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 "mutex_linux.inline.hpp"
39 #include "oops/oop.inline.hpp"
40 #include "os_linux.inline.hpp"
41 #include "os_share_linux.hpp"
42 #include "prims/jniFastGetField.hpp"
43 #include "prims/jvm.h"
44 #include "prims/jvm_misc.hpp"
45 #include "runtime/arguments.hpp"
46 #include "runtime/atomic.inline.hpp"
47 #include "runtime/extendedPC.hpp"
48 #include "runtime/globals.hpp"
49 #include "runtime/interfaceSupport.hpp"
50 #include "runtime/init.hpp"
51 #include "runtime/java.hpp"
52 #include "runtime/javaCalls.hpp"
53 #include "runtime/mutexLocker.hpp"
54 #include "runtime/objectMonitor.hpp"
55 #include "runtime/orderAccess.inline.hpp"
56 #include "runtime/osThread.hpp"
57 #include "runtime/perfMemory.hpp"
58 #include "runtime/sharedRuntime.hpp"
59 #include "runtime/statSampler.hpp"
60 #include "runtime/stubRoutines.hpp"
61 #include "runtime/thread.inline.hpp"
62 #include "runtime/threadCritical.hpp"
63 #include "runtime/timer.hpp"
64 #include "semaphore_posix.hpp"
65 #include "services/attachListener.hpp"
66 #include "services/memTracker.hpp"
67 #include "services/runtimeService.hpp"
68 #include "utilities/decoder.hpp"
69 #include "utilities/defaultStream.hpp"
70 #include "utilities/events.hpp"
71 #include "utilities/elfFile.hpp"
72 #include "utilities/growableArray.hpp"
73 #include "utilities/macros.hpp"
74 #include "utilities/vmError.hpp"
75
76 // put OS-includes here
77 # include <sys/types.h>
78 # include <sys/mman.h>
79 # include <sys/stat.h>
80 # include <sys/select.h>
81 # include <pthread.h>
82 # include <signal.h>
83 # include <errno.h>
84 # include <dlfcn.h>
85 # include <stdio.h>
86 # include <unistd.h>
87 # include <sys/resource.h>
88 # include <pthread.h>
89 # include <sys/stat.h>
90 # include <sys/time.h>
91 # include <sys/times.h>
92 # include <sys/utsname.h>
93 # include <sys/socket.h>
94 # include <sys/wait.h>
95 # include <pwd.h>
96 # include <poll.h>
97 # include <semaphore.h>
98 # include <fcntl.h>
99 # include <string.h>
100 # include <syscall.h>
101 # include <sys/sysinfo.h>
102 # include <gnu/libc-version.h>
103 # include <sys/ipc.h>
104 # include <sys/shm.h>
105 # include <link.h>
106 # include <stdint.h>
107 # include <inttypes.h>
108 # include <sys/ioctl.h>
109
110 #ifndef _GNU_SOURCE
111 #define _GNU_SOURCE
112 #include <sched.h>
113 #undef _GNU_SOURCE
114 #else
115 #include <sched.h>
116 #endif
117
118 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
119 // getrusage() is prepared to handle the associated failure.
120 #ifndef RUSAGE_THREAD
121 #define RUSAGE_THREAD (1) /* only the calling thread */
122 #endif
123
124 #define MAX_PATH (2 * K)
125
126 #define MAX_SECS 100000000
127
128 // for timer info max values which include all bits
129 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
130
131 #define LARGEPAGES_BIT (1 << 6)
132 ////////////////////////////////////////////////////////////////////////////////
133 // global variables
134 julong os::Linux::_physical_memory = 0;
135
136 address os::Linux::_initial_thread_stack_bottom = NULL;
137 uintptr_t os::Linux::_initial_thread_stack_size = 0;
138
139 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
140 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
141 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
142 Mutex* os::Linux::_createThread_lock = NULL;
143 pthread_t os::Linux::_main_thread;
144 int os::Linux::_page_size = -1;
145 const int os::Linux::_vm_default_page_size = (8 * K);
146 bool os::Linux::_supports_fast_thread_cpu_time = false;
147 uint32_t os::Linux::_os_version = 0;
148 const char * os::Linux::_glibc_version = NULL;
149 const char * os::Linux::_libpthread_version = NULL;
150 pthread_condattr_t os::Linux::_condattr[1];
151
152 static jlong initial_time_count=0;
153
154 static int clock_tics_per_sec = 100;
155
156 // For diagnostics to print a message once. see run_periodic_checks
157 static sigset_t check_signal_done;
158 static bool check_signals = true;
159
160 // Signal number used to suspend/resume a thread
161
162 // do not use any signal number less than SIGSEGV, see 4355769
163 static int SR_signum = SIGUSR2;
164 sigset_t SR_sigset;
165
166 // Declarations
167 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
168
169 // utility functions
170
171 static int SR_initialize();
172
173 julong os::available_memory() {
174 return Linux::available_memory();
175 }
176
177 julong os::Linux::available_memory() {
178 // values in struct sysinfo are "unsigned long"
179 struct sysinfo si;
180 sysinfo(&si);
181
182 return (julong)si.freeram * si.mem_unit;
183 }
184
185 julong os::physical_memory() {
186 return Linux::physical_memory();
187 }
188
189 // Return true if user is running as root.
190
191 bool os::have_special_privileges() {
192 static bool init = false;
193 static bool privileges = false;
194 if (!init) {
195 privileges = (getuid() != geteuid()) || (getgid() != getegid());
196 init = true;
197 }
198 return privileges;
199 }
200
201
202 #ifndef SYS_gettid
203 // i386: 224, ia64: 1105, amd64: 186, sparc 143
204 #ifdef __ia64__
205 #define SYS_gettid 1105
206 #else
207 #ifdef __i386__
208 #define SYS_gettid 224
209 #else
210 #ifdef __amd64__
211 #define SYS_gettid 186
212 #else
213 #ifdef __sparc__
214 #define SYS_gettid 143
215 #else
216 #error define gettid for the arch
217 #endif
218 #endif
219 #endif
220 #endif
221 #endif
222
223 // Cpu architecture string
224 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
225
226
227 // pid_t gettid()
228 //
229 // Returns the kernel thread id of the currently running thread. Kernel
230 // thread id is used to access /proc.
231 pid_t os::Linux::gettid() {
232 int rslt = syscall(SYS_gettid);
233 assert(rslt != -1, "must be."); // old linuxthreads implementation?
234 return (pid_t)rslt;
235 }
236
237 // Most versions of linux have a bug where the number of processors are
238 // determined by looking at the /proc file system. In a chroot environment,
239 // the system call returns 1. This causes the VM to act as if it is
240 // a single processor and elide locking (see is_MP() call).
241 static bool unsafe_chroot_detected = false;
242 static const char *unstable_chroot_error = "/proc file system not found.\n"
243 "Java may be unstable running multithreaded in a chroot "
244 "environment on Linux when /proc filesystem is not mounted.";
245
246 void os::Linux::initialize_system_info() {
247 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
248 if (processor_count() == 1) {
249 pid_t pid = os::Linux::gettid();
250 char fname[32];
251 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
252 FILE *fp = fopen(fname, "r");
253 if (fp == NULL) {
254 unsafe_chroot_detected = true;
255 } else {
256 fclose(fp);
257 }
258 }
259 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
260 assert(processor_count() > 0, "linux error");
261 }
262
263 void os::init_system_properties_values() {
264 // The next steps are taken in the product version:
265 //
266 // Obtain the JAVA_HOME value from the location of libjvm.so.
267 // This library should be located at:
268 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
269 //
270 // If "/jre/lib/" appears at the right place in the path, then we
271 // assume libjvm.so is installed in a JDK and we use this path.
272 //
273 // Otherwise exit with message: "Could not create the Java virtual machine."
274 //
275 // The following extra steps are taken in the debugging version:
276 //
277 // If "/jre/lib/" does NOT appear at the right place in the path
278 // instead of exit check for $JAVA_HOME environment variable.
279 //
280 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
281 // then we append a fake suffix "hotspot/libjvm.so" to this path so
282 // it looks like libjvm.so is installed there
283 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
284 //
285 // Otherwise exit.
286 //
287 // Important note: if the location of libjvm.so changes this
288 // code needs to be changed accordingly.
289
290 // See ld(1):
291 // The linker uses the following search paths to locate required
292 // shared libraries:
293 // 1: ...
294 // ...
295 // 7: The default directories, normally /lib and /usr/lib.
296 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
297 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
298 #else
299 #define DEFAULT_LIBPATH "/lib:/usr/lib"
300 #endif
301
302 // Base path of extensions installed on the system.
303 #define SYS_EXT_DIR "/usr/java/packages"
304 #define EXTENSIONS_DIR "/lib/ext"
305
306 // Buffer that fits several sprintfs.
307 // Note that the space for the colon and the trailing null are provided
308 // by the nulls included by the sizeof operator.
309 const size_t bufsize =
310 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
311 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
312 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
313
314 // sysclasspath, java_home, dll_dir
315 {
316 char *pslash;
317 os::jvm_path(buf, bufsize);
318
319 // Found the full path to libjvm.so.
320 // Now cut the path to <java_home>/jre if we can.
321 pslash = strrchr(buf, '/');
322 if (pslash != NULL) {
323 *pslash = '\0'; // Get rid of /libjvm.so.
324 }
325 pslash = strrchr(buf, '/');
326 if (pslash != NULL) {
327 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
328 }
329 Arguments::set_dll_dir(buf);
330
331 if (pslash != NULL) {
332 pslash = strrchr(buf, '/');
333 if (pslash != NULL) {
334 *pslash = '\0'; // Get rid of /<arch>.
335 pslash = strrchr(buf, '/');
336 if (pslash != NULL) {
337 *pslash = '\0'; // Get rid of /lib.
338 }
339 }
340 }
341 Arguments::set_java_home(buf);
342 set_boot_path('/', ':');
343 }
344
345 // Where to look for native libraries.
346 //
347 // Note: Due to a legacy implementation, most of the library path
348 // is set in the launcher. This was to accomodate linking restrictions
349 // on legacy Linux implementations (which are no longer supported).
350 // Eventually, all the library path setting will be done here.
351 //
352 // However, to prevent the proliferation of improperly built native
353 // libraries, the new path component /usr/java/packages is added here.
354 // Eventually, all the library path setting will be done here.
355 {
356 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
357 // should always exist (until the legacy problem cited above is
358 // addressed).
359 const char *v = ::getenv("LD_LIBRARY_PATH");
360 const char *v_colon = ":";
361 if (v == NULL) { v = ""; v_colon = ""; }
362 // That's +1 for the colon and +1 for the trailing '\0'.
363 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
364 strlen(v) + 1 +
365 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
366 mtInternal);
367 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
368 Arguments::set_library_path(ld_library_path);
369 FREE_C_HEAP_ARRAY(char, ld_library_path);
370 }
371
372 // Extensions directories.
373 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
374 Arguments::set_ext_dirs(buf);
375
376 FREE_C_HEAP_ARRAY(char, buf);
377
378 #undef DEFAULT_LIBPATH
379 #undef SYS_EXT_DIR
380 #undef EXTENSIONS_DIR
381 }
382
383 ////////////////////////////////////////////////////////////////////////////////
384 // breakpoint support
385
386 void os::breakpoint() {
387 BREAKPOINT;
388 }
389
390 extern "C" void breakpoint() {
391 // use debugger to set breakpoint here
392 }
393
394 ////////////////////////////////////////////////////////////////////////////////
395 // signal support
396
397 debug_only(static bool signal_sets_initialized = false);
398 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
399
400 bool os::Linux::is_sig_ignored(int sig) {
401 struct sigaction oact;
402 sigaction(sig, (struct sigaction*)NULL, &oact);
403 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
404 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
405 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
406 return true;
407 } else {
408 return false;
409 }
410 }
411
412 void os::Linux::signal_sets_init() {
413 // Should also have an assertion stating we are still single-threaded.
414 assert(!signal_sets_initialized, "Already initialized");
415 // Fill in signals that are necessarily unblocked for all threads in
416 // the VM. Currently, we unblock the following signals:
417 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
418 // by -Xrs (=ReduceSignalUsage));
419 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
420 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
421 // the dispositions or masks wrt these signals.
422 // Programs embedding the VM that want to use the above signals for their
423 // own purposes must, at this time, use the "-Xrs" option to prevent
424 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
425 // (See bug 4345157, and other related bugs).
426 // In reality, though, unblocking these signals is really a nop, since
427 // these signals are not blocked by default.
428 sigemptyset(&unblocked_sigs);
429 sigemptyset(&allowdebug_blocked_sigs);
430 sigaddset(&unblocked_sigs, SIGILL);
431 sigaddset(&unblocked_sigs, SIGSEGV);
432 sigaddset(&unblocked_sigs, SIGBUS);
433 sigaddset(&unblocked_sigs, SIGFPE);
434 #if defined(PPC64)
435 sigaddset(&unblocked_sigs, SIGTRAP);
436 #endif
437 sigaddset(&unblocked_sigs, SR_signum);
438
439 if (!ReduceSignalUsage) {
440 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
441 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
442 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
443 }
444 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
445 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
446 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
447 }
448 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
449 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
450 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
451 }
452 }
453 // Fill in signals that are blocked by all but the VM thread.
454 sigemptyset(&vm_sigs);
455 if (!ReduceSignalUsage) {
456 sigaddset(&vm_sigs, BREAK_SIGNAL);
457 }
458 debug_only(signal_sets_initialized = true);
459
460 }
461
462 // These are signals that are unblocked while a thread is running Java.
463 // (For some reason, they get blocked by default.)
464 sigset_t* os::Linux::unblocked_signals() {
465 assert(signal_sets_initialized, "Not initialized");
466 return &unblocked_sigs;
467 }
468
469 // These are the signals that are blocked while a (non-VM) thread is
470 // running Java. Only the VM thread handles these signals.
471 sigset_t* os::Linux::vm_signals() {
472 assert(signal_sets_initialized, "Not initialized");
473 return &vm_sigs;
474 }
475
476 // These are signals that are blocked during cond_wait to allow debugger in
477 sigset_t* os::Linux::allowdebug_blocked_signals() {
478 assert(signal_sets_initialized, "Not initialized");
479 return &allowdebug_blocked_sigs;
480 }
481
482 void os::Linux::hotspot_sigmask(Thread* thread) {
483
484 //Save caller's signal mask before setting VM signal mask
485 sigset_t caller_sigmask;
486 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
487
488 OSThread* osthread = thread->osthread();
489 osthread->set_caller_sigmask(caller_sigmask);
490
491 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
492
493 if (!ReduceSignalUsage) {
494 if (thread->is_VM_thread()) {
495 // Only the VM thread handles BREAK_SIGNAL ...
496 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
497 } else {
498 // ... all other threads block BREAK_SIGNAL
499 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
500 }
501 }
502 }
503
504 //////////////////////////////////////////////////////////////////////////////
505 // detecting pthread library
506
507 void os::Linux::libpthread_init() {
508 // Save glibc and pthread version strings.
509 #if !defined(_CS_GNU_LIBC_VERSION) || \
510 !defined(_CS_GNU_LIBPTHREAD_VERSION)
511 #error "glibc too old (< 2.3.2)"
512 #endif
513
514 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
515 assert(n > 0, "cannot retrieve glibc version");
516 char *str = (char *)malloc(n, mtInternal);
517 confstr(_CS_GNU_LIBC_VERSION, str, n);
518 os::Linux::set_glibc_version(str);
519
520 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
521 assert(n > 0, "cannot retrieve pthread version");
522 str = (char *)malloc(n, mtInternal);
523 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
524 os::Linux::set_libpthread_version(str);
525 }
526
527 /////////////////////////////////////////////////////////////////////////////
528 // thread stack expansion
529
530 // os::Linux::manually_expand_stack() takes care of expanding the thread
531 // stack. Note that this is normally not needed: pthread stacks allocate
532 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
533 // committed. Therefore it is not necessary to expand the stack manually.
534 //
535 // Manually expanding the stack was historically needed on LinuxThreads
536 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
537 // it is kept to deal with very rare corner cases:
538 //
539 // For one, user may run the VM on an own implementation of threads
540 // whose stacks are - like the old LinuxThreads - implemented using
541 // mmap(MAP_GROWSDOWN).
542 //
543 // Also, this coding may be needed if the VM is running on the primordial
544 // thread. Normally we avoid running on the primordial thread; however,
545 // user may still invoke the VM on the primordial thread.
546 //
547 // The following historical comment describes the details about running
548 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
549
550
551 // Force Linux kernel to expand current thread stack. If "bottom" is close
552 // to the stack guard, caller should block all signals.
553 //
554 // MAP_GROWSDOWN:
555 // A special mmap() flag that is used to implement thread stacks. It tells
556 // kernel that the memory region should extend downwards when needed. This
557 // allows early versions of LinuxThreads to only mmap the first few pages
558 // when creating a new thread. Linux kernel will automatically expand thread
559 // stack as needed (on page faults).
560 //
561 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
562 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
563 // region, it's hard to tell if the fault is due to a legitimate stack
564 // access or because of reading/writing non-exist memory (e.g. buffer
565 // overrun). As a rule, if the fault happens below current stack pointer,
566 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
567 // application (see Linux kernel fault.c).
568 //
569 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
570 // stack overflow detection.
571 //
572 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
573 // not use MAP_GROWSDOWN.
574 //
575 // To get around the problem and allow stack banging on Linux, we need to
576 // manually expand thread stack after receiving the SIGSEGV.
577 //
578 // There are two ways to expand thread stack to address "bottom", we used
579 // both of them in JVM before 1.5:
580 // 1. adjust stack pointer first so that it is below "bottom", and then
581 // touch "bottom"
582 // 2. mmap() the page in question
583 //
584 // Now alternate signal stack is gone, it's harder to use 2. For instance,
585 // if current sp is already near the lower end of page 101, and we need to
586 // call mmap() to map page 100, it is possible that part of the mmap() frame
587 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
588 // That will destroy the mmap() frame and cause VM to crash.
589 //
590 // The following code works by adjusting sp first, then accessing the "bottom"
591 // page to force a page fault. Linux kernel will then automatically expand the
592 // stack mapping.
593 //
594 // _expand_stack_to() assumes its frame size is less than page size, which
595 // should always be true if the function is not inlined.
596
597 static void NOINLINE _expand_stack_to(address bottom) {
598 address sp;
599 size_t size;
600 volatile char *p;
601
602 // Adjust bottom to point to the largest address within the same page, it
603 // gives us a one-page buffer if alloca() allocates slightly more memory.
604 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
605 bottom += os::Linux::page_size() - 1;
606
607 // sp might be slightly above current stack pointer; if that's the case, we
608 // will alloca() a little more space than necessary, which is OK. Don't use
609 // os::current_stack_pointer(), as its result can be slightly below current
610 // stack pointer, causing us to not alloca enough to reach "bottom".
611 sp = (address)&sp;
612
613 if (sp > bottom) {
614 size = sp - bottom;
615 p = (volatile char *)alloca(size);
616 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
617 p[0] = '\0';
618 }
619 }
620
621 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
622 assert(t!=NULL, "just checking");
623 assert(t->osthread()->expanding_stack(), "expand should be set");
624 assert(t->stack_base() != NULL, "stack_base was not initialized");
625
626 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
627 sigset_t mask_all, old_sigset;
628 sigfillset(&mask_all);
629 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
630 _expand_stack_to(addr);
631 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
632 return true;
633 }
634 return false;
635 }
636
637 //////////////////////////////////////////////////////////////////////////////
638 // create new thread
639
640 // Thread start routine for all newly created threads
641 static void *java_start(Thread *thread) {
642 // Try to randomize the cache line index of hot stack frames.
643 // This helps when threads of the same stack traces evict each other's
644 // cache lines. The threads can be either from the same JVM instance, or
645 // from different JVM instances. The benefit is especially true for
646 // processors with hyperthreading technology.
