check if addr is inside libjvm.so
1479 bool os::address_is_in_vm(address addr) {
1480 static address libjvm_base_addr;
1481 Dl_info dlinfo;
1482
1483 if (libjvm_base_addr == NULL) {
1484 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1485 libjvm_base_addr = (address)dlinfo.dli_fbase;
1486 }
1487 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1488 }
1489
1490 if (dladdr((void *)addr, &dlinfo) != 0) {
1491 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1492 }
1493
1494 return false;
1495 }
1496
1497 bool os::dll_address_to_function_name(address addr, char *buf,
1498 int buflen, int *offset,
1499 bool demangle) {
1500 // buf is not optional, but offset is optional
1501 assert(buf != NULL, "sanity check");
1502
1503 Dl_info dlinfo;
1504
1505 if (dladdr((void*)addr, &dlinfo) != 0) {
1506 // see if we have a matching symbol
1507 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1508 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1509 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1510 }
1511 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1512 return true;
1513 }
1514 // no matching symbol so try for just file info
1515 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1516 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1517 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1518 return true;
1519 }
1520 }
1521 }
1522
1523 buf[0] = '\0';
1524 if (offset != NULL) *offset = -1;
1525 return false;
1526 }
1527
1528 struct _address_to_library_name {
1529 address addr; // input : memory address
1530 size_t buflen; // size of fname
1531 char* fname; // output: library name
1532 address base; // library base addr
1533 };
1534
1535 static int address_to_library_name_callback(struct dl_phdr_info *info,
1536 size_t size, void *data) {
1537 int i;
1538 bool found = false;
1539 address libbase = NULL;
1540 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1541
1542 // iterate through all loadable segments
1543 for (i = 0; i < info->dlpi_phnum; i++) {
1544 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1545 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1546 // base address of a library is the lowest address of its loaded
1547 // segments.
1548 if (libbase == NULL || libbase > segbase) {
1549 libbase = segbase;
1550 }
1551 // see if 'addr' is within current segment
1552 if (segbase <= d->addr &&
1553 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1554 found = true;
1555 }
1556 }
1557 }
1558
1559 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1560 // so dll_address_to_library_name() can fall through to use dladdr() which
1561 // can figure out executable name from argv[0].
1562 if (found && info->dlpi_name && info->dlpi_name[0]) {
1563 d->base = libbase;
1564 if (d->fname) {
1565 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1566 }
1567 return 1;
1568 }
1569 return 0;
1570 }
1571
1572 bool os::dll_address_to_library_name(address addr, char* buf,
1573 int buflen, int* offset) {
1574 // buf is not optional, but offset is optional
1575 assert(buf != NULL, "sanity check");
1576
1577 Dl_info dlinfo;
1578 struct _address_to_library_name data;
1579
1580 // There is a bug in old glibc dladdr() implementation that it could resolve
1581 // to wrong library name if the .so file has a base address != NULL. Here
1582 // we iterate through the program headers of all loaded libraries to find
1583 // out which library 'addr' really belongs to. This workaround can be
1584 // removed once the minimum requirement for glibc is moved to 2.3.x.
1585 data.addr = addr;
1586 data.fname = buf;
1587 data.buflen = buflen;
1588 data.base = NULL;
1589 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1590
1591 if (rslt) {
1592 // buf already contains library name
1593 if (offset) *offset = addr - data.base;
1594 return true;
1595 }
1596 if (dladdr((void*)addr, &dlinfo) != 0) {
1597 if (dlinfo.dli_fname != NULL) {
1598 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1599 }
1600 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1601 *offset = addr - (address)dlinfo.dli_fbase;
1602 }
1603 return true;
1604 }
1605
1606 buf[0] = '\0';
1607 if (offset) *offset = -1;
1608 return false;
1609 }
1610
1611 // Loads .dll/.so and
1612 // in case of error it checks if .dll/.so was built for the
1613 // same architecture as Hotspot is running on
1614
1615
1616 // Remember the stack's state. The Linux dynamic linker will change
1617 // the stack to 'executable' at most once, so we must safepoint only once.
1618 bool os::Linux::_stack_is_executable = false;
1619
1620 // VM operation that loads a library. This is necessary if stack protection
1621 // of the Java stacks can be lost during loading the library. If we
1622 // do not stop the Java threads, they can stack overflow before the stacks
1623 // are protected again.
1624 class VM_LinuxDllLoad: public VM_Operation {
1625 private:
1626 const char *_filename;
1627 char *_ebuf;
1628 int _ebuflen;
1629 void *_lib;
1630 public:
1631 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1632 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1633 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1634 void doit() {
1635 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1636 os::Linux::_stack_is_executable = true;
1637 }
1638 void* loaded_library() { return _lib; }
1639 };
1640
1641 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1642 void * result = NULL;
1643 bool load_attempted = false;
1644
1645 // Check whether the library to load might change execution rights
1646 // of the stack. If they are changed, the protection of the stack
1647 // guard pages will be lost. We need a safepoint to fix this.
1648 //
1649 // See Linux man page execstack(8) for more info.
1650 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1651 ElfFile ef(filename);
1652 if (!ef.specifies_noexecstack()) {
1653 if (!is_init_completed()) {
1654 os::Linux::_stack_is_executable = true;
1655 // This is OK - No Java threads have been created yet, and hence no
1656 // stack guard pages to fix.
1657 //
1658 // This should happen only when you are building JDK7 using a very
1659 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1660 //
1661 // Dynamic loader will make all stacks executable after
1662 // this function returns, and will not do that again.
1663 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1664 } else {
1665 warning("You have loaded library %s which might have disabled stack guard. "
1666 "The VM will try to fix the stack guard now.\n"
1667 "It's highly recommended that you fix the library with "
1668 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1669 filename);
1670
1671 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1672 JavaThread *jt = JavaThread::current();
1673 if (jt->thread_state() != _thread_in_native) {
1674 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1675 // that requires ExecStack. Cannot enter safe point. Let's give up.
1676 warning("Unable to fix stack guard. Giving up.");
1677 } else {
1678 if (!LoadExecStackDllInVMThread) {
1679 // This is for the case where the DLL has an static
1680 // constructor function that executes JNI code. We cannot
1681 // load such DLLs in the VMThread.
1682 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1683 }
1684
1685 ThreadInVMfromNative tiv(jt);
1686 debug_only(VMNativeEntryWrapper vew;)
1687
1688 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1689 VMThread::execute(&op);
1690 if (LoadExecStackDllInVMThread) {
1691 result = op.loaded_library();
1692 }
1693 load_attempted = true;
1694 }
1695 }
1696 }
1697 }
1698
1699 if (!load_attempted) {
1700 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1701 }
1702
1703 if (result != NULL) {
1704 // Successful loading
1705 return result;
1706 }
1707
1708 Elf32_Ehdr elf_head;
1709 int diag_msg_max_length=ebuflen-strlen(ebuf);
1710 char* diag_msg_buf=ebuf+strlen(ebuf);
1711
1712 if (diag_msg_max_length==0) {
1713 // No more space in ebuf for additional diagnostics message
1714 return NULL;
1715 }
1716
1717
1718 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1719
1720 if (file_descriptor < 0) {
1721 // Can't open library, report dlerror() message
1722 return NULL;
1723 }
1724
1725 bool failed_to_read_elf_head=
1726 (sizeof(elf_head)!=
1727 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1728
1729 ::close(file_descriptor);
1730 if (failed_to_read_elf_head) {
1731 // file i/o error - report dlerror() msg
1732 return NULL;
1733 }
1734
1735 typedef struct {
1736 Elf32_Half code; // Actual value as defined in elf.h
1737 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1738 unsigned char elf_class; // 32 or 64 bit
1739 unsigned char endianess; // MSB or LSB
1740 char* name; // String representation
1741 } arch_t;
1742
1743 #ifndef EM_486
1744 #define EM_486 6 /* Intel 80486 */
1745 #endif
1746 #ifndef EM_AARCH64
1747 #define EM_AARCH64 183 /* ARM AARCH64 */
1748 #endif
1749
1750 static const arch_t arch_array[]={
1751 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1752 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1753 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1754 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1755 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1756 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1757 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1758 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1759 #if defined(VM_LITTLE_ENDIAN)
1760 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1761 #else
1762 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64 LE"},
1763 #endif
1764 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1765 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1766 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1767 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1768 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1769 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1770 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1771 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1772 };
1773
1774 #if (defined IA32)
1775 static Elf32_Half running_arch_code=EM_386;
1776 #elif (defined AMD64)
1777 static Elf32_Half running_arch_code=EM_X86_64;
1778 #elif (defined IA64)
1779 static Elf32_Half running_arch_code=EM_IA_64;
1780 #elif (defined __sparc) && (defined _LP64)
1781 static Elf32_Half running_arch_code=EM_SPARCV9;
1782 #elif (defined __sparc) && (!defined _LP64)
1783 static Elf32_Half running_arch_code=EM_SPARC;
1784 #elif (defined __powerpc64__)
1785 static Elf32_Half running_arch_code=EM_PPC64;
1786 #elif (defined __powerpc__)
1787 static Elf32_Half running_arch_code=EM_PPC;
1788 #elif (defined AARCH64)
1789 static Elf32_Half running_arch_code=EM_AARCH64;
1790 #elif (defined ARM)
1791 static Elf32_Half running_arch_code=EM_ARM;
1792 #elif (defined S390)
1793 static Elf32_Half running_arch_code=EM_S390;
1794 #elif (defined ALPHA)
1795 static Elf32_Half running_arch_code=EM_ALPHA;
1796 #elif (defined MIPSEL)
1797 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1798 #elif (defined PARISC)
1799 static Elf32_Half running_arch_code=EM_PARISC;
1800 #elif (defined MIPS)
1801 static Elf32_Half running_arch_code=EM_MIPS;
1802 #elif (defined M68K)
1803 static Elf32_Half running_arch_code=EM_68K;
1804 #else
1805 #error Method os::dll_load requires that one of following is defined:\
1806 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, __sparc
1807 #endif
1808
1809 // Identify compatability class for VM's architecture and library's architecture
1810 // Obtain string descriptions for architectures
1811
1812 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1813 int running_arch_index=-1;
1814
1815 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1816 if (running_arch_code == arch_array[i].code) {
1817 running_arch_index = i;
1818 }
1819 if (lib_arch.code == arch_array[i].code) {
1820 lib_arch.compat_class = arch_array[i].compat_class;
1821 lib_arch.name = arch_array[i].name;
1822 }
1823 }
1824
1825 assert(running_arch_index != -1,
1826 "Didn't find running architecture code (running_arch_code) in arch_array");
1827 if (running_arch_index == -1) {
1828 // Even though running architecture detection failed
1829 // we may still continue with reporting dlerror() message
1830 return NULL;
1831 }
1832
1833 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1834 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1835 return NULL;
1836 }
1837
1838 #ifndef S390
1839 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1840 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1841 return NULL;
1842 }
1843 #endif // !S390
1844
1845 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1846 if (lib_arch.name!=NULL) {
1847 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1848 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1849 lib_arch.name, arch_array[running_arch_index].name);
1850 } else {
1851 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1852 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1853 lib_arch.code,
1854 arch_array[running_arch_index].name);
1855 }
1856 }
1857
1858 return NULL;
1859 }
1860
1861 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1862 int ebuflen) {
1863 void * result = ::dlopen(filename, RTLD_LAZY);
1864 if (result == NULL) {
1865 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1866 ebuf[ebuflen-1] = '\0';
1867 }
1868 return result;
1869 }
1870
1871 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1872 int ebuflen) {
1873 void * result = NULL;
1874 if (LoadExecStackDllInVMThread) {
1875 result = dlopen_helper(filename, ebuf, ebuflen);
1876 }
1877
1878 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1879 // library that requires an executable stack, or which does not have this
1880 // stack attribute set, dlopen changes the stack attribute to executable. The
1881 // read protection of the guard pages gets lost.
1882 //
1883 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1884 // may have been queued at the same time.
1885
1886 if (!_stack_is_executable) {
1887 JavaThread *jt = Threads::first();
1888
1889 while (jt) {
1890 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1891 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1892 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1893 warning("Attempt to reguard stack yellow zone failed.");
1894 }
1895 }
1896 jt = jt->next();
1897 }
1898 }
1899
1900 return result;
1901 }
1902
1903 void* os::dll_lookup(void* handle, const char* name) {
1904 void* res = dlsym(handle, name);
1905 return res;
1906 }
1907
1908 void* os::get_default_process_handle() {
1909 return (void*)::dlopen(NULL, RTLD_LAZY);
1910 }
1911
1912 static bool _print_ascii_file(const char* filename, outputStream* st) {
1913 int fd = ::open(filename, O_RDONLY);
1914 if (fd == -1) {
1915 return false;
1916 }
1917
1918 char buf[33];
1919 int bytes;
1920 buf[32] = '\0';
1921 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1922 st->print_raw(buf, bytes);
1923 }
1924
1925 ::close(fd);
1926
1927 return true;
1928 }
1929
1930 void os::print_dll_info(outputStream *st) {
1931 st->print_cr("Dynamic libraries:");
1932
1933 char fname[32];
1934 pid_t pid = os::Linux::gettid();
1935
1936 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1937
1938 if (!_print_ascii_file(fname, st)) {
1939 st->print("Can not get library information for pid = %d\n", pid);
1940 }
1941 }
1942
1943 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1944 FILE *procmapsFile = NULL;
1945
1946 // Open the procfs maps file for the current process
1947 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1948 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1949 char line[PATH_MAX + 100];
1950
1951 // Read line by line from 'file'
1952 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1953 u8 base, top, offset, inode;
1954 char permissions[5];
1955 char device[6];
1956 char name[PATH_MAX + 1];
1957
1958 // Parse fields from line
1959 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1960 &base, &top, permissions, &offset, device, &inode, name);
1961
1962 // Filter by device id '00:00' so that we only get file system mapped files.
1963 if (strcmp(device, "00:00") != 0) {
1964
1965 // Call callback with the fields of interest
1966 if(callback(name, (address)base, (address)top, param)) {
1967 // Oops abort, callback aborted
1968 fclose(procmapsFile);
1969 return 1;
1970 }
1971 }
1972 }
1973 fclose(procmapsFile);
1974 }
1975 return 0;
1976 }
1977
1978 void os::print_os_info_brief(outputStream* st) {
1979 os::Linux::print_distro_info(st);
1980
1981 os::Posix::print_uname_info(st);
1982
1983 os::Linux::print_libversion_info(st);
1984
1985 }
1986
1987 void os::print_os_info(outputStream* st) {
1988 st->print("OS:");
1989
1990 os::Linux::print_distro_info(st);
1991
1992 os::Posix::print_uname_info(st);
1993
1994 // Print warning if unsafe chroot environment detected
1995 if (unsafe_chroot_detected) {
1996 st->print("WARNING!! ");
1997 st->print_cr("%s", unstable_chroot_error);
1998 }
1999
2000 os::Linux::print_libversion_info(st);
2001
2002 os::Posix::print_rlimit_info(st);
2003
2004 os::Posix::print_load_average(st);
2005
2006 os::Linux::print_full_memory_info(st);
2007 }
2008
2009 // Try to identify popular distros.
2010 // Most Linux distributions have a /etc/XXX-release file, which contains
2011 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2012 // file that also contains the OS version string. Some have more than one
2013 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2014 // /etc/redhat-release.), so the order is important.
2015 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2016 // their own specific XXX-release file as well as a redhat-release file.
2017 // Because of this the XXX-release file needs to be searched for before the
2018 // redhat-release file.
2019 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2020 // search for redhat-release / SuSE-release needs to be before lsb-release.
2021 // Since the lsb-release file is the new standard it needs to be searched
2022 // before the older style release files.
2023 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2024 // next to last resort. The os-release file is a new standard that contains
2025 // distribution information and the system-release file seems to be an old
2026 // standard that has been replaced by the lsb-release and os-release files.
2027 // Searching for the debian_version file is the last resort. It contains
2028 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2029 // "Debian " is printed before the contents of the debian_version file.
2030
2031 const char* distro_files[] = {
2032 "/etc/oracle-release",
2033 "/etc/mandriva-release",
2034 "/etc/mandrake-release",
2035 "/etc/sun-release",
2036 "/etc/redhat-release",
2037 "/etc/SuSE-release",
2038 "/etc/lsb-release",
2039 "/etc/turbolinux-release",
2040 "/etc/gentoo-release",
2041 "/etc/ltib-release",
2042 "/etc/angstrom-version",
2043 "/etc/system-release",
2044 "/etc/os-release",
2045 NULL };
2046
2047 void os::Linux::print_distro_info(outputStream* st) {
2048 for (int i = 0;; i++) {
2049 const char* file = distro_files[i];
2050 if (file == NULL) {
2051 break; // done
2052 }
2053 // If file prints, we found it.
2054 if (_print_ascii_file(file, st)) {
2055 return;
2056 }
2057 }
2058
2059 if (file_exists("/etc/debian_version")) {
2060 st->print("Debian ");
2061 _print_ascii_file("/etc/debian_version", st);
2062 } else {
2063 st->print("Linux");
2064 }
2065 st->cr();
2066 }
2067
2068 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2069 char buf[256];
2070 while (fgets(buf, sizeof(buf), fp)) {
2071 // Edit out extra stuff in expected format
2072 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2073 char* ptr = strstr(buf, "\""); // the name is in quotes
2074 if (ptr != NULL) {
2075 ptr++; // go beyond first quote
2076 char* nl = strchr(ptr, '\"');
2077 if (nl != NULL) *nl = '\0';
2078 strncpy(distro, ptr, length);
2079 } else {
2080 ptr = strstr(buf, "=");
2081 ptr++; // go beyond equals then
2082 char* nl = strchr(ptr, '\n');
2083 if (nl != NULL) *nl = '\0';
2084 strncpy(distro, ptr, length);
2085 }
2086 return;
2087 } else if (get_first_line) {
2088 char* nl = strchr(buf, '\n');
2089 if (nl != NULL) *nl = '\0';
2090 strncpy(distro, buf, length);
2091 return;
2092 }
2093 }
2094 // print last line and close
2095 char* nl = strchr(buf, '\n');
2096 if (nl != NULL) *nl = '\0';
2097 strncpy(distro, buf, length);
2098 }
2099
2100 static void parse_os_info(char* distro, size_t length, const char* file) {
2101 FILE* fp = fopen(file, "r");
2102 if (fp != NULL) {
2103 // if suse format, print out first line
2104 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2105 parse_os_info_helper(fp, distro, length, get_first_line);
2106 fclose(fp);
2107 }
2108 }
2109
2110 void os::get_summary_os_info(char* buf, size_t buflen) {
2111 for (int i = 0;; i++) {
2112 const char* file = distro_files[i];
2113 if (file == NULL) {
2114 break; // ran out of distro_files
2115 }
2116 if (file_exists(file)) {
2117 parse_os_info(buf, buflen, file);
2118 return;
2119 }
2120 }
2121 // special case for debian
2122 if (file_exists("/etc/debian_version")) {
2123 strncpy(buf, "Debian ", buflen);
2124 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2125 } else {
2126 strncpy(buf, "Linux", buflen);
2127 }
2128 }
2129
2130 void os::Linux::print_libversion_info(outputStream* st) {
2131 // libc, pthread
2132 st->print("libc:");
2133 st->print("%s ", os::Linux::glibc_version());
2134 st->print("%s ", os::Linux::libpthread_version());
2135 st->cr();
2136 }
2137
2138 void os::Linux::print_full_memory_info(outputStream* st) {
2139 st->print("\n/proc/meminfo:\n");
2140 _print_ascii_file("/proc/meminfo", st);
2141 st->cr();
2142 }
2143
2144 void os::print_memory_info(outputStream* st) {
2145
2146 st->print("Memory:");
2147 st->print(" %dk page", os::vm_page_size()>>10);
2148
2149 // values in struct sysinfo are "unsigned long"
2150 struct sysinfo si;
2151 sysinfo(&si);
2152
2153 st->print(", physical " UINT64_FORMAT "k",
2154 os::physical_memory() >> 10);
2155 st->print("(" UINT64_FORMAT "k free)",
2156 os::available_memory() >> 10);
2157 st->print(", swap " UINT64_FORMAT "k",
2158 ((jlong)si.totalswap * si.mem_unit) >> 10);
2159 st->print("(" UINT64_FORMAT "k free)",
2160 ((jlong)si.freeswap * si.mem_unit) >> 10);
2161 st->cr();
2162 }
2163
2164 // Print the first "model name" line and the first "flags" line
2165 // that we find and nothing more. We assume "model name" comes
2166 // before "flags" so if we find a second "model name", then the
2167 // "flags" field is considered missing.
2168 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2169 #if defined(IA32) || defined(AMD64)
2170 // Other platforms have less repetitive cpuinfo files
2171 FILE *fp = fopen("/proc/cpuinfo", "r");
2172 if (fp) {
2173 while (!feof(fp)) {
2174 if (fgets(buf, buflen, fp)) {
2175 // Assume model name comes before flags
2176 bool model_name_printed = false;
2177 if (strstr(buf, "model name") != NULL) {
2178 if (!model_name_printed) {
2179 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2180 st->print_raw(buf);
2181 model_name_printed = true;
2182 } else {
2183 // model name printed but not flags? Odd, just return
2184 fclose(fp);
2185 return true;
2186 }
2187 }
2188 // print the flags line too
2189 if (strstr(buf, "flags") != NULL) {
2190 st->print_raw(buf);
2191 fclose(fp);
2192 return true;
2193 }
2194 }
2195 }
2196 fclose(fp);
2197 }
2198 #endif // x86 platforms
2199 return false;
2200 }
2201
2202 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2203 // Only print the model name if the platform provides this as a summary
2204 if (!print_model_name_and_flags(st, buf, buflen)) {
2205 st->print("\n/proc/cpuinfo:\n");
2206 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2207 st->print_cr(" <Not Available>");
2208 }
2209 }
2210 }
2211
2212 #if defined(AMD64) || defined(IA32) || defined(X32)
2213 const char* search_string = "model name";
2214 #elif defined(PPC64)
2215 const char* search_string = "cpu";
2216 #elif defined(S390)
2217 const char* search_string = "processor";
2218 #elif defined(SPARC)
2219 const char* search_string = "cpu";
2220 #else
2221 const char* search_string = "Processor";
2222 #endif
2223
2224 // Parses the cpuinfo file for string representing the model name.
2225 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2226 FILE* fp = fopen("/proc/cpuinfo", "r");
2227 if (fp != NULL) {
2228 while (!feof(fp)) {
2229 char buf[256];
2230 if (fgets(buf, sizeof(buf), fp)) {
2231 char* start = strstr(buf, search_string);
2232 if (start != NULL) {
2233 char *ptr = start + strlen(search_string);
2234 char *end = buf + strlen(buf);
2235 while (ptr != end) {
2236 // skip whitespace and colon for the rest of the name.
2237 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2238 break;
2239 }
2240 ptr++;
2241 }
2242 if (ptr != end) {
2243 // reasonable string, get rid of newline and keep the rest
2244 char* nl = strchr(buf, '\n');
2245 if (nl != NULL) *nl = '\0';
2246 strncpy(cpuinfo, ptr, length);
2247 fclose(fp);
2248 return;
2249 }
2250 }
2251 }
2252 }
2253 fclose(fp);
2254 }
2255 // cpuinfo not found or parsing failed, just print generic string. The entire
2256 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2257 #if defined(AARCH64)
2258 strncpy(cpuinfo, "AArch64", length);
2259 #elif defined(AMD64)
2260 strncpy(cpuinfo, "x86_64", length);
2261 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2262 strncpy(cpuinfo, "ARM", length);
2263 #elif defined(IA32)
2264 strncpy(cpuinfo, "x86_32", length);
2265 #elif defined(IA64)
2266 strncpy(cpuinfo, "IA64", length);
2267 #elif defined(PPC)
2268 strncpy(cpuinfo, "PPC64", length);
2269 #elif defined(S390)
2270 strncpy(cpuinfo, "S390", length);
2271 #elif defined(SPARC)
2272 strncpy(cpuinfo, "sparcv9", length);
2273 #elif defined(ZERO_LIBARCH)
2274 strncpy(cpuinfo, ZERO_LIBARCH, length);
2275 #else
2276 strncpy(cpuinfo, "unknown", length);
2277 #endif
2278 }
2279
2280 static void print_signal_handler(outputStream* st, int sig,
2281 char* buf, size_t buflen);
2282
2283 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2284 st->print_cr("Signal Handlers:");
2285 print_signal_handler(st, SIGSEGV, buf, buflen);
2286 print_signal_handler(st, SIGBUS , buf, buflen);
2287 print_signal_handler(st, SIGFPE , buf, buflen);
2288 print_signal_handler(st, SIGPIPE, buf, buflen);
2289 print_signal_handler(st, SIGXFSZ, buf, buflen);
2290 print_signal_handler(st, SIGILL , buf, buflen);
2291 print_signal_handler(st, SR_signum, buf, buflen);
2292 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2293 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2294 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2295 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2296 #if defined(PPC64)
2297 print_signal_handler(st, SIGTRAP, buf, buflen);
2298 #endif
2299 }
2300
2301 static char saved_jvm_path[MAXPATHLEN] = {0};
2302
2303 // Find the full path to the current module, libjvm.so
2304 void os::jvm_path(char *buf, jint buflen) {
2305 // Error checking.
2306 if (buflen < MAXPATHLEN) {
2307 assert(false, "must use a large-enough buffer");
2308 buf[0] = '\0';
2309 return;
2310 }
2311 // Lazy resolve the path to current module.
2312 if (saved_jvm_path[0] != 0) {
2313 strcpy(buf, saved_jvm_path);
2314 return;
2315 }
2316
2317 char dli_fname[MAXPATHLEN];
2318 bool ret = dll_address_to_library_name(
2319 CAST_FROM_FN_PTR(address, os::jvm_path),
2320 dli_fname, sizeof(dli_fname), NULL);
2321 assert(ret, "cannot locate libjvm");
2322 char *rp = NULL;
2323 if (ret && dli_fname[0] != '\0') {
2324 rp = realpath(dli_fname, buf);
2325 }
2326 if (rp == NULL) {
2327 return;
2328 }
2329
2330 if (Arguments::sun_java_launcher_is_altjvm()) {
2331 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2332 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2333 // If "/jre/lib/" appears at the right place in the string, then
2334 // assume we are installed in a JDK and we're done. Otherwise, check
2335 // for a JAVA_HOME environment variable and fix up the path so it
2336 // looks like libjvm.so is installed there (append a fake suffix
2337 // hotspot/libjvm.so).
2338 const char *p = buf + strlen(buf) - 1;
2339 for (int count = 0; p > buf && count < 5; ++count) {
2340 for (--p; p > buf && *p != '/'; --p)
2341 /* empty */ ;
2342 }
2343
2344 if (strncmp(p, "/jre/lib/", 9) != 0) {
2345 // Look for JAVA_HOME in the environment.
2346 char* java_home_var = ::getenv("JAVA_HOME");
2347 if (java_home_var != NULL && java_home_var[0] != 0) {
2348 char* jrelib_p;
2349 int len;
2350
2351 // Check the current module name "libjvm.so".
2352 p = strrchr(buf, '/');
2353 if (p == NULL) {
2354 return;
2355 }
2356 assert(strstr(p, "/libjvm") == p, "invalid library name");
2357
2358 rp = realpath(java_home_var, buf);
2359 if (rp == NULL) {
2360 return;
2361 }
2362
2363 // determine if this is a legacy image or modules image
2364 // modules image doesn't have "jre" subdirectory
2365 len = strlen(buf);
2366 assert(len < buflen, "Ran out of buffer room");
2367 jrelib_p = buf + len;
2368 snprintf(jrelib_p, buflen-len, "/jre/lib");
2369 if (0 != access(buf, F_OK)) {
2370 snprintf(jrelib_p, buflen-len, "/lib");
2371 }
2372
2373 if (0 == access(buf, F_OK)) {
2374 // Use current module name "libjvm.so"
2375 len = strlen(buf);
2376 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2377 } else {
2378 // Go back to path of .so
2379 rp = realpath(dli_fname, buf);
2380 if (rp == NULL) {
2381 return;
2382 }
2383 }
2384 }
2385 }
2386 }
2387
2388 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2389 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2390 }
2391
2392 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2393 // no prefix required, not even "_"
2394 }
2395
2396 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2397 // no suffix required
2398 }
2399
2400 ////////////////////////////////////////////////////////////////////////////////
2401 // sun.misc.Signal support
2402
2403 static volatile jint sigint_count = 0;
2404
2405 static void UserHandler(int sig, void *siginfo, void *context) {
2406 // 4511530 - sem_post is serialized and handled by the manager thread. When
2407 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2408 // don't want to flood the manager thread with sem_post requests.
2409 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2410 return;
2411 }
2412
2413 // Ctrl-C is pressed during error reporting, likely because the error
2414 // handler fails to abort. Let VM die immediately.
2415 if (sig == SIGINT && is_error_reported()) {
2416 os::die();
2417 }
2418
2419 os::signal_notify(sig);
2420 }
2421
2422 void* os::user_handler() {
2423 return CAST_FROM_FN_PTR(void*, UserHandler);
2424 }
2425
2426 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2427 struct timespec ts;
2428 // Semaphore's are always associated with CLOCK_REALTIME
2429 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2430 // see unpackTime for discussion on overflow checking
2431 if (sec >= MAX_SECS) {
2432 ts.tv_sec += MAX_SECS;
2433 ts.tv_nsec = 0;
2434 } else {
2435 ts.tv_sec += sec;
2436 ts.tv_nsec += nsec;
2437 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2438 ts.tv_nsec -= NANOSECS_PER_SEC;
2439 ++ts.tv_sec; // note: this must be <= max_secs
2440 }
2441 }
2442
2443 return ts;
2444 }
2445
2446 extern "C" {
2447 typedef void (*sa_handler_t)(int);
2448 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2449 }
2450
2451 void* os::signal(int signal_number, void* handler) {
2452 struct sigaction sigAct, oldSigAct;
2453
2454 sigfillset(&(sigAct.sa_mask));
2455 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2456 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2457
2458 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2459 // -1 means registration failed
2460 return (void *)-1;
2461 }
2462
2463 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2464 }
2465
2466 void os::signal_raise(int signal_number) {
2467 ::raise(signal_number);
2468 }
2469
2470 // The following code is moved from os.cpp for making this
2471 // code platform specific, which it is by its very nature.
2472
2473 // Will be modified when max signal is changed to be dynamic
2474 int os::sigexitnum_pd() {
2475 return NSIG;
2476 }
2477
2478 // a counter for each possible signal value
2479 static volatile jint pending_signals[NSIG+1] = { 0 };
2480
2481 // Linux(POSIX) specific hand shaking semaphore.
2482 static sem_t sig_sem;
2483 static PosixSemaphore sr_semaphore;
2484
2485 void os::signal_init_pd() {
2486 // Initialize signal structures
2487 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2488
2489 // Initialize signal semaphore
2490 ::sem_init(&sig_sem, 0, 0);
2491 }
2492
2493 void os::signal_notify(int sig) {
2494 Atomic::inc(&pending_signals[sig]);
2495 ::sem_post(&sig_sem);
2496 }
2497
2498 static int check_pending_signals(bool wait) {
2499 Atomic::store(0, &sigint_count);
2500 for (;;) {
2501 for (int i = 0; i < NSIG + 1; i++) {
2502 jint n = pending_signals[i];
2503 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2504 return i;
2505 }
2506 }
2507 if (!wait) {
2508 return -1;
2509 }
2510 JavaThread *thread = JavaThread::current();
2511 ThreadBlockInVM tbivm(thread);
2512
2513 bool threadIsSuspended;
2514 do {
2515 thread->set_suspend_equivalent();
2516 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2517 ::sem_wait(&sig_sem);
2518
2519 // were we externally suspended while we were waiting?
2520 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2521 if (threadIsSuspended) {
2522 // The semaphore has been incremented, but while we were waiting
2523 // another thread suspended us. We don't want to continue running
2524 // while suspended because that would surprise the thread that
2525 // suspended us.
2526 ::sem_post(&sig_sem);
2527
2528 thread->java_suspend_self();
2529 }
2530 } while (threadIsSuspended);
2531 }
2532 }
2533
2534 int os::signal_lookup() {
2535 return check_pending_signals(false);
2536 }
2537
2538 int os::signal_wait() {
2539 return check_pending_signals(true);
2540 }
2541
2542 ////////////////////////////////////////////////////////////////////////////////
2543 // Virtual Memory
2544
2545 int os::vm_page_size() {
2546 // Seems redundant as all get out
2547 assert(os::Linux::page_size() != -1, "must call os::init");
2548 return os::Linux::page_size();
2549 }
2550
2551 // Solaris allocates memory by pages.
2552 int os::vm_allocation_granularity() {
2553 assert(os::Linux::page_size() != -1, "must call os::init");
2554 return os::Linux::page_size();
2555 }
2556
2557 // Rationale behind this function:
2558 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2559 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2560 // samples for JITted code. Here we create private executable mapping over the code cache
2561 // and then we can use standard (well, almost, as mapping can change) way to provide
2562 // info for the reporting script by storing timestamp and location of symbol
2563 void linux_wrap_code(char* base, size_t size) {
2564 static volatile jint cnt = 0;
2565
2566 if (!UseOprofile) {
2567 return;
2568 }
2569
2570 char buf[PATH_MAX+1];
2571 int num = Atomic::add(1, &cnt);
2572
2573 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2574 os::get_temp_directory(), os::current_process_id(), num);
2575 unlink(buf);
2576
2577 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2578
2579 if (fd != -1) {
2580 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2581 if (rv != (off_t)-1) {
2582 if (::write(fd, "", 1) == 1) {
2583 mmap(base, size,
2584 PROT_READ|PROT_WRITE|PROT_EXEC,
2585 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2586 }
2587 }
2588 ::close(fd);
2589 unlink(buf);
2590 }
2591 }
2592
2593 static bool recoverable_mmap_error(int err) {
2594 // See if the error is one we can let the caller handle. This
2595 // list of errno values comes from JBS-6843484. I can't find a
2596 // Linux man page that documents this specific set of errno
2597 // values so while this list currently matches Solaris, it may
2598 // change as we gain experience with this failure mode.
2599 switch (err) {
2600 case EBADF:
2601 case EINVAL:
2602 case ENOTSUP:
2603 // let the caller deal with these errors
2604 return true;
2605
2606 default:
2607 // Any remaining errors on this OS can cause our reserved mapping
2608 // to be lost. That can cause confusion where different data
2609 // structures think they have the same memory mapped. The worst
2610 // scenario is if both the VM and a library think they have the
2611 // same memory mapped.
2612 return false;
2613 }
2614 }
2615
2616 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2617 int err) {
2618 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2619 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2620 os::strerror(err), err);
2621 }
2622
2623 static void warn_fail_commit_memory(char* addr, size_t size,
2624 size_t alignment_hint, bool exec,
2625 int err) {
2626 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2627 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2628 alignment_hint, exec, os::strerror(err), err);
2629 }
2630
2631 // NOTE: Linux kernel does not really reserve the pages for us.
2632 // All it does is to check if there are enough free pages
2633 // left at the time of mmap(). This could be a potential
2634 // problem.
2635 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2636 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2637 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2638 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2639 if (res != (uintptr_t) MAP_FAILED) {
2640 if (UseNUMAInterleaving) {
2641 numa_make_global(addr, size);
2642 }
2643 return 0;
2644 }
2645
2646 int err = errno; // save errno from mmap() call above
2647
2648 if (!recoverable_mmap_error(err)) {
2649 warn_fail_commit_memory(addr, size, exec, err);
2650 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2651 }
2652
2653 return err;
2654 }
2655
2656 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2657 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2658 }
2659
2660 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2661 const char* mesg) {
2662 assert(mesg != NULL, "mesg must be specified");
2663 int err = os::Linux::commit_memory_impl(addr, size, exec);
2664 if (err != 0) {
2665 // the caller wants all commit errors to exit with the specified mesg:
2666 warn_fail_commit_memory(addr, size, exec, err);
2667 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2668 }
2669 }
2670
2671 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2672 #ifndef MAP_HUGETLB
2673 #define MAP_HUGETLB 0x40000
2674 #endif
2675
2676 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2677 #ifndef MADV_HUGEPAGE
2678 #define MADV_HUGEPAGE 14
2679 #endif
2680
2681 int os::Linux::commit_memory_impl(char* addr, size_t size,
2682 size_t alignment_hint, bool exec) {
2683 int err = os::Linux::commit_memory_impl(addr, size, exec);
2684 if (err == 0) {
2685 realign_memory(addr, size, alignment_hint);
2686 }
2687 return err;
2688 }
2689
2690 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2691 bool exec) {
2692 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2693 }
2694
2695 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2696 size_t alignment_hint, bool exec,
2697 const char* mesg) {
2698 assert(mesg != NULL, "mesg must be specified");
2699 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2700 if (err != 0) {
2701 // the caller wants all commit errors to exit with the specified mesg:
2702 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2703 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2704 }
2705 }
2706
2707 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2708 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2709 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2710 // be supported or the memory may already be backed by huge pages.
2711 ::madvise(addr, bytes, MADV_HUGEPAGE);
2712 }
2713 }
2714
2715 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2716 // This method works by doing an mmap over an existing mmaping and effectively discarding
2717 // the existing pages. However it won't work for SHM-based large pages that cannot be
2718 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2719 // small pages on top of the SHM segment. This method always works for small pages, so we
2720 // allow that in any case.
2721 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2722 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2723 }
2724 }
2725
2726 void os::numa_make_global(char *addr, size_t bytes) {
2727 Linux::numa_interleave_memory(addr, bytes);
2728 }
2729
2730 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2731 // bind policy to MPOL_PREFERRED for the current thread.