647 static int counter = 0;
648 int pid = os::current_process_id();
649 alloca(((pid ^ counter++) & 7) * 128);
650
651 thread->initialize_thread_current();
652
653 OSThread* osthread = thread->osthread();
654 Monitor* sync = osthread->startThread_lock();
655
656 osthread->set_thread_id(os::current_thread_id());
657
658 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
659 os::current_thread_id(), (uintx) pthread_self());
660
661 if (UseNUMA) {
662 int lgrp_id = os::numa_get_group_id();
663 if (lgrp_id != -1) {
664 thread->set_lgrp_id(lgrp_id);
665 }
666 }
667 // initialize signal mask for this thread
668 os::Linux::hotspot_sigmask(thread);
669
670 // initialize floating point control register
671 os::Linux::init_thread_fpu_state();
672
673 // handshaking with parent thread
674 {
675 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
676
677 // notify parent thread
678 osthread->set_state(INITIALIZED);
679 sync->notify_all();
680
681 // wait until os::start_thread()
682 while (osthread->get_state() == INITIALIZED) {
683 sync->wait(Mutex::_no_safepoint_check_flag);
684 }
685 }
686
687 // call one more level start routine
688 thread->run();
689
690 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
691 os::current_thread_id(), (uintx) pthread_self());
692
693 return 0;
694 }
695
696 bool os::create_thread(Thread* thread, ThreadType thr_type,
697 size_t stack_size) {
698 assert(thread->osthread() == NULL, "caller responsible");
699
700 // Allocate the OSThread object
701 OSThread* osthread = new OSThread(NULL, NULL);
702 if (osthread == NULL) {
703 return false;
704 }
705
706 // set the correct thread state
707 osthread->set_thread_type(thr_type);
708
709 // Initial state is ALLOCATED but not INITIALIZED
710 osthread->set_state(ALLOCATED);
711
712 thread->set_osthread(osthread);
713
714 // init thread attributes
715 pthread_attr_t attr;
716 pthread_attr_init(&attr);
717 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
718
719 // stack size
720 // calculate stack size if it's not specified by caller
721 if (stack_size == 0) {
722 stack_size = os::Linux::default_stack_size(thr_type);
723
724 switch (thr_type) {
725 case os::java_thread:
726 // Java threads use ThreadStackSize which default value can be
727 // changed with the flag -Xss
728 assert(JavaThread::stack_size_at_create() > 0, "this should be set");
729 stack_size = JavaThread::stack_size_at_create();
730 break;
731 case os::compiler_thread:
732 if (CompilerThreadStackSize > 0) {
733 stack_size = (size_t)(CompilerThreadStackSize * K);
734 break;
735 } // else fall through:
736 // use VMThreadStackSize if CompilerThreadStackSize is not defined
737 case os::vm_thread:
738 case os::pgc_thread:
739 case os::cgc_thread:
740 case os::watcher_thread:
741 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
742 break;
743 }
744 }
745
746 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
747 pthread_attr_setstacksize(&attr, stack_size);
748
749 // glibc guard page
750 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
751
752 ThreadState state;
753
754 {
755 pthread_t tid;
756 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
757
758 char buf[64];
759 if (ret == 0) {
760 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
761 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
762 } else {
763 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
764 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
765 }
766
767 pthread_attr_destroy(&attr);
768
769 if (ret != 0) {
770 // Need to clean up stuff we've allocated so far
771 thread->set_osthread(NULL);
772 delete osthread;
773 return false;
774 }
775
776 // Store pthread info into the OSThread
777 osthread->set_pthread_id(tid);
778
779 // Wait until child thread is either initialized or aborted
780 {
781 Monitor* sync_with_child = osthread->startThread_lock();
782 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
783 while ((state = osthread->get_state()) == ALLOCATED) {
784 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
785 }
786 }
787 }
788
789 // Aborted due to thread limit being reached
790 if (state == ZOMBIE) {
791 thread->set_osthread(NULL);
792 delete osthread;
793 return false;
794 }
795
796 // The thread is returned suspended (in state INITIALIZED),
797 // and is started higher up in the call chain
798 assert(state == INITIALIZED, "race condition");
799 return true;
800 }
801
802 /////////////////////////////////////////////////////////////////////////////
803 // attach existing thread
804
805 // bootstrap the main thread
806 bool os::create_main_thread(JavaThread* thread) {
807 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
808 return create_attached_thread(thread);
809 }
810
811 bool os::create_attached_thread(JavaThread* thread) {
812 #ifdef ASSERT
813 thread->verify_not_published();
814 #endif
815
816 // Allocate the OSThread object
817 OSThread* osthread = new OSThread(NULL, NULL);
818
819 if (osthread == NULL) {
820 return false;
821 }
822
823 // Store pthread info into the OSThread
824 osthread->set_thread_id(os::Linux::gettid());
825 osthread->set_pthread_id(::pthread_self());
826
827 // initialize floating point control register
828 os::Linux::init_thread_fpu_state();
829
830 // Initial thread state is RUNNABLE
831 osthread->set_state(RUNNABLE);
832
833 thread->set_osthread(osthread);
834
835 if (UseNUMA) {
836 int lgrp_id = os::numa_get_group_id();
837 if (lgrp_id != -1) {
838 thread->set_lgrp_id(lgrp_id);
839 }
840 }
841
842 if (os::Linux::is_initial_thread()) {
843 // If current thread is initial thread, its stack is mapped on demand,
844 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
845 // the entire stack region to avoid SEGV in stack banging.
846 // It is also useful to get around the heap-stack-gap problem on SuSE
847 // kernel (see 4821821 for details). We first expand stack to the top
848 // of yellow zone, then enable stack yellow zone (order is significant,
849 // enabling yellow zone first will crash JVM on SuSE Linux), so there
850 // is no gap between the last two virtual memory regions.
851
852 JavaThread *jt = (JavaThread *)thread;
853 address addr = jt->stack_reserved_zone_base();
854 assert(addr != NULL, "initialization problem?");
855 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
856
857 osthread->set_expanding_stack();
858 os::Linux::manually_expand_stack(jt, addr);
859 osthread->clear_expanding_stack();
860 }
861
862 // initialize signal mask for this thread
863 // and save the caller's signal mask
864 os::Linux::hotspot_sigmask(thread);
865
866 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
867 os::current_thread_id(), (uintx) pthread_self());
868
869 return true;
870 }
871
872 void os::pd_start_thread(Thread* thread) {
873 OSThread * osthread = thread->osthread();
874 assert(osthread->get_state() != INITIALIZED, "just checking");
875 Monitor* sync_with_child = osthread->startThread_lock();
876 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
877 sync_with_child->notify();
878 }
879
880 // Free Linux resources related to the OSThread
881 void os::free_thread(OSThread* osthread) {
882 assert(osthread != NULL, "osthread not set");
883
884 if (Thread::current()->osthread() == osthread) {
885 #ifdef ASSERT
886 sigset_t current;
887 sigemptyset(¤t);
888 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
889 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
890 #endif
891
892 // Restore caller's signal mask
893 sigset_t sigmask = osthread->caller_sigmask();
894 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
895 }
896
897 delete osthread;
898 }
899
900 //////////////////////////////////////////////////////////////////////////////
901 // initial thread
902
903 // Check if current thread is the initial thread, similar to Solaris thr_main.
904 bool os::Linux::is_initial_thread(void) {
905 char dummy;
906 // If called before init complete, thread stack bottom will be null.
907 // Can be called if fatal error occurs before initialization.
908 if (initial_thread_stack_bottom() == NULL) return false;
909 assert(initial_thread_stack_bottom() != NULL &&
910 initial_thread_stack_size() != 0,
911 "os::init did not locate initial thread's stack region");
912 if ((address)&dummy >= initial_thread_stack_bottom() &&
913 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
914 return true;
915 } else {
916 return false;
917 }
918 }
919
920 // Find the virtual memory area that contains addr
921 static bool find_vma(address addr, address* vma_low, address* vma_high) {
922 FILE *fp = fopen("/proc/self/maps", "r");
923 if (fp) {
924 address low, high;
925 while (!feof(fp)) {
926 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
927 if (low <= addr && addr < high) {
928 if (vma_low) *vma_low = low;
929 if (vma_high) *vma_high = high;
930 fclose(fp);
931 return true;
932 }
933 }
934 for (;;) {
935 int ch = fgetc(fp);
936 if (ch == EOF || ch == (int)'\n') break;
937 }
938 }
939 fclose(fp);
940 }
941 return false;
942 }
943
944 // Locate initial thread stack. This special handling of initial thread stack
945 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
946 // bogus value for initial thread.
947 void os::Linux::capture_initial_stack(size_t max_size) {
948 // stack size is the easy part, get it from RLIMIT_STACK
949 size_t stack_size;
950 struct rlimit rlim;
951 getrlimit(RLIMIT_STACK, &rlim);
952 stack_size = rlim.rlim_cur;
953
954 // 6308388: a bug in ld.so will relocate its own .data section to the
955 // lower end of primordial stack; reduce ulimit -s value a little bit
956 // so we won't install guard page on ld.so's data section.
957 stack_size -= 2 * page_size();
958
959 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
960 // 7.1, in both cases we will get 2G in return value.
961 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
962 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
963 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
964 // in case other parts in glibc still assumes 2M max stack size.
965 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
966 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
967 if (stack_size > 2 * K * K IA64_ONLY(*2)) {
968 stack_size = 2 * K * K IA64_ONLY(*2);
969 }
970 // Try to figure out where the stack base (top) is. This is harder.
971 //
972 // When an application is started, glibc saves the initial stack pointer in
973 // a global variable "__libc_stack_end", which is then used by system
974 // libraries. __libc_stack_end should be pretty close to stack top. The
975 // variable is available since the very early days. However, because it is
976 // a private interface, it could disappear in the future.
977 //
978 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
979 // to __libc_stack_end, it is very close to stack top, but isn't the real
980 // stack top. Note that /proc may not exist if VM is running as a chroot
981 // program, so reading /proc/<pid>/stat could fail. Also the contents of
982 // /proc/<pid>/stat could change in the future (though unlikely).
983 //
984 // We try __libc_stack_end first. If that doesn't work, look for
985 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
986 // as a hint, which should work well in most cases.
987
988 uintptr_t stack_start;
989
990 // try __libc_stack_end first
991 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
992 if (p && *p) {
993 stack_start = *p;
994 } else {
995 // see if we can get the start_stack field from /proc/self/stat
996 FILE *fp;
997 int pid;
998 char state;
999 int ppid;
1000 int pgrp;
1001 int session;
1002 int nr;
1003 int tpgrp;
1004 unsigned long flags;
1005 unsigned long minflt;
1006 unsigned long cminflt;
1007 unsigned long majflt;
1008 unsigned long cmajflt;
1009 unsigned long utime;
1010 unsigned long stime;
1011 long cutime;
1012 long cstime;
1013 long prio;
1014 long nice;
1015 long junk;
1016 long it_real;
1017 uintptr_t start;
1018 uintptr_t vsize;
1019 intptr_t rss;
1020 uintptr_t rsslim;
1021 uintptr_t scodes;
1022 uintptr_t ecode;
1023 int i;
1024
1025 // Figure what the primordial thread stack base is. Code is inspired
1026 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1027 // followed by command name surrounded by parentheses, state, etc.
1028 char stat[2048];
1029 int statlen;
1030
1031 fp = fopen("/proc/self/stat", "r");
1032 if (fp) {
1033 statlen = fread(stat, 1, 2047, fp);
1034 stat[statlen] = '\0';
1035 fclose(fp);
1036
1037 // Skip pid and the command string. Note that we could be dealing with
1038 // weird command names, e.g. user could decide to rename java launcher
1039 // to "java 1.4.2 :)", then the stat file would look like
1040 // 1234 (java 1.4.2 :)) R ... ...
1041 // We don't really need to know the command string, just find the last
1042 // occurrence of ")" and then start parsing from there. See bug 4726580.
1043 char * s = strrchr(stat, ')');
1044
1045 i = 0;
1046 if (s) {
1047 // Skip blank chars
1048 do { s++; } while (s && isspace(*s));
1049
1050 #define _UFM UINTX_FORMAT
1051 #define _DFM INTX_FORMAT
1052
1053 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1054 // 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
1055 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,
1056 &state, // 3 %c
1057 &ppid, // 4 %d
1058 &pgrp, // 5 %d
1059 &session, // 6 %d
1060 &nr, // 7 %d
1061 &tpgrp, // 8 %d
1062 &flags, // 9 %lu
1063 &minflt, // 10 %lu
1064 &cminflt, // 11 %lu
1065 &majflt, // 12 %lu
1066 &cmajflt, // 13 %lu
1067 &utime, // 14 %lu
1068 &stime, // 15 %lu
1069 &cutime, // 16 %ld
1070 &cstime, // 17 %ld
1071 &prio, // 18 %ld
1072 &nice, // 19 %ld
1073 &junk, // 20 %ld
1074 &it_real, // 21 %ld
1075 &start, // 22 UINTX_FORMAT
1076 &vsize, // 23 UINTX_FORMAT
1077 &rss, // 24 INTX_FORMAT
1078 &rsslim, // 25 UINTX_FORMAT
1079 &scodes, // 26 UINTX_FORMAT
1080 &ecode, // 27 UINTX_FORMAT
1081 &stack_start); // 28 UINTX_FORMAT
1082 }
1083
1084 #undef _UFM
1085 #undef _DFM
1086
1087 if (i != 28 - 2) {
1088 assert(false, "Bad conversion from /proc/self/stat");
1089 // product mode - assume we are the initial thread, good luck in the
1090 // embedded case.
1091 warning("Can't detect initial thread stack location - bad conversion");
1092 stack_start = (uintptr_t) &rlim;
1093 }
1094 } else {
1095 // For some reason we can't open /proc/self/stat (for example, running on
1096 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1097 // most cases, so don't abort:
1098 warning("Can't detect initial thread stack location - no /proc/self/stat");
1099 stack_start = (uintptr_t) &rlim;
1100 }
1101 }
1102
1103 // Now we have a pointer (stack_start) very close to the stack top, the
1104 // next thing to do is to figure out the exact location of stack top. We
1105 // can find out the virtual memory area that contains stack_start by
1106 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1107 // and its upper limit is the real stack top. (again, this would fail if
1108 // running inside chroot, because /proc may not exist.)
1109
1110 uintptr_t stack_top;
1111 address low, high;
1112 if (find_vma((address)stack_start, &low, &high)) {
1113 // success, "high" is the true stack top. (ignore "low", because initial
1114 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1115 stack_top = (uintptr_t)high;
1116 } else {
1117 // failed, likely because /proc/self/maps does not exist
1118 warning("Can't detect initial thread stack location - find_vma failed");
1119 // best effort: stack_start is normally within a few pages below the real
1120 // stack top, use it as stack top, and reduce stack size so we won't put
1121 // guard page outside stack.
1122 stack_top = stack_start;
1123 stack_size -= 16 * page_size();
1124 }
1125
1126 // stack_top could be partially down the page so align it
1127 stack_top = align_size_up(stack_top, page_size());
1128
1129 if (max_size && stack_size > max_size) {
1130 _initial_thread_stack_size = max_size;
1131 } else {
1132 _initial_thread_stack_size = stack_size;
1133 }
1134
1135 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1136 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1137 }
1138
1139 ////////////////////////////////////////////////////////////////////////////////
1140 // time support
1141
1142 // Time since start-up in seconds to a fine granularity.
1143 // Used by VMSelfDestructTimer and the MemProfiler.
1144 double os::elapsedTime() {
1145
1146 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1147 }
1148
1149 jlong os::elapsed_counter() {
1150 return javaTimeNanos() - initial_time_count;
1151 }
1152
1153 jlong os::elapsed_frequency() {
1154 return NANOSECS_PER_SEC; // nanosecond resolution
1155 }
1156
1157 bool os::supports_vtime() { return true; }
1158 bool os::enable_vtime() { return false; }
1159 bool os::vtime_enabled() { return false; }
1160
1161 double os::elapsedVTime() {
1162 struct rusage usage;
1163 int retval = getrusage(RUSAGE_THREAD, &usage);
1164 if (retval == 0) {
1165 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);
1166 } else {
1167 // better than nothing, but not much
1168 return elapsedTime();
1169 }
1170 }
1171
1172 jlong os::javaTimeMillis() {
1173 timeval time;
1174 int status = gettimeofday(&time, NULL);
1175 assert(status != -1, "linux error");
1176 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1177 }
1178
1179 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1180 timeval time;
1181 int status = gettimeofday(&time, NULL);
1182 assert(status != -1, "linux error");
1183 seconds = jlong(time.tv_sec);
1184 nanos = jlong(time.tv_usec) * 1000;
1185 }
1186
1187
1188 #ifndef CLOCK_MONOTONIC
1189 #define CLOCK_MONOTONIC (1)
1190 #endif
1191
1192 void os::Linux::clock_init() {
1193 // we do dlopen's in this particular order due to bug in linux
1194 // dynamical loader (see 6348968) leading to crash on exit
1195 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1196 if (handle == NULL) {
1197 handle = dlopen("librt.so", RTLD_LAZY);
1198 }
1199
1200 if (handle) {
1201 int (*clock_getres_func)(clockid_t, struct timespec*) =
1202 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1203 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1204 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1205 if (clock_getres_func && clock_gettime_func) {
1206 // See if monotonic clock is supported by the kernel. Note that some
1207 // early implementations simply return kernel jiffies (updated every
1208 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1209 // for nano time (though the monotonic property is still nice to have).
1210 // It's fixed in newer kernels, however clock_getres() still returns
1211 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1212 // resolution for now. Hopefully as people move to new kernels, this
1213 // won't be a problem.
1214 struct timespec res;
1215 struct timespec tp;
1216 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1217 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1218 // yes, monotonic clock is supported
1219 _clock_gettime = clock_gettime_func;
1220 return;
1221 } else {
1222 // close librt if there is no monotonic clock
1223 dlclose(handle);
1224 }
1225 }
1226 }
1227 warning("No monotonic clock was available - timed services may " \
1228 "be adversely affected if the time-of-day clock changes");
1229 }
1230
1231 #ifndef SYS_clock_getres
1232 #if defined(IA32) || defined(AMD64)
1233 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1234 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1235 #else
1236 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1237 #define sys_clock_getres(x,y) -1
1238 #endif
1239 #else
1240 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1241 #endif
1242
1243 void os::Linux::fast_thread_clock_init() {
1244 if (!UseLinuxPosixThreadCPUClocks) {
1245 return;
1246 }
1247 clockid_t clockid;
1248 struct timespec tp;
1249 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1250 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1251
1252 // Switch to using fast clocks for thread cpu time if
1253 // the sys_clock_getres() returns 0 error code.
1254 // Note, that some kernels may support the current thread
1255 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1256 // returned by the pthread_getcpuclockid().
1257 // If the fast Posix clocks are supported then the sys_clock_getres()
1258 // must return at least tp.tv_sec == 0 which means a resolution
1259 // better than 1 sec. This is extra check for reliability.
1260
1261 if (pthread_getcpuclockid_func &&
1262 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1263 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1264 _supports_fast_thread_cpu_time = true;
1265 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1266 }
1267 }
1268
1269 jlong os::javaTimeNanos() {
1270 if (os::supports_monotonic_clock()) {
1271 struct timespec tp;
1272 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1273 assert(status == 0, "gettime error");
1274 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1275 return result;
1276 } else {
1277 timeval time;
1278 int status = gettimeofday(&time, NULL);
1279 assert(status != -1, "linux error");
1280 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1281 return 1000 * usecs;
1282 }
1283 }
1284
1285 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1286 if (os::supports_monotonic_clock()) {
1287 info_ptr->max_value = ALL_64_BITS;
1288
1289 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1290 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1291 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1292 } else {
1293 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1294 info_ptr->max_value = ALL_64_BITS;
1295
1296 // gettimeofday is a real time clock so it skips
1297 info_ptr->may_skip_backward = true;
1298 info_ptr->may_skip_forward = true;
1299 }
1300
1301 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1302 }
1303
1304 // Return the real, user, and system times in seconds from an
1305 // arbitrary fixed point in the past.
1306 bool os::getTimesSecs(double* process_real_time,
1307 double* process_user_time,
1308 double* process_system_time) {
1309 struct tms ticks;
1310 clock_t real_ticks = times(&ticks);
1311
1312 if (real_ticks == (clock_t) (-1)) {
1313 return false;
1314 } else {
1315 double ticks_per_second = (double) clock_tics_per_sec;
1316 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1317 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1318 *process_real_time = ((double) real_ticks) / ticks_per_second;
1319
1320 return true;
1321 }
1322 }
1323
1324
1325 char * os::local_time_string(char *buf, size_t buflen) {
1326 struct tm t;
1327 time_t long_time;
1328 time(&long_time);
1329 localtime_r(&long_time, &t);
1330 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1331 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1332 t.tm_hour, t.tm_min, t.tm_sec);
1333 return buf;
1334 }
1335
1336 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1337 return localtime_r(clock, res);
1338 }
1339
1340 ////////////////////////////////////////////////////////////////////////////////
1341 // runtime exit support
1342
1343 // Note: os::shutdown() might be called very early during initialization, or
1344 // called from signal handler. Before adding something to os::shutdown(), make
1345 // sure it is async-safe and can handle partially initialized VM.
1346 void os::shutdown() {
1347
1348 // allow PerfMemory to attempt cleanup of any persistent resources
1349 perfMemory_exit();
1350
1351 // needs to remove object in file system
1352 AttachListener::abort();
1353
1354 // flush buffered output, finish log files
1355 ostream_abort();
1356
1357 // Check for abort hook
1358 abort_hook_t abort_hook = Arguments::abort_hook();
1359 if (abort_hook != NULL) {
1360 abort_hook();
1361 }
1362
1363 }
1364
1365 // Note: os::abort() might be called very early during initialization, or
1366 // called from signal handler. Before adding something to os::abort(), make
1367 // sure it is async-safe and can handle partially initialized VM.
1368 void os::abort(bool dump_core, void* siginfo, const void* context) {
1369 os::shutdown();
1370 if (dump_core) {
1371 #ifndef PRODUCT
1372 fdStream out(defaultStream::output_fd());
1373 out.print_raw("Current thread is ");
1374 char buf[16];
1375 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1376 out.print_raw_cr(buf);
1377 out.print_raw_cr("Dumping core ...");
1378 #endif
1379 ::abort(); // dump core
1380 }
1381
1382 ::exit(1);
1383 }
1384
1385 // Die immediately, no exit hook, no abort hook, no cleanup.
1386 void os::die() {
1387 ::abort();
1388 }
1389
1390
1391 // This method is a copy of JDK's sysGetLastErrorString
1392 // from src/solaris/hpi/src/system_md.c
1393
1394 size_t os::lasterror(char *buf, size_t len) {
1395 if (errno == 0) return 0;
1396
1397 const char *s = os::strerror(errno);
1398 size_t n = ::strlen(s);
1399 if (n >= len) {
1400 n = len - 1;
1401 }
1402 ::strncpy(buf, s, n);
1403 buf[n] = '\0';
1404 return n;
1405 }
1406
1407 // thread_id is kernel thread id (similar to Solaris LWP id)
1408 intx os::current_thread_id() { return os::Linux::gettid(); }
1409 int os::current_process_id() {
1410 return ::getpid();
1411 }
1412
1413 // DLL functions
1414
1415 const char* os::dll_file_extension() { return ".so"; }
1416
1417 // This must be hard coded because it's the system's temporary
1418 // directory not the java application's temp directory, ala java.io.tmpdir.
1419 const char* os::get_temp_directory() { return "/tmp"; }
1420
1421 static bool file_exists(const char* filename) {
1422 struct stat statbuf;
1423 if (filename == NULL || strlen(filename) == 0) {
1424 return false;
1425 }
1426 return os::stat(filename, &statbuf) == 0;
1427 }
1428
1429 bool os::dll_build_name(char* buffer, size_t buflen,
1430 const char* pname, const char* fname) {
1431 bool retval = false;
1432 // Copied from libhpi
1433 const size_t pnamelen = pname ? strlen(pname) : 0;
1434
1435 // Return error on buffer overflow.
1436 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1437 return retval;
1438 }
1439
1440 if (pnamelen == 0) {
1441 snprintf(buffer, buflen, "lib%s.so", fname);
1442 retval = true;
1443 } else if (strchr(pname, *os::path_separator()) != NULL) {
1444 int n;
1445 char** pelements = split_path(pname, &n);
1446 if (pelements == NULL) {
1447 return false;
1448 }
1449 for (int i = 0; i < n; i++) {
1450 // Really shouldn't be NULL, but check can't hurt
1451 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1452 continue; // skip the empty path values
1453 }
1454 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1455 if (file_exists(buffer)) {
1456 retval = true;
1457 break;
1458 }
1459 }
1460 // release the storage
1461 for (int i = 0; i < n; i++) {
1462 if (pelements[i] != NULL) {
1463 FREE_C_HEAP_ARRAY(char, pelements[i]);
1464 }
1465 }
1466 if (pelements != NULL) {
1467 FREE_C_HEAP_ARRAY(char*, pelements);
1468 }
1469 } else {
1470 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1471 retval = true;
1472 }
1473 return retval;
1474 }
1475
1476 // check if addr is inside libjvm.so
1477 bool os::address_is_in_vm(address addr) {
1478 static address libjvm_base_addr;
1479 Dl_info dlinfo;
1480
1481 if (libjvm_base_addr == NULL) {
1482 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1483 libjvm_base_addr = (address)dlinfo.dli_fbase;
1484 }
1485 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1486 }
1487
1488 if (dladdr((void *)addr, &dlinfo) != 0) {
1489 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1490 }
1491
1492 return false;
1493 }
1494
1495 bool os::dll_address_to_function_name(address addr, char *buf,
1496 int buflen, int *offset,
1497 bool demangle) {
1498 // buf is not optional, but offset is optional
1499 assert(buf != NULL, "sanity check");
1500
1501 Dl_info dlinfo;
1502
1503 if (dladdr((void*)addr, &dlinfo) != 0) {
1504 // see if we have a matching symbol
1505 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1506 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1507 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1508 }
1509 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1510 return true;
1511 }
1512 // no matching symbol so try for just file info
1513 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1514 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1515 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1516 return true;
1517 }
1518 }
1519 }
1520
1521 buf[0] = '\0';
1522 if (offset != NULL) *offset = -1;
1523 return false;
1524 }
1525
1526 struct _address_to_library_name {
1527 address addr; // input : memory address
1528 size_t buflen; // size of fname
1529 char* fname; // output: library name
1530 address base; // library base addr
1531 };
1532
1533 static int address_to_library_name_callback(struct dl_phdr_info *info,
1534 size_t size, void *data) {
1535 int i;
1536 bool found = false;
1537 address libbase = NULL;
1538 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1539
1540 // iterate through all loadable segments
1541 for (i = 0; i < info->dlpi_phnum; i++) {
1542 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1543 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1544 // base address of a library is the lowest address of its loaded
1545 // segments.
1546 if (libbase == NULL || libbase > segbase) {
1547 libbase = segbase;
1548 }
1549 // see if 'addr' is within current segment
1550 if (segbase <= d->addr &&
1551 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1552 found = true;
1553 }
1554 }
1555 }
1556
1557 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1558 // so dll_address_to_library_name() can fall through to use dladdr() which
1559 // can figure out executable name from argv[0].
1560 if (found && info->dlpi_name && info->dlpi_name[0]) {
1561 d->base = libbase;
1562 if (d->fname) {
1563 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1564 }
1565 return 1;
1566 }
1567 return 0;
1568 }
1569
1570 bool os::dll_address_to_library_name(address addr, char* buf,
1571 int buflen, int* offset) {
1572 // buf is not optional, but offset is optional
1573 assert(buf != NULL, "sanity check");
1574
1575 Dl_info dlinfo;
1576 struct _address_to_library_name data;
1577
1578 // There is a bug in old glibc dladdr() implementation that it could resolve
1579 // to wrong library name if the .so file has a base address != NULL. Here
1580 // we iterate through the program headers of all loaded libraries to find
1581 // out which library 'addr' really belongs to. This workaround can be
1582 // removed once the minimum requirement for glibc is moved to 2.3.x.
1583 data.addr = addr;
1584 data.fname = buf;
1585 data.buflen = buflen;
1586 data.base = NULL;
1587 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1588
1589 if (rslt) {
1590 // buf already contains library name
1591 if (offset) *offset = addr - data.base;
1592 return true;
1593 }
1594 if (dladdr((void*)addr, &dlinfo) != 0) {
1595 if (dlinfo.dli_fname != NULL) {
1596 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1597 }
1598 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1599 *offset = addr - (address)dlinfo.dli_fbase;
1600 }
1601 return true;
1602 }
1603
1604 buf[0] = '\0';
1605 if (offset) *offset = -1;
1606 return false;
1607 }
1608
1609 // Loads .dll/.so and
1610 // in case of error it checks if .dll/.so was built for the
1611 // same architecture as Hotspot is running on
1612
1613
1614 // Remember the stack's state. The Linux dynamic linker will change
1615 // the stack to 'executable' at most once, so we must safepoint only once.
1616 bool os::Linux::_stack_is_executable = false;
1617
1618 // VM operation that loads a library. This is necessary if stack protection
1619 // of the Java stacks can be lost during loading the library. If we
1620 // do not stop the Java threads, they can stack overflow before the stacks
1621 // are protected again.
1622 class VM_LinuxDllLoad: public VM_Operation {
1623 private:
1624 const char *_filename;
1625 char *_ebuf;
1626 int _ebuflen;
1627 void *_lib;
1628 public:
1629 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1630 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1631 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1632 void doit() {
1633 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1634 os::Linux::_stack_is_executable = true;
1635 }
1636 void* loaded_library() { return _lib; }
1637 };
1638
1639 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1640 void * result = NULL;
1641 bool load_attempted = false;
1642
1643 // Check whether the library to load might change execution rights
1644 // of the stack. If they are changed, the protection of the stack
1645 // guard pages will be lost. We need a safepoint to fix this.
1646 //
1647 // See Linux man page execstack(8) for more info.
1648 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1649 ElfFile ef(filename);
1650 if (!ef.specifies_noexecstack()) {
1651 if (!is_init_completed()) {
1652 os::Linux::_stack_is_executable = true;
1653 // This is OK - No Java threads have been created yet, and hence no
1654 // stack guard pages to fix.
1655 //
1656 // This should happen only when you are building JDK7 using a very
1657 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1658 //
1659 // Dynamic loader will make all stacks executable after
1660 // this function returns, and will not do that again.
1661 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1662 } else {
1663 warning("You have loaded library %s which might have disabled stack guard. "
1664 "The VM will try to fix the stack guard now.\n"
1665 "It's highly recommended that you fix the library with "
1666 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1667 filename);
1668
1669 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1670 JavaThread *jt = JavaThread::current();
1671 if (jt->thread_state() != _thread_in_native) {
1672 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1673 // that requires ExecStack. Cannot enter safe point. Let's give up.
1674 warning("Unable to fix stack guard. Giving up.");
1675 } else {
1676 if (!LoadExecStackDllInVMThread) {
1677 // This is for the case where the DLL has an static
1678 // constructor function that executes JNI code. We cannot
1679 // load such DLLs in the VMThread.
1680 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1681 }
1682
1683 ThreadInVMfromNative tiv(jt);
1684 debug_only(VMNativeEntryWrapper vew;)
1685
1686 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1687 VMThread::execute(&op);
1688 if (LoadExecStackDllInVMThread) {
1689 result = op.loaded_library();
1690 }
1691 load_attempted = true;
1692 }
1693 }
1694 }
1695 }
1696
1697 if (!load_attempted) {
1698 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1699 }
1700
1701 if (result != NULL) {
1702 // Successful loading
1703 return result;
1704 }
1705
1706 Elf32_Ehdr elf_head;
1707 int diag_msg_max_length=ebuflen-strlen(ebuf);
1708 char* diag_msg_buf=ebuf+strlen(ebuf);
1709
1710 if (diag_msg_max_length==0) {
1711 // No more space in ebuf for additional diagnostics message
1712 return NULL;
1713 }
1714
1715
1716 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1717
1718 if (file_descriptor < 0) {
1719 // Can't open library, report dlerror() message
1720 return NULL;
1721 }
1722
1723 bool failed_to_read_elf_head=
1724 (sizeof(elf_head)!=
1725 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1726
1727 ::close(file_descriptor);
1728 if (failed_to_read_elf_head) {
1729 // file i/o error - report dlerror() msg
1730 return NULL;
1731 }
1732
1733 typedef struct {
1734 Elf32_Half code; // Actual value as defined in elf.h
1735 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1736 char elf_class; // 32 or 64 bit
1737 char endianess; // MSB or LSB
1738 char* name; // String representation
1739 } arch_t;
1740
1741 #ifndef EM_486
1742 #define EM_486 6 /* Intel 80486 */
1743 #endif
1744 #ifndef EM_AARCH64
1745 #define EM_AARCH64 183 /* ARM AARCH64 */
1746 #endif
1747
1748 static const arch_t arch_array[]={
1749 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1750 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1751 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1752 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1753 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1754 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1755 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1756 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1757 #if defined(VM_LITTLE_ENDIAN)
1758 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1759 #else
1760 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64 LE"},
1761 #endif
1762 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1763 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1764 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1765 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1766 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1767 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1768 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1769 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1770 };
1771
1772 #if (defined IA32)
1773 static Elf32_Half running_arch_code=EM_386;
1774 #elif (defined AMD64)
1775 static Elf32_Half running_arch_code=EM_X86_64;
1776 #elif (defined IA64)
1777 static Elf32_Half running_arch_code=EM_IA_64;
1778 #elif (defined __sparc) && (defined _LP64)
1779 static Elf32_Half running_arch_code=EM_SPARCV9;
1780 #elif (defined __sparc) && (!defined _LP64)
1781 static Elf32_Half running_arch_code=EM_SPARC;
1782 #elif (defined __powerpc64__)
1783 static Elf32_Half running_arch_code=EM_PPC64;
1784 #elif (defined __powerpc__)
1785 static Elf32_Half running_arch_code=EM_PPC;
1786 #elif (defined ARM)
1787 static Elf32_Half running_arch_code=EM_ARM;
1788 #elif (defined S390)
1789 static Elf32_Half running_arch_code=EM_S390;
1790 #elif (defined ALPHA)
1791 static Elf32_Half running_arch_code=EM_ALPHA;
1792 #elif (defined MIPSEL)
1793 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1794 #elif (defined PARISC)
1795 static Elf32_Half running_arch_code=EM_PARISC;
1796 #elif (defined MIPS)
1797 static Elf32_Half running_arch_code=EM_MIPS;
1798 #elif (defined M68K)
1799 static Elf32_Half running_arch_code=EM_68K;
1800 #elif (defined AARCH64)
1801 static Elf32_Half running_arch_code=EM_AARCH64;
1802 #else
1803 #error Method os::dll_load requires that one of following is defined:\
1804 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1805 #endif
1806
1807 // Identify compatability class for VM's architecture and library's architecture
1808 // Obtain string descriptions for architectures
1809
1810 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1811 int running_arch_index=-1;
1812
1813 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1814 if (running_arch_code == arch_array[i].code) {
1815 running_arch_index = i;
1816 }
1817 if (lib_arch.code == arch_array[i].code) {
1818 lib_arch.compat_class = arch_array[i].compat_class;
1819 lib_arch.name = arch_array[i].name;
1820 }
1821 }
1822
1823 assert(running_arch_index != -1,
1824 "Didn't find running architecture code (running_arch_code) in arch_array");
1825 if (running_arch_index == -1) {
1826 // Even though running architecture detection failed
1827 // we may still continue with reporting dlerror() message
1828 return NULL;
1829 }
1830
1831 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1832 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1833 return NULL;
1834 }
1835
1836 #ifndef S390
1837 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1838 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1839 return NULL;
1840 }
1841 #endif // !S390
1842
1843 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1844 if (lib_arch.name!=NULL) {
1845 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1846 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1847 lib_arch.name, arch_array[running_arch_index].name);
1848 } else {
1849 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1850 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1851 lib_arch.code,
1852 arch_array[running_arch_index].name);
1853 }
1854 }
1855
1856 return NULL;
1857 }
1858
1859 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1860 int ebuflen) {
1861 void * result = ::dlopen(filename, RTLD_LAZY);
1862 if (result == NULL) {
1863 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1864 ebuf[ebuflen-1] = '\0';
1865 }
1866 return result;
1867 }
1868
1869 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1870 int ebuflen) {
1871 void * result = NULL;
1872 if (LoadExecStackDllInVMThread) {
1873 result = dlopen_helper(filename, ebuf, ebuflen);
1874 }
1875
1876 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1877 // library that requires an executable stack, or which does not have this
1878 // stack attribute set, dlopen changes the stack attribute to executable. The
1879 // read protection of the guard pages gets lost.
1880 //
1881 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1882 // may have been queued at the same time.
1883
1884 if (!_stack_is_executable) {
1885 JavaThread *jt = Threads::first();
1886
1887 while (jt) {
1888 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1889 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1890 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1891 warning("Attempt to reguard stack yellow zone failed.");
1892 }
1893 }
1894 jt = jt->next();
1895 }
1896 }
1897
1898 return result;
1899 }
1900
1901 void* os::dll_lookup(void* handle, const char* name) {
1902 void* res = dlsym(handle, name);
1903 return res;
1904 }
1905
1906 void* os::get_default_process_handle() {
1907 return (void*)::dlopen(NULL, RTLD_LAZY);
1908 }
1909
1910 static bool _print_ascii_file(const char* filename, outputStream* st) {
1911 int fd = ::open(filename, O_RDONLY);
1912 if (fd == -1) {
1913 return false;
1914 }
1915
1916 char buf[32];
1917 int bytes;
1918 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1919 st->print_raw(buf, bytes);
1920 }
1921
1922 ::close(fd);
1923
1924 return true;
1925 }
1926
1927 void os::print_dll_info(outputStream *st) {
1928 st->print_cr("Dynamic libraries:");
1929
1930 char fname[32];
1931 pid_t pid = os::Linux::gettid();
1932
1933 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1934
1935 if (!_print_ascii_file(fname, st)) {
1936 st->print("Can not get library information for pid = %d\n", pid);
1937 }
1938 }
1939
1940 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1941 FILE *procmapsFile = NULL;
1942
1943 // Open the procfs maps file for the current process
1944 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1945 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1946 char line[PATH_MAX + 100];
1947
1948 // Read line by line from 'file'
1949 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1950 u8 base, top, offset, inode;
1951 char permissions[5];
1952 char device[6];
1953 char name[PATH_MAX + 1];
1954
1955 // Parse fields from line
1956 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1957 &base, &top, permissions, &offset, device, &inode, name);
1958
1959 // Filter by device id '00:00' so that we only get file system mapped files.
1960 if (strcmp(device, "00:00") != 0) {
1961
1962 // Call callback with the fields of interest
1963 if(callback(name, (address)base, (address)top, param)) {
1964 // Oops abort, callback aborted
1965 fclose(procmapsFile);
1966 return 1;
1967 }
1968 }
1969 }
1970 fclose(procmapsFile);
1971 }
1972 return 0;
1973 }
1974
1975 void os::print_os_info_brief(outputStream* st) {
1976 os::Linux::print_distro_info(st);
1977
1978 os::Posix::print_uname_info(st);
1979
1980 os::Linux::print_libversion_info(st);
1981
1982 }
1983
1984 void os::print_os_info(outputStream* st) {
1985 st->print("OS:");
1986
1987 os::Linux::print_distro_info(st);
1988
1989 os::Posix::print_uname_info(st);
1990
1991 // Print warning if unsafe chroot environment detected
1992 if (unsafe_chroot_detected) {
1993 st->print("WARNING!! ");
1994 st->print_cr("%s", unstable_chroot_error);
1995 }
1996
1997 os::Linux::print_libversion_info(st);
1998
1999 os::Posix::print_rlimit_info(st);
2000
2001 os::Posix::print_load_average(st);
2002
2003 os::Linux::print_full_memory_info(st);
2004 }
2005
2006 // Try to identify popular distros.
2007 // Most Linux distributions have a /etc/XXX-release file, which contains
2008 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2009 // file that also contains the OS version string. Some have more than one
2010 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2011 // /etc/redhat-release.), so the order is important.
2012 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2013 // their own specific XXX-release file as well as a redhat-release file.
2014 // Because of this the XXX-release file needs to be searched for before the
2015 // redhat-release file.
2016 // Since Red Hat has a lsb-release file that is not very descriptive the
2017 // search for redhat-release needs to be before lsb-release.
2018 // Since the lsb-release file is the new standard it needs to be searched
2019 // before the older style release files.
2020 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2021 // next to last resort. The os-release file is a new standard that contains
2022 // distribution information and the system-release file seems to be an old
2023 // standard that has been replaced by the lsb-release and os-release files.
2024 // Searching for the debian_version file is the last resort. It contains
2025 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2026 // "Debian " is printed before the contents of the debian_version file.
2027
2028 const char* distro_files[] = {
2029 "/etc/oracle-release",
2030 "/etc/mandriva-release",
2031 "/etc/mandrake-release",
2032 "/etc/sun-release",
2033 "/etc/redhat-release",
2034 "/etc/lsb-release",
2035 "/etc/SuSE-release",
2036 "/etc/turbolinux-release",
2037 "/etc/gentoo-release",
2038 "/etc/ltib-release",
2039 "/etc/angstrom-version",
2040 "/etc/system-release",
2041 "/etc/os-release",
2042 NULL };
2043
2044 void os::Linux::print_distro_info(outputStream* st) {
2045 for (int i = 0;; i++) {
2046 const char* file = distro_files[i];
2047 if (file == NULL) {
2048 break; // done
2049 }
2050 // If file prints, we found it.
2051 if (_print_ascii_file(file, st)) {
2052 return;
2053 }
2054 }
2055
2056 if (file_exists("/etc/debian_version")) {
2057 st->print("Debian ");
2058 _print_ascii_file("/etc/debian_version", st);
2059 } else {
2060 st->print("Linux");
2061 }
2062 st->cr();
2063 }
2064
2065 static void parse_os_info(char* distro, size_t length, const char* file) {
2066 FILE* fp = fopen(file, "r");
2067 if (fp != NULL) {
2068 char buf[256];
2069 // get last line of the file.