2732 #define USE_MPOL_PREFERRED 0
2733
2734 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2735 // To make NUMA and large pages more robust when both enabled, we need to ease
2736 // the requirements on where the memory should be allocated. MPOL_BIND is the
2737 // default policy and it will force memory to be allocated on the specified
2738 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2739 // the specified node, but will not force it. Using this policy will prevent
2740 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2741 // free large pages.
2742 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2743 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2744 }
2745
2746 bool os::numa_topology_changed() { return false; }
2747
2748 size_t os::numa_get_groups_num() {
2749 int max_node = Linux::numa_max_node();
2750 return max_node > 0 ? max_node + 1 : 1;
2751 }
2752
2753 int os::numa_get_group_id() {
2754 int cpu_id = Linux::sched_getcpu();
2755 if (cpu_id != -1) {
2756 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2757 if (lgrp_id != -1) {
2758 return lgrp_id;
2759 }
2760 }
2761 return 0;
2762 }
2763
2764 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2765 for (size_t i = 0; i < size; i++) {
2766 ids[i] = i;
2767 }
2768 return size;
2769 }
2770
2771 bool os::get_page_info(char *start, page_info* info) {
2772 return false;
2773 }
2774
2775 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2776 page_info* page_found) {
2777 return end;
2778 }
2779
2780
2781 int os::Linux::sched_getcpu_syscall(void) {
2782 unsigned int cpu = 0;
2783 int retval = -1;
2784
2785 #if defined(IA32)
2786 #ifndef SYS_getcpu
2787 #define SYS_getcpu 318
2788 #endif
2789 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2790 #elif defined(AMD64)
2791 // Unfortunately we have to bring all these macros here from vsyscall.h
2792 // to be able to compile on old linuxes.
2793 #define __NR_vgetcpu 2
2794 #define VSYSCALL_START (-10UL << 20)
2795 #define VSYSCALL_SIZE 1024
2796 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2797 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2798 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2799 retval = vgetcpu(&cpu, NULL, NULL);
2800 #endif
2801
2802 return (retval == -1) ? retval : cpu;
2803 }
2804
2805 // Something to do with the numa-aware allocator needs these symbols
2806 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2807 extern "C" JNIEXPORT void numa_error(char *where) { }
2808
2809
2810 // If we are running with libnuma version > 2, then we should
2811 // be trying to use symbols with versions 1.1
2812 // If we are running with earlier version, which did not have symbol versions,
2813 // we should use the base version.
2814 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2815 typedef void* (*dlvsym_func_type)(void* handle, const char* name, const char* version);
2816 static dlvsym_func_type dlvsym_func;
2817 static bool initialized = false;
2818
2819 if (!initialized) {
2820 dlvsym_func = (dlvsym_func_type)dlsym(RTLD_NEXT, "dlvsym");
2821 initialized = true;
2822 }
2823
2824 if (dlvsym_func != NULL) {
2825 void *f = dlvsym_func(handle, name, "libnuma_1.1");
2826 if (f != NULL) {
2827 return f;
2828 }
2829 }
2830
2831 return dlsym(handle, name);
2832 }
2833
2834 bool os::Linux::libnuma_init() {
2835 // sched_getcpu() should be in libc.
2836 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2837 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2838
2839 // If it's not, try a direct syscall.
2840 if (sched_getcpu() == -1) {
2841 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2842 (void*)&sched_getcpu_syscall));
2843 }
2844
2845 if (sched_getcpu() != -1) { // Does it work?
2846 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2847 if (handle != NULL) {
2848 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2849 libnuma_dlsym(handle, "numa_node_to_cpus")));
2850 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2851 libnuma_dlsym(handle, "numa_max_node")));
2852 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2853 libnuma_dlsym(handle, "numa_available")));
2854 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2855 libnuma_dlsym(handle, "numa_tonode_memory")));
2856 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2857 libnuma_dlsym(handle, "numa_interleave_memory")));
2858 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2859 libnuma_dlsym(handle, "numa_set_bind_policy")));
2860
2861
2862 if (numa_available() != -1) {
2863 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2864 // Create a cpu -> node mapping
2865 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2866 rebuild_cpu_to_node_map();
2867 return true;
2868 }
2869 }
2870 }
2871 return false;
2872 }
2873
2874 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2875 // Creating guard page is very expensive. Java thread has HotSpot
2876 // guard pages, only enable glibc guard page for non-Java threads.
2877 // (Remember: compiler thread is a Java thread, too!)
2878 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2879 }
2880
2881 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2882 // The table is later used in get_node_by_cpu().
2883 void os::Linux::rebuild_cpu_to_node_map() {
2884 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2885 // in libnuma (possible values are starting from 16,
2886 // and continuing up with every other power of 2, but less
2887 // than the maximum number of CPUs supported by kernel), and
2888 // is a subject to change (in libnuma version 2 the requirements
2889 // are more reasonable) we'll just hardcode the number they use
2890 // in the library.
2891 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2892
2893 size_t cpu_num = processor_count();
2894 size_t cpu_map_size = NCPUS / BitsPerCLong;
2895 size_t cpu_map_valid_size =
2896 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2897
2898 cpu_to_node()->clear();
2899 cpu_to_node()->at_grow(cpu_num - 1);
2900 size_t node_num = numa_get_groups_num();
2901
2902 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2903 for (size_t i = 0; i < node_num; i++) {
2904 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2905 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2906 if (cpu_map[j] != 0) {
2907 for (size_t k = 0; k < BitsPerCLong; k++) {
2908 if (cpu_map[j] & (1UL << k)) {
2909 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2910 }
2911 }
2912 }
2913 }
2914 }
2915 }
2916 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2917 }
2918
2919 int os::Linux::get_node_by_cpu(int cpu_id) {
2920 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2921 return cpu_to_node()->at(cpu_id);
2922 }
2923 return -1;
2924 }
2925
2926 GrowableArray<int>* os::Linux::_cpu_to_node;
2927 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2928 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2929 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2930 os::Linux::numa_available_func_t os::Linux::_numa_available;
2931 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2932 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2933 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2934 unsigned long* os::Linux::_numa_all_nodes;
2935
2936 bool os::pd_uncommit_memory(char* addr, size_t size) {
2937 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2938 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2939 return res != (uintptr_t) MAP_FAILED;
2940 }
2941
2942 static address get_stack_commited_bottom(address bottom, size_t size) {
2943 address nbot = bottom;
2944 address ntop = bottom + size;
2945
2946 size_t page_sz = os::vm_page_size();
2947 unsigned pages = size / page_sz;
2948
2949 unsigned char vec[1];
2950 unsigned imin = 1, imax = pages + 1, imid;
2951 int mincore_return_value = 0;
2952
2953 assert(imin <= imax, "Unexpected page size");
2954
2955 while (imin < imax) {
2956 imid = (imax + imin) / 2;
2957 nbot = ntop - (imid * page_sz);
2958
2959 // Use a trick with mincore to check whether the page is mapped or not.
2960 // mincore sets vec to 1 if page resides in memory and to 0 if page
2961 // is swapped output but if page we are asking for is unmapped
2962 // it returns -1,ENOMEM
2963 mincore_return_value = mincore(nbot, page_sz, vec);
2964
2965 if (mincore_return_value == -1) {
2966 // Page is not mapped go up
2967 // to find first mapped page
2968 if (errno != EAGAIN) {
2969 assert(errno == ENOMEM, "Unexpected mincore errno");
2970 imax = imid;
2971 }
2972 } else {
2973 // Page is mapped go down
2974 // to find first not mapped page
2975 imin = imid + 1;
2976 }
2977 }
2978
2979 nbot = nbot + page_sz;
2980
2981 // Adjust stack bottom one page up if last checked page is not mapped
2982 if (mincore_return_value == -1) {
2983 nbot = nbot + page_sz;
2984 }
2985
2986 return nbot;
2987 }
2988
2989
2990 // Linux uses a growable mapping for the stack, and if the mapping for
2991 // the stack guard pages is not removed when we detach a thread the
2992 // stack cannot grow beyond the pages where the stack guard was
2993 // mapped. If at some point later in the process the stack expands to
2994 // that point, the Linux kernel cannot expand the stack any further
2995 // because the guard pages are in the way, and a segfault occurs.
2996 //
2997 // However, it's essential not to split the stack region by unmapping
2998 // a region (leaving a hole) that's already part of the stack mapping,
2999 // so if the stack mapping has already grown beyond the guard pages at
3000 // the time we create them, we have to truncate the stack mapping.
3001 // So, we need to know the extent of the stack mapping when
3002 // create_stack_guard_pages() is called.
3003
3004 // We only need this for stacks that are growable: at the time of
3005 // writing thread stacks don't use growable mappings (i.e. those
3006 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3007 // only applies to the main thread.
3008
3009 // If the (growable) stack mapping already extends beyond the point
3010 // where we're going to put our guard pages, truncate the mapping at
3011 // that point by munmap()ping it. This ensures that when we later
3012 // munmap() the guard pages we don't leave a hole in the stack
3013 // mapping. This only affects the main/initial thread
3014
3015 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3016 if (os::Linux::is_initial_thread()) {
3017 // As we manually grow stack up to bottom inside create_attached_thread(),
3018 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3019 // we don't need to do anything special.
3020 // Check it first, before calling heavy function.
3021 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3022 unsigned char vec[1];
3023
3024 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3025 // Fallback to slow path on all errors, including EAGAIN
3026 stack_extent = (uintptr_t) get_stack_commited_bottom(
3027 os::Linux::initial_thread_stack_bottom(),
3028 (size_t)addr - stack_extent);
3029 }
3030
3031 if (stack_extent < (uintptr_t)addr) {
3032 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3033 }
3034 }
3035
3036 return os::commit_memory(addr, size, !ExecMem);
3037 }
3038
3039 // If this is a growable mapping, remove the guard pages entirely by
3040 // munmap()ping them. If not, just call uncommit_memory(). This only
3041 // affects the main/initial thread, but guard against future OS changes
3042 // It's safe to always unmap guard pages for initial thread because we
3043 // always place it right after end of the mapped region
3044
3045 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3046 uintptr_t stack_extent, stack_base;
3047
3048 if (os::Linux::is_initial_thread()) {
3049 return ::munmap(addr, size) == 0;
3050 }
3051
3052 return os::uncommit_memory(addr, size);
3053 }
3054
3055 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3056 // at 'requested_addr'. If there are existing memory mappings at the same
3057 // location, however, they will be overwritten. If 'fixed' is false,
3058 // 'requested_addr' is only treated as a hint, the return value may or
3059 // may not start from the requested address. Unlike Linux mmap(), this
3060 // function returns NULL to indicate failure.
3061 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3062 char * addr;
3063 int flags;
3064
3065 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3066 if (fixed) {
3067 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3068 flags |= MAP_FIXED;
3069 }
3070
3071 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3072 // touch an uncommitted page. Otherwise, the read/write might
3073 // succeed if we have enough swap space to back the physical page.
3074 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3075 flags, -1, 0);
3076
3077 return addr == MAP_FAILED ? NULL : addr;
3078 }
3079
3080 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3081 // (req_addr != NULL) or with a given alignment.
3082 // - bytes shall be a multiple of alignment.
3083 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3084 // - alignment sets the alignment at which memory shall be allocated.
3085 // It must be a multiple of allocation granularity.
3086 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3087 // req_addr or NULL.
3088 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3089
3090 size_t extra_size = bytes;
3091 if (req_addr == NULL && alignment > 0) {
3092 extra_size += alignment;
3093 }
3094
3095 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3096 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3097 -1, 0);
3098 if (start == MAP_FAILED) {
3099 start = NULL;
3100 } else {
3101 if (req_addr != NULL) {
3102 if (start != req_addr) {
3103 ::munmap(start, extra_size);
3104 start = NULL;
3105 }
3106 } else {
3107 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3108 char* const end_aligned = start_aligned + bytes;
3109 char* const end = start + extra_size;
3110 if (start_aligned > start) {
3111 ::munmap(start, start_aligned - start);
3112 }
3113 if (end_aligned < end) {
3114 ::munmap(end_aligned, end - end_aligned);
3115 }
3116 start = start_aligned;
3117 }
3118 }
3119 return start;
3120 }
3121
3122 static int anon_munmap(char * addr, size_t size) {
3123 return ::munmap(addr, size) == 0;
3124 }
3125
3126 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3127 size_t alignment_hint) {
3128 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3129 }
3130
3131 bool os::pd_release_memory(char* addr, size_t size) {
3132 return anon_munmap(addr, size);
3133 }
3134
3135 static bool linux_mprotect(char* addr, size_t size, int prot) {
3136 // Linux wants the mprotect address argument to be page aligned.
3137 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3138
3139 // According to SUSv3, mprotect() should only be used with mappings
3140 // established by mmap(), and mmap() always maps whole pages. Unaligned
3141 // 'addr' likely indicates problem in the VM (e.g. trying to change
3142 // protection of malloc'ed or statically allocated memory). Check the
3143 // caller if you hit this assert.
3144 assert(addr == bottom, "sanity check");
3145
3146 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3147 return ::mprotect(bottom, size, prot) == 0;
3148 }
3149
3150 // Set protections specified
3151 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3152 bool is_committed) {
3153 unsigned int p = 0;
3154 switch (prot) {
3155 case MEM_PROT_NONE: p = PROT_NONE; break;
3156 case MEM_PROT_READ: p = PROT_READ; break;
3157 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3158 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3159 default:
3160 ShouldNotReachHere();
3161 }
3162 // is_committed is unused.
3163 return linux_mprotect(addr, bytes, p);
3164 }
3165
3166 bool os::guard_memory(char* addr, size_t size) {
3167 return linux_mprotect(addr, size, PROT_NONE);
3168 }
3169
3170 bool os::unguard_memory(char* addr, size_t size) {
3171 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3172 }
3173
3174 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3175 size_t page_size) {
3176 bool result = false;
3177 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3178 MAP_ANONYMOUS|MAP_PRIVATE,
3179 -1, 0);
3180 if (p != MAP_FAILED) {
3181 void *aligned_p = align_ptr_up(p, page_size);
3182
3183 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3184
3185 munmap(p, page_size * 2);
3186 }
3187
3188 if (warn && !result) {
3189 warning("TransparentHugePages is not supported by the operating system.");
3190 }
3191
3192 return result;
3193 }
3194
3195 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3196 bool result = false;
3197 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3198 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3199 -1, 0);
3200
3201 if (p != MAP_FAILED) {
3202 // We don't know if this really is a huge page or not.
3203 FILE *fp = fopen("/proc/self/maps", "r");
3204 if (fp) {
3205 while (!feof(fp)) {
3206 char chars[257];
3207 long x = 0;
3208 if (fgets(chars, sizeof(chars), fp)) {
3209 if (sscanf(chars, "%lx-%*x", &x) == 1
3210 && x == (long)p) {
3211 if (strstr (chars, "hugepage")) {
3212 result = true;
3213 break;
3214 }
3215 }
3216 }
3217 }
3218 fclose(fp);
3219 }
3220 munmap(p, page_size);
3221 }
3222
3223 if (warn && !result) {
3224 warning("HugeTLBFS is not supported by the operating system.");
3225 }
3226
3227 return result;
3228 }
3229
3230 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3231 //
3232 // From the coredump_filter documentation:
3233 //
3234 // - (bit 0) anonymous private memory
3235 // - (bit 1) anonymous shared memory
3236 // - (bit 2) file-backed private memory
3237 // - (bit 3) file-backed shared memory
3238 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3239 // effective only if the bit 2 is cleared)
3240 // - (bit 5) hugetlb private memory
3241 // - (bit 6) hugetlb shared memory
3242 //
3243 static void set_coredump_filter(void) {
3244 FILE *f;
3245 long cdm;
3246
3247 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3248 return;
3249 }
3250
3251 if (fscanf(f, "%lx", &cdm) != 1) {
3252 fclose(f);
3253 return;
3254 }
3255
3256 rewind(f);
3257
3258 if ((cdm & LARGEPAGES_BIT) == 0) {
3259 cdm |= LARGEPAGES_BIT;
3260 fprintf(f, "%#lx", cdm);
3261 }
3262
3263 fclose(f);
3264 }
3265
3266 // Large page support
3267
3268 static size_t _large_page_size = 0;
3269
3270 size_t os::Linux::find_large_page_size() {
3271 size_t large_page_size = 0;
3272
3273 // large_page_size on Linux is used to round up heap size. x86 uses either
3274 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3275 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3276 // page as large as 256M.
3277 //
3278 // Here we try to figure out page size by parsing /proc/meminfo and looking
3279 // for a line with the following format:
3280 // Hugepagesize: 2048 kB
3281 //
3282 // If we can't determine the value (e.g. /proc is not mounted, or the text
3283 // format has been changed), we'll use the largest page size supported by
3284 // the processor.
3285
3286 #ifndef ZERO
3287 large_page_size =
3288 AARCH64_ONLY(2 * M)
3289 AMD64_ONLY(2 * M)
3290 ARM32_ONLY(2 * M)
3291 IA32_ONLY(4 * M)
3292 IA64_ONLY(256 * M)
3293 PPC_ONLY(4 * M)
3294 S390_ONLY(1 * M)
3295 SPARC_ONLY(4 * M);
3296 #endif // ZERO
3297
3298 FILE *fp = fopen("/proc/meminfo", "r");
3299 if (fp) {
3300 while (!feof(fp)) {
3301 int x = 0;
3302 char buf[16];
3303 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3304 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3305 large_page_size = x * K;
3306 break;
3307 }
3308 } else {
3309 // skip to next line
3310 for (;;) {
3311 int ch = fgetc(fp);
3312 if (ch == EOF || ch == (int)'\n') break;
3313 }
3314 }
3315 }
3316 fclose(fp);
3317 }
3318
3319 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3320 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3321 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3322 proper_unit_for_byte_size(large_page_size));
3323 }
3324
3325 return large_page_size;
3326 }
3327
3328 size_t os::Linux::setup_large_page_size() {
3329 _large_page_size = Linux::find_large_page_size();
3330 const size_t default_page_size = (size_t)Linux::page_size();
3331 if (_large_page_size > default_page_size) {
3332 _page_sizes[0] = _large_page_size;
3333 _page_sizes[1] = default_page_size;
3334 _page_sizes[2] = 0;
3335 }
3336
3337 return _large_page_size;
3338 }
3339
3340 bool os::Linux::setup_large_page_type(size_t page_size) {
3341 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3342 FLAG_IS_DEFAULT(UseSHM) &&
3343 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3344
3345 // The type of large pages has not been specified by the user.
3346
3347 // Try UseHugeTLBFS and then UseSHM.
3348 UseHugeTLBFS = UseSHM = true;
3349
3350 // Don't try UseTransparentHugePages since there are known
3351 // performance issues with it turned on. This might change in the future.