2070 while (fgets(buf, sizeof(buf), fp)) { }
2071 // Edit out extra stuff in expected ubuntu format
2072 if (strstr(buf, "DISTRIB_DESCRIPTION=") != 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 } else {
2087 // if not in expected Ubuntu format, print out whole line minus \n
2088 char* nl = strchr(buf, '\n');
2089 if (nl != NULL) *nl = '\0';
2090 strncpy(distro, buf, length);
2091 }
2092 // close distro file
2093 fclose(fp);
2094 }
2095 }
2096
2097 void os::get_summary_os_info(char* buf, size_t buflen) {
2098 for (int i = 0;; i++) {
2099 const char* file = distro_files[i];
2100 if (file == NULL) {
2101 break; // ran out of distro_files
2102 }
2103 if (file_exists(file)) {
2104 parse_os_info(buf, buflen, file);
2105 return;
2106 }
2107 }
2108 // special case for debian
2109 if (file_exists("/etc/debian_version")) {
2110 strncpy(buf, "Debian ", buflen);
2111 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2112 } else {
2113 strncpy(buf, "Linux", buflen);
2114 }
2115 }
2116
2117 void os::Linux::print_libversion_info(outputStream* st) {
2118 // libc, pthread
2119 st->print("libc:");
2120 st->print("%s ", os::Linux::glibc_version());
2121 st->print("%s ", os::Linux::libpthread_version());
2122 st->cr();
2123 }
2124
2125 void os::Linux::print_full_memory_info(outputStream* st) {
2126 st->print("\n/proc/meminfo:\n");
2127 _print_ascii_file("/proc/meminfo", st);
2128 st->cr();
2129 }
2130
2131 void os::print_memory_info(outputStream* st) {
2132
2133 st->print("Memory:");
2134 st->print(" %dk page", os::vm_page_size()>>10);
2135
2136 // values in struct sysinfo are "unsigned long"
2137 struct sysinfo si;
2138 sysinfo(&si);
2139
2140 st->print(", physical " UINT64_FORMAT "k",
2141 os::physical_memory() >> 10);
2142 st->print("(" UINT64_FORMAT "k free)",
2143 os::available_memory() >> 10);
2144 st->print(", swap " UINT64_FORMAT "k",
2145 ((jlong)si.totalswap * si.mem_unit) >> 10);
2146 st->print("(" UINT64_FORMAT "k free)",
2147 ((jlong)si.freeswap * si.mem_unit) >> 10);
2148 st->cr();
2149 }
2150
2151 // Print the first "model name" line and the first "flags" line
2152 // that we find and nothing more. We assume "model name" comes
2153 // before "flags" so if we find a second "model name", then the
2154 // "flags" field is considered missing.
2155 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2156 #if defined(IA32) || defined(AMD64)
2157 // Other platforms have less repetitive cpuinfo files
2158 FILE *fp = fopen("/proc/cpuinfo", "r");
2159 if (fp) {
2160 while (!feof(fp)) {
2161 if (fgets(buf, buflen, fp)) {
2162 // Assume model name comes before flags
2163 bool model_name_printed = false;
2164 if (strstr(buf, "model name") != NULL) {
2165 if (!model_name_printed) {
2166 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2167 st->print_raw(buf);
2168 model_name_printed = true;
2169 } else {
2170 // model name printed but not flags? Odd, just return
2171 fclose(fp);
2172 return true;
2173 }
2174 }
2175 // print the flags line too
2176 if (strstr(buf, "flags") != NULL) {
2177 st->print_raw(buf);
2178 fclose(fp);
2179 return true;
2180 }
2181 }
2182 }
2183 fclose(fp);
2184 }
2185 #endif // x86 platforms
2186 return false;
2187 }
2188
2189 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2190 // Only print the model name if the platform provides this as a summary
2191 if (!print_model_name_and_flags(st, buf, buflen)) {
2192 st->print("\n/proc/cpuinfo:\n");
2193 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2194 st->print_cr(" <Not Available>");
2195 }
2196 }
2197 }
2198
2199 #if defined(AMD64) || defined(IA32) || defined(X32)
2200 const char* search_string = "model name";
2201 #elif defined(SPARC)
2202 const char* search_string = "cpu";
2203 #elif defined(PPC64)
2204 const char* search_string = "cpu";
2205 #else
2206 const char* search_string = "Processor";
2207 #endif
2208
2209 // Parses the cpuinfo file for string representing the model name.
2210 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2211 FILE* fp = fopen("/proc/cpuinfo", "r");
2212 if (fp != NULL) {
2213 while (!feof(fp)) {
2214 char buf[256];
2215 if (fgets(buf, sizeof(buf), fp)) {
2216 char* start = strstr(buf, search_string);
2217 if (start != NULL) {
2218 char *ptr = start + strlen(search_string);
2219 char *end = buf + strlen(buf);
2220 while (ptr != end) {
2221 // skip whitespace and colon for the rest of the name.
2222 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2223 break;
2224 }
2225 ptr++;
2226 }
2227 if (ptr != end) {
2228 // reasonable string, get rid of newline and keep the rest
2229 char* nl = strchr(buf, '\n');
2230 if (nl != NULL) *nl = '\0';
2231 strncpy(cpuinfo, ptr, length);
2232 fclose(fp);
2233 return;
2234 }
2235 }
2236 }
2237 }
2238 fclose(fp);
2239 }
2240 // cpuinfo not found or parsing failed, just print generic string. The entire
2241 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2242 #if defined(AMD64)
2243 strncpy(cpuinfo, "x86_64", length);
2244 #elif defined(IA32)
2245 strncpy(cpuinfo, "x86_32", length);
2246 #elif defined(IA64)
2247 strncpy(cpuinfo, "IA64", length);
2248 #elif defined(SPARC)
2249 strncpy(cpuinfo, "sparcv9", length);
2250 #elif defined(AARCH64)
2251 strncpy(cpuinfo, "AArch64", length);
2252 #elif defined(ARM)
2253 strncpy(cpuinfo, "ARM", length);
2254 #elif defined(PPC)
2255 strncpy(cpuinfo, "PPC64", length);
2256 #elif defined(ZERO_LIBARCH)
2257 strncpy(cpuinfo, ZERO_LIBARCH, length);
2258 #else
2259 strncpy(cpuinfo, "unknown", length);
2260 #endif
2261 }
2262
2263 static void print_signal_handler(outputStream* st, int sig,
2264 char* buf, size_t buflen);
2265
2266 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2267 st->print_cr("Signal Handlers:");
2268 print_signal_handler(st, SIGSEGV, buf, buflen);
2269 print_signal_handler(st, SIGBUS , buf, buflen);
2270 print_signal_handler(st, SIGFPE , buf, buflen);
2271 print_signal_handler(st, SIGPIPE, buf, buflen);
2272 print_signal_handler(st, SIGXFSZ, buf, buflen);
2273 print_signal_handler(st, SIGILL , buf, buflen);
2274 print_signal_handler(st, SR_signum, buf, buflen);
2275 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2276 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2277 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2278 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2279 #if defined(PPC64)
2280 print_signal_handler(st, SIGTRAP, buf, buflen);
2281 #endif
2282 }
2283
2284 static char saved_jvm_path[MAXPATHLEN] = {0};
2285
2286 // Find the full path to the current module, libjvm.so
2287 void os::jvm_path(char *buf, jint buflen) {
2288 // Error checking.
2289 if (buflen < MAXPATHLEN) {
2290 assert(false, "must use a large-enough buffer");
2291 buf[0] = '\0';
2292 return;
2293 }
2294 // Lazy resolve the path to current module.
2295 if (saved_jvm_path[0] != 0) {
2296 strcpy(buf, saved_jvm_path);
2297 return;
2298 }
2299
2300 char dli_fname[MAXPATHLEN];
2301 bool ret = dll_address_to_library_name(
2302 CAST_FROM_FN_PTR(address, os::jvm_path),
2303 dli_fname, sizeof(dli_fname), NULL);
2304 assert(ret, "cannot locate libjvm");
2305 char *rp = NULL;
2306 if (ret && dli_fname[0] != '\0') {
2307 rp = realpath(dli_fname, buf);
2308 }
2309 if (rp == NULL) {
2310 return;
2311 }
2312
2313 if (Arguments::sun_java_launcher_is_altjvm()) {
2314 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2315 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2316 // If "/jre/lib/" appears at the right place in the string, then
2317 // assume we are installed in a JDK and we're done. Otherwise, check
2318 // for a JAVA_HOME environment variable and fix up the path so it
2319 // looks like libjvm.so is installed there (append a fake suffix
2320 // hotspot/libjvm.so).
2321 const char *p = buf + strlen(buf) - 1;
2322 for (int count = 0; p > buf && count < 5; ++count) {
2323 for (--p; p > buf && *p != '/'; --p)
2324 /* empty */ ;
2325 }
2326
2327 if (strncmp(p, "/jre/lib/", 9) != 0) {
2328 // Look for JAVA_HOME in the environment.
2329 char* java_home_var = ::getenv("JAVA_HOME");
2330 if (java_home_var != NULL && java_home_var[0] != 0) {
2331 char* jrelib_p;
2332 int len;
2333
2334 // Check the current module name "libjvm.so".
2335 p = strrchr(buf, '/');
2336 if (p == NULL) {
2337 return;
2338 }
2339 assert(strstr(p, "/libjvm") == p, "invalid library name");
2340
2341 rp = realpath(java_home_var, buf);
2342 if (rp == NULL) {
2343 return;
2344 }
2345
2346 // determine if this is a legacy image or modules image
2347 // modules image doesn't have "jre" subdirectory
2348 len = strlen(buf);
2349 assert(len < buflen, "Ran out of buffer room");
2350 jrelib_p = buf + len;
2351 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2352 if (0 != access(buf, F_OK)) {
2353 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2354 }
2355
2356 if (0 == access(buf, F_OK)) {
2357 // Use current module name "libjvm.so"
2358 len = strlen(buf);
2359 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2360 } else {
2361 // Go back to path of .so
2362 rp = realpath(dli_fname, buf);
2363 if (rp == NULL) {
2364 return;
2365 }
2366 }
2367 }
2368 }
2369 }
2370
2371 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2372 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2373 }
2374
2375 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2376 // no prefix required, not even "_"
2377 }
2378
2379 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2380 // no suffix required
2381 }
2382
2383 ////////////////////////////////////////////////////////////////////////////////
2384 // sun.misc.Signal support
2385
2386 static volatile jint sigint_count = 0;
2387
2388 static void UserHandler(int sig, void *siginfo, void *context) {
2389 // 4511530 - sem_post is serialized and handled by the manager thread. When
2390 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2391 // don't want to flood the manager thread with sem_post requests.
2392 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2393 return;
2394 }
2395
2396 // Ctrl-C is pressed during error reporting, likely because the error
2397 // handler fails to abort. Let VM die immediately.
2398 if (sig == SIGINT && is_error_reported()) {
2399 os::die();
2400 }
2401
2402 os::signal_notify(sig);
2403 }
2404
2405 void* os::user_handler() {
2406 return CAST_FROM_FN_PTR(void*, UserHandler);
2407 }
2408
2409 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2410 struct timespec ts;
2411 // Semaphore's are always associated with CLOCK_REALTIME
2412 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2413 // see unpackTime for discussion on overflow checking
2414 if (sec >= MAX_SECS) {
2415 ts.tv_sec += MAX_SECS;
2416 ts.tv_nsec = 0;
2417 } else {
2418 ts.tv_sec += sec;
2419 ts.tv_nsec += nsec;
2420 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2421 ts.tv_nsec -= NANOSECS_PER_SEC;
2422 ++ts.tv_sec; // note: this must be <= max_secs
2423 }
2424 }
2425
2426 return ts;
2427 }
2428
2429 extern "C" {
2430 typedef void (*sa_handler_t)(int);
2431 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2432 }
2433
2434 void* os::signal(int signal_number, void* handler) {
2435 struct sigaction sigAct, oldSigAct;
2436
2437 sigfillset(&(sigAct.sa_mask));
2438 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2439 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2440
2441 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2442 // -1 means registration failed
2443 return (void *)-1;
2444 }
2445
2446 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2447 }
2448
2449 void os::signal_raise(int signal_number) {
2450 ::raise(signal_number);
2451 }
2452
2453 // The following code is moved from os.cpp for making this
2454 // code platform specific, which it is by its very nature.
2455
2456 // Will be modified when max signal is changed to be dynamic
2457 int os::sigexitnum_pd() {
2458 return NSIG;
2459 }
2460
2461 // a counter for each possible signal value
2462 static volatile jint pending_signals[NSIG+1] = { 0 };
2463
2464 // Linux(POSIX) specific hand shaking semaphore.
2465 static sem_t sig_sem;
2466 static PosixSemaphore sr_semaphore;
2467
2468 void os::signal_init_pd() {
2469 // Initialize signal structures
2470 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2471
2472 // Initialize signal semaphore
2473 ::sem_init(&sig_sem, 0, 0);
2474 }
2475
2476 void os::signal_notify(int sig) {
2477 Atomic::inc(&pending_signals[sig]);
2478 ::sem_post(&sig_sem);
2479 }
2480
2481 static int check_pending_signals(bool wait) {
2482 Atomic::store(0, &sigint_count);
2483 for (;;) {
2484 for (int i = 0; i < NSIG + 1; i++) {
2485 jint n = pending_signals[i];
2486 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2487 return i;
2488 }
2489 }
2490 if (!wait) {
2491 return -1;
2492 }
2493 JavaThread *thread = JavaThread::current();
2494 ThreadBlockInVM tbivm(thread);
2495
2496 bool threadIsSuspended;
2497 do {
2498 thread->set_suspend_equivalent();
2499 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2500 ::sem_wait(&sig_sem);
2501
2502 // were we externally suspended while we were waiting?
2503 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2504 if (threadIsSuspended) {
2505 // The semaphore has been incremented, but while we were waiting
2506 // another thread suspended us. We don't want to continue running
2507 // while suspended because that would surprise the thread that
2508 // suspended us.
2509 ::sem_post(&sig_sem);
2510
2511 thread->java_suspend_self();
2512 }
2513 } while (threadIsSuspended);
2514 }
2515 }
2516
2517 int os::signal_lookup() {
2518 return check_pending_signals(false);
2519 }
2520
2521 int os::signal_wait() {
2522 return check_pending_signals(true);
2523 }
2524
2525 ////////////////////////////////////////////////////////////////////////////////
2526 // Virtual Memory
2527
2528 int os::vm_page_size() {
2529 // Seems redundant as all get out
2530 assert(os::Linux::page_size() != -1, "must call os::init");
2531 return os::Linux::page_size();
2532 }
2533
2534 // Solaris allocates memory by pages.
2535 int os::vm_allocation_granularity() {
2536 assert(os::Linux::page_size() != -1, "must call os::init");
2537 return os::Linux::page_size();
2538 }
2539
2540 // Rationale behind this function:
2541 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2542 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2543 // samples for JITted code. Here we create private executable mapping over the code cache
2544 // and then we can use standard (well, almost, as mapping can change) way to provide
2545 // info for the reporting script by storing timestamp and location of symbol
2546 void linux_wrap_code(char* base, size_t size) {
2547 static volatile jint cnt = 0;
2548
2549 if (!UseOprofile) {
2550 return;
2551 }
2552
2553 char buf[PATH_MAX+1];
2554 int num = Atomic::add(1, &cnt);
2555
2556 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2557 os::get_temp_directory(), os::current_process_id(), num);
2558 unlink(buf);
2559
2560 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2561
2562 if (fd != -1) {
2563 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2564 if (rv != (off_t)-1) {
2565 if (::write(fd, "", 1) == 1) {
2566 mmap(base, size,
2567 PROT_READ|PROT_WRITE|PROT_EXEC,
2568 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2569 }
2570 }
2571 ::close(fd);
2572 unlink(buf);
2573 }
2574 }
2575
2576 static bool recoverable_mmap_error(int err) {
2577 // See if the error is one we can let the caller handle. This
2578 // list of errno values comes from JBS-6843484. I can't find a
2579 // Linux man page that documents this specific set of errno
2580 // values so while this list currently matches Solaris, it may
2581 // change as we gain experience with this failure mode.
2582 switch (err) {
2583 case EBADF:
2584 case EINVAL:
2585 case ENOTSUP:
2586 // let the caller deal with these errors
2587 return true;
2588
2589 default:
2590 // Any remaining errors on this OS can cause our reserved mapping
2591 // to be lost. That can cause confusion where different data
2592 // structures think they have the same memory mapped. The worst
2593 // scenario is if both the VM and a library think they have the
2594 // same memory mapped.
2595 return false;
2596 }
2597 }
2598
2599 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2600 int err) {
2601 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2602 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2603 os::strerror(err), err);
2604 }
2605
2606 static void warn_fail_commit_memory(char* addr, size_t size,
2607 size_t alignment_hint, bool exec,
2608 int err) {
2609 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2610 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2611 alignment_hint, exec, os::strerror(err), err);
2612 }
2613
2614 // NOTE: Linux kernel does not really reserve the pages for us.
2615 // All it does is to check if there are enough free pages
2616 // left at the time of mmap(). This could be a potential
2617 // problem.
2618 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2619 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2620 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2621 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2622 if (res != (uintptr_t) MAP_FAILED) {
2623 if (UseNUMAInterleaving) {
2624 numa_make_global(addr, size);
2625 }
2626 return 0;
2627 }
2628
2629 int err = errno; // save errno from mmap() call above
2630
2631 if (!recoverable_mmap_error(err)) {
2632 warn_fail_commit_memory(addr, size, exec, err);
2633 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2634 }
2635
2636 return err;
2637 }
2638
2639 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2640 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2641 }
2642
2643 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2644 const char* mesg) {
2645 assert(mesg != NULL, "mesg must be specified");
2646 int err = os::Linux::commit_memory_impl(addr, size, exec);
2647 if (err != 0) {
2648 // the caller wants all commit errors to exit with the specified mesg:
2649 warn_fail_commit_memory(addr, size, exec, err);
2650 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2651 }
2652 }
2653
2654 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2655 #ifndef MAP_HUGETLB
2656 #define MAP_HUGETLB 0x40000
2657 #endif
2658
2659 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2660 #ifndef MADV_HUGEPAGE
2661 #define MADV_HUGEPAGE 14
2662 #endif
2663
2664 int os::Linux::commit_memory_impl(char* addr, size_t size,
2665 size_t alignment_hint, bool exec) {
2666 int err = os::Linux::commit_memory_impl(addr, size, exec);
2667 if (err == 0) {
2668 realign_memory(addr, size, alignment_hint);
2669 }
2670 return err;
2671 }
2672
2673 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2674 bool exec) {
2675 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2676 }
2677
2678 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2679 size_t alignment_hint, bool exec,
2680 const char* mesg) {
2681 assert(mesg != NULL, "mesg must be specified");
2682 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2683 if (err != 0) {
2684 // the caller wants all commit errors to exit with the specified mesg:
2685 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2686 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2687 }
2688 }
2689
2690 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2691 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2692 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2693 // be supported or the memory may already be backed by huge pages.
2694 ::madvise(addr, bytes, MADV_HUGEPAGE);
2695 }
2696 }
2697
2698 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2699 // This method works by doing an mmap over an existing mmaping and effectively discarding
2700 // the existing pages. However it won't work for SHM-based large pages that cannot be
2701 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2702 // small pages on top of the SHM segment. This method always works for small pages, so we
2703 // allow that in any case.
2704 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2705 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2706 }
2707 }
2708
2709 void os::numa_make_global(char *addr, size_t bytes) {
2710 Linux::numa_interleave_memory(addr, bytes);
2711 }
2712
2713 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2714 // bind policy to MPOL_PREFERRED for the current thread.
2715 #define USE_MPOL_PREFERRED 0
2716
2717 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2718 // To make NUMA and large pages more robust when both enabled, we need to ease
2719 // the requirements on where the memory should be allocated. MPOL_BIND is the
2720 // default policy and it will force memory to be allocated on the specified
2721 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2722 // the specified node, but will not force it. Using this policy will prevent
2723 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2724 // free large pages.
2725 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2726 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2727 }
2728
2729 bool os::numa_topology_changed() { return false; }
2730
2731 size_t os::numa_get_groups_num() {
2732 int max_node = Linux::numa_max_node();
2733 return max_node > 0 ? max_node + 1 : 1;
2734 }
2735
2736 int os::numa_get_group_id() {
2737 int cpu_id = Linux::sched_getcpu();
2738 if (cpu_id != -1) {
2739 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2740 if (lgrp_id != -1) {
2741 return lgrp_id;
2742 }
2743 }
2744 return 0;
2745 }
2746
2747 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2748 for (size_t i = 0; i < size; i++) {
2749 ids[i] = i;
2750 }
2751 return size;
2752 }
2753
2754 bool os::get_page_info(char *start, page_info* info) {
2755 return false;
2756 }
2757
2758 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2759 page_info* page_found) {
2760 return end;
2761 }
2762
2763
2764 int os::Linux::sched_getcpu_syscall(void) {
2765 unsigned int cpu = 0;
2766 int retval = -1;
2767
2768 #if defined(IA32)
2769 #ifndef SYS_getcpu
2770 #define SYS_getcpu 318
2771 #endif
2772 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2773 #elif defined(AMD64)
2774 // Unfortunately we have to bring all these macros here from vsyscall.h
2775 // to be able to compile on old linuxes.
2776 #define __NR_vgetcpu 2
2777 #define VSYSCALL_START (-10UL << 20)
2778 #define VSYSCALL_SIZE 1024
2779 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2780 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2781 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2782 retval = vgetcpu(&cpu, NULL, NULL);
2783 #endif
2784
2785 return (retval == -1) ? retval : cpu;
2786 }
2787
2788 // Something to do with the numa-aware allocator needs these symbols
2789 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2790 extern "C" JNIEXPORT void numa_error(char *where) { }
2791
2792
2793 // If we are running with libnuma version > 2, then we should
2794 // be trying to use symbols with versions 1.1
2795 // If we are running with earlier version, which did not have symbol versions,
2796 // we should use the base version.
2797 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2798 void *f = dlvsym(handle, name, "libnuma_1.1");
2799 if (f == NULL) {
2800 f = dlsym(handle, name);
2801 }
2802 return f;
2803 }
2804
2805 bool os::Linux::libnuma_init() {
2806 // sched_getcpu() should be in libc.
2807 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2808 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2809
2810 // If it's not, try a direct syscall.
2811 if (sched_getcpu() == -1) {
2812 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2813 (void*)&sched_getcpu_syscall));
2814 }
2815
2816 if (sched_getcpu() != -1) { // Does it work?
2817 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2818 if (handle != NULL) {
2819 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2820 libnuma_dlsym(handle, "numa_node_to_cpus")));
2821 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2822 libnuma_dlsym(handle, "numa_max_node")));
2823 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2824 libnuma_dlsym(handle, "numa_available")));
2825 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2826 libnuma_dlsym(handle, "numa_tonode_memory")));
2827 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2828 libnuma_dlsym(handle, "numa_interleave_memory")));
2829 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2830 libnuma_dlsym(handle, "numa_set_bind_policy")));
2831
2832
2833 if (numa_available() != -1) {
2834 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2835 // Create a cpu -> node mapping
2836 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2837 rebuild_cpu_to_node_map();
2838 return true;
2839 }
2840 }
2841 }
2842 return false;
2843 }
2844
2845 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2846 // The table is later used in get_node_by_cpu().
2847 void os::Linux::rebuild_cpu_to_node_map() {
2848 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2849 // in libnuma (possible values are starting from 16,
2850 // and continuing up with every other power of 2, but less
2851 // than the maximum number of CPUs supported by kernel), and
2852 // is a subject to change (in libnuma version 2 the requirements
2853 // are more reasonable) we'll just hardcode the number they use
2854 // in the library.
2855 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2856
2857 size_t cpu_num = os::active_processor_count();
2858 size_t cpu_map_size = NCPUS / BitsPerCLong;
2859 size_t cpu_map_valid_size =
2860 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2861
2862 cpu_to_node()->clear();
2863 cpu_to_node()->at_grow(cpu_num - 1);
2864 size_t node_num = numa_get_groups_num();
2865
2866 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2867 for (size_t i = 0; i < node_num; i++) {
2868 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2869 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2870 if (cpu_map[j] != 0) {
2871 for (size_t k = 0; k < BitsPerCLong; k++) {
2872 if (cpu_map[j] & (1UL << k)) {
2873 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2874 }
2875 }
2876 }
2877 }
2878 }
2879 }
2880 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2881 }
2882
2883 int os::Linux::get_node_by_cpu(int cpu_id) {
2884 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2885 return cpu_to_node()->at(cpu_id);
2886 }
2887 return -1;
2888 }
2889
2890 GrowableArray<int>* os::Linux::_cpu_to_node;
2891 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2892 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2893 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2894 os::Linux::numa_available_func_t os::Linux::_numa_available;
2895 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2896 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2897 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2898 unsigned long* os::Linux::_numa_all_nodes;
2899
2900 bool os::pd_uncommit_memory(char* addr, size_t size) {
2901 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2902 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2903 return res != (uintptr_t) MAP_FAILED;
2904 }
2905
2906 static address get_stack_commited_bottom(address bottom, size_t size) {
2907 address nbot = bottom;
2908 address ntop = bottom + size;
2909
2910 size_t page_sz = os::vm_page_size();
2911 unsigned pages = size / page_sz;
2912
2913 unsigned char vec[1];
2914 unsigned imin = 1, imax = pages + 1, imid;
2915 int mincore_return_value = 0;
2916
2917 assert(imin <= imax, "Unexpected page size");
2918
2919 while (imin < imax) {
2920 imid = (imax + imin) / 2;
2921 nbot = ntop - (imid * page_sz);
2922
2923 // Use a trick with mincore to check whether the page is mapped or not.
2924 // mincore sets vec to 1 if page resides in memory and to 0 if page
2925 // is swapped output but if page we are asking for is unmapped
2926 // it returns -1,ENOMEM
2927 mincore_return_value = mincore(nbot, page_sz, vec);
2928
2929 if (mincore_return_value == -1) {
2930 // Page is not mapped go up
2931 // to find first mapped page
2932 if (errno != EAGAIN) {
2933 assert(errno == ENOMEM, "Unexpected mincore errno");
2934 imax = imid;
2935 }
2936 } else {
2937 // Page is mapped go down
2938 // to find first not mapped page
2939 imin = imid + 1;
2940 }
2941 }
2942
2943 nbot = nbot + page_sz;
2944
2945 // Adjust stack bottom one page up if last checked page is not mapped
2946 if (mincore_return_value == -1) {
2947 nbot = nbot + page_sz;
2948 }
2949
2950 return nbot;
2951 }
2952
2953
2954 // Linux uses a growable mapping for the stack, and if the mapping for
2955 // the stack guard pages is not removed when we detach a thread the
2956 // stack cannot grow beyond the pages where the stack guard was
2957 // mapped. If at some point later in the process the stack expands to
2958 // that point, the Linux kernel cannot expand the stack any further
2959 // because the guard pages are in the way, and a segfault occurs.
2960 //
2961 // However, it's essential not to split the stack region by unmapping
2962 // a region (leaving a hole) that's already part of the stack mapping,
2963 // so if the stack mapping has already grown beyond the guard pages at
2964 // the time we create them, we have to truncate the stack mapping.
2965 // So, we need to know the extent of the stack mapping when
2966 // create_stack_guard_pages() is called.
2967
2968 // We only need this for stacks that are growable: at the time of
2969 // writing thread stacks don't use growable mappings (i.e. those
2970 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2971 // only applies to the main thread.
2972
2973 // If the (growable) stack mapping already extends beyond the point
2974 // where we're going to put our guard pages, truncate the mapping at
2975 // that point by munmap()ping it. This ensures that when we later
2976 // munmap() the guard pages we don't leave a hole in the stack
2977 // mapping. This only affects the main/initial thread
2978
2979 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2980 if (os::Linux::is_initial_thread()) {
2981 // As we manually grow stack up to bottom inside create_attached_thread(),
2982 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
2983 // we don't need to do anything special.
2984 // Check it first, before calling heavy function.
2985 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
2986 unsigned char vec[1];
2987
2988 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
2989 // Fallback to slow path on all errors, including EAGAIN
2990 stack_extent = (uintptr_t) get_stack_commited_bottom(
2991 os::Linux::initial_thread_stack_bottom(),
2992 (size_t)addr - stack_extent);
2993 }
2994
2995 if (stack_extent < (uintptr_t)addr) {
2996 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
2997 }
2998 }
2999
3000 return os::commit_memory(addr, size, !ExecMem);
3001 }
3002
3003 // If this is a growable mapping, remove the guard pages entirely by
3004 // munmap()ping them. If not, just call uncommit_memory(). This only
3005 // affects the main/initial thread, but guard against future OS changes
3006 // It's safe to always unmap guard pages for initial thread because we
3007 // always place it right after end of the mapped region
3008
3009 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3010 uintptr_t stack_extent, stack_base;
3011
3012 if (os::Linux::is_initial_thread()) {
3013 return ::munmap(addr, size) == 0;
3014 }
3015
3016 return os::uncommit_memory(addr, size);
3017 }
3018
3019 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3020 // at 'requested_addr'. If there are existing memory mappings at the same
3021 // location, however, they will be overwritten. If 'fixed' is false,
3022 // 'requested_addr' is only treated as a hint, the return value may or
3023 // may not start from the requested address. Unlike Linux mmap(), this
3024 // function returns NULL to indicate failure.
3025 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3026 char * addr;
3027 int flags;
3028
3029 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3030 if (fixed) {
3031 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3032 flags |= MAP_FIXED;
3033 }
3034
3035 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3036 // touch an uncommitted page. Otherwise, the read/write might
3037 // succeed if we have enough swap space to back the physical page.
3038 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3039 flags, -1, 0);
3040
3041 return addr == MAP_FAILED ? NULL : addr;
3042 }
3043
3044 static int anon_munmap(char * addr, size_t size) {
3045 return ::munmap(addr, size) == 0;
3046 }
3047
3048 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3049 size_t alignment_hint) {
3050 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3051 }
3052
3053 bool os::pd_release_memory(char* addr, size_t size) {
3054 return anon_munmap(addr, size);
3055 }
3056
3057 static bool linux_mprotect(char* addr, size_t size, int prot) {
3058 // Linux wants the mprotect address argument to be page aligned.
3059 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3060
3061 // According to SUSv3, mprotect() should only be used with mappings
3062 // established by mmap(), and mmap() always maps whole pages. Unaligned
3063 // 'addr' likely indicates problem in the VM (e.g. trying to change
3064 // protection of malloc'ed or statically allocated memory). Check the
3065 // caller if you hit this assert.
3066 assert(addr == bottom, "sanity check");
3067
3068 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3069 return ::mprotect(bottom, size, prot) == 0;
3070 }
3071
3072 // Set protections specified
3073 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3074 bool is_committed) {
3075 unsigned int p = 0;
3076 switch (prot) {
3077 case MEM_PROT_NONE: p = PROT_NONE; break;
3078 case MEM_PROT_READ: p = PROT_READ; break;
3079 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3080 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3081 default:
3082 ShouldNotReachHere();
3083 }
3084 // is_committed is unused.
3085 return linux_mprotect(addr, bytes, p);
3086 }
3087
3088 bool os::guard_memory(char* addr, size_t size) {
3089 return linux_mprotect(addr, size, PROT_NONE);
3090 }
3091
3092 bool os::unguard_memory(char* addr, size_t size) {
3093 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3094 }
3095
3096 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3097 size_t page_size) {
3098 bool result = false;
3099 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3100 MAP_ANONYMOUS|MAP_PRIVATE,
3101 -1, 0);
3102 if (p != MAP_FAILED) {
3103 void *aligned_p = align_ptr_up(p, page_size);
3104
3105 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3106
3107 munmap(p, page_size * 2);
3108 }
3109
3110 if (warn && !result) {
3111 warning("TransparentHugePages is not supported by the operating system.");
3112 }
3113
3114 return result;
3115 }
3116
3117 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3118 bool result = false;
3119 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3120 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3121 -1, 0);
3122
3123 if (p != MAP_FAILED) {
3124 // We don't know if this really is a huge page or not.
3125 FILE *fp = fopen("/proc/self/maps", "r");
3126 if (fp) {
3127 while (!feof(fp)) {
3128 char chars[257];
3129 long x = 0;
3130 if (fgets(chars, sizeof(chars), fp)) {
3131 if (sscanf(chars, "%lx-%*x", &x) == 1
3132 && x == (long)p) {
3133 if (strstr (chars, "hugepage")) {
3134 result = true;
3135 break;
3136 }
3137 }
3138 }
3139 }
3140 fclose(fp);
3141 }
3142 munmap(p, page_size);
3143 }
3144
3145 if (warn && !result) {
3146 warning("HugeTLBFS is not supported by the operating system.");
3147 }
3148
3149 return result;
3150 }
3151
3152 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3153 //
3154 // From the coredump_filter documentation:
3155 //
3156 // - (bit 0) anonymous private memory
3157 // - (bit 1) anonymous shared memory
3158 // - (bit 2) file-backed private memory
3159 // - (bit 3) file-backed shared memory
3160 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3161 // effective only if the bit 2 is cleared)
3162 // - (bit 5) hugetlb private memory
3163 // - (bit 6) hugetlb shared memory
3164 //
3165 static void set_coredump_filter(void) {
3166 FILE *f;
3167 long cdm;
3168
3169 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3170 return;
3171 }
3172
3173 if (fscanf(f, "%lx", &cdm) != 1) {
3174 fclose(f);
3175 return;
3176 }
3177
3178 rewind(f);
3179
3180 if ((cdm & LARGEPAGES_BIT) == 0) {
3181 cdm |= LARGEPAGES_BIT;
3182 fprintf(f, "%#lx", cdm);
3183 }
3184
3185 fclose(f);
3186 }
3187
3188 // Large page support
3189
3190 static size_t _large_page_size = 0;
3191
3192 size_t os::Linux::find_large_page_size() {
3193 size_t large_page_size = 0;
3194
3195 // large_page_size on Linux is used to round up heap size. x86 uses either
3196 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3197 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3198 // page as large as 256M.
3199 //
3200 // Here we try to figure out page size by parsing /proc/meminfo and looking
3201 // for a line with the following format:
3202 // Hugepagesize: 2048 kB
3203 //
3204 // If we can't determine the value (e.g. /proc is not mounted, or the text
3205 // format has been changed), we'll use the largest page size supported by
3206 // the processor.
3207
3208 #ifndef ZERO
3209 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3210 ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3211 #endif // ZERO
3212
3213 FILE *fp = fopen("/proc/meminfo", "r");
3214 if (fp) {
3215 while (!feof(fp)) {
3216 int x = 0;
3217 char buf[16];
3218 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3219 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3220 large_page_size = x * K;
3221 break;
3222 }
3223 } else {
3224 // skip to next line
3225 for (;;) {
3226 int ch = fgetc(fp);
3227 if (ch == EOF || ch == (int)'\n') break;
3228 }
3229 }
3230 }
3231 fclose(fp);
3232 }
3233
3234 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3235 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3236 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3237 proper_unit_for_byte_size(large_page_size));
3238 }
3239
3240 return large_page_size;
3241 }
3242
3243 size_t os::Linux::setup_large_page_size() {
3244 _large_page_size = Linux::find_large_page_size();
3245 const size_t default_page_size = (size_t)Linux::page_size();
3246 if (_large_page_size > default_page_size) {
3247 _page_sizes[0] = _large_page_size;
3248 _page_sizes[1] = default_page_size;
3249 _page_sizes[2] = 0;
3250 }
3251
3252 return _large_page_size;
3253 }
3254
3255 bool os::Linux::setup_large_page_type(size_t page_size) {
3256 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3257 FLAG_IS_DEFAULT(UseSHM) &&
3258 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3259
3260 // The type of large pages has not been specified by the user.
3261
3262 // Try UseHugeTLBFS and then UseSHM.
3263 UseHugeTLBFS = UseSHM = true;
3264
3265 // Don't try UseTransparentHugePages since there are known
3266 // performance issues with it turned on. This might change in the future.
3267 UseTransparentHugePages = false;
3268 }
3269
3270 if (UseTransparentHugePages) {
3271 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3272 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3273 UseHugeTLBFS = false;
3274 UseSHM = false;
3275 return true;
3276 }
3277 UseTransparentHugePages = false;
3278 }
3279
3280 if (UseHugeTLBFS) {
3281 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3282 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3283 UseSHM = false;
3284 return true;
3285 }
3286 UseHugeTLBFS = false;
3287 }
3288
3289 return UseSHM;
3290 }
3291
3292 void os::large_page_init() {
3293 if (!UseLargePages &&
3294 !UseTransparentHugePages &&
3295 !UseHugeTLBFS &&
3296 !UseSHM) {
3297 // Not using large pages.
3298 return;
3299 }
3300
3301 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3302 // The user explicitly turned off large pages.
3303 // Ignore the rest of the large pages flags.
3304 UseTransparentHugePages = false;
3305 UseHugeTLBFS = false;
3306 UseSHM = false;
3307 return;
3308 }
3309
3310 size_t large_page_size = Linux::setup_large_page_size();
3311 UseLargePages = Linux::setup_large_page_type(large_page_size);
3312
3313 set_coredump_filter();
3314 }
3315
3316 #ifndef SHM_HUGETLB
3317 #define SHM_HUGETLB 04000
3318 #endif
3319
3320 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3321 char* req_addr, bool exec) {
3322 // "exec" is passed in but not used. Creating the shared image for
3323 // the code cache doesn't have an SHM_X executable permission to check.
3324 assert(UseLargePages && UseSHM, "only for SHM large pages");
3325 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3326
3327 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3328 return NULL; // Fallback to small pages.
3329 }
3330
3331 key_t key = IPC_PRIVATE;
3332 char *addr;
3333
3334 bool warn_on_failure = UseLargePages &&
3335 (!FLAG_IS_DEFAULT(UseLargePages) ||
3336 !FLAG_IS_DEFAULT(UseSHM) ||
3337 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3338 char msg[128];
3339
3340 // Create a large shared memory region to attach to based on size.
3341 // Currently, size is the total size of the heap
3342 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3343 if (shmid == -1) {
3344 // Possible reasons for shmget failure:
3345 // 1. shmmax is too small for Java heap.
3346 // > check shmmax value: cat /proc/sys/kernel/shmmax
3347 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3348 // 2. not enough large page memory.
3349 // > check available large pages: cat /proc/meminfo
3350 // > increase amount of large pages:
3351 // echo new_value > /proc/sys/vm/nr_hugepages
3352 // Note 1: different Linux may use different name for this property,
3353 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3354 // Note 2: it's possible there's enough physical memory available but
3355 // they are so fragmented after a long run that they can't
3356 // coalesce into large pages. Try to reserve large pages when
3357 // the system is still "fresh".
3358 if (warn_on_failure) {
3359 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3360 warning("%s", msg);
3361 }
3362 return NULL;
3363 }
3364
3365 // attach to the region
3366 addr = (char*)shmat(shmid, req_addr, 0);
3367 int err = errno;
3368
3369 // Remove shmid. If shmat() is successful, the actual shared memory segment
3370 // will be deleted when it's detached by shmdt() or when the process
3371 // terminates. If shmat() is not successful this will remove the shared
3372 // segment immediately.
3373 shmctl(shmid, IPC_RMID, NULL);
3374
3375 if ((intptr_t)addr == -1) {
3376 if (warn_on_failure) {
3377 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3378 warning("%s", msg);
3379 }
3380 return NULL;
3381 }
3382
3383 return addr;
3384 }
3385
3386 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3387 int error) {
3388 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3389
3390 bool warn_on_failure = UseLargePages &&
3391 (!FLAG_IS_DEFAULT(UseLargePages) ||
3392 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3393 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3394
3395 if (warn_on_failure) {
3396 char msg[128];
3397 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3398 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3399 warning("%s", msg);
3400 }
3401 }
3402
3403 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3404 char* req_addr,
3405 bool exec) {
3406 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3407 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3408 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3409
3410 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3411 char* addr = (char*)::mmap(req_addr, bytes, prot,
3412 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3413 -1, 0);
3414
3415 if (addr == MAP_FAILED) {
3416 warn_on_large_pages_failure(req_addr, bytes, errno);
3417 return NULL;
3418 }
3419
3420 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3421
3422 return addr;
3423 }
3424
3425 // Helper for os::Linux::reserve_memory_special_huge_tlbfs_mixed().
3426 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3427 // (req_addr != NULL) or with a given alignment.
3428 // - bytes shall be a multiple of alignment.
3429 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3430 // - alignment sets the alignment at which memory shall be allocated.
3431 // It must be a multiple of allocation granularity.
3432 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3433 // req_addr or NULL.
3434 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3435
3436 size_t extra_size = bytes;
3437 if (req_addr == NULL && alignment > 0) {
3438 extra_size += alignment;
3439 }
3440
3441 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3442 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3443 -1, 0);
3444 if (start == MAP_FAILED) {
3445 start = NULL;
3446 } else {
3447 if (req_addr != NULL) {
3448 if (start != req_addr) {
3449 ::munmap(start, extra_size);
3450 start = NULL;
3451 }
3452 } else {
3453 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3454 char* const end_aligned = start_aligned + bytes;
3455 char* const end = start + extra_size;
3456 if (start_aligned > start) {
3457 ::munmap(start, start_aligned - start);
3458 }
3459 if (end_aligned < end) {
3460 ::munmap(end_aligned, end - end_aligned);
3461 }
3462 start = start_aligned;
3463 }
3464 }
3465 return start;
3466
3467 }
3468
3469 // Helper function to create a temp file in the given directory
3470 int os::Linux::create_tmpfile(const char* dir, size_t size, bool exec) {
3471
3472 static char name_template[] = "/jvmheap.XXXXXX";
3473
3474 char fullname[strlen(dir) + sizeof(name_template)];
3475 (void)strcpy(fullname, dir);
3476 (void)strcat(fullname, name_template);
3477
3478 sigset_t set, oldset;
3479 sigfillset(&set);
3480
3481 // block all signals while we do the file operation
3482 (void)sigprocmask(SIG_BLOCK, &set, &oldset);
3483
3484 // set the file creation mask
3485 mode_t new_mask = exec ? (S_IRUSR | S_IWUSR | S_IXUSR) : (S_IRUSR | S_IWUSR);
3486 mode_t prev_umask = umask(new_mask);
3487
3488 // create a new file
3489 int fd = mkstemp(fullname);
3490
3491 // reset the file creation mask
3492 umask(prev_umask);
3493
3494 if (fd < 0) {
3495 warning("Could not create file for heap");
3496 return -1;
3497 }
3498
3499 // delete the name from the filesystem. When 'fd' is closed, the file (and space) will be deleted
3500 (void)unlink(fullname);
3501
3502 // reset the signal mask
3503 (void)sigprocmask(SIG_SETMASK, &oldset, NULL);
3504
3505 // allocate space for the file
3506 if ((errno = posix_fallocate(fd, 0, (off_t)size)) != 0) {
3507 warning("Could not allocate sufficient disk space for heap");
3508 return -1;
3509 }
3510
3511 return fd;
3512 }
3513
3514 // Map the given address range to a temporary file created at the specified directory.