3352 UseTransparentHugePages = false;
3353 }
3354
3355 if (UseTransparentHugePages) {
3356 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3357 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3358 UseHugeTLBFS = false;
3359 UseSHM = false;
3360 return true;
3361 }
3362 UseTransparentHugePages = false;
3363 }
3364
3365 if (UseHugeTLBFS) {
3366 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3367 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3368 UseSHM = false;
3369 return true;
3370 }
3371 UseHugeTLBFS = false;
3372 }
3373
3374 return UseSHM;
3375 }
3376
3377 void os::large_page_init() {
3378 if (!UseLargePages &&
3379 !UseTransparentHugePages &&
3380 !UseHugeTLBFS &&
3381 !UseSHM) {
3382 // Not using large pages.
3383 return;
3384 }
3385
3386 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3387 // The user explicitly turned off large pages.
3388 // Ignore the rest of the large pages flags.
3389 UseTransparentHugePages = false;
3390 UseHugeTLBFS = false;
3391 UseSHM = false;
3392 return;
3393 }
3394
3395 size_t large_page_size = Linux::setup_large_page_size();
3396 UseLargePages = Linux::setup_large_page_type(large_page_size);
3397
3398 set_coredump_filter();
3399 }
3400
3401 #ifndef SHM_HUGETLB
3402 #define SHM_HUGETLB 04000
3403 #endif
3404
3405 #define shm_warning_format(format, ...) \
3406 do { \
3407 if (UseLargePages && \
3408 (!FLAG_IS_DEFAULT(UseLargePages) || \
3409 !FLAG_IS_DEFAULT(UseSHM) || \
3410 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3411 warning(format, __VA_ARGS__); \
3412 } \
3413 } while (0)
3414
3415 #define shm_warning(str) shm_warning_format("%s", str)
3416
3417 #define shm_warning_with_errno(str) \
3418 do { \
3419 int err = errno; \
3420 shm_warning_format(str " (error = %d)", err); \
3421 } while (0)
3422
3423 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3424 assert(is_size_aligned(bytes, alignment), "Must be divisible by the alignment");
3425
3426 if (!is_size_aligned(alignment, SHMLBA)) {
3427 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3428 return NULL;
3429 }
3430
3431 // To ensure that we get 'alignment' aligned memory from shmat,
3432 // we pre-reserve aligned virtual memory and then attach to that.
3433
3434 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3435 if (pre_reserved_addr == NULL) {
3436 // Couldn't pre-reserve aligned memory.
3437 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3438 return NULL;
3439 }
3440
3441 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3442 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3443
3444 if ((intptr_t)addr == -1) {
3445 int err = errno;
3446 shm_warning_with_errno("Failed to attach shared memory.");
3447
3448 assert(err != EACCES, "Unexpected error");
3449 assert(err != EIDRM, "Unexpected error");
3450 assert(err != EINVAL, "Unexpected error");
3451
3452 // Since we don't know if the kernel unmapped the pre-reserved memory area
3453 // we can't unmap it, since that would potentially unmap memory that was
3454 // mapped from other threads.
3455 return NULL;
3456 }
3457
3458 return addr;
3459 }
3460
3461 static char* shmat_at_address(int shmid, char* req_addr) {
3462 if (!is_ptr_aligned(req_addr, SHMLBA)) {
3463 assert(false, "Requested address needs to be SHMLBA aligned");
3464 return NULL;
3465 }
3466
3467 char* addr = (char*)shmat(shmid, req_addr, 0);
3468
3469 if ((intptr_t)addr == -1) {
3470 shm_warning_with_errno("Failed to attach shared memory.");
3471 return NULL;
3472 }
3473
3474 return addr;
3475 }
3476
3477 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3478 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3479 if (req_addr != NULL) {
3480 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3481 assert(is_ptr_aligned(req_addr, alignment), "Must be divisible by given alignment");
3482 return shmat_at_address(shmid, req_addr);
3483 }
3484
3485 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3486 // return large page size aligned memory addresses when req_addr == NULL.
3487 // However, if the alignment is larger than the large page size, we have
3488 // to manually ensure that the memory returned is 'alignment' aligned.
3489 if (alignment > os::large_page_size()) {
3490 assert(is_size_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3491 return shmat_with_alignment(shmid, bytes, alignment);
3492 } else {
3493 return shmat_at_address(shmid, NULL);
3494 }
3495 }
3496
3497 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3498 char* req_addr, bool exec) {
3499 // "exec" is passed in but not used. Creating the shared image for
3500 // the code cache doesn't have an SHM_X executable permission to check.
3501 assert(UseLargePages && UseSHM, "only for SHM large pages");
3502 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3503 assert(is_ptr_aligned(req_addr, alignment), "Unaligned address");
3504
3505 if (!is_size_aligned(bytes, os::large_page_size())) {
3506 return NULL; // Fallback to small pages.
3507 }
3508
3509 // Create a large shared memory region to attach to based on size.
3510 // Currently, size is the total size of the heap.
3511 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3512 if (shmid == -1) {
3513 // Possible reasons for shmget failure:
3514 // 1. shmmax is too small for Java heap.
3515 // > check shmmax value: cat /proc/sys/kernel/shmmax
3516 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3517 // 2. not enough large page memory.
3518 // > check available large pages: cat /proc/meminfo
3519 // > increase amount of large pages:
3520 // echo new_value > /proc/sys/vm/nr_hugepages
3521 // Note 1: different Linux may use different name for this property,
3522 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3523 // Note 2: it's possible there's enough physical memory available but
3524 // they are so fragmented after a long run that they can't
3525 // coalesce into large pages. Try to reserve large pages when
3526 // the system is still "fresh".
3527 shm_warning_with_errno("Failed to reserve shared memory.");
3528 return NULL;
3529 }
3530
3531 // Attach to the region.
3532 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3533
3534 // Remove shmid. If shmat() is successful, the actual shared memory segment
3535 // will be deleted when it's detached by shmdt() or when the process
3536 // terminates. If shmat() is not successful this will remove the shared
3537 // segment immediately.
3538 shmctl(shmid, IPC_RMID, NULL);
3539
3540 return addr;
3541 }
3542
3543 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3544 int error) {
3545 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3546
3547 bool warn_on_failure = UseLargePages &&
3548 (!FLAG_IS_DEFAULT(UseLargePages) ||
3549 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3550 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3551
3552 if (warn_on_failure) {
3553 char msg[128];
3554 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3555 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3556 warning("%s", msg);
3557 }
3558 }
3559
3560 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3561 char* req_addr,
3562 bool exec) {
3563 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3564 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3565 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3566
3567 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3568 char* addr = (char*)::mmap(req_addr, bytes, prot,
3569 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3570 -1, 0);
3571
3572 if (addr == MAP_FAILED) {
3573 warn_on_large_pages_failure(req_addr, bytes, errno);
3574 return NULL;
3575 }
3576
3577 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3578
3579 return addr;
3580 }
3581
3582 // Reserve memory using mmap(MAP_HUGETLB).
3583 // - bytes shall be a multiple of alignment.
3584 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3585 // - alignment sets the alignment at which memory shall be allocated.
3586 // It must be a multiple of allocation granularity.
3587 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3588 // req_addr or NULL.
3589 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3590 size_t alignment,
3591 char* req_addr,
3592 bool exec) {
3593 size_t large_page_size = os::large_page_size();
3594 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3595
3596 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3597 assert(is_size_aligned(bytes, alignment), "Must be");
3598
3599 // First reserve - but not commit - the address range in small pages.
3600 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3601
3602 if (start == NULL) {
3603 return NULL;
3604 }
3605
3606 assert(is_ptr_aligned(start, alignment), "Must be");
3607
3608 char* end = start + bytes;
3609
3610 // Find the regions of the allocated chunk that can be promoted to large pages.
3611 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3612 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3613
3614 size_t lp_bytes = lp_end - lp_start;
3615
3616 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3617
3618 if (lp_bytes == 0) {
3619 // The mapped region doesn't even span the start and the end of a large page.
3620 // Fall back to allocate a non-special area.
3621 ::munmap(start, end - start);
3622 return NULL;
3623 }
3624
3625 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3626
3627 void* result;
3628
3629 // Commit small-paged leading area.
3630 if (start != lp_start) {
3631 result = ::mmap(start, lp_start - start, prot,
3632 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3633 -1, 0);
3634 if (result == MAP_FAILED) {
3635 ::munmap(lp_start, end - lp_start);
3636 return NULL;
3637 }
3638 }
3639
3640 // Commit large-paged area.
3641 result = ::mmap(lp_start, lp_bytes, prot,
3642 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3643 -1, 0);
3644 if (result == MAP_FAILED) {
3645 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3646 // If the mmap above fails, the large pages region will be unmapped and we
3647 // have regions before and after with small pages. Release these regions.
3648 //
3649 // | mapped | unmapped | mapped |
3650 // ^ ^ ^ ^
3651 // start lp_start lp_end end
3652 //
3653 ::munmap(start, lp_start - start);
3654 ::munmap(lp_end, end - lp_end);
3655 return NULL;
3656 }
3657
3658 // Commit small-paged trailing area.
3659 if (lp_end != end) {
3660 result = ::mmap(lp_end, end - lp_end, prot,
3661 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3662 -1, 0);
3663 if (result == MAP_FAILED) {
3664 ::munmap(start, lp_end - start);
3665 return NULL;
3666 }
3667 }
3668
3669 return start;
3670 }
3671
3672 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3673 size_t alignment,
3674 char* req_addr,
3675 bool exec) {
3676 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3677 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3678 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3679 assert(is_power_of_2(os::large_page_size()), "Must be");
3680 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3681
3682 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3683 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3684 } else {
3685 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3686 }
3687 }
3688
3689 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3690 char* req_addr, bool exec) {
3691 assert(UseLargePages, "only for large pages");
3692
3693 char* addr;
3694 if (UseSHM) {
3695 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3696 } else {
3697 assert(UseHugeTLBFS, "must be");
3698 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3699 }
3700
3701 if (addr != NULL) {
3702 if (UseNUMAInterleaving) {
3703 numa_make_global(addr, bytes);
3704 }
3705
3706 // The memory is committed
3707 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3708 }
3709
3710 return addr;
3711 }
3712
3713 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3714 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3715 return shmdt(base) == 0;
3716 }
3717
3718 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3719 return pd_release_memory(base, bytes);
3720 }
3721
3722 bool os::release_memory_special(char* base, size_t bytes) {
3723 bool res;
3724 if (MemTracker::tracking_level() > NMT_minimal) {
3725 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3726 res = os::Linux::release_memory_special_impl(base, bytes);
3727 if (res) {
3728 tkr.record((address)base, bytes);
3729 }
3730
3731 } else {
3732 res = os::Linux::release_memory_special_impl(base, bytes);
3733 }
3734 return res;
3735 }
3736
3737 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3738 assert(UseLargePages, "only for large pages");
3739 bool res;
3740
3741 if (UseSHM) {
3742 res = os::Linux::release_memory_special_shm(base, bytes);
3743 } else {
3744 assert(UseHugeTLBFS, "must be");
3745 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3746 }
3747 return res;
3748 }
3749
3750 size_t os::large_page_size() {
3751 return _large_page_size;
3752 }
3753
3754 // With SysV SHM the entire memory region must be allocated as shared
3755 // memory.
3756 // HugeTLBFS allows application to commit large page memory on demand.
3757 // However, when committing memory with HugeTLBFS fails, the region
3758 // that was supposed to be committed will lose the old reservation
3759 // and allow other threads to steal that memory region. Because of this
3760 // behavior we can't commit HugeTLBFS memory.
3761 bool os::can_commit_large_page_memory() {
3762 return UseTransparentHugePages;
3763 }
3764
3765 bool os::can_execute_large_page_memory() {
3766 return UseTransparentHugePages || UseHugeTLBFS;
3767 }
3768
3769 // Reserve memory at an arbitrary address, only if that area is
3770 // available (and not reserved for something else).
3771
3772 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3773 const int max_tries = 10;
3774 char* base[max_tries];
3775 size_t size[max_tries];
3776 const size_t gap = 0x000000;
3777
3778 // Assert only that the size is a multiple of the page size, since
3779 // that's all that mmap requires, and since that's all we really know
3780 // about at this low abstraction level. If we need higher alignment,
3781 // we can either pass an alignment to this method or verify alignment
3782 // in one of the methods further up the call chain. See bug 5044738.
3783 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3784
3785 // Repeatedly allocate blocks until the block is allocated at the
3786 // right spot.
3787
3788 // Linux mmap allows caller to pass an address as hint; give it a try first,
3789 // if kernel honors the hint then we can return immediately.
3790 char * addr = anon_mmap(requested_addr, bytes, false);
3791 if (addr == requested_addr) {
3792 return requested_addr;
3793 }
3794
3795 if (addr != NULL) {
3796 // mmap() is successful but it fails to reserve at the requested address
3797 anon_munmap(addr, bytes);
3798 }
3799
3800 int i;
3801 for (i = 0; i < max_tries; ++i) {
3802 base[i] = reserve_memory(bytes);
3803
3804 if (base[i] != NULL) {
3805 // Is this the block we wanted?
3806 if (base[i] == requested_addr) {
3807 size[i] = bytes;
3808 break;
3809 }
3810
3811 // Does this overlap the block we wanted? Give back the overlapped
3812 // parts and try again.
3813
3814 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3815 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3816 unmap_memory(base[i], top_overlap);
3817 base[i] += top_overlap;
3818 size[i] = bytes - top_overlap;
3819 } else {
3820 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3821 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3822 unmap_memory(requested_addr, bottom_overlap);
3823 size[i] = bytes - bottom_overlap;
3824 } else {
3825 size[i] = bytes;
3826 }
3827 }
3828 }
3829 }
3830
3831 // Give back the unused reserved pieces.
3832
3833 for (int j = 0; j < i; ++j) {
3834 if (base[j] != NULL) {
3835 unmap_memory(base[j], size[j]);
3836 }
3837 }
3838
3839 if (i < max_tries) {
3840 return requested_addr;
3841 } else {
3842 return NULL;
3843 }
3844 }
3845
3846 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3847 return ::read(fd, buf, nBytes);
3848 }
3849
3850 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3851 return ::pread(fd, buf, nBytes, offset);
3852 }
3853
3854 // Short sleep, direct OS call.
3855 //
3856 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3857 // sched_yield(2) will actually give up the CPU:
3858 //
3859 // * Alone on this pariticular CPU, keeps running.
3860 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3861 // (pre 2.6.39).
3862 //
3863 // So calling this with 0 is an alternative.
3864 //
3865 void os::naked_short_sleep(jlong ms) {
3866 struct timespec req;
3867
3868 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3869 req.tv_sec = 0;
3870 if (ms > 0) {
3871 req.tv_nsec = (ms % 1000) * 1000000;
3872 } else {
3873 req.tv_nsec = 1;
3874 }
3875
3876 nanosleep(&req, NULL);
3877
3878 return;
3879 }
3880
3881 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3882 void os::infinite_sleep() {
3883 while (true) { // sleep forever ...
3884 ::sleep(100); // ... 100 seconds at a time
3885 }
3886 }
3887
3888 // Used to convert frequent JVM_Yield() to nops
3889 bool os::dont_yield() {
3890 return DontYieldALot;
3891 }
3892
3893 void os::naked_yield() {
3894 sched_yield();
3895 }
3896
3897 ////////////////////////////////////////////////////////////////////////////////
3898 // thread priority support
3899
3900 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3901 // only supports dynamic priority, static priority must be zero. For real-time
3902 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3903 // However, for large multi-threaded applications, SCHED_RR is not only slower
3904 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3905 // of 5 runs - Sep 2005).
3906 //
3907 // The following code actually changes the niceness of kernel-thread/LWP. It
3908 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3909 // not the entire user process, and user level threads are 1:1 mapped to kernel
3910 // threads. It has always been the case, but could change in the future. For
3911 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3912 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3913
3914 int os::java_to_os_priority[CriticalPriority + 1] = {
3915 19, // 0 Entry should never be used
3916
3917 4, // 1 MinPriority
3918 3, // 2
3919 2, // 3
3920
3921 1, // 4
3922 0, // 5 NormPriority
3923 -1, // 6
3924
3925 -2, // 7
3926 -3, // 8
3927 -4, // 9 NearMaxPriority
3928
3929 -5, // 10 MaxPriority
3930
3931 -5 // 11 CriticalPriority
3932 };
3933
3934 static int prio_init() {
3935 if (ThreadPriorityPolicy == 1) {
3936 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3937 // if effective uid is not root. Perhaps, a more elegant way of doing
3938 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3939 if (geteuid() != 0) {
3940 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3941 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3942 }
3943 ThreadPriorityPolicy = 0;
3944 }
3945 }
3946 if (UseCriticalJavaThreadPriority) {
3947 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3948 }
3949 return 0;
3950 }
3951
3952 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3953 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3954
3955 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3956 return (ret == 0) ? OS_OK : OS_ERR;
3957 }
3958
3959 OSReturn os::get_native_priority(const Thread* const thread,
3960 int *priority_ptr) {
3961 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3962 *priority_ptr = java_to_os_priority[NormPriority];
3963 return OS_OK;
3964 }
3965
3966 errno = 0;
3967 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3968 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3969 }
3970
3971 // Hint to the underlying OS that a task switch would not be good.
3972 // Void return because it's a hint and can fail.
3973 void os::hint_no_preempt() {}
3974
3975 ////////////////////////////////////////////////////////////////////////////////
3976 // suspend/resume support
3977
3978 // the low-level signal-based suspend/resume support is a remnant from the
3979 // old VM-suspension that used to be for java-suspension, safepoints etc,
3980 // within hotspot. Now there is a single use-case for this:
3981 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3982 // that runs in the watcher thread.
3983 // The remaining code is greatly simplified from the more general suspension
3984 // code that used to be used.
3985 //
3986 // The protocol is quite simple:
3987 // - suspend:
3988 // - sends a signal to the target thread
3989 // - polls the suspend state of the osthread using a yield loop
3990 // - target thread signal handler (SR_handler) sets suspend state
3991 // and blocks in sigsuspend until continued
3992 // - resume:
3993 // - sets target osthread state to continue
3994 // - sends signal to end the sigsuspend loop in the SR_handler
3995 //
3996 // Note that the SR_lock plays no role in this suspend/resume protocol,
3997 // but is checked for NULL in SR_handler as a thread termination indicator.
3998
3999 static void resume_clear_context(OSThread *osthread) {
4000 osthread->set_ucontext(NULL);
4001 osthread->set_siginfo(NULL);
4002 }
4003
4004 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4005 ucontext_t* context) {
4006 osthread->set_ucontext(context);
4007 osthread->set_siginfo(siginfo);
4008 }
4009
4010 // Handler function invoked when a thread's execution is suspended or
4011 // resumed. We have to be careful that only async-safe functions are
4012 // called here (Note: most pthread functions are not async safe and
4013 // should be avoided.)
4014 //
4015 // Note: sigwait() is a more natural fit than sigsuspend() from an
4016 // interface point of view, but sigwait() prevents the signal hander
4017 // from being run. libpthread would get very confused by not having
4018 // its signal handlers run and prevents sigwait()'s use with the
4019 // mutex granting granting signal.
4020 //
4021 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4022 //
4023 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4024 // Save and restore errno to avoid confusing native code with EINTR
4025 // after sigsuspend.
4026 int old_errno = errno;
4027
4028 Thread* thread = Thread::current_or_null_safe();
4029 assert(thread != NULL, "Missing current thread in SR_handler");
4030
4031 // On some systems we have seen signal delivery get "stuck" until the signal
4032 // mask is changed as part of thread termination. Check that the current thread
4033 // has not already terminated (via SR_lock()) - else the following assertion
4034 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4035 // destructor has completed.
4036
4037 if (thread->SR_lock() == NULL) {
4038 return;
4039 }
4040
4041 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4042
4043 OSThread* osthread = thread->osthread();
4044
4045 os::SuspendResume::State current = osthread->sr.state();
4046 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4047 suspend_save_context(osthread, siginfo, context);
4048
4049 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4050 os::SuspendResume::State state = osthread->sr.suspended();
4051 if (state == os::SuspendResume::SR_SUSPENDED) {
4052 sigset_t suspend_set; // signals for sigsuspend()
4053 sigemptyset(&suspend_set);
4054 // get current set of blocked signals and unblock resume signal
4055 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4056 sigdelset(&suspend_set, SR_signum);
4057
4058 sr_semaphore.signal();
4059 // wait here until we are resumed
4060 while (1) {
4061 sigsuspend(&suspend_set);
4062
4063 os::SuspendResume::State result = osthread->sr.running();
4064 if (result == os::SuspendResume::SR_RUNNING) {
4065 sr_semaphore.signal();
4066 break;
4067 }
4068 }
4069
4070 } else if (state == os::SuspendResume::SR_RUNNING) {
4071 // request was cancelled, continue
4072 } else {
4073 ShouldNotReachHere();
4074 }
4075
4076 resume_clear_context(osthread);
4077 } else if (current == os::SuspendResume::SR_RUNNING) {
4078 // request was cancelled, continue
4079 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4080 // ignore
4081 } else {
4082 // ignore
4083 }
4084
4085 errno = old_errno;
4086 }
4087
4088 static int SR_initialize() {
4089 struct sigaction act;
4090 char *s;
4091
4092 // Get signal number to use for suspend/resume
4093 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4094 int sig = ::strtol(s, 0, 10);
4095 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4096 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4097 SR_signum = sig;
4098 } else {
4099 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4100 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4101 }
4102 }
4103
4104 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4105 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4106
4107 sigemptyset(&SR_sigset);
4108 sigaddset(&SR_sigset, SR_signum);
4109
4110 // Set up signal handler for suspend/resume
4111 act.sa_flags = SA_RESTART|SA_SIGINFO;
4112 act.sa_handler = (void (*)(int)) SR_handler;
4113
4114 // SR_signum is blocked by default.
4115 // 4528190 - We also need to block pthread restart signal (32 on all
4116 // supported Linux platforms). Note that LinuxThreads need to block
4117 // this signal for all threads to work properly. So we don't have
4118 // to use hard-coded signal number when setting up the mask.
4119 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4120
4121 if (sigaction(SR_signum, &act, 0) == -1) {
4122 return -1;
4123 }
4124
4125 // Save signal flag
4126 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4127 return 0;
4128 }
4129
4130 static int sr_notify(OSThread* osthread) {
4131 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4132 assert_status(status == 0, status, "pthread_kill");
4133 return status;
4134 }
4135
4136 // "Randomly" selected value for how long we want to spin
4137 // before bailing out on suspending a thread, also how often
4138 // we send a signal to a thread we want to resume
4139 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4140 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4141
4142 // returns true on success and false on error - really an error is fatal
4143 // but this seems the normal response to library errors
4144 static bool do_suspend(OSThread* osthread) {
4145 assert(osthread->sr.is_running(), "thread should be running");
4146 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4147
4148 // mark as suspended and send signal
4149 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4150 // failed to switch, state wasn't running?
4151 ShouldNotReachHere();
4152 return false;
4153 }
4154
4155 if (sr_notify(osthread) != 0) {
4156 ShouldNotReachHere();
4157 }
4158
4159 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4160 while (true) {
4161 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4162 break;
4163 } else {
4164 // timeout
4165 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4166 if (cancelled == os::SuspendResume::SR_RUNNING) {
4167 return false;
4168 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4169 // make sure that we consume the signal on the semaphore as well
4170 sr_semaphore.wait();
4171 break;
4172 } else {
4173 ShouldNotReachHere();
4174 return false;
4175 }
4176 }
4177 }
4178
4179 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4180 return true;
4181 }
4182
4183 static void do_resume(OSThread* osthread) {
4184 assert(osthread->sr.is_suspended(), "thread should be suspended");
4185 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4186
4187 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4188 // failed to switch to WAKEUP_REQUEST
4189 ShouldNotReachHere();
4190 return;
4191 }
4192
4193 while (true) {
4194 if (sr_notify(osthread) == 0) {
4195 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4196 if (osthread->sr.is_running()) {
4197 return;
4198 }
4199 }
4200 } else {
4201 ShouldNotReachHere();
4202 }
4203 }
4204
4205 guarantee(osthread->sr.is_running(), "Must be running!");
4206 }
4207
4208 ///////////////////////////////////////////////////////////////////////////////////
4209 // signal handling (except suspend/resume)
4210
4211 // This routine may be used by user applications as a "hook" to catch signals.
4212 // The user-defined signal handler must pass unrecognized signals to this
4213 // routine, and if it returns true (non-zero), then the signal handler must
4214 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4215 // routine will never retun false (zero), but instead will execute a VM panic
4216 // routine kill the process.
4217 //
4218 // If this routine returns false, it is OK to call it again. This allows
4219 // the user-defined signal handler to perform checks either before or after
4220 // the VM performs its own checks. Naturally, the user code would be making
4221 // a serious error if it tried to handle an exception (such as a null check
4222 // or breakpoint) that the VM was generating for its own correct operation.
4223 //
4224 // This routine may recognize any of the following kinds of signals:
4225 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4226 // It should be consulted by handlers for any of those signals.
4227 //
4228 // The caller of this routine must pass in the three arguments supplied
4229 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4230 // field of the structure passed to sigaction(). This routine assumes that
4231 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4232 //
4233 // Note that the VM will print warnings if it detects conflicting signal
4234 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4235 //
4236 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4237 siginfo_t* siginfo,
4238 void* ucontext,
4239 int abort_if_unrecognized);
4240
4241 void signalHandler(int sig, siginfo_t* info, void* uc) {
4242 assert(info != NULL && uc != NULL, "it must be old kernel");
4243 int orig_errno = errno; // Preserve errno value over signal handler.
4244 JVM_handle_linux_signal(sig, info, uc, true);
4245 errno = orig_errno;
4246 }
4247
4248
4249 // This boolean allows users to forward their own non-matching signals
4250 // to JVM_handle_linux_signal, harmlessly.
4251 bool os::Linux::signal_handlers_are_installed = false;
4252
4253 // For signal-chaining
4254 struct sigaction sigact[NSIG];
4255 uint64_t sigs = 0;
4256 #if (64 < NSIG-1)
4257 #error "Not all signals can be encoded in sigs. Adapt its type!"
4258 #endif
4259 bool os::Linux::libjsig_is_loaded = false;
4260 typedef struct sigaction *(*get_signal_t)(int);
4261 get_signal_t os::Linux::get_signal_action = NULL;
4262
4263 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4264 struct sigaction *actp = NULL;
4265
4266 if (libjsig_is_loaded) {
4267 // Retrieve the old signal handler from libjsig
4268 actp = (*get_signal_action)(sig);
4269 }
4270 if (actp == NULL) {
4271 // Retrieve the preinstalled signal handler from jvm
4272 actp = get_preinstalled_handler(sig);
4273 }
4274
4275 return actp;
4276 }
4277
4278 static bool call_chained_handler(struct sigaction *actp, int sig,
4279 siginfo_t *siginfo, void *context) {
4280 // Call the old signal handler
4281 if (actp->sa_handler == SIG_DFL) {
4282 // It's more reasonable to let jvm treat it as an unexpected exception
4283 // instead of taking the default action.
4284 return false;
4285 } else if (actp->sa_handler != SIG_IGN) {
4286 if ((actp->sa_flags & SA_NODEFER) == 0) {
4287 // automaticlly block the signal
4288 sigaddset(&(actp->sa_mask), sig);
4289 }
4290
4291 sa_handler_t hand = NULL;
4292 sa_sigaction_t sa = NULL;
4293 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4294 // retrieve the chained handler
4295 if (siginfo_flag_set) {
4296 sa = actp->sa_sigaction;
4297 } else {
4298 hand = actp->sa_handler;
4299 }
4300
4301 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4302 actp->sa_handler = SIG_DFL;
4303 }
4304
4305 // try to honor the signal mask
4306 sigset_t oset;
4307 sigemptyset(&oset);
4308 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4309
4310 // call into the chained handler
4311 if (siginfo_flag_set) {
4312 (*sa)(sig, siginfo, context);
4313 } else {
4314 (*hand)(sig);
4315 }
4316
4317 // restore the signal mask
4318 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4319 }
4320 // Tell jvm's signal handler the signal is taken care of.
4321 return true;
4322 }
4323
4324 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4325 bool chained = false;
4326 // signal-chaining
4327 if (UseSignalChaining) {
4328 struct sigaction *actp = get_chained_signal_action(sig);
4329 if (actp != NULL) {
4330 chained = call_chained_handler(actp, sig, siginfo, context);
4331 }
4332 }
4333 return chained;
4334 }
4335
4336 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4337 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4338 return &sigact[sig];
4339 }
4340 return NULL;
4341 }
4342
4343 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4344 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4345 sigact[sig] = oldAct;
4346 sigs |= (uint64_t)1 << (sig-1);
4347 }
4348
4349 // for diagnostic
4350 int sigflags[NSIG];
4351
4352 int os::Linux::get_our_sigflags(int sig) {
4353 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4354 return sigflags[sig];
4355 }
4356
4357 void os::Linux::set_our_sigflags(int sig, int flags) {
4358 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4359 if (sig > 0 && sig < NSIG) {
4360 sigflags[sig] = flags;
4361 }
4362 }
4363
4364 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4365 // Check for overwrite.
4366 struct sigaction oldAct;
4367 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4368
4369 void* oldhand = oldAct.sa_sigaction
4370 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4371 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4372 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4373 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4374 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4375 if (AllowUserSignalHandlers || !set_installed) {
4376 // Do not overwrite; user takes responsibility to forward to us.
4377 return;
4378 } else if (UseSignalChaining) {
4379 // save the old handler in jvm
4380 save_preinstalled_handler(sig, oldAct);
4381 // libjsig also interposes the sigaction() call below and saves the
4382 // old sigaction on it own.
4383 } else {
4384 fatal("Encountered unexpected pre-existing sigaction handler "
4385 "%#lx for signal %d.", (long)oldhand, sig);
4386 }
4387 }
4388
4389 struct sigaction sigAct;
4390 sigfillset(&(sigAct.sa_mask));
4391 sigAct.sa_handler = SIG_DFL;
4392 if (!set_installed) {
4393 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4394 } else {
4395 sigAct.sa_sigaction = signalHandler;
4396 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4397 }
4398 // Save flags, which are set by ours
4399 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4400 sigflags[sig] = sigAct.sa_flags;
4401
4402 int ret = sigaction(sig, &sigAct, &oldAct);
4403 assert(ret == 0, "check");
4404
4405 void* oldhand2 = oldAct.sa_sigaction
4406 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4407 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4408 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4409 }
4410
4411 // install signal handlers for signals that HotSpot needs to
4412 // handle in order to support Java-level exception handling.
4413
4414 void os::Linux::install_signal_handlers() {
4415 if (!signal_handlers_are_installed) {
4416 signal_handlers_are_installed = true;
4417
4418 // signal-chaining
4419 typedef void (*signal_setting_t)();
4420 signal_setting_t begin_signal_setting = NULL;
4421 signal_setting_t end_signal_setting = NULL;
4422 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4423 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4424 if (begin_signal_setting != NULL) {
4425 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4426 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4427 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4428 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4429 libjsig_is_loaded = true;
4430 assert(UseSignalChaining, "should enable signal-chaining");
4431 }
4432 if (libjsig_is_loaded) {
4433 // Tell libjsig jvm is setting signal handlers
4434 (*begin_signal_setting)();
4435 }
4436
4437 set_signal_handler(SIGSEGV, true);
4438 set_signal_handler(SIGPIPE, true);
4439 set_signal_handler(SIGBUS, true);
4440 set_signal_handler(SIGILL, true);
4441 set_signal_handler(SIGFPE, true);
4442 #if defined(PPC64)
4443 set_signal_handler(SIGTRAP, true);
4444 #endif
4445 set_signal_handler(SIGXFSZ, true);
4446
4447 if (libjsig_is_loaded) {
4448 // Tell libjsig jvm finishes setting signal handlers
4449 (*end_signal_setting)();
4450 }
4451
4452 // We don't activate signal checker if libjsig is in place, we trust ourselves
4453 // and if UserSignalHandler is installed all bets are off.
4454 // Log that signal checking is off only if -verbose:jni is specified.
4455 if (CheckJNICalls) {
4456 if (libjsig_is_loaded) {
4457 if (PrintJNIResolving) {
4458 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4459 }
4460 check_signals = false;
4461 }
4462 if (AllowUserSignalHandlers) {
4463 if (PrintJNIResolving) {
4464 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4465 }
4466 check_signals = false;
4467 }
4468 }
4469 }
4470 }
4471
4472 // This is the fastest way to get thread cpu time on Linux.
4473 // Returns cpu time (user+sys) for any thread, not only for current.
4474 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4475 // It might work on 2.6.10+ with a special kernel/glibc patch.
4476 // For reference, please, see IEEE Std 1003.1-2004:
4477 // http://www.unix.org/single_unix_specification
4478
4479 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4480 struct timespec tp;
4481 int rc = os::Linux::clock_gettime(clockid, &tp);
4482 assert(rc == 0, "clock_gettime is expected to return 0 code");
4483
4484 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4485 }
4486
4487 void os::Linux::initialize_os_info() {
4488 assert(_os_version == 0, "OS info already initialized");
4489
4490 struct utsname _uname;
4491
4492 uint32_t major;
4493 uint32_t minor;
4494 uint32_t fix;
4495
4496 int rc;
4497
4498 // Kernel version is unknown if
4499 // verification below fails.
4500 _os_version = 0x01000000;
4501
4502 rc = uname(&_uname);
4503 if (rc != -1) {
4504
4505 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4506 if (rc == 3) {
4507
4508 if (major < 256 && minor < 256 && fix < 256) {
4509 // Kernel version format is as expected,
4510 // set it overriding unknown state.
4511 _os_version = (major << 16) |
4512 (minor << 8 ) |
4513 (fix << 0 ) ;
4514 }
4515 }
4516 }
4517 }
4518
4519 uint32_t os::Linux::os_version() {
4520 assert(_os_version != 0, "not initialized");
4521 return _os_version & 0x00FFFFFF;
4522 }
4523
4524 bool os::Linux::os_version_is_known() {
4525 assert(_os_version != 0, "not initialized");
4526 return _os_version & 0x01000000 ? false : true;
4527 }
4528
4529 /////
4530 // glibc on Linux platform uses non-documented flag
4531 // to indicate, that some special sort of signal
4532 // trampoline is used.
4533 // We will never set this flag, and we should
4534 // ignore this flag in our diagnostic
4535 #ifdef SIGNIFICANT_SIGNAL_MASK
4536 #undef SIGNIFICANT_SIGNAL_MASK
4537 #endif
4538 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4539
4540 static const char* get_signal_handler_name(address handler,
4541 char* buf, int buflen) {
4542 int offset = 0;
4543 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4544 if (found) {
4545 // skip directory names
4546 const char *p1, *p2;
4547 p1 = buf;
4548 size_t len = strlen(os::file_separator());
4549 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4550 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4551 } else {
4552 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4553 }
4554 return buf;
4555 }
4556
4557 static void print_signal_handler(outputStream* st, int sig,
4558 char* buf, size_t buflen) {
4559 struct sigaction sa;
4560
4561 sigaction(sig, NULL, &sa);
4562
4563 // See comment for SIGNIFICANT_SIGNAL_MASK define
4564 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4565
4566 st->print("%s: ", os::exception_name(sig, buf, buflen));
4567
4568 address handler = (sa.sa_flags & SA_SIGINFO)
4569 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4570 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4571
4572 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4573 st->print("SIG_DFL");
4574 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4575 st->print("SIG_IGN");
4576 } else {
4577 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4578 }
4579
4580 st->print(", sa_mask[0]=");
4581 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4582
4583 address rh = VMError::get_resetted_sighandler(sig);
4584 // May be, handler was resetted by VMError?
4585 if (rh != NULL) {
4586 handler = rh;
4587 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4588 }
4589
4590 st->print(", sa_flags=");
4591 os::Posix::print_sa_flags(st, sa.sa_flags);
4592
4593 // Check: is it our handler?
4594 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4595 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4596 // It is our signal handler
4597 // check for flags, reset system-used one!