3515 // The address range must already be reserved for guaranteed success. If it not reserved, their could be an error while mapping leading to JVM shutdown
3516 bool os::map_memory_to_file(char* base, size_t size, const char* backingFileDir) {
3517
3518 int fd = os::Linux::create_tmpfile(backingFileDir, size, false);
3519 if (fd == -1) {
3520 vm_exit_during_initialization(err_msg("Could not create temporary file in %s for object heap", backingFileDir));
3521 return false;
3522 }
3523 int prot = PROT_READ | PROT_WRITE;
3524 char* addr = (char*)mmap(base, size, prot, MAP_SHARED | MAP_FIXED, fd, 0);
3525
3526 if (addr == MAP_FAILED || addr != base) {
3527 close(fd);
3528 vm_exit_during_initialization(err_msg("Error in mapping object heap at the given filesystem dir %s", backingFileDir));
3529 return false;
3530 }
3531
3532 return true;
3533 }
3534
3535 // Reserve memory using mmap(MAP_HUGETLB).
3536 // - bytes shall be a multiple of alignment.
3537 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3538 // - alignment sets the alignment at which memory shall be allocated.
3539 // It must be a multiple of allocation granularity.
3540 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3541 // req_addr or NULL.
3542 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3543 size_t alignment,
3544 char* req_addr,
3545 bool exec) {
3546 size_t large_page_size = os::large_page_size();
3547 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3548
3549 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3550 assert(is_size_aligned(bytes, alignment), "Must be");
3551
3552 // First reserve - but not commit - the address range in small pages.
3553 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3554
3555 if (start == NULL) {
3556 return NULL;
3557 }
3558
3559 assert(is_ptr_aligned(start, alignment), "Must be");
3560
3561 char* end = start + bytes;
3562
3563 // Find the regions of the allocated chunk that can be promoted to large pages.
3564 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3565 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3566
3567 size_t lp_bytes = lp_end - lp_start;
3568
3569 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3570
3571 if (lp_bytes == 0) {
3572 // The mapped region doesn't even span the start and the end of a large page.
3573 // Fall back to allocate a non-special area.
3574 ::munmap(start, end - start);
3575 return NULL;
3576 }
3577
3578 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3579
3580 void* result;
3581
3582 // Commit small-paged leading area.
3583 if (start != lp_start) {
3584 result = ::mmap(start, lp_start - start, prot,
3585 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3586 -1, 0);
3587 if (result == MAP_FAILED) {
3588 ::munmap(lp_start, end - lp_start);
3589 return NULL;
3590 }
3591 }
3592
3593 // Commit large-paged area.
3594 result = ::mmap(lp_start, lp_bytes, prot,
3595 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3596 -1, 0);
3597 if (result == MAP_FAILED) {
3598 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3599 // If the mmap above fails, the large pages region will be unmapped and we
3600 // have regions before and after with small pages. Release these regions.
3601 //
3602 // | mapped | unmapped | mapped |
3603 // ^ ^ ^ ^
3604 // start lp_start lp_end end
3605 //
3606 ::munmap(start, lp_start - start);
3607 ::munmap(lp_end, end - lp_end);
3608 return NULL;
3609 }
3610
3611 // Commit small-paged trailing area.
3612 if (lp_end != end) {
3613 result = ::mmap(lp_end, end - lp_end, prot,
3614 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3615 -1, 0);
3616 if (result == MAP_FAILED) {
3617 ::munmap(start, lp_end - start);
3618 return NULL;
3619 }
3620 }
3621
3622 return start;
3623 }
3624
3625 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3626 size_t alignment,
3627 char* req_addr,
3628 bool exec) {
3629 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3630 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3631 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3632 assert(is_power_of_2(os::large_page_size()), "Must be");
3633 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3634
3635 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3636 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3637 } else {
3638 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3639 }
3640 }
3641
3642 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3643 char* req_addr, bool exec) {
3644 assert(UseLargePages, "only for large pages");
3645
3646 char* addr;
3647 if (UseSHM) {
3648 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3649 } else {
3650 assert(UseHugeTLBFS, "must be");
3651 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3652 }
3653
3654 if (addr != NULL) {
3655 if (UseNUMAInterleaving) {
3656 numa_make_global(addr, bytes);
3657 }
3658
3659 // The memory is committed
3660 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3661 }
3662
3663 return addr;
3664 }
3665
3666 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3667 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3668 return shmdt(base) == 0;
3669 }
3670
3671 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3672 return pd_release_memory(base, bytes);
3673 }
3674
3675 bool os::release_memory_special(char* base, size_t bytes) {
3676 bool res;
3677 if (MemTracker::tracking_level() > NMT_minimal) {
3678 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3679 res = os::Linux::release_memory_special_impl(base, bytes);
3680 if (res) {
3681 tkr.record((address)base, bytes);
3682 }
3683
3684 } else {
3685 res = os::Linux::release_memory_special_impl(base, bytes);
3686 }
3687 return res;
3688 }
3689
3690 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3691 assert(UseLargePages, "only for large pages");
3692 bool res;
3693
3694 if (UseSHM) {
3695 res = os::Linux::release_memory_special_shm(base, bytes);
3696 } else {
3697 assert(UseHugeTLBFS, "must be");
3698 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3699 }
3700 return res;
3701 }
3702
3703 size_t os::large_page_size() {
3704 return _large_page_size;
3705 }
3706
3707 // With SysV SHM the entire memory region must be allocated as shared
3708 // memory.
3709 // HugeTLBFS allows application to commit large page memory on demand.
3710 // However, when committing memory with HugeTLBFS fails, the region
3711 // that was supposed to be committed will lose the old reservation
3712 // and allow other threads to steal that memory region. Because of this
3713 // behavior we can't commit HugeTLBFS memory.
3714 bool os::can_commit_large_page_memory() {
3715 return UseTransparentHugePages;
3716 }
3717
3718 bool os::can_execute_large_page_memory() {
3719 return UseTransparentHugePages || UseHugeTLBFS;
3720 }
3721
3722 // Reserve memory at an arbitrary address, only if that area is
3723 // available (and not reserved for something else).
3724
3725 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3726 const int max_tries = 10;
3727 char* base[max_tries];
3728 size_t size[max_tries];
3729 const size_t gap = 0x000000;
3730
3731 // Assert only that the size is a multiple of the page size, since
3732 // that's all that mmap requires, and since that's all we really know
3733 // about at this low abstraction level. If we need higher alignment,
3734 // we can either pass an alignment to this method or verify alignment
3735 // in one of the methods further up the call chain. See bug 5044738.
3736 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3737
3738 // Repeatedly allocate blocks until the block is allocated at the
3739 // right spot.
3740
3741 // Linux mmap allows caller to pass an address as hint; give it a try first,
3742 // if kernel honors the hint then we can return immediately.
3743 char * addr = anon_mmap(requested_addr, bytes, false);
3744 if (addr == requested_addr) {
3745 return requested_addr;
3746 }
3747
3748 if (addr != NULL) {
3749 // mmap() is successful but it fails to reserve at the requested address
3750 anon_munmap(addr, bytes);
3751 }
3752
3753 int i;
3754 for (i = 0; i < max_tries; ++i) {
3755 base[i] = reserve_memory(bytes);
3756
3757 if (base[i] != NULL) {
3758 // Is this the block we wanted?
3759 if (base[i] == requested_addr) {
3760 size[i] = bytes;
3761 break;
3762 }
3763
3764 // Does this overlap the block we wanted? Give back the overlapped
3765 // parts and try again.
3766
3767 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3768 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3769 unmap_memory(base[i], top_overlap);
3770 base[i] += top_overlap;
3771 size[i] = bytes - top_overlap;
3772 } else {
3773 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3774 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3775 unmap_memory(requested_addr, bottom_overlap);
3776 size[i] = bytes - bottom_overlap;
3777 } else {
3778 size[i] = bytes;
3779 }
3780 }
3781 }
3782 }
3783
3784 // Give back the unused reserved pieces.
3785
3786 for (int j = 0; j < i; ++j) {
3787 if (base[j] != NULL) {
3788 unmap_memory(base[j], size[j]);
3789 }
3790 }
3791
3792 if (i < max_tries) {
3793 return requested_addr;
3794 } else {
3795 return NULL;
3796 }
3797 }
3798
3799 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3800 return ::read(fd, buf, nBytes);
3801 }
3802
3803 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3804 return ::pread(fd, buf, nBytes, offset);
3805 }
3806
3807 // Short sleep, direct OS call.
3808 //
3809 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3810 // sched_yield(2) will actually give up the CPU:
3811 //
3812 // * Alone on this pariticular CPU, keeps running.
3813 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3814 // (pre 2.6.39).
3815 //
3816 // So calling this with 0 is an alternative.
3817 //
3818 void os::naked_short_sleep(jlong ms) {
3819 struct timespec req;
3820
3821 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3822 req.tv_sec = 0;
3823 if (ms > 0) {
3824 req.tv_nsec = (ms % 1000) * 1000000;
3825 } else {
3826 req.tv_nsec = 1;
3827 }
3828
3829 nanosleep(&req, NULL);
3830
3831 return;
3832 }
3833
3834 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3835 void os::infinite_sleep() {
3836 while (true) { // sleep forever ...
3837 ::sleep(100); // ... 100 seconds at a time
3838 }
3839 }
3840
3841 // Used to convert frequent JVM_Yield() to nops
3842 bool os::dont_yield() {
3843 return DontYieldALot;
3844 }
3845
3846 void os::naked_yield() {
3847 sched_yield();
3848 }
3849
3850 ////////////////////////////////////////////////////////////////////////////////
3851 // thread priority support
3852
3853 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3854 // only supports dynamic priority, static priority must be zero. For real-time
3855 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3856 // However, for large multi-threaded applications, SCHED_RR is not only slower
3857 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3858 // of 5 runs - Sep 2005).
3859 //
3860 // The following code actually changes the niceness of kernel-thread/LWP. It
3861 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3862 // not the entire user process, and user level threads are 1:1 mapped to kernel
3863 // threads. It has always been the case, but could change in the future. For
3864 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3865 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3866
3867 int os::java_to_os_priority[CriticalPriority + 1] = {
3868 19, // 0 Entry should never be used
3869
3870 4, // 1 MinPriority
3871 3, // 2
3872 2, // 3
3873
3874 1, // 4
3875 0, // 5 NormPriority
3876 -1, // 6
3877
3878 -2, // 7
3879 -3, // 8
3880 -4, // 9 NearMaxPriority
3881
3882 -5, // 10 MaxPriority
3883
3884 -5 // 11 CriticalPriority
3885 };
3886
3887 static int prio_init() {
3888 if (ThreadPriorityPolicy == 1) {
3889 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3890 // if effective uid is not root. Perhaps, a more elegant way of doing
3891 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3892 if (geteuid() != 0) {
3893 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3894 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3895 }
3896 ThreadPriorityPolicy = 0;
3897 }
3898 }
3899 if (UseCriticalJavaThreadPriority) {
3900 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3901 }
3902 return 0;
3903 }
3904
3905 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3906 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3907
3908 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3909 return (ret == 0) ? OS_OK : OS_ERR;
3910 }
3911
3912 OSReturn os::get_native_priority(const Thread* const thread,
3913 int *priority_ptr) {
3914 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3915 *priority_ptr = java_to_os_priority[NormPriority];
3916 return OS_OK;
3917 }
3918
3919 errno = 0;
3920 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3921 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3922 }
3923
3924 // Hint to the underlying OS that a task switch would not be good.
3925 // Void return because it's a hint and can fail.
3926 void os::hint_no_preempt() {}
3927
3928 ////////////////////////////////////////////////////////////////////////////////
3929 // suspend/resume support
3930
3931 // the low-level signal-based suspend/resume support is a remnant from the
3932 // old VM-suspension that used to be for java-suspension, safepoints etc,
3933 // within hotspot. Now there is a single use-case for this:
3934 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3935 // that runs in the watcher thread.
3936 // The remaining code is greatly simplified from the more general suspension
3937 // code that used to be used.
3938 //
3939 // The protocol is quite simple:
3940 // - suspend:
3941 // - sends a signal to the target thread
3942 // - polls the suspend state of the osthread using a yield loop
3943 // - target thread signal handler (SR_handler) sets suspend state
3944 // and blocks in sigsuspend until continued
3945 // - resume:
3946 // - sets target osthread state to continue
3947 // - sends signal to end the sigsuspend loop in the SR_handler
3948 //
3949 // Note that the SR_lock plays no role in this suspend/resume protocol.
3950
3951 static void resume_clear_context(OSThread *osthread) {
3952 osthread->set_ucontext(NULL);
3953 osthread->set_siginfo(NULL);
3954 }
3955
3956 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3957 ucontext_t* context) {
3958 osthread->set_ucontext(context);
3959 osthread->set_siginfo(siginfo);
3960 }
3961
3962 // Handler function invoked when a thread's execution is suspended or
3963 // resumed. We have to be careful that only async-safe functions are
3964 // called here (Note: most pthread functions are not async safe and
3965 // should be avoided.)
3966 //
3967 // Note: sigwait() is a more natural fit than sigsuspend() from an
3968 // interface point of view, but sigwait() prevents the signal hander
3969 // from being run. libpthread would get very confused by not having
3970 // its signal handlers run and prevents sigwait()'s use with the
3971 // mutex granting granting signal.
3972 //
3973 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3974 //
3975 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3976 // Save and restore errno to avoid confusing native code with EINTR
3977 // after sigsuspend.
3978 int old_errno = errno;
3979
3980 Thread* thread = Thread::current_or_null_safe();
3981 assert(thread != NULL, "Missing current thread in SR_handler");
3982 OSThread* osthread = thread->osthread();
3983 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3984
3985 os::SuspendResume::State current = osthread->sr.state();
3986 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3987 suspend_save_context(osthread, siginfo, context);
3988
3989 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3990 os::SuspendResume::State state = osthread->sr.suspended();
3991 if (state == os::SuspendResume::SR_SUSPENDED) {
3992 sigset_t suspend_set; // signals for sigsuspend()
3993 sigemptyset(&suspend_set);
3994 // get current set of blocked signals and unblock resume signal
3995 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3996 sigdelset(&suspend_set, SR_signum);
3997
3998 sr_semaphore.signal();
3999 // wait here until we are resumed
4000 while (1) {
4001 sigsuspend(&suspend_set);
4002
4003 os::SuspendResume::State result = osthread->sr.running();
4004 if (result == os::SuspendResume::SR_RUNNING) {
4005 sr_semaphore.signal();
4006 break;
4007 }
4008 }
4009
4010 } else if (state == os::SuspendResume::SR_RUNNING) {
4011 // request was cancelled, continue
4012 } else {
4013 ShouldNotReachHere();
4014 }
4015
4016 resume_clear_context(osthread);
4017 } else if (current == os::SuspendResume::SR_RUNNING) {
4018 // request was cancelled, continue
4019 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4020 // ignore
4021 } else {
4022 // ignore
4023 }
4024
4025 errno = old_errno;
4026 }
4027
4028 static int SR_initialize() {
4029 struct sigaction act;
4030 char *s;
4031
4032 // Get signal number to use for suspend/resume
4033 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4034 int sig = ::strtol(s, 0, 10);
4035 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4036 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4037 SR_signum = sig;
4038 } else {
4039 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4040 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4041 }
4042 }
4043
4044 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4045 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4046
4047 sigemptyset(&SR_sigset);
4048 sigaddset(&SR_sigset, SR_signum);
4049
4050 // Set up signal handler for suspend/resume
4051 act.sa_flags = SA_RESTART|SA_SIGINFO;
4052 act.sa_handler = (void (*)(int)) SR_handler;
4053
4054 // SR_signum is blocked by default.
4055 // 4528190 - We also need to block pthread restart signal (32 on all
4056 // supported Linux platforms). Note that LinuxThreads need to block
4057 // this signal for all threads to work properly. So we don't have
4058 // to use hard-coded signal number when setting up the mask.
4059 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4060
4061 if (sigaction(SR_signum, &act, 0) == -1) {
4062 return -1;
4063 }
4064
4065 // Save signal flag
4066 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4067 return 0;
4068 }
4069
4070 static int sr_notify(OSThread* osthread) {
4071 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4072 assert_status(status == 0, status, "pthread_kill");
4073 return status;
4074 }
4075
4076 // "Randomly" selected value for how long we want to spin
4077 // before bailing out on suspending a thread, also how often
4078 // we send a signal to a thread we want to resume
4079 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4080 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4081
4082 // returns true on success and false on error - really an error is fatal
4083 // but this seems the normal response to library errors
4084 static bool do_suspend(OSThread* osthread) {
4085 assert(osthread->sr.is_running(), "thread should be running");
4086 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4087
4088 // mark as suspended and send signal
4089 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4090 // failed to switch, state wasn't running?
4091 ShouldNotReachHere();
4092 return false;
4093 }
4094
4095 if (sr_notify(osthread) != 0) {
4096 ShouldNotReachHere();
4097 }
4098
4099 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4100 while (true) {
4101 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4102 break;
4103 } else {
4104 // timeout
4105 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4106 if (cancelled == os::SuspendResume::SR_RUNNING) {
4107 return false;
4108 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4109 // make sure that we consume the signal on the semaphore as well
4110 sr_semaphore.wait();
4111 break;
4112 } else {
4113 ShouldNotReachHere();
4114 return false;
4115 }
4116 }
4117 }
4118
4119 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4120 return true;
4121 }
4122
4123 static void do_resume(OSThread* osthread) {
4124 assert(osthread->sr.is_suspended(), "thread should be suspended");
4125 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4126
4127 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4128 // failed to switch to WAKEUP_REQUEST
4129 ShouldNotReachHere();
4130 return;
4131 }
4132
4133 while (true) {
4134 if (sr_notify(osthread) == 0) {
4135 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4136 if (osthread->sr.is_running()) {
4137 return;
4138 }
4139 }
4140 } else {
4141 ShouldNotReachHere();
4142 }
4143 }
4144
4145 guarantee(osthread->sr.is_running(), "Must be running!");
4146 }
4147
4148 ///////////////////////////////////////////////////////////////////////////////////
4149 // signal handling (except suspend/resume)
4150
4151 // This routine may be used by user applications as a "hook" to catch signals.
4152 // The user-defined signal handler must pass unrecognized signals to this
4153 // routine, and if it returns true (non-zero), then the signal handler must
4154 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4155 // routine will never retun false (zero), but instead will execute a VM panic
4156 // routine kill the process.
4157 //
4158 // If this routine returns false, it is OK to call it again. This allows
4159 // the user-defined signal handler to perform checks either before or after
4160 // the VM performs its own checks. Naturally, the user code would be making
4161 // a serious error if it tried to handle an exception (such as a null check
4162 // or breakpoint) that the VM was generating for its own correct operation.
4163 //
4164 // This routine may recognize any of the following kinds of signals:
4165 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4166 // It should be consulted by handlers for any of those signals.
4167 //
4168 // The caller of this routine must pass in the three arguments supplied
4169 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4170 // field of the structure passed to sigaction(). This routine assumes that
4171 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4172 //
4173 // Note that the VM will print warnings if it detects conflicting signal
4174 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4175 //
4176 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4177 siginfo_t* siginfo,
4178 void* ucontext,
4179 int abort_if_unrecognized);
4180
4181 void signalHandler(int sig, siginfo_t* info, void* uc) {
4182 assert(info != NULL && uc != NULL, "it must be old kernel");
4183 int orig_errno = errno; // Preserve errno value over signal handler.
4184 JVM_handle_linux_signal(sig, info, uc, true);
4185 errno = orig_errno;
4186 }
4187
4188
4189 // This boolean allows users to forward their own non-matching signals
4190 // to JVM_handle_linux_signal, harmlessly.
4191 bool os::Linux::signal_handlers_are_installed = false;
4192
4193 // For signal-chaining
4194 struct sigaction sigact[NSIG];
4195 uint64_t sigs = 0;
4196 #if (64 < NSIG-1)
4197 #error "Not all signals can be encoded in sigs. Adapt its type!"