4598 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4599 st->print(
4600 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4601 os::Linux::get_our_sigflags(sig));
4602 }
4603 }
4604 st->cr();
4605 }
4606
4607
4608 #define DO_SIGNAL_CHECK(sig) \
4609 do { \
4610 if (!sigismember(&check_signal_done, sig)) { \
4611 os::Linux::check_signal_handler(sig); \
4612 } \
4613 } while (0)
4614
4615 // This method is a periodic task to check for misbehaving JNI applications
4616 // under CheckJNI, we can add any periodic checks here
4617
4618 void os::run_periodic_checks() {
4619 if (check_signals == false) return;
4620
4621 // SEGV and BUS if overridden could potentially prevent
4622 // generation of hs*.log in the event of a crash, debugging
4623 // such a case can be very challenging, so we absolutely
4624 // check the following for a good measure:
4625 DO_SIGNAL_CHECK(SIGSEGV);
4626 DO_SIGNAL_CHECK(SIGILL);
4627 DO_SIGNAL_CHECK(SIGFPE);
4628 DO_SIGNAL_CHECK(SIGBUS);
4629 DO_SIGNAL_CHECK(SIGPIPE);
4630 DO_SIGNAL_CHECK(SIGXFSZ);
4631 #if defined(PPC64)
4632 DO_SIGNAL_CHECK(SIGTRAP);
4633 #endif
4634
4635 // ReduceSignalUsage allows the user to override these handlers
4636 // see comments at the very top and jvm_solaris.h
4637 if (!ReduceSignalUsage) {
4638 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4639 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4640 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4641 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4642 }
4643
4644 DO_SIGNAL_CHECK(SR_signum);
4645 }
4646
4647 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4648
4649 static os_sigaction_t os_sigaction = NULL;
4650
4651 void os::Linux::check_signal_handler(int sig) {
4652 char buf[O_BUFLEN];
4653 address jvmHandler = NULL;
4654
4655
4656 struct sigaction act;
4657 if (os_sigaction == NULL) {
4658 // only trust the default sigaction, in case it has been interposed
4659 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4660 if (os_sigaction == NULL) return;
4661 }
4662
4663 os_sigaction(sig, (struct sigaction*)NULL, &act);
4664
4665
4666 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4667
4668 address thisHandler = (act.sa_flags & SA_SIGINFO)
4669 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4670 : CAST_FROM_FN_PTR(address, act.sa_handler);
4671
4672
4673 switch (sig) {
4674 case SIGSEGV:
4675 case SIGBUS:
4676 case SIGFPE:
4677 case SIGPIPE:
4678 case SIGILL:
4679 case SIGXFSZ:
4680 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4681 break;
4682
4683 case SHUTDOWN1_SIGNAL:
4684 case SHUTDOWN2_SIGNAL:
4685 case SHUTDOWN3_SIGNAL:
4686 case BREAK_SIGNAL:
4687 jvmHandler = (address)user_handler();
4688 break;
4689
4690 default:
4691 if (sig == SR_signum) {
4692 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4693 } else {
4694 return;
4695 }
4696 break;
4697 }
4698
4699 if (thisHandler != jvmHandler) {
4700 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4701 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4702 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4703 // No need to check this sig any longer
4704 sigaddset(&check_signal_done, sig);
4705 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4706 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4707 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4708 exception_name(sig, buf, O_BUFLEN));
4709 }
4710 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4711 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4712 tty->print("expected:");
4713 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4714 tty->cr();
4715 tty->print(" found:");
4716 os::Posix::print_sa_flags(tty, act.sa_flags);
4717 tty->cr();
4718 // No need to check this sig any longer
4719 sigaddset(&check_signal_done, sig);
4720 }
4721
4722 // Dump all the signal
4723 if (sigismember(&check_signal_done, sig)) {
4724 print_signal_handlers(tty, buf, O_BUFLEN);
4725 }
4726 }
4727
4728 extern void report_error(char* file_name, int line_no, char* title,
4729 char* format, ...);
4730
4731 // Some linux distributions (notably: Alpine Linux) include the
4732 // grsecurity in the kernel by default. Of particular interest from a
4733 // JVM perspective is PaX (https://pax.grsecurity.net/), which adds
4734 // some security features related to page attributes. Specifically,
4735 // the MPROTECT PaX functionality
4736 // (https://pax.grsecurity.net/docs/mprotect.txt) prevents dynamic
4737 // code generation by disallowing a (previously) writable page to be
4738 // marked as executable. This is, of course, exactly what HotSpot does
4739 // for both JIT compiled method, as well as for stubs, adapters, etc.
4740 //
4741 // Instead of crashing "lazily" when trying to make a page executable,
4742 // this code probes for the presence of PaX and reports the failure
4743 // eagerly.
4744 static void check_pax(void) {
4745 // Zero doesn't generate code dynamically, so no need to perform the PaX check
4746 #ifndef ZERO
4747 size_t size = os::Linux::page_size();
4748
4749 void* p = ::mmap(NULL, size, PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4750 if (p == MAP_FAILED) {
4751 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "failed to allocate memory for PaX check.");
4752 }
4753
4754 int res = ::mprotect(p, size, PROT_WRITE|PROT_EXEC);
4755 if (res == -1) {
4756 vm_exit_during_initialization("Failed to mark memory page as executable",
4757 "Please check if grsecurity/PaX is enabled in your kernel.\n"
4758 "\n"
4759 "For example, you can do this by running (note: you may need root privileges):\n"
4760 "\n"
4761 " sysctl kernel.pax.softmode\n"
4762 "\n"
4763 "If PaX is included in the kernel you will see something like this:\n"
4764 "\n"
4765 " kernel.pax.softmode = 0\n"
4766 "\n"
4767 "In particular, if the value is 0 (zero), then PaX is enabled.\n"
4768 "\n"
4769 "PaX includes security functionality which interferes with the dynamic code\n"
4770 "generation the JVM relies on. Specifically, the MPROTECT functionality as\n"
4771 "described on https://pax.grsecurity.net/docs/mprotect.txt is not compatible\n"
4772 "with the JVM. If you want to allow the JVM to run you will have to disable PaX.\n"
4773 "You can do this on a per-executable basis using the paxctl tool, for example:\n"
4774 "\n"
4775 " paxctl -cm bin/java\n"
4776 "\n"
4777 "Please note that this modifies the executable binary in-place, so may want\n"
4778 "to make a backup of it first. Also note that you have to repeat this for other\n"
4779 "executables like javac, jar, jcmd, etc.\n"
4780 );
4781
4782 }
4783
4784 ::munmap(p, size);
4785 #endif
4786 }
4787
4788 // this is called _before_ the most of global arguments have been parsed
4789 void os::init(void) {
4790 char dummy; // used to get a guess on initial stack address
4791 // first_hrtime = gethrtime();
4792
4793 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4794
4795 init_random(1234567);
4796
4797 ThreadCritical::initialize();
4798
4799 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4800 if (Linux::page_size() == -1) {
4801 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4802 os::strerror(errno));
4803 }
4804 init_page_sizes((size_t) Linux::page_size());
4805
4806 Linux::initialize_system_info();
4807
4808 Linux::initialize_os_info();
4809
4810 // main_thread points to the aboriginal thread
4811 Linux::_main_thread = pthread_self();
4812
4813 Linux::clock_init();
4814 initial_time_count = javaTimeNanos();
4815
4816 // pthread_condattr initialization for monotonic clock
4817 int status;
4818 pthread_condattr_t* _condattr = os::Linux::condAttr();
4819 if ((status = pthread_condattr_init(_condattr)) != 0) {
4820 fatal("pthread_condattr_init: %s", os::strerror(status));
4821 }
4822 // Only set the clock if CLOCK_MONOTONIC is available
4823 if (os::supports_monotonic_clock()) {
4824 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4825 if (status == EINVAL) {
4826 warning("Unable to use monotonic clock with relative timed-waits" \
4827 " - changes to the time-of-day clock may have adverse affects");
4828 } else {
4829 fatal("pthread_condattr_setclock: %s", os::strerror(status));
4830 }
4831 }
4832 }
4833 // else it defaults to CLOCK_REALTIME
4834
4835 // retrieve entry point for pthread_setname_np
4836 Linux::_pthread_setname_np =
4837 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4838
4839 check_pax();
4840 }
4841
4842 // To install functions for atexit system call
4843 extern "C" {
4844 static void perfMemory_exit_helper() {
4845 perfMemory_exit();
4846 }
4847 }
4848
4849 // this is called _after_ the global arguments have been parsed
4850 jint os::init_2(void) {
4851 Linux::fast_thread_clock_init();
4852
4853 // Allocate a single page and mark it as readable for safepoint polling
4854 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4855 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4856
4857 os::set_polling_page(polling_page);
4858 log_info(os)("SafePoint Polling address: " INTPTR_FORMAT, p2i(polling_page));
4859
4860 if (!UseMembar) {
4861 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4862 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4863 os::set_memory_serialize_page(mem_serialize_page);
4864 log_info(os)("Memory Serialize Page address: " INTPTR_FORMAT, p2i(mem_serialize_page));
4865 }
4866
4867 // initialize suspend/resume support - must do this before signal_sets_init()
4868 if (SR_initialize() != 0) {
4869 perror("SR_initialize failed");
4870 return JNI_ERR;
4871 }
4872
4873 Linux::signal_sets_init();
4874 Linux::install_signal_handlers();
4875
4876 // Check and sets minimum stack sizes against command line options
4877 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4878 return JNI_ERR;
4879 }
4880 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4881
4882 #if defined(IA32)
4883 workaround_expand_exec_shield_cs_limit();
4884 #endif
4885
4886 Linux::libpthread_init();
4887 log_info(os)("HotSpot is running with %s, %s",
4888 Linux::glibc_version(), Linux::libpthread_version());
4889
4890 if (UseNUMA) {
4891 if (!Linux::libnuma_init()) {
4892 UseNUMA = false;
4893 } else {
4894 if ((Linux::numa_max_node() < 1)) {
4895 // There's only one node(they start from 0), disable NUMA.
4896 UseNUMA = false;
4897 }
4898 }
4899 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4900 // we can make the adaptive lgrp chunk resizing work. If the user specified
4901 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4902 // disable adaptive resizing.
4903 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4904 if (FLAG_IS_DEFAULT(UseNUMA)) {
4905 UseNUMA = false;
4906 } else {
4907 if (FLAG_IS_DEFAULT(UseLargePages) &&
4908 FLAG_IS_DEFAULT(UseSHM) &&
4909 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4910 UseLargePages = false;
4911 } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4912 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4913 UseAdaptiveSizePolicy = false;
4914 UseAdaptiveNUMAChunkSizing = false;
4915 }
4916 }
4917 }
4918 if (!UseNUMA && ForceNUMA) {
4919 UseNUMA = true;
4920 }
4921 }
4922
4923 if (MaxFDLimit) {
4924 // set the number of file descriptors to max. print out error
4925 // if getrlimit/setrlimit fails but continue regardless.
4926 struct rlimit nbr_files;
4927 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4928 if (status != 0) {
4929 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4930 } else {
4931 nbr_files.rlim_cur = nbr_files.rlim_max;
4932 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4933 if (status != 0) {
4934 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4935 }
4936 }
4937 }
4938
4939 // Initialize lock used to serialize thread creation (see os::create_thread)
4940 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4941
4942 // at-exit methods are called in the reverse order of their registration.
4943 // atexit functions are called on return from main or as a result of a
4944 // call to exit(3C). There can be only 32 of these functions registered
4945 // and atexit() does not set errno.
4946
4947 if (PerfAllowAtExitRegistration) {
4948 // only register atexit functions if PerfAllowAtExitRegistration is set.
4949 // atexit functions can be delayed until process exit time, which
4950 // can be problematic for embedded VM situations. Embedded VMs should
4951 // call DestroyJavaVM() to assure that VM resources are released.
4952
4953 // note: perfMemory_exit_helper atexit function may be removed in
4954 // the future if the appropriate cleanup code can be added to the
4955 // VM_Exit VMOperation's doit method.
4956 if (atexit(perfMemory_exit_helper) != 0) {
4957 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4958 }
4959 }
4960
4961 // initialize thread priority policy
4962 prio_init();
4963
4964 return JNI_OK;
4965 }
4966
4967 // Mark the polling page as unreadable
4968 void os::make_polling_page_unreadable(void) {
4969 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4970 fatal("Could not disable polling page");
4971 }
4972 }
4973
4974 // Mark the polling page as readable
4975 void os::make_polling_page_readable(void) {
4976 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4977 fatal("Could not enable polling page");
4978 }
4979 }
4980
4981 // older glibc versions don't have this macro (which expands to
4982 // an optimized bit-counting function) so we have to roll our own
4983 #ifndef CPU_COUNT
4984
4985 static int _cpu_count(const cpu_set_t* cpus) {
4986 int count = 0;
4987 // only look up to the number of configured processors
4988 for (int i = 0; i < os::processor_count(); i++) {
4989 if (CPU_ISSET(i, cpus)) {
4990 count++;
4991 }
4992 }
4993 return count;
4994 }
4995
4996 #define CPU_COUNT(cpus) _cpu_count(cpus)
4997
4998 #endif // CPU_COUNT
4999
5000 // Get the current number of available processors for this process.
5001 // This value can change at any time during a process's lifetime.
5002 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5003 // If it appears there may be more than 1024 processors then we do a
5004 // dynamic check - see 6515172 for details.
5005 // If anything goes wrong we fallback to returning the number of online
5006 // processors - which can be greater than the number available to the process.
5007 int os::active_processor_count() {
5008 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5009 cpu_set_t* cpus_p = &cpus;
5010 int cpus_size = sizeof(cpu_set_t);
5011
5012 int configured_cpus = processor_count(); // upper bound on available cpus
5013 int cpu_count = 0;
5014
5015 // old build platforms may not support dynamic cpu sets
5016 #ifdef CPU_ALLOC
5017
5018 // To enable easy testing of the dynamic path on different platforms we
5019 // introduce a diagnostic flag: UseCpuAllocPath
5020 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5021 // kernel may use a mask bigger than cpu_set_t
5022 log_trace(os)("active_processor_count: using dynamic path %s"
5023 "- configured processors: %d",
5024 UseCpuAllocPath ? "(forced) " : "",
5025 configured_cpus);
5026 cpus_p = CPU_ALLOC(configured_cpus);
5027 if (cpus_p != NULL) {
5028 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5029 // zero it just to be safe
5030 CPU_ZERO_S(cpus_size, cpus_p);
5031 }
5032 else {
5033 // failed to allocate so fallback to online cpus
5034 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5035 log_trace(os)("active_processor_count: "
5036 "CPU_ALLOC failed (%s) - using "
5037 "online processor count: %d",
5038 os::strerror(errno), online_cpus);
5039 return online_cpus;
5040 }
5041 }
5042 else {
5043 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5044 configured_cpus);
5045 }
5046 #else // CPU_ALLOC
5047 // these stubs won't be executed
5048 #define CPU_COUNT_S(size, cpus) -1
5049 #define CPU_FREE(cpus)
5050
5051 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5052 configured_cpus);
5053 #endif // CPU_ALLOC
5054
5055 // pid 0 means the current thread - which we have to assume represents the process
5056 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5057 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5058 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5059 }
5060 else {
5061 cpu_count = CPU_COUNT(cpus_p);
5062 }
5063 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5064 }
5065 else {
5066 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5067 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5068 "which may exceed available processors", os::strerror(errno), cpu_count);
5069 }
5070
5071 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5072 CPU_FREE(cpus_p);
5073 }
5074
5075 assert(cpu_count > 0 && cpu_count <= processor_count(), "sanity check");
5076 return cpu_count;
5077 }
5078
5079 void os::set_native_thread_name(const char *name) {
5080 if (Linux::_pthread_setname_np) {
5081 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5082 snprintf(buf, sizeof(buf), "%s", name);
5083 buf[sizeof(buf) - 1] = '\0';
5084 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5085 // ERANGE should not happen; all other errors should just be ignored.
5086 assert(rc != ERANGE, "pthread_setname_np failed");
5087 }
5088 }
5089
5090 bool os::distribute_processes(uint length, uint* distribution) {
5091 // Not yet implemented.
5092 return false;
5093 }
5094
5095 bool os::bind_to_processor(uint processor_id) {
5096 // Not yet implemented.
5097 return false;
5098 }
5099
5100 ///
5101
5102 void os::SuspendedThreadTask::internal_do_task() {
5103 if (do_suspend(_thread->osthread())) {
5104 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5105 do_task(context);
5106 do_resume(_thread->osthread());
5107 }
5108 }
5109
5110 class PcFetcher : public os::SuspendedThreadTask {
5111 public:
5112 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5113 ExtendedPC result();
5114 protected:
5115 void do_task(const os::SuspendedThreadTaskContext& context);
5116 private:
5117 ExtendedPC _epc;
5118 };
5119
5120 ExtendedPC PcFetcher::result() {
5121 guarantee(is_done(), "task is not done yet.");
5122 return _epc;
5123 }
5124
5125 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5126 Thread* thread = context.thread();
5127 OSThread* osthread = thread->osthread();
5128 if (osthread->ucontext() != NULL) {
5129 _epc = os::Linux::ucontext_get_pc((const ucontext_t *) context.ucontext());
5130 } else {
5131 // NULL context is unexpected, double-check this is the VMThread
5132 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5133 }
5134 }
5135
5136 // Suspends the target using the signal mechanism and then grabs the PC before
5137 // resuming the target. Used by the flat-profiler only
5138 ExtendedPC os::get_thread_pc(Thread* thread) {
5139 // Make sure that it is called by the watcher for the VMThread
5140 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5141 assert(thread->is_VM_thread(), "Can only be called for VMThread");
5142
5143 PcFetcher fetcher(thread);
5144 fetcher.run();
5145 return fetcher.result();
5146 }
5147
5148 ////////////////////////////////////////////////////////////////////////////////
5149 // debug support
5150
5151 bool os::find(address addr, outputStream* st) {
5152 Dl_info dlinfo;
5153 memset(&dlinfo, 0, sizeof(dlinfo));
5154 if (dladdr(addr, &dlinfo) != 0) {
5155 st->print(PTR_FORMAT ": ", p2i(addr));
5156 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5157 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5158 p2i(addr) - p2i(dlinfo.dli_saddr));
5159 } else if (dlinfo.dli_fbase != NULL) {
5160 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5161 } else {
5162 st->print("<absolute address>");
5163 }
5164 if (dlinfo.dli_fname != NULL) {
5165 st->print(" in %s", dlinfo.dli_fname);
5166 }
5167 if (dlinfo.dli_fbase != NULL) {
5168 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5169 }
5170 st->cr();
5171
5172 if (Verbose) {
5173 // decode some bytes around the PC
5174 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5175 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5176 address lowest = (address) dlinfo.dli_sname;
5177 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5178 if (begin < lowest) begin = lowest;
5179 Dl_info dlinfo2;
5180 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5181 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5182 end = (address) dlinfo2.dli_saddr;
5183 }
5184 Disassembler::decode(begin, end, st);
5185 }
5186 return true;
5187 }
5188 return false;
5189 }
5190
5191 ////////////////////////////////////////////////////////////////////////////////
5192 // misc
5193
5194 // This does not do anything on Linux. This is basically a hook for being
5195 // able to use structured exception handling (thread-local exception filters)
5196 // on, e.g., Win32.
5197 void
5198 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5199 JavaCallArguments* args, Thread* thread) {
5200 f(value, method, args, thread);
5201 }
5202
5203 void os::print_statistics() {
5204 }
5205
5206 bool os::message_box(const char* title, const char* message) {
5207 int i;
5208 fdStream err(defaultStream::error_fd());
5209 for (i = 0; i < 78; i++) err.print_raw("=");
5210 err.cr();
5211 err.print_raw_cr(title);
5212 for (i = 0; i < 78; i++) err.print_raw("-");
5213 err.cr();
5214 err.print_raw_cr(message);
5215 for (i = 0; i < 78; i++) err.print_raw("=");
5216 err.cr();
5217
5218 char buf[16];
5219 // Prevent process from exiting upon "read error" without consuming all CPU
5220 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5221
5222 return buf[0] == 'y' || buf[0] == 'Y';
5223 }
5224
5225 int os::stat(const char *path, struct stat *sbuf) {
5226 char pathbuf[MAX_PATH];
5227 if (strlen(path) > MAX_PATH - 1) {
5228 errno = ENAMETOOLONG;
5229 return -1;
5230 }
5231 os::native_path(strcpy(pathbuf, path));
5232 return ::stat(pathbuf, sbuf);
5233 }
5234
5235 // Is a (classpath) directory empty?
5236 bool os::dir_is_empty(const char* path) {
5237 DIR *dir = NULL;
5238 struct dirent *ptr;
5239
5240 dir = opendir(path);
5241 if (dir == NULL) return true;
5242
5243 // Scan the directory
5244 bool result = true;
5245 char buf[sizeof(struct dirent) + MAX_PATH];
5246 while (result && (ptr = ::readdir(dir)) != NULL) {
5247 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5248 result = false;
5249 }
5250 }
5251 closedir(dir);
5252 return result;
5253 }
5254
5255 // This code originates from JDK's sysOpen and open64_w
5256 // from src/solaris/hpi/src/system_md.c
5257
5258 int os::open(const char *path, int oflag, int mode) {
5259 if (strlen(path) > MAX_PATH - 1) {
5260 errno = ENAMETOOLONG;
5261 return -1;
5262 }
5263
5264 // All file descriptors that are opened in the Java process and not
5265 // specifically destined for a subprocess should have the close-on-exec
5266 // flag set. If we don't set it, then careless 3rd party native code
5267 // might fork and exec without closing all appropriate file descriptors
5268 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5269 // turn might:
5270 //
5271 // - cause end-of-file to fail to be detected on some file
5272 // descriptors, resulting in mysterious hangs, or
5273 //
5274 // - might cause an fopen in the subprocess to fail on a system
5275 // suffering from bug 1085341.
5276 //
5277 // (Yes, the default setting of the close-on-exec flag is a Unix
5278 // design flaw)
5279 //
5280 // See:
5281 // 1085341: 32-bit stdio routines should support file descriptors >255
5282 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5283 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5284 //
5285 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5286 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5287 // because it saves a system call and removes a small window where the flag
5288 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5289 // and we fall back to using FD_CLOEXEC (see below).
5290 #ifdef O_CLOEXEC
5291 oflag |= O_CLOEXEC;
5292 #endif
5293
5294 int fd = ::open64(path, oflag, mode);
5295 if (fd == -1) return -1;
5296
5297 //If the open succeeded, the file might still be a directory
5298 {
5299 struct stat64 buf64;
5300 int ret = ::fstat64(fd, &buf64);
5301 int st_mode = buf64.st_mode;
5302
5303 if (ret != -1) {
5304 if ((st_mode & S_IFMT) == S_IFDIR) {
5305 errno = EISDIR;
5306 ::close(fd);
5307 return -1;
5308 }
5309 } else {
5310 ::close(fd);
5311 return -1;
5312 }
5313 }
5314
5315 #ifdef FD_CLOEXEC
5316 // Validate that the use of the O_CLOEXEC flag on open above worked.
5317 // With recent kernels, we will perform this check exactly once.
5318 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5319 if (!O_CLOEXEC_is_known_to_work) {
5320 int flags = ::fcntl(fd, F_GETFD);
5321 if (flags != -1) {
5322 if ((flags & FD_CLOEXEC) != 0)
5323 O_CLOEXEC_is_known_to_work = 1;
5324 else
5325 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5326 }
5327 }
5328 #endif
5329
5330 return fd;
5331 }
5332
5333
5334 // create binary file, rewriting existing file if required
5335 int os::create_binary_file(const char* path, bool rewrite_existing) {
5336 int oflags = O_WRONLY | O_CREAT;
5337 if (!rewrite_existing) {
5338 oflags |= O_EXCL;
5339 }
5340 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5341 }
5342
5343 // return current position of file pointer
5344 jlong os::current_file_offset(int fd) {
5345 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5346 }
5347
5348 // move file pointer to the specified offset
5349 jlong os::seek_to_file_offset(int fd, jlong offset) {
5350 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5351 }
5352
5353 // This code originates from JDK's sysAvailable
5354 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5355
5356 int os::available(int fd, jlong *bytes) {
5357 jlong cur, end;
5358 int mode;
5359 struct stat64 buf64;
5360
5361 if (::fstat64(fd, &buf64) >= 0) {
5362 mode = buf64.st_mode;
5363 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5364 int n;
5365 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5366 *bytes = n;
5367 return 1;
5368 }
5369 }
5370 }
5371 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5372 return 0;
5373 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5374 return 0;
5375 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5376 return 0;
5377 }
5378 *bytes = end - cur;
5379 return 1;
5380 }
5381
5382 // Map a block of memory.
5383 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5384 char *addr, size_t bytes, bool read_only,
5385 bool allow_exec) {
5386 int prot;
5387 int flags = MAP_PRIVATE;
5388
5389 if (read_only) {
5390 prot = PROT_READ;
5391 } else {
5392 prot = PROT_READ | PROT_WRITE;
5393 }
5394
5395 if (allow_exec) {
5396 prot |= PROT_EXEC;
5397 }
5398
5399 if (addr != NULL) {
5400 flags |= MAP_FIXED;
5401 }
5402
5403 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5404 fd, file_offset);
5405 if (mapped_address == MAP_FAILED) {
5406 return NULL;
5407 }
5408 return mapped_address;
5409 }
5410
5411
5412 // Remap a block of memory.
5413 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5414 char *addr, size_t bytes, bool read_only,
5415 bool allow_exec) {
5416 // same as map_memory() on this OS
5417 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5418 allow_exec);
5419 }
5420
5421
5422 // Unmap a block of memory.
5423 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5424 return munmap(addr, bytes) == 0;
5425 }
5426
5427 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5428
5429 static clockid_t thread_cpu_clockid(Thread* thread) {
5430 pthread_t tid = thread->osthread()->pthread_id();
5431 clockid_t clockid;
5432
5433 // Get thread clockid
5434 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5435 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5436 return clockid;
5437 }
5438
5439 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5440 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5441 // of a thread.
5442 //
5443 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5444 // the fast estimate available on the platform.
5445
5446 jlong os::current_thread_cpu_time() {
5447 if (os::Linux::supports_fast_thread_cpu_time()) {
5448 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5449 } else {
5450 // return user + sys since the cost is the same
5451 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5452 }
5453 }
5454
5455 jlong os::thread_cpu_time(Thread* thread) {
5456 // consistent with what current_thread_cpu_time() returns
5457 if (os::Linux::supports_fast_thread_cpu_time()) {
5458 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5459 } else {
5460 return slow_thread_cpu_time(thread, true /* user + sys */);
5461 }
5462 }
5463
5464 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5465 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5466 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5467 } else {
5468 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5469 }
5470 }
5471
5472 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5473 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5474 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5475 } else {
5476 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5477 }
5478 }
5479
5480 // -1 on error.
5481 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5482 pid_t tid = thread->osthread()->thread_id();
5483 char *s;
5484 char stat[2048];
5485 int statlen;
5486 char proc_name[64];
5487 int count;
5488 long sys_time, user_time;
5489 char cdummy;
5490 int idummy;
5491 long ldummy;
5492 FILE *fp;
5493
5494 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5495 fp = fopen(proc_name, "r");
5496 if (fp == NULL) return -1;
5497 statlen = fread(stat, 1, 2047, fp);
5498 stat[statlen] = '\0';
5499 fclose(fp);
5500
5501 // Skip pid and the command string. Note that we could be dealing with
5502 // weird command names, e.g. user could decide to rename java launcher
5503 // to "java 1.4.2 :)", then the stat file would look like
5504 // 1234 (java 1.4.2 :)) R ... ...
5505 // We don't really need to know the command string, just find the last
5506 // occurrence of ")" and then start parsing from there. See bug 4726580.
5507 s = strrchr(stat, ')');
5508 if (s == NULL) return -1;
5509
5510 // Skip blank chars
5511 do { s++; } while (s && isspace(*s));
5512
5513 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5514 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5515 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5516 &user_time, &sys_time);
5517 if (count != 13) return -1;
5518 if (user_sys_cpu_time) {
5519 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5520 } else {
5521 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5522 }
5523 }
5524
5525 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5526 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5527 info_ptr->may_skip_backward = false; // elapsed time not wall time
5528 info_ptr->may_skip_forward = false; // elapsed time not wall time
5529 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5530 }
5531
5532 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5533 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5534 info_ptr->may_skip_backward = false; // elapsed time not wall time
5535 info_ptr->may_skip_forward = false; // elapsed time not wall time
5536 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5537 }
5538
5539 bool os::is_thread_cpu_time_supported() {
5540 return true;
5541 }
5542
5543 // System loadavg support. Returns -1 if load average cannot be obtained.
5544 // Linux doesn't yet have a (official) notion of processor sets,
5545 // so just return the system wide load average.
5546 int os::loadavg(double loadavg[], int nelem) {
5547 return ::getloadavg(loadavg, nelem);
5548 }
5549
5550 void os::pause() {
5551 char filename[MAX_PATH];
5552 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5553 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5554 } else {
5555 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5556 }
5557
5558 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5559 if (fd != -1) {
5560 struct stat buf;
5561 ::close(fd);
5562 while (::stat(filename, &buf) == 0) {
5563 (void)::poll(NULL, 0, 100);
5564 }
5565 } else {
5566 jio_fprintf(stderr,
5567 "Could not open pause file '%s', continuing immediately.\n", filename);
5568 }
5569 }
5570
5571
5572 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5573 // comment paragraphs are worth repeating here:
5574 //
5575 // Assumption:
5576 // Only one parker can exist on an event, which is why we allocate
5577 // them per-thread. Multiple unparkers can coexist.
5578 //
5579 // _Event serves as a restricted-range semaphore.
5580 // -1 : thread is blocked, i.e. there is a waiter
5581 // 0 : neutral: thread is running or ready,
5582 // could have been signaled after a wait started
5583 // 1 : signaled - thread is running or ready
5584 //
5585
5586 // utility to compute the abstime argument to timedwait:
5587 // millis is the relative timeout time
5588 // abstime will be the absolute timeout time
5589 // TODO: replace compute_abstime() with unpackTime()
5590
5591 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5592 if (millis < 0) millis = 0;
5593
5594 jlong seconds = millis / 1000;
5595 millis %= 1000;
5596 if (seconds > 50000000) { // see man cond_timedwait(3T)
5597 seconds = 50000000;
5598 }
5599
5600 if (os::supports_monotonic_clock()) {
5601 struct timespec now;
5602 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5603 assert_status(status == 0, status, "clock_gettime");
5604 abstime->tv_sec = now.tv_sec + seconds;
5605 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5606 if (nanos >= NANOSECS_PER_SEC) {
5607 abstime->tv_sec += 1;
5608 nanos -= NANOSECS_PER_SEC;
5609 }
5610 abstime->tv_nsec = nanos;
5611 } else {
5612 struct timeval now;
5613 int status = gettimeofday(&now, NULL);
5614 assert(status == 0, "gettimeofday");
5615 abstime->tv_sec = now.tv_sec + seconds;
5616 long usec = now.tv_usec + millis * 1000;
5617 if (usec >= 1000000) {
5618 abstime->tv_sec += 1;
5619 usec -= 1000000;
5620 }
5621 abstime->tv_nsec = usec * 1000;
5622 }
5623 return abstime;
5624 }
5625
5626 void os::PlatformEvent::park() { // AKA "down()"
5627 // Transitions for _Event:
5628 // -1 => -1 : illegal
5629 // 1 => 0 : pass - return immediately
5630 // 0 => -1 : block; then set _Event to 0 before returning
5631
5632 // Invariant: Only the thread associated with the Event/PlatformEvent
5633 // may call park().
5634 // TODO: assert that _Assoc != NULL or _Assoc == Self
5635 assert(_nParked == 0, "invariant");
5636
5637 int v;
5638 for (;;) {
5639 v = _Event;
5640 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5641 }
5642 guarantee(v >= 0, "invariant");
5643 if (v == 0) {
5644 // Do this the hard way by blocking ...
5645 int status = pthread_mutex_lock(_mutex);
5646 assert_status(status == 0, status, "mutex_lock");
5647 guarantee(_nParked == 0, "invariant");
5648 ++_nParked;
5649 while (_Event < 0) {
5650 status = pthread_cond_wait(_cond, _mutex);
5651 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5652 // Treat this the same as if the wait was interrupted
5653 if (status == ETIME) { status = EINTR; }
5654 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5655 }
5656 --_nParked;
5657
5658 _Event = 0;
5659 status = pthread_mutex_unlock(_mutex);
5660 assert_status(status == 0, status, "mutex_unlock");
5661 // Paranoia to ensure our locked and lock-free paths interact
5662 // correctly with each other.
5663 OrderAccess::fence();
5664 }
5665 guarantee(_Event >= 0, "invariant");
5666 }
5667
5668 int os::PlatformEvent::park(jlong millis) {
5669 // Transitions for _Event:
5670 // -1 => -1 : illegal
5671 // 1 => 0 : pass - return immediately
5672 // 0 => -1 : block; then set _Event to 0 before returning
5673
5674 guarantee(_nParked == 0, "invariant");
5675
5676 int v;
5677 for (;;) {
5678 v = _Event;
5679 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5680 }
5681 guarantee(v >= 0, "invariant");
5682 if (v != 0) return OS_OK;
5683
5684 // We do this the hard way, by blocking the thread.
5685 // Consider enforcing a minimum timeout value.
5686 struct timespec abst;
5687 compute_abstime(&abst, millis);
5688
5689 int ret = OS_TIMEOUT;
5690 int status = pthread_mutex_lock(_mutex);
5691 assert_status(status == 0, status, "mutex_lock");
5692 guarantee(_nParked == 0, "invariant");
5693 ++_nParked;
5694
5695 // Object.wait(timo) will return because of
5696 // (a) notification
5697 // (b) timeout
5698 // (c) thread.interrupt
5699 //
5700 // Thread.interrupt and object.notify{All} both call Event::set.
5701 // That is, we treat thread.interrupt as a special case of notification.
5702 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5703 // We assume all ETIME returns are valid.
5704 //
5705 // TODO: properly differentiate simultaneous notify+interrupt.
5706 // In that case, we should propagate the notify to another waiter.
5707
5708 while (_Event < 0) {
5709 status = pthread_cond_timedwait(_cond, _mutex, &abst);
5710 assert_status(status == 0 || status == EINTR ||
5711 status == ETIME || status == ETIMEDOUT,
5712 status, "cond_timedwait");
5713 if (!FilterSpuriousWakeups) break; // previous semantics
5714 if (status == ETIME || status == ETIMEDOUT) break;
5715 // We consume and ignore EINTR and spurious wakeups.
5716 }
5717 --_nParked;
5718 if (_Event >= 0) {
5719 ret = OS_OK;
5720 }
5721 _Event = 0;
5722 status = pthread_mutex_unlock(_mutex);
5723 assert_status(status == 0, status, "mutex_unlock");
5724 assert(_nParked == 0, "invariant");
5725 // Paranoia to ensure our locked and lock-free paths interact
5726 // correctly with each other.
5727 OrderAccess::fence();
5728 return ret;
5729 }
5730
5731 void os::PlatformEvent::unpark() {
5732 // Transitions for _Event:
5733 // 0 => 1 : just return
5734 // 1 => 1 : just return
5735 // -1 => either 0 or 1; must signal target thread
5736 // That is, we can safely transition _Event from -1 to either
5737 // 0 or 1.
5738 // See also: "Semaphores in Plan 9" by Mullender & Cox
5739 //
5740 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5741 // that it will take two back-to-back park() calls for the owning
5742 // thread to block. This has the benefit of forcing a spurious return
5743 // from the first park() call after an unpark() call which will help
5744 // shake out uses of park() and unpark() without condition variables.
5745
5746 if (Atomic::xchg(1, &_Event) >= 0) return;
5747
5748 // Wait for the thread associated with the event to vacate
5749 int status = pthread_mutex_lock(_mutex);
5750 assert_status(status == 0, status, "mutex_lock");
5751 int AnyWaiters = _nParked;
5752 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5753 status = pthread_mutex_unlock(_mutex);
5754 assert_status(status == 0, status, "mutex_unlock");
5755 if (AnyWaiters != 0) {
5756 // Note that we signal() *after* dropping the lock for "immortal" Events.
5757 // This is safe and avoids a common class of futile wakeups. In rare
5758 // circumstances this can cause a thread to return prematurely from
5759 // cond_{timed}wait() but the spurious wakeup is benign and the victim
5760 // will simply re-test the condition and re-park itself.
5761 // This provides particular benefit if the underlying platform does not
5762 // provide wait morphing.
5763 status = pthread_cond_signal(_cond);
5764 assert_status(status == 0, status, "cond_signal");
5765 }
5766 }
5767
5768
5769 // JSR166
5770 // -------------------------------------------------------
5771
5772 // The solaris and linux implementations of park/unpark are fairly
5773 // conservative for now, but can be improved. They currently use a
5774 // mutex/condvar pair, plus a a count.
5775 // Park decrements count if > 0, else does a condvar wait. Unpark
5776 // sets count to 1 and signals condvar. Only one thread ever waits
5777 // on the condvar. Contention seen when trying to park implies that someone
5778 // is unparking you, so don't wait. And spurious returns are fine, so there
5779 // is no need to track notifications.
5780
5781 // This code is common to linux and solaris and will be moved to a
5782 // common place in dolphin.
5783 //
5784 // The passed in time value is either a relative time in nanoseconds
5785 // or an absolute time in milliseconds. Either way it has to be unpacked
5786 // into suitable seconds and nanoseconds components and stored in the
5787 // given timespec structure.
5788 // Given time is a 64-bit value and the time_t used in the timespec is only
5789 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5790 // overflow if times way in the future are given. Further on Solaris versions
5791 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5792 // number of seconds, in abstime, is less than current_time + 100,000,000.
5793 // As it will be 28 years before "now + 100000000" will overflow we can
5794 // ignore overflow and just impose a hard-limit on seconds using the value
5795 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5796 // years from "now".
5797
5798 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5799 assert(time > 0, "convertTime");
5800 time_t max_secs = 0;
5801
5802 if (!os::supports_monotonic_clock() || isAbsolute) {
5803 struct timeval now;
5804 int status = gettimeofday(&now, NULL);
5805 assert(status == 0, "gettimeofday");
5806
5807 max_secs = now.tv_sec + MAX_SECS;
5808
5809 if (isAbsolute) {
5810 jlong secs = time / 1000;
5811 if (secs > max_secs) {
5812 absTime->tv_sec = max_secs;
5813 } else {
5814 absTime->tv_sec = secs;
5815 }
5816 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5817 } else {
5818 jlong secs = time / NANOSECS_PER_SEC;
5819 if (secs >= MAX_SECS) {
5820 absTime->tv_sec = max_secs;
5821 absTime->tv_nsec = 0;
5822 } else {
5823 absTime->tv_sec = now.tv_sec + secs;
5824 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5825 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5826 absTime->tv_nsec -= NANOSECS_PER_SEC;
5827 ++absTime->tv_sec; // note: this must be <= max_secs
5828 }
5829 }
5830 }
5831 } else {
5832 // must be relative using monotonic clock
5833 struct timespec now;
5834 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5835 assert_status(status == 0, status, "clock_gettime");
5836 max_secs = now.tv_sec + MAX_SECS;
5837 jlong secs = time / NANOSECS_PER_SEC;
5838 if (secs >= MAX_SECS) {
5839 absTime->tv_sec = max_secs;
5840 absTime->tv_nsec = 0;
5841 } else {
5842 absTime->tv_sec = now.tv_sec + secs;
5843 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5844 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5845 absTime->tv_nsec -= NANOSECS_PER_SEC;
5846 ++absTime->tv_sec; // note: this must be <= max_secs
5847 }
5848 }
5849 }
5850 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5851 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5852 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5853 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5854 }
5855
5856 void Parker::park(bool isAbsolute, jlong time) {
5857 // Ideally we'd do something useful while spinning, such
5858 // as calling unpackTime().