4198 #endif
4199 bool os::Linux::libjsig_is_loaded = false;
4200 typedef struct sigaction *(*get_signal_t)(int);
4201 get_signal_t os::Linux::get_signal_action = NULL;
4202
4203 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4204 struct sigaction *actp = NULL;
4205
4206 if (libjsig_is_loaded) {
4207 // Retrieve the old signal handler from libjsig
4208 actp = (*get_signal_action)(sig);
4209 }
4210 if (actp == NULL) {
4211 // Retrieve the preinstalled signal handler from jvm
4212 actp = get_preinstalled_handler(sig);
4213 }
4214
4215 return actp;
4216 }
4217
4218 static bool call_chained_handler(struct sigaction *actp, int sig,
4219 siginfo_t *siginfo, void *context) {
4220 // Call the old signal handler
4221 if (actp->sa_handler == SIG_DFL) {
4222 // It's more reasonable to let jvm treat it as an unexpected exception
4223 // instead of taking the default action.
4224 return false;
4225 } else if (actp->sa_handler != SIG_IGN) {
4226 if ((actp->sa_flags & SA_NODEFER) == 0) {
4227 // automaticlly block the signal
4228 sigaddset(&(actp->sa_mask), sig);
4229 }
4230
4231 sa_handler_t hand = NULL;
4232 sa_sigaction_t sa = NULL;
4233 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4234 // retrieve the chained handler
4235 if (siginfo_flag_set) {
4236 sa = actp->sa_sigaction;
4237 } else {
4238 hand = actp->sa_handler;
4239 }
4240
4241 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4242 actp->sa_handler = SIG_DFL;
4243 }
4244
4245 // try to honor the signal mask
4246 sigset_t oset;
4247 sigemptyset(&oset);
4248 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4249
4250 // call into the chained handler
4251 if (siginfo_flag_set) {
4252 (*sa)(sig, siginfo, context);
4253 } else {
4254 (*hand)(sig);
4255 }
4256
4257 // restore the signal mask
4258 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4259 }
4260 // Tell jvm's signal handler the signal is taken care of.
4261 return true;
4262 }
4263
4264 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4265 bool chained = false;
4266 // signal-chaining
4267 if (UseSignalChaining) {
4268 struct sigaction *actp = get_chained_signal_action(sig);
4269 if (actp != NULL) {
4270 chained = call_chained_handler(actp, sig, siginfo, context);
4271 }
4272 }
4273 return chained;
4274 }
4275
4276 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4277 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4278 return &sigact[sig];
4279 }
4280 return NULL;
4281 }
4282
4283 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4284 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4285 sigact[sig] = oldAct;
4286 sigs |= (uint64_t)1 << (sig-1);
4287 }
4288
4289 // for diagnostic
4290 int sigflags[NSIG];
4291
4292 int os::Linux::get_our_sigflags(int sig) {
4293 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4294 return sigflags[sig];
4295 }
4296
4297 void os::Linux::set_our_sigflags(int sig, int flags) {
4298 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4299 if (sig > 0 && sig < NSIG) {
4300 sigflags[sig] = flags;
4301 }
4302 }
4303
4304 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4305 // Check for overwrite.
4306 struct sigaction oldAct;
4307 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4308
4309 void* oldhand = oldAct.sa_sigaction
4310 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4311 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4312 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4313 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4314 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4315 if (AllowUserSignalHandlers || !set_installed) {
4316 // Do not overwrite; user takes responsibility to forward to us.
4317 return;
4318 } else if (UseSignalChaining) {
4319 // save the old handler in jvm
4320 save_preinstalled_handler(sig, oldAct);
4321 // libjsig also interposes the sigaction() call below and saves the
4322 // old sigaction on it own.
4323 } else {
4324 fatal("Encountered unexpected pre-existing sigaction handler "
4325 "%#lx for signal %d.", (long)oldhand, sig);
4326 }
4327 }
4328
4329 struct sigaction sigAct;
4330 sigfillset(&(sigAct.sa_mask));
4331 sigAct.sa_handler = SIG_DFL;
4332 if (!set_installed) {
4333 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4334 } else {
4335 sigAct.sa_sigaction = signalHandler;
4336 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4337 }
4338 // Save flags, which are set by ours
4339 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4340 sigflags[sig] = sigAct.sa_flags;
4341
4342 int ret = sigaction(sig, &sigAct, &oldAct);
4343 assert(ret == 0, "check");
4344
4345 void* oldhand2 = oldAct.sa_sigaction
4346 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4347 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4348 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4349 }
4350
4351 // install signal handlers for signals that HotSpot needs to
4352 // handle in order to support Java-level exception handling.
4353
4354 void os::Linux::install_signal_handlers() {
4355 if (!signal_handlers_are_installed) {
4356 signal_handlers_are_installed = true;
4357
4358 // signal-chaining
4359 typedef void (*signal_setting_t)();
4360 signal_setting_t begin_signal_setting = NULL;
4361 signal_setting_t end_signal_setting = NULL;
4362 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4363 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4364 if (begin_signal_setting != NULL) {
4365 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4366 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4367 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4368 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4369 libjsig_is_loaded = true;
4370 assert(UseSignalChaining, "should enable signal-chaining");
4371 }
4372 if (libjsig_is_loaded) {
4373 // Tell libjsig jvm is setting signal handlers
4374 (*begin_signal_setting)();
4375 }
4376
4377 set_signal_handler(SIGSEGV, true);
4378 set_signal_handler(SIGPIPE, true);
4379 set_signal_handler(SIGBUS, true);
4380 set_signal_handler(SIGILL, true);
4381 set_signal_handler(SIGFPE, true);
4382 #if defined(PPC64)
4383 set_signal_handler(SIGTRAP, true);
4384 #endif
4385 set_signal_handler(SIGXFSZ, true);
4386
4387 if (libjsig_is_loaded) {
4388 // Tell libjsig jvm finishes setting signal handlers
4389 (*end_signal_setting)();
4390 }
4391
4392 // We don't activate signal checker if libjsig is in place, we trust ourselves
4393 // and if UserSignalHandler is installed all bets are off.
4394 // Log that signal checking is off only if -verbose:jni is specified.
4395 if (CheckJNICalls) {
4396 if (libjsig_is_loaded) {
4397 if (PrintJNIResolving) {
4398 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4399 }
4400 check_signals = false;
4401 }
4402 if (AllowUserSignalHandlers) {
4403 if (PrintJNIResolving) {
4404 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4405 }
4406 check_signals = false;
4407 }
4408 }
4409 }
4410 }
4411
4412 // This is the fastest way to get thread cpu time on Linux.
4413 // Returns cpu time (user+sys) for any thread, not only for current.
4414 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4415 // It might work on 2.6.10+ with a special kernel/glibc patch.
4416 // For reference, please, see IEEE Std 1003.1-2004:
4417 // http://www.unix.org/single_unix_specification
4418
4419 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4420 struct timespec tp;
4421 int rc = os::Linux::clock_gettime(clockid, &tp);
4422 assert(rc == 0, "clock_gettime is expected to return 0 code");
4423
4424 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4425 }
4426
4427 void os::Linux::initialize_os_info() {
4428 assert(_os_version == 0, "OS info already initialized");
4429
4430 struct utsname _uname;
4431
4432 uint32_t major;
4433 uint32_t minor;
4434 uint32_t fix;
4435
4436 int rc;
4437
4438 // Kernel version is unknown if
4439 // verification below fails.
4440 _os_version = 0x01000000;
4441
4442 rc = uname(&_uname);
4443 if (rc != -1) {
4444
4445 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4446 if (rc == 3) {
4447
4448 if (major < 256 && minor < 256 && fix < 256) {
4449 // Kernel version format is as expected,
4450 // set it overriding unknown state.
4451 _os_version = (major << 16) |
4452 (minor << 8 ) |
4453 (fix << 0 ) ;
4454 }
4455 }
4456 }
4457 }
4458
4459 uint32_t os::Linux::os_version() {
4460 assert(_os_version != 0, "not initialized");
4461 return _os_version & 0x00FFFFFF;
4462 }
4463
4464 bool os::Linux::os_version_is_known() {
4465 assert(_os_version != 0, "not initialized");
4466 return _os_version & 0x01000000 ? false : true;
4467 }
4468
4469 /////
4470 // glibc on Linux platform uses non-documented flag
4471 // to indicate, that some special sort of signal
4472 // trampoline is used.
4473 // We will never set this flag, and we should
4474 // ignore this flag in our diagnostic
4475 #ifdef SIGNIFICANT_SIGNAL_MASK
4476 #undef SIGNIFICANT_SIGNAL_MASK
4477 #endif
4478 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4479
4480 static const char* get_signal_handler_name(address handler,
4481 char* buf, int buflen) {
4482 int offset = 0;
4483 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4484 if (found) {
4485 // skip directory names
4486 const char *p1, *p2;
4487 p1 = buf;
4488 size_t len = strlen(os::file_separator());
4489 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4490 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4491 } else {
4492 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4493 }
4494 return buf;
4495 }
4496
4497 static void print_signal_handler(outputStream* st, int sig,
4498 char* buf, size_t buflen) {
4499 struct sigaction sa;
4500
4501 sigaction(sig, NULL, &sa);
4502
4503 // See comment for SIGNIFICANT_SIGNAL_MASK define
4504 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4505
4506 st->print("%s: ", os::exception_name(sig, buf, buflen));
4507
4508 address handler = (sa.sa_flags & SA_SIGINFO)
4509 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4510 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4511
4512 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4513 st->print("SIG_DFL");
4514 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4515 st->print("SIG_IGN");
4516 } else {
4517 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4518 }
4519
4520 st->print(", sa_mask[0]=");
4521 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4522
4523 address rh = VMError::get_resetted_sighandler(sig);
4524 // May be, handler was resetted by VMError?
4525 if (rh != NULL) {
4526 handler = rh;
4527 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4528 }
4529
4530 st->print(", sa_flags=");
4531 os::Posix::print_sa_flags(st, sa.sa_flags);
4532
4533 // Check: is it our handler?
4534 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4535 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4536 // It is our signal handler
4537 // check for flags, reset system-used one!
4538 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4539 st->print(
4540 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4541 os::Linux::get_our_sigflags(sig));
4542 }
4543 }
4544 st->cr();
4545 }
4546
4547
4548 #define DO_SIGNAL_CHECK(sig) \
4549 do { \
4550 if (!sigismember(&check_signal_done, sig)) { \
4551 os::Linux::check_signal_handler(sig); \
4552 } \
4553 } while (0)
4554
4555 // This method is a periodic task to check for misbehaving JNI applications
4556 // under CheckJNI, we can add any periodic checks here
4557
4558 void os::run_periodic_checks() {
4559 if (check_signals == false) return;
4560
4561 // SEGV and BUS if overridden could potentially prevent
4562 // generation of hs*.log in the event of a crash, debugging
4563 // such a case can be very challenging, so we absolutely
4564 // check the following for a good measure:
4565 DO_SIGNAL_CHECK(SIGSEGV);
4566 DO_SIGNAL_CHECK(SIGILL);
4567 DO_SIGNAL_CHECK(SIGFPE);
4568 DO_SIGNAL_CHECK(SIGBUS);
4569 DO_SIGNAL_CHECK(SIGPIPE);
4570 DO_SIGNAL_CHECK(SIGXFSZ);
4571 #if defined(PPC64)
4572 DO_SIGNAL_CHECK(SIGTRAP);
4573 #endif
4574
4575 // ReduceSignalUsage allows the user to override these handlers
4576 // see comments at the very top and jvm_solaris.h
4577 if (!ReduceSignalUsage) {
4578 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4579 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4580 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4581 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4582 }
4583
4584 DO_SIGNAL_CHECK(SR_signum);
4585 }
4586
4587 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4588
4589 static os_sigaction_t os_sigaction = NULL;
4590
4591 void os::Linux::check_signal_handler(int sig) {
4592 char buf[O_BUFLEN];
4593 address jvmHandler = NULL;
4594
4595
4596 struct sigaction act;
4597 if (os_sigaction == NULL) {
4598 // only trust the default sigaction, in case it has been interposed
4599 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4600 if (os_sigaction == NULL) return;
4601 }
4602
4603 os_sigaction(sig, (struct sigaction*)NULL, &act);
4604
4605
4606 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4607
4608 address thisHandler = (act.sa_flags & SA_SIGINFO)
4609 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4610 : CAST_FROM_FN_PTR(address, act.sa_handler);
4611
4612
4613 switch (sig) {
4614 case SIGSEGV:
4615 case SIGBUS:
4616 case SIGFPE:
4617 case SIGPIPE:
4618 case SIGILL:
4619 case SIGXFSZ:
4620 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4621 break;
4622
4623 case SHUTDOWN1_SIGNAL:
4624 case SHUTDOWN2_SIGNAL:
4625 case SHUTDOWN3_SIGNAL:
4626 case BREAK_SIGNAL:
4627 jvmHandler = (address)user_handler();
4628 break;
4629
4630 default:
4631 if (sig == SR_signum) {
4632 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4633 } else {
4634 return;
4635 }
4636 break;
4637 }
4638
4639 if (thisHandler != jvmHandler) {
4640 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4641 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4642 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4643 // No need to check this sig any longer
4644 sigaddset(&check_signal_done, sig);
4645 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4646 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4647 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4648 exception_name(sig, buf, O_BUFLEN));
4649 }
4650 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4651 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4652 tty->print("expected:");
4653 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4654 tty->cr();
4655 tty->print(" found:");
4656 os::Posix::print_sa_flags(tty, act.sa_flags);
4657 tty->cr();
4658 // No need to check this sig any longer
4659 sigaddset(&check_signal_done, sig);
4660 }
4661
4662 // Dump all the signal
4663 if (sigismember(&check_signal_done, sig)) {
4664 print_signal_handlers(tty, buf, O_BUFLEN);
4665 }
4666 }
4667
4668 extern void report_error(char* file_name, int line_no, char* title,
4669 char* format, ...);
4670
4671 // this is called _before_ the most of global arguments have been parsed
4672 void os::init(void) {
4673 char dummy; // used to get a guess on initial stack address
4674 // first_hrtime = gethrtime();
4675
4676 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4677
4678 init_random(1234567);
4679
4680 ThreadCritical::initialize();
4681
4682 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4683 if (Linux::page_size() == -1) {
4684 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4685 os::strerror(errno));
4686 }
4687 init_page_sizes((size_t) Linux::page_size());
4688
4689 Linux::initialize_system_info();
4690
4691 Linux::initialize_os_info();
4692
4693 // main_thread points to the aboriginal thread
4694 Linux::_main_thread = pthread_self();
4695
4696 Linux::clock_init();
4697 initial_time_count = javaTimeNanos();
4698
4699 // pthread_condattr initialization for monotonic clock
4700 int status;
4701 pthread_condattr_t* _condattr = os::Linux::condAttr();
4702 if ((status = pthread_condattr_init(_condattr)) != 0) {
4703 fatal("pthread_condattr_init: %s", os::strerror(status));
4704 }
4705 // Only set the clock if CLOCK_MONOTONIC is available
4706 if (os::supports_monotonic_clock()) {
4707 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4708 if (status == EINVAL) {
4709 warning("Unable to use monotonic clock with relative timed-waits" \
4710 " - changes to the time-of-day clock may have adverse affects");
4711 } else {
4712 fatal("pthread_condattr_setclock: %s", os::strerror(status));
4713 }
4714 }
4715 }
4716 // else it defaults to CLOCK_REALTIME
4717
4718 // retrieve entry point for pthread_setname_np
4719 Linux::_pthread_setname_np =
4720 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4721
4722 }
4723
4724 // To install functions for atexit system call
4725 extern "C" {
4726 static void perfMemory_exit_helper() {
4727 perfMemory_exit();
4728 }
4729 }
4730
4731 // this is called _after_ the global arguments have been parsed
4732 jint os::init_2(void) {
4733 Linux::fast_thread_clock_init();
4734
4735 // Allocate a single page and mark it as readable for safepoint polling
4736 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4737 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4738
4739 os::set_polling_page(polling_page);
4740 log_info(os)("SafePoint Polling address: " INTPTR_FORMAT, p2i(polling_page));
4741
4742 if (!UseMembar) {
4743 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4744 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4745 os::set_memory_serialize_page(mem_serialize_page);
4746 log_info(os)("Memory Serialize Page address: " INTPTR_FORMAT, p2i(mem_serialize_page));
4747 }
4748
4749 // initialize suspend/resume support - must do this before signal_sets_init()
4750 if (SR_initialize() != 0) {
4751 perror("SR_initialize failed");
4752 return JNI_ERR;
4753 }
4754
4755 Linux::signal_sets_init();
4756 Linux::install_signal_handlers();
4757
4758 // Check minimum allowable stack size for thread creation and to initialize
4759 // the java system classes, including StackOverflowError - depends on page
4760 // size. Add a page for compiler2 recursion in main thread.
4761 // Add in 2*BytesPerWord times page size to account for VM stack during
4762 // class initialization depending on 32 or 64 bit VM.
4763 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4764 JavaThread::stack_guard_zone_size() +
4765 JavaThread::stack_shadow_zone_size() +
4766 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4767
4768 size_t threadStackSizeInBytes = ThreadStackSize * K;
4769 if (threadStackSizeInBytes != 0 &&
4770 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4771 tty->print_cr("\nThe stack size specified is too small, "
4772 "Specify at least " SIZE_FORMAT "k",
4773 os::Linux::min_stack_allowed/ K);
4774 return JNI_ERR;
4775 }
4776
4777 // Make the stack size a multiple of the page size so that
4778 // the yellow/red zones can be guarded.
4779 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4780 vm_page_size()));
4781
4782 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4783
4784 #if defined(IA32)
4785 workaround_expand_exec_shield_cs_limit();
4786 #endif
4787
4788 Linux::libpthread_init();
4789 log_info(os)("HotSpot is running with %s, %s",
4790 Linux::glibc_version(), Linux::libpthread_version());
4791
4792 if (UseNUMA) {
4793 if (!Linux::libnuma_init()) {
4794 UseNUMA = false;
4795 } else {
4796 if ((Linux::numa_max_node() < 1)) {
4797 // There's only one node(they start from 0), disable NUMA.
4798 UseNUMA = false;
4799 }
4800 }
4801 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4802 // we can make the adaptive lgrp chunk resizing work. If the user specified
4803 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4804 // disable adaptive resizing.
4805 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4806 if (FLAG_IS_DEFAULT(UseNUMA)) {
4807 UseNUMA = false;
4808 } else {
4809 if (FLAG_IS_DEFAULT(UseLargePages) &&
4810 FLAG_IS_DEFAULT(UseSHM) &&
4811 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4812 UseLargePages = false;
4813 } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4814 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4815 UseAdaptiveSizePolicy = false;
4816 UseAdaptiveNUMAChunkSizing = false;
4817 }
4818 }
4819 }
4820 if (!UseNUMA && ForceNUMA) {
4821 UseNUMA = true;
4822 }
4823 }
4824
4825 if (MaxFDLimit) {
4826 // set the number of file descriptors to max. print out error
4827 // if getrlimit/setrlimit fails but continue regardless.
4828 struct rlimit nbr_files;
4829 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4830 if (status != 0) {
4831 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4832 } else {
4833 nbr_files.rlim_cur = nbr_files.rlim_max;
4834 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4835 if (status != 0) {
4836 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4837 }
4838 }
4839 }
4840
4841 // Initialize lock used to serialize thread creation (see os::create_thread)
4842 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4843
4844 // at-exit methods are called in the reverse order of their registration.
4845 // atexit functions are called on return from main or as a result of a
4846 // call to exit(3C). There can be only 32 of these functions registered
4847 // and atexit() does not set errno.
4848
4849 if (PerfAllowAtExitRegistration) {
4850 // only register atexit functions if PerfAllowAtExitRegistration is set.
4851 // atexit functions can be delayed until process exit time, which
4852 // can be problematic for embedded VM situations. Embedded VMs should
4853 // call DestroyJavaVM() to assure that VM resources are released.
4854
4855 // note: perfMemory_exit_helper atexit function may be removed in
4856 // the future if the appropriate cleanup code can be added to the
4857 // VM_Exit VMOperation's doit method.
4858 if (atexit(perfMemory_exit_helper) != 0) {
4859 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4860 }
4861 }
4862
4863 // initialize thread priority policy
4864 prio_init();
4865
4866 return JNI_OK;
4867 }
4868
4869 // Mark the polling page as unreadable
4870 void os::make_polling_page_unreadable(void) {
4871 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4872 fatal("Could not disable polling page");
4873 }
4874 }
4875
4876 // Mark the polling page as readable
4877 void os::make_polling_page_readable(void) {
4878 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4879 fatal("Could not enable polling page");
4880 }
4881 }
4882
4883 // older glibc versions don't have this macro (which expands to
4884 // an optimized bit-counting function) so we have to roll our own
4885 #ifndef CPU_COUNT
4886
4887 static int _cpu_count(const cpu_set_t* cpus) {
4888 int count = 0;
4889 // only look up to the number of configured processors
4890 for (int i = 0; i < os::processor_count(); i++) {
4891 if (CPU_ISSET(i, cpus)) {
4892 count++;
4893 }
4894 }
4895 return count;
4896 }
4897
4898 #define CPU_COUNT(cpus) _cpu_count(cpus)
4899
4900 #endif // CPU_COUNT
4901
4902 // Get the current number of available processors for this process.