5859
5860 // Optional fast-path check:
5861 // Return immediately if a permit is available.
5862 // We depend on Atomic::xchg() having full barrier semantics
5863 // since we are doing a lock-free update to _counter.
5864 if (Atomic::xchg(0, &_counter) > 0) return;
5865
5866 Thread* thread = Thread::current();
5867 assert(thread->is_Java_thread(), "Must be JavaThread");
5868 JavaThread *jt = (JavaThread *)thread;
5869
5870 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5871 // Check interrupt before trying to wait
5872 if (Thread::is_interrupted(thread, false)) {
5873 return;
5874 }
5875
5876 // Next, demultiplex/decode time arguments
5877 timespec absTime;
5878 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5879 return;
5880 }
5881 if (time > 0) {
5882 unpackTime(&absTime, isAbsolute, time);
5883 }
5884
5885
5886 // Enter safepoint region
5887 // Beware of deadlocks such as 6317397.
5888 // The per-thread Parker:: mutex is a classic leaf-lock.
5889 // In particular a thread must never block on the Threads_lock while
5890 // holding the Parker:: mutex. If safepoints are pending both the
5891 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5892 ThreadBlockInVM tbivm(jt);
5893
5894 // Don't wait if cannot get lock since interference arises from
5895 // unblocking. Also. check interrupt before trying wait
5896 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5897 return;
5898 }
5899
5900 int status;
5901 if (_counter > 0) { // no wait needed
5902 _counter = 0;
5903 status = pthread_mutex_unlock(_mutex);
5904 assert_status(status == 0, status, "invariant");
5905 // Paranoia to ensure our locked and lock-free paths interact
5906 // correctly with each other and Java-level accesses.
5907 OrderAccess::fence();
5908 return;
5909 }
5910
5911 #ifdef ASSERT
5912 // Don't catch signals while blocked; let the running threads have the signals.
5913 // (This allows a debugger to break into the running thread.)
5914 sigset_t oldsigs;
5915 sigemptyset(&oldsigs);
5916 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5917 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5918 #endif
5919
5920 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5921 jt->set_suspend_equivalent();
5922 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5923
5924 assert(_cur_index == -1, "invariant");
5925 if (time == 0) {
5926 _cur_index = REL_INDEX; // arbitrary choice when not timed
5927 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5928 } else {
5929 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5930 status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5931 }
5932 _cur_index = -1;
5933 assert_status(status == 0 || status == EINTR ||
5934 status == ETIME || status == ETIMEDOUT,
5935 status, "cond_timedwait");
5936
5937 #ifdef ASSERT
5938 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5939 #endif
5940
5941 _counter = 0;
5942 status = pthread_mutex_unlock(_mutex);
5943 assert_status(status == 0, status, "invariant");
5944 // Paranoia to ensure our locked and lock-free paths interact
5945 // correctly with each other and Java-level accesses.
5946 OrderAccess::fence();
5947
5948 // If externally suspended while waiting, re-suspend
5949 if (jt->handle_special_suspend_equivalent_condition()) {
5950 jt->java_suspend_self();
5951 }
5952 }
5953
5954 void Parker::unpark() {
5955 int status = pthread_mutex_lock(_mutex);
5956 assert_status(status == 0, status, "invariant");
5957 const int s = _counter;
5958 _counter = 1;
5959 // must capture correct index before unlocking
5960 int index = _cur_index;
5961 status = pthread_mutex_unlock(_mutex);
5962 assert_status(status == 0, status, "invariant");
5963 if (s < 1 && index != -1) {
5964 // thread is definitely parked
5965 status = pthread_cond_signal(&_cond[index]);
5966 assert_status(status == 0, status, "invariant");
5967 }
5968 }
5969
5970
5971 extern char** environ;
5972
5973 // Run the specified command in a separate process. Return its exit value,
5974 // or -1 on failure (e.g. can't fork a new process).
5975 // Unlike system(), this function can be called from signal handler. It
5976 // doesn't block SIGINT et al.
5977 int os::fork_and_exec(char* cmd) {
5978 const char * argv[4] = {"sh", "-c", cmd, NULL};
5979
5980 pid_t pid = fork();
5981
5982 if (pid < 0) {
5983 // fork failed
5984 return -1;
5985
5986 } else if (pid == 0) {
5987 // child process
5988
5989 execve("/bin/sh", (char* const*)argv, environ);
5990
5991 // execve failed
5992 _exit(-1);
5993
5994 } else {
5995 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5996 // care about the actual exit code, for now.
5997
5998 int status;
5999
6000 // Wait for the child process to exit. This returns immediately if
6001 // the child has already exited. */
6002 while (waitpid(pid, &status, 0) < 0) {
6003 switch (errno) {
6004 case ECHILD: return 0;
6005 case EINTR: break;
6006 default: return -1;
6007 }
6008 }
6009
6010 if (WIFEXITED(status)) {
6011 // The child exited normally; get its exit code.
6012 return WEXITSTATUS(status);
6013 } else if (WIFSIGNALED(status)) {
6014 // The child exited because of a signal
6015 // The best value to return is 0x80 + signal number,
6016 // because that is what all Unix shells do, and because
6017 // it allows callers to distinguish between process exit and
6018 // process death by signal.
6019 return 0x80 + WTERMSIG(status);
6020 } else {
6021 // Unknown exit code; pass it through
6022 return status;
6023 }
6024 }
6025 }
6026
6027 // is_headless_jre()
6028 //
6029 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6030 // in order to report if we are running in a headless jre
6031 //
6032 // Since JDK8 xawt/libmawt.so was moved into the same directory
6033 // as libawt.so, and renamed libawt_xawt.so
6034 //
6035 bool os::is_headless_jre() {
6036 struct stat statbuf;
6037 char buf[MAXPATHLEN];
6038 char libmawtpath[MAXPATHLEN];
6039 const char *xawtstr = "/xawt/libmawt.so";
6040 const char *new_xawtstr = "/libawt_xawt.so";
6041 char *p;
6042
6043 // Get path to libjvm.so
6044 os::jvm_path(buf, sizeof(buf));
6045
6046 // Get rid of libjvm.so
6047 p = strrchr(buf, '/');
6048 if (p == NULL) {
6049 return false;
6050 } else {
6051 *p = '\0';
6052 }
6053
6054 // Get rid of client or server
6055 p = strrchr(buf, '/');
6056 if (p == NULL) {
6057 return false;
6058 } else {
6059 *p = '\0';
6060 }
6061
6062 // check xawt/libmawt.so
6063 strcpy(libmawtpath, buf);
6064 strcat(libmawtpath, xawtstr);
6065 if (::stat(libmawtpath, &statbuf) == 0) return false;
6066
6067 // check libawt_xawt.so
6068 strcpy(libmawtpath, buf);
6069 strcat(libmawtpath, new_xawtstr);
6070 if (::stat(libmawtpath, &statbuf) == 0) return false;
6071
6072 return true;
6073 }
6074
6075 // Get the default path to the core file
6076 // Returns the length of the string
6077 int os::get_core_path(char* buffer, size_t bufferSize) {
6078 /*
6079 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
6080 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
6081 */
6082 const int core_pattern_len = 129;
6083 char core_pattern[core_pattern_len] = {0};
6084
6085 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
6086 if (core_pattern_file == -1) {
6087 return -1;
6088 }
6089
6090 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
6091 ::close(core_pattern_file);
6092 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
6093 return -1;
6094 }
6095 if (core_pattern[ret-1] == '\n') {
6096 core_pattern[ret-1] = '\0';
6097 } else {
6098 core_pattern[ret] = '\0';
6099 }
6100
6101 char *pid_pos = strstr(core_pattern, "%p");
6102 int written;
6103
6104 if (core_pattern[0] == '/') {
6105 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
6106 } else {
6107 char cwd[PATH_MAX];
6108
6109 const char* p = get_current_directory(cwd, PATH_MAX);
6110 if (p == NULL) {
6111 return -1;
6112 }
6113
6114 if (core_pattern[0] == '|') {
6115 written = jio_snprintf(buffer, bufferSize,
6116 "\"%s\" (or dumping to %s/core.%d)",
6117 &core_pattern[1], p, current_process_id());
6118 } else {
6119 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6120 }
6121 }
6122
6123 if (written < 0) {
6124 return -1;
6125 }
6126
6127 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6128 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6129
6130 if (core_uses_pid_file != -1) {
6131 char core_uses_pid = 0;
6132 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6133 ::close(core_uses_pid_file);
6134
6135 if (core_uses_pid == '1') {
6136 jio_snprintf(buffer + written, bufferSize - written,
6137 ".%d", current_process_id());
6138 }
6139 }
6140 }
6141
6142 return strlen(buffer);
6143 }
6144
6145 bool os::start_debugging(char *buf, int buflen) {
6146 int len = (int)strlen(buf);
6147 char *p = &buf[len];
6148
6149 jio_snprintf(p, buflen-len,
6150 "\n\n"
6151 "Do you want to debug the problem?\n\n"
6152 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
6153 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
6154 "Otherwise, press RETURN to abort...",
6155 os::current_process_id(), os::current_process_id(),
6156 os::current_thread_id(), os::current_thread_id());
6157
6158 bool yes = os::message_box("Unexpected Error", buf);
6159
6160 if (yes) {
6161 // yes, user asked VM to launch debugger
6162 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
6163 os::current_process_id(), os::current_process_id());
6164
6165 os::fork_and_exec(buf);
6166 yes = false;
6167 }
6168 return yes;
6169 }
6170
6171
6172 // Java/Compiler thread:
6173 //
6174 // Low memory addresses
6175 // P0 +------------------------+
6176 // | |\ Java thread created by VM does not have glibc
6177 // | glibc guard page | - guard page, attached Java thread usually has
6178 // | |/ 1 glibc guard page.
6179 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6180 // | |\
6181 // | HotSpot Guard Pages | - red, yellow and reserved pages
6182 // | |/
6183 // +------------------------+ JavaThread::stack_reserved_zone_base()
6184 // | |\
6185 // | Normal Stack | -
6186 // | |/
6187 // P2 +------------------------+ Thread::stack_base()
6188 //
6189 // Non-Java thread:
6190 //
6191 // Low memory addresses
6192 // P0 +------------------------+
6193 // | |\
6194 // | glibc guard page | - usually 1 page
6195 // | |/
6196 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6197 // | |\
6198 // | Normal Stack | -
6199 // | |/
6200 // P2 +------------------------+ Thread::stack_base()
6201 //
6202 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
6203 // returned from pthread_attr_getstack().
6204 // ** Due to NPTL implementation error, linux takes the glibc guard page out
6205 // of the stack size given in pthread_attr. We work around this for
6206 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
6207 //
6208 #ifndef ZERO
6209 static void current_stack_region(address * bottom, size_t * size) {
6210 if (os::Linux::is_initial_thread()) {
6211 // initial thread needs special handling because pthread_getattr_np()
6212 // may return bogus value.
6213 *bottom = os::Linux::initial_thread_stack_bottom();
6214 *size = os::Linux::initial_thread_stack_size();
6215 } else {
6216 pthread_attr_t attr;
6217
6218 int rslt = pthread_getattr_np(pthread_self(), &attr);
6219
6220 // JVM needs to know exact stack location, abort if it fails
6221 if (rslt != 0) {
6222 if (rslt == ENOMEM) {
6223 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
6224 } else {
6225 fatal("pthread_getattr_np failed with error = %d", rslt);
6226 }
6227 }
6228
6229 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
6230 fatal("Cannot locate current stack attributes!");
6231 }
6232
6233 // Work around NPTL stack guard error.
6234 size_t guard_size = 0;
6235 rslt = pthread_attr_getguardsize(&attr, &guard_size);
6236 if (rslt != 0) {
6237 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
6238 }
6239 *bottom += guard_size;
6240 *size -= guard_size;
6241
6242 pthread_attr_destroy(&attr);
6243
6244 }
6245 assert(os::current_stack_pointer() >= *bottom &&
6246 os::current_stack_pointer() < *bottom + *size, "just checking");
6247 }
6248
6249 address os::current_stack_base() {
6250 address bottom;
6251 size_t size;
6252 current_stack_region(&bottom, &size);
6253 return (bottom + size);
6254 }
6255
6256 size_t os::current_stack_size() {
6257 // This stack size includes the usable stack and HotSpot guard pages
6258 // (for the threads that have Hotspot guard pages).
6259 address bottom;
6260 size_t size;
6261 current_stack_region(&bottom, &size);
6262 return size;
6263 }
6264 #endif
6265
6266 static inline struct timespec get_mtime(const char* filename) {
6267 struct stat st;
6268 int ret = os::stat(filename, &st);
6269 assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
6270 return st.st_mtim;
6271 }
6272
6273 int os::compare_file_modified_times(const char* file1, const char* file2) {
6274 struct timespec filetime1 = get_mtime(file1);
6275 struct timespec filetime2 = get_mtime(file2);
6276 int diff = filetime1.tv_sec - filetime2.tv_sec;
6277 if (diff == 0) {
6278 return filetime1.tv_nsec - filetime2.tv_nsec;
6279 }
6280 return diff;
6281 }
6282
6283 /////////////// Unit tests ///////////////
6284
6285 #ifndef PRODUCT
6286
6287 #define test_log(...) \
6288 do { \
6289 if (VerboseInternalVMTests) { \
6290 tty->print_cr(__VA_ARGS__); \
6291 tty->flush(); \
6292 } \
6293 } while (false)
6294
6295 class TestReserveMemorySpecial : AllStatic {
6296 public:
6297 static void small_page_write(void* addr, size_t size) {
6298 size_t page_size = os::vm_page_size();
6299
6300 char* end = (char*)addr + size;
6301 for (char* p = (char*)addr; p < end; p += page_size) {
6302 *p = 1;
6303 }
6304 }
6305
6306 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6307 if (!UseHugeTLBFS) {
6308 return;
6309 }
6310
6311 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6312
6313 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6314
6315 if (addr != NULL) {
6316 small_page_write(addr, size);
6317
6318 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6319 }
6320 }
6321
6322 static void test_reserve_memory_special_huge_tlbfs_only() {
6323 if (!UseHugeTLBFS) {
6324 return;
6325 }
6326
6327 size_t lp = os::large_page_size();
6328
6329 for (size_t size = lp; size <= lp * 10; size += lp) {
6330 test_reserve_memory_special_huge_tlbfs_only(size);
6331 }
6332 }
6333
6334 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6335 size_t lp = os::large_page_size();
6336 size_t ag = os::vm_allocation_granularity();
6337
6338 // sizes to test
6339 const size_t sizes[] = {
6340 lp, lp + ag, lp + lp / 2, lp * 2,
6341 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6342 lp * 10, lp * 10 + lp / 2
6343 };
6344 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6345
6346 // For each size/alignment combination, we test three scenarios:
6347 // 1) with req_addr == NULL
6348 // 2) with a non-null req_addr at which we expect to successfully allocate
6349 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6350 // expect the allocation to either fail or to ignore req_addr
6351
6352 // Pre-allocate two areas; they shall be as large as the largest allocation
6353 // and aligned to the largest alignment we will be testing.
6354 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6355 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6356 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6357 -1, 0);
6358 assert(mapping1 != MAP_FAILED, "should work");
6359
6360 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6361 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6362 -1, 0);
6363 assert(mapping2 != MAP_FAILED, "should work");
6364
6365 // Unmap the first mapping, but leave the second mapping intact: the first
6366 // mapping will serve as a value for a "good" req_addr (case 2). The second
6367 // mapping, still intact, as "bad" req_addr (case 3).
6368 ::munmap(mapping1, mapping_size);
6369
6370 // Case 1
6371 test_log("%s, req_addr NULL:", __FUNCTION__);
6372 test_log("size align result");
6373
6374 for (int i = 0; i < num_sizes; i++) {
6375 const size_t size = sizes[i];
6376 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6377 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6378 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6379 size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6380 if (p != NULL) {
6381 assert(is_ptr_aligned(p, alignment), "must be");
6382 small_page_write(p, size);
6383 os::Linux::release_memory_special_huge_tlbfs(p, size);
6384 }
6385 }
6386 }
6387
6388 // Case 2
6389 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6390 test_log("size align req_addr result");
6391
6392 for (int i = 0; i < num_sizes; i++) {
6393 const size_t size = sizes[i];
6394 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6395 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6396 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6397 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6398 size, alignment, p2i(req_addr), p2i(p),
6399 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6400 if (p != NULL) {
6401 assert(p == req_addr, "must be");
6402 small_page_write(p, size);
6403 os::Linux::release_memory_special_huge_tlbfs(p, size);
6404 }
6405 }
6406 }
6407
6408 // Case 3
6409 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6410 test_log("size align req_addr result");
6411
6412 for (int i = 0; i < num_sizes; i++) {
6413 const size_t size = sizes[i];
6414 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6415 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6416 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6417 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6418 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6419 // as the area around req_addr contains already existing mappings, the API should always
6420 // return NULL (as per contract, it cannot return another address)
6421 assert(p == NULL, "must be");
6422 }
6423 }
6424
6425 ::munmap(mapping2, mapping_size);
6426
6427 }
6428
6429 static void test_reserve_memory_special_huge_tlbfs() {
6430 if (!UseHugeTLBFS) {
6431 return;
6432 }
6433
6434 test_reserve_memory_special_huge_tlbfs_only();
6435 test_reserve_memory_special_huge_tlbfs_mixed();
6436 }
6437
6438 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6439 if (!UseSHM) {
6440 return;
6441 }
6442
6443 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6444
6445 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6446
6447 if (addr != NULL) {
6448 assert(is_ptr_aligned(addr, alignment), "Check");
6449 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6450
6451 small_page_write(addr, size);
6452
6453 os::Linux::release_memory_special_shm(addr, size);
6454 }
6455 }
6456
6457 static void test_reserve_memory_special_shm() {
6458 size_t lp = os::large_page_size();
6459 size_t ag = os::vm_allocation_granularity();
6460
6461 for (size_t size = ag; size < lp * 3; size += ag) {
6462 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6463 test_reserve_memory_special_shm(size, alignment);
6464 }
6465 }
6466 }
6467
6468 static void test() {
6469 test_reserve_memory_special_huge_tlbfs();
6470 test_reserve_memory_special_shm();
6471 }
6472 };
6473
6474 void TestReserveMemorySpecial_test() {
6475 TestReserveMemorySpecial::test();
6476 }
6477
6478 #endif