4903 // This value can change at any time during a process's lifetime.
4904 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
4905 // If it appears there may be more than 1024 processors then we do a
4906 // dynamic check - see 6515172 for details.
4907 // If anything goes wrong we fallback to returning the number of online
4908 // processors - which can be greater than the number available to the process.
4909 int os::active_processor_count() {
4910 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
4911 cpu_set_t* cpus_p = &cpus;
4912 int cpus_size = sizeof(cpu_set_t);
4913
4914 int configured_cpus = processor_count(); // upper bound on available cpus
4915 int cpu_count = 0;
4916
4917 // old build platforms may not support dynamic cpu sets
4918 #ifdef CPU_ALLOC
4919
4920 // To enable easy testing of the dynamic path on different platforms we
4921 // introduce a diagnostic flag: UseCpuAllocPath
4922 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
4923 // kernel may use a mask bigger than cpu_set_t
4924 log_trace(os)("active_processor_count: using dynamic path %s"
4925 "- configured processors: %d",
4926 UseCpuAllocPath ? "(forced) " : "",
4927 configured_cpus);
4928 cpus_p = CPU_ALLOC(configured_cpus);
4929 if (cpus_p != NULL) {
4930 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
4931 // zero it just to be safe
4932 CPU_ZERO_S(cpus_size, cpus_p);
4933 }
4934 else {
4935 // failed to allocate so fallback to online cpus
4936 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4937 log_trace(os)("active_processor_count: "
4938 "CPU_ALLOC failed (%s) - using "
4939 "online processor count: %d",
4940 os::strerror(errno), online_cpus);
4941 return online_cpus;
4942 }
4943 }
4944 else {
4945 log_trace(os)("active_processor_count: using static path - configured processors: %d",
4946 configured_cpus);
4947 }
4948 #else // CPU_ALLOC
4949 // these stubs won't be executed
4950 #define CPU_COUNT_S(size, cpus) -1
4951 #define CPU_FREE(cpus)
4952
4953 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
4954 configured_cpus);
4955 #endif // CPU_ALLOC
4956
4957 // pid 0 means the current thread - which we have to assume represents the process
4958 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
4959 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
4960 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
4961 }
4962 else {
4963 cpu_count = CPU_COUNT(cpus_p);
4964 }
4965 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
4966 }
4967 else {
4968 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
4969 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
4970 "which may exceed available processors", os::strerror(errno), cpu_count);
4971 }
4972
4973 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
4974 CPU_FREE(cpus_p);
4975 }
4976
4977 assert(cpu_count > 0 && cpu_count <= processor_count(), "sanity check");
4978 return cpu_count;
4979 }
4980
4981 void os::set_native_thread_name(const char *name) {
4982 if (Linux::_pthread_setname_np) {
4983 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4984 snprintf(buf, sizeof(buf), "%s", name);
4985 buf[sizeof(buf) - 1] = '\0';
4986 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4987 // ERANGE should not happen; all other errors should just be ignored.
4988 assert(rc != ERANGE, "pthread_setname_np failed");
4989 }
4990 }
4991
4992 bool os::distribute_processes(uint length, uint* distribution) {
4993 // Not yet implemented.
4994 return false;
4995 }
4996
4997 bool os::bind_to_processor(uint processor_id) {
4998 // Not yet implemented.
4999 return false;
5000 }
5001
5002 ///
5003
5004 void os::SuspendedThreadTask::internal_do_task() {
5005 if (do_suspend(_thread->osthread())) {
5006 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5007 do_task(context);
5008 do_resume(_thread->osthread());
5009 }
5010 }
5011
5012 class PcFetcher : public os::SuspendedThreadTask {
5013 public:
5014 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5015 ExtendedPC result();
5016 protected:
5017 void do_task(const os::SuspendedThreadTaskContext& context);
5018 private:
5019 ExtendedPC _epc;
5020 };
5021
5022 ExtendedPC PcFetcher::result() {
5023 guarantee(is_done(), "task is not done yet.");
5024 return _epc;
5025 }
5026
5027 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5028 Thread* thread = context.thread();
5029 OSThread* osthread = thread->osthread();
5030 if (osthread->ucontext() != NULL) {
5031 _epc = os::Linux::ucontext_get_pc((const ucontext_t *) context.ucontext());
5032 } else {
5033 // NULL context is unexpected, double-check this is the VMThread
5034 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5035 }
5036 }
5037
5038 // Suspends the target using the signal mechanism and then grabs the PC before
5039 // resuming the target. Used by the flat-profiler only
5040 ExtendedPC os::get_thread_pc(Thread* thread) {
5041 // Make sure that it is called by the watcher for the VMThread
5042 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5043 assert(thread->is_VM_thread(), "Can only be called for VMThread");
5044
5045 PcFetcher fetcher(thread);
5046 fetcher.run();
5047 return fetcher.result();
5048 }
5049
5050 ////////////////////////////////////////////////////////////////////////////////
5051 // debug support
5052
5053 bool os::find(address addr, outputStream* st) {
5054 Dl_info dlinfo;
5055 memset(&dlinfo, 0, sizeof(dlinfo));
5056 if (dladdr(addr, &dlinfo) != 0) {
5057 st->print(PTR_FORMAT ": ", p2i(addr));
5058 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5059 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5060 p2i(addr) - p2i(dlinfo.dli_saddr));
5061 } else if (dlinfo.dli_fbase != NULL) {
5062 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5063 } else {
5064 st->print("<absolute address>");
5065 }
5066 if (dlinfo.dli_fname != NULL) {
5067 st->print(" in %s", dlinfo.dli_fname);
5068 }
5069 if (dlinfo.dli_fbase != NULL) {
5070 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5071 }
5072 st->cr();
5073
5074 if (Verbose) {
5075 // decode some bytes around the PC
5076 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5077 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5078 address lowest = (address) dlinfo.dli_sname;
5079 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5080 if (begin < lowest) begin = lowest;
5081 Dl_info dlinfo2;
5082 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5083 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5084 end = (address) dlinfo2.dli_saddr;
5085 }
5086 Disassembler::decode(begin, end, st);
5087 }
5088 return true;
5089 }
5090 return false;
5091 }
5092
5093 ////////////////////////////////////////////////////////////////////////////////
5094 // misc
5095
5096 // This does not do anything on Linux. This is basically a hook for being
5097 // able to use structured exception handling (thread-local exception filters)
5098 // on, e.g., Win32.
5099 void
5100 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5101 JavaCallArguments* args, Thread* thread) {
5102 f(value, method, args, thread);
5103 }
5104
5105 void os::print_statistics() {
5106 }
5107
5108 bool os::message_box(const char* title, const char* message) {
5109 int i;
5110 fdStream err(defaultStream::error_fd());
5111 for (i = 0; i < 78; i++) err.print_raw("=");
5112 err.cr();
5113 err.print_raw_cr(title);
5114 for (i = 0; i < 78; i++) err.print_raw("-");
5115 err.cr();
5116 err.print_raw_cr(message);
5117 for (i = 0; i < 78; i++) err.print_raw("=");
5118 err.cr();
5119
5120 char buf[16];
5121 // Prevent process from exiting upon "read error" without consuming all CPU
5122 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5123
5124 return buf[0] == 'y' || buf[0] == 'Y';
5125 }
5126
5127 int os::stat(const char *path, struct stat *sbuf) {
5128 char pathbuf[MAX_PATH];
5129 if (strlen(path) > MAX_PATH - 1) {
5130 errno = ENAMETOOLONG;
5131 return -1;
5132 }
5133 os::native_path(strcpy(pathbuf, path));
5134 return ::stat(pathbuf, sbuf);
5135 }
5136
5137 bool os::check_heap(bool force) {
5138 return true;
5139 }
5140
5141 // Is a (classpath) directory empty?
5142 bool os::dir_is_empty(const char* path) {
5143 DIR *dir = NULL;
5144 struct dirent *ptr;
5145
5146 dir = opendir(path);
5147 if (dir == NULL) return true;
5148
5149 // Scan the directory
5150 bool result = true;
5151 char buf[sizeof(struct dirent) + MAX_PATH];
5152 while (result && (ptr = ::readdir(dir)) != NULL) {
5153 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5154 result = false;
5155 }
5156 }
5157 closedir(dir);
5158 return result;
5159 }
5160
5161 // This code originates from JDK's sysOpen and open64_w
5162 // from src/solaris/hpi/src/system_md.c
5163
5164 int os::open(const char *path, int oflag, int mode) {
5165 if (strlen(path) > MAX_PATH - 1) {
5166 errno = ENAMETOOLONG;
5167 return -1;
5168 }
5169
5170 // All file descriptors that are opened in the Java process and not
5171 // specifically destined for a subprocess should have the close-on-exec
5172 // flag set. If we don't set it, then careless 3rd party native code
5173 // might fork and exec without closing all appropriate file descriptors
5174 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5175 // turn might:
5176 //
5177 // - cause end-of-file to fail to be detected on some file
5178 // descriptors, resulting in mysterious hangs, or
5179 //
5180 // - might cause an fopen in the subprocess to fail on a system
5181 // suffering from bug 1085341.
5182 //
5183 // (Yes, the default setting of the close-on-exec flag is a Unix
5184 // design flaw)
5185 //
5186 // See:
5187 // 1085341: 32-bit stdio routines should support file descriptors >255
5188 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5189 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5190 //
5191 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5192 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5193 // because it saves a system call and removes a small window where the flag
5194 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5195 // and we fall back to using FD_CLOEXEC (see below).
5196 #ifdef O_CLOEXEC
5197 oflag |= O_CLOEXEC;
5198 #endif
5199
5200 int fd = ::open64(path, oflag, mode);
5201 if (fd == -1) return -1;
5202
5203 //If the open succeeded, the file might still be a directory
5204 {
5205 struct stat64 buf64;
5206 int ret = ::fstat64(fd, &buf64);
5207 int st_mode = buf64.st_mode;
5208
5209 if (ret != -1) {
5210 if ((st_mode & S_IFMT) == S_IFDIR) {
5211 errno = EISDIR;
5212 ::close(fd);
5213 return -1;
5214 }
5215 } else {
5216 ::close(fd);
5217 return -1;
5218 }
5219 }
5220
5221 #ifdef FD_CLOEXEC
5222 // Validate that the use of the O_CLOEXEC flag on open above worked.
5223 // With recent kernels, we will perform this check exactly once.
5224 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5225 if (!O_CLOEXEC_is_known_to_work) {
5226 int flags = ::fcntl(fd, F_GETFD);
5227 if (flags != -1) {
5228 if ((flags & FD_CLOEXEC) != 0)
5229 O_CLOEXEC_is_known_to_work = 1;
5230 else
5231 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5232 }
5233 }
5234 #endif
5235
5236 return fd;
5237 }
5238
5239
5240 // create binary file, rewriting existing file if required
5241 int os::create_binary_file(const char* path, bool rewrite_existing) {
5242 int oflags = O_WRONLY | O_CREAT;
5243 if (!rewrite_existing) {
5244 oflags |= O_EXCL;
5245 }
5246 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5247 }
5248
5249 // return current position of file pointer
5250 jlong os::current_file_offset(int fd) {
5251 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5252 }
5253
5254 // move file pointer to the specified offset
5255 jlong os::seek_to_file_offset(int fd, jlong offset) {
5256 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5257 }
5258
5259 // This code originates from JDK's sysAvailable
5260 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5261
5262 int os::available(int fd, jlong *bytes) {
5263 jlong cur, end;
5264 int mode;
5265 struct stat64 buf64;
5266
5267 if (::fstat64(fd, &buf64) >= 0) {
5268 mode = buf64.st_mode;
5269 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5270 int n;
5271 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5272 *bytes = n;
5273 return 1;
5274 }
5275 }
5276 }
5277 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5278 return 0;
5279 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5280 return 0;
5281 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5282 return 0;
5283 }
5284 *bytes = end - cur;
5285 return 1;
5286 }
5287
5288 // Map a block of memory.
5289 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5290 char *addr, size_t bytes, bool read_only,
5291 bool allow_exec) {
5292 int prot;
5293 int flags = MAP_PRIVATE;
5294
5295 if (read_only) {
5296 prot = PROT_READ;
5297 } else {
5298 prot = PROT_READ | PROT_WRITE;
5299 }
5300
5301 if (allow_exec) {
5302 prot |= PROT_EXEC;
5303 }
5304
5305 if (addr != NULL) {
5306 flags |= MAP_FIXED;
5307 }
5308
5309 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5310 fd, file_offset);
5311 if (mapped_address == MAP_FAILED) {
5312 return NULL;
5313 }
5314 return mapped_address;
5315 }
5316
5317
5318 // Remap a block of memory.
5319 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5320 char *addr, size_t bytes, bool read_only,
5321 bool allow_exec) {
5322 // same as map_memory() on this OS
5323 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5324 allow_exec);
5325 }
5326
5327
5328 // Unmap a block of memory.
5329 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5330 return munmap(addr, bytes) == 0;
5331 }
5332
5333 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5334
5335 static clockid_t thread_cpu_clockid(Thread* thread) {
5336 pthread_t tid = thread->osthread()->pthread_id();
5337 clockid_t clockid;
5338
5339 // Get thread clockid
5340 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5341 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5342 return clockid;
5343 }
5344
5345 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5346 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5347 // of a thread.
5348 //
5349 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5350 // the fast estimate available on the platform.
5351
5352 jlong os::current_thread_cpu_time() {
5353 if (os::Linux::supports_fast_thread_cpu_time()) {
5354 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5355 } else {
5356 // return user + sys since the cost is the same
5357 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5358 }
5359 }
5360
5361 jlong os::thread_cpu_time(Thread* thread) {
5362 // consistent with what current_thread_cpu_time() returns
5363 if (os::Linux::supports_fast_thread_cpu_time()) {
5364 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5365 } else {
5366 return slow_thread_cpu_time(thread, true /* user + sys */);
5367 }
5368 }
5369
5370 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5371 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5372 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5373 } else {
5374 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5375 }
5376 }
5377
5378 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5379 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5380 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5381 } else {
5382 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5383 }
5384 }
5385
5386 // -1 on error.
5387 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5388 pid_t tid = thread->osthread()->thread_id();
5389 char *s;
5390 char stat[2048];
5391 int statlen;
5392 char proc_name[64];
5393 int count;
5394 long sys_time, user_time;
5395 char cdummy;
5396 int idummy;
5397 long ldummy;
5398 FILE *fp;
5399
5400 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5401 fp = fopen(proc_name, "r");
5402 if (fp == NULL) return -1;
5403 statlen = fread(stat, 1, 2047, fp);
5404 stat[statlen] = '\0';
5405 fclose(fp);
5406
5407 // Skip pid and the command string. Note that we could be dealing with
5408 // weird command names, e.g. user could decide to rename java launcher
5409 // to "java 1.4.2 :)", then the stat file would look like
5410 // 1234 (java 1.4.2 :)) R ... ...
5411 // We don't really need to know the command string, just find the last
5412 // occurrence of ")" and then start parsing from there. See bug 4726580.
5413 s = strrchr(stat, ')');
5414 if (s == NULL) return -1;
5415
5416 // Skip blank chars
5417 do { s++; } while (s && isspace(*s));
5418
5419 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5420 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5421 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5422 &user_time, &sys_time);
5423 if (count != 13) return -1;
5424 if (user_sys_cpu_time) {
5425 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5426 } else {
5427 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5428 }
5429 }
5430
5431 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5432 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5433 info_ptr->may_skip_backward = false; // elapsed time not wall time
5434 info_ptr->may_skip_forward = false; // elapsed time not wall time
5435 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5436 }
5437
5438 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5439 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5440 info_ptr->may_skip_backward = false; // elapsed time not wall time
5441 info_ptr->may_skip_forward = false; // elapsed time not wall time
5442 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5443 }
5444
5445 bool os::is_thread_cpu_time_supported() {
5446 return true;
5447 }
5448
5449 // System loadavg support. Returns -1 if load average cannot be obtained.
5450 // Linux doesn't yet have a (official) notion of processor sets,
5451 // so just return the system wide load average.
5452 int os::loadavg(double loadavg[], int nelem) {
5453 return ::getloadavg(loadavg, nelem);
5454 }
5455
5456 void os::pause() {
5457 char filename[MAX_PATH];
5458 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5459 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5460 } else {
5461 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5462 }
5463
5464 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5465 if (fd != -1) {
5466 struct stat buf;
5467 ::close(fd);
5468 while (::stat(filename, &buf) == 0) {
5469 (void)::poll(NULL, 0, 100);
5470 }
5471 } else {
5472 jio_fprintf(stderr,
5473 "Could not open pause file '%s', continuing immediately.\n", filename);
5474 }
5475 }
5476
5477
5478 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5479 // comment paragraphs are worth repeating here:
5480 //
5481 // Assumption:
5482 // Only one parker can exist on an event, which is why we allocate
5483 // them per-thread. Multiple unparkers can coexist.
5484 //
5485 // _Event serves as a restricted-range semaphore.
5486 // -1 : thread is blocked, i.e. there is a waiter
5487 // 0 : neutral: thread is running or ready,
5488 // could have been signaled after a wait started
5489 // 1 : signaled - thread is running or ready
5490 //
5491
5492 // utility to compute the abstime argument to timedwait:
5493 // millis is the relative timeout time
5494 // abstime will be the absolute timeout time
5495 // TODO: replace compute_abstime() with unpackTime()
5496
5497 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5498 if (millis < 0) millis = 0;
5499
5500 jlong seconds = millis / 1000;
5501 millis %= 1000;
5502 if (seconds > 50000000) { // see man cond_timedwait(3T)
5503 seconds = 50000000;
5504 }
5505
5506 if (os::supports_monotonic_clock()) {
5507 struct timespec now;
5508 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5509 assert_status(status == 0, status, "clock_gettime");
5510 abstime->tv_sec = now.tv_sec + seconds;
5511 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5512 if (nanos >= NANOSECS_PER_SEC) {
5513 abstime->tv_sec += 1;
5514 nanos -= NANOSECS_PER_SEC;
5515 }
5516 abstime->tv_nsec = nanos;
5517 } else {
5518 struct timeval now;
5519 int status = gettimeofday(&now, NULL);
5520 assert(status == 0, "gettimeofday");
5521 abstime->tv_sec = now.tv_sec + seconds;
5522 long usec = now.tv_usec + millis * 1000;
5523 if (usec >= 1000000) {
5524 abstime->tv_sec += 1;
5525 usec -= 1000000;
5526 }
5527 abstime->tv_nsec = usec * 1000;
5528 }
5529 return abstime;
5530 }
5531
5532 void os::PlatformEvent::park() { // AKA "down()"
5533 // Transitions for _Event:
5534 // -1 => -1 : illegal
5535 // 1 => 0 : pass - return immediately
5536 // 0 => -1 : block; then set _Event to 0 before returning
5537
5538 // Invariant: Only the thread associated with the Event/PlatformEvent
5539 // may call park().
5540 // TODO: assert that _Assoc != NULL or _Assoc == Self
5541 assert(_nParked == 0, "invariant");
5542
5543 int v;
5544 for (;;) {
5545 v = _Event;
5546 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5547 }
5548 guarantee(v >= 0, "invariant");
5549 if (v == 0) {
5550 // Do this the hard way by blocking ...
5551 int status = pthread_mutex_lock(_mutex);
5552 assert_status(status == 0, status, "mutex_lock");
5553 guarantee(_nParked == 0, "invariant");
5554 ++_nParked;
5555 while (_Event < 0) {
5556 status = pthread_cond_wait(_cond, _mutex);
5557 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5558 // Treat this the same as if the wait was interrupted
5559 if (status == ETIME) { status = EINTR; }
5560 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5561 }
5562 --_nParked;
5563
5564 _Event = 0;
5565 status = pthread_mutex_unlock(_mutex);
5566 assert_status(status == 0, status, "mutex_unlock");
5567 // Paranoia to ensure our locked and lock-free paths interact
5568 // correctly with each other.
5569 OrderAccess::fence();
5570 }
5571 guarantee(_Event >= 0, "invariant");
5572 }
5573
5574 int os::PlatformEvent::park(jlong millis) {
5575 // Transitions for _Event:
5576 // -1 => -1 : illegal
5577 // 1 => 0 : pass - return immediately
5578 // 0 => -1 : block; then set _Event to 0 before returning
5579
5580 guarantee(_nParked == 0, "invariant");
5581
5582 int v;
5583 for (;;) {
5584 v = _Event;
5585 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5586 }
5587 guarantee(v >= 0, "invariant");
5588 if (v != 0) return OS_OK;
5589
5590 // We do this the hard way, by blocking the thread.
5591 // Consider enforcing a minimum timeout value.
5592 struct timespec abst;
5593 compute_abstime(&abst, millis);
5594
5595 int ret = OS_TIMEOUT;
5596 int status = pthread_mutex_lock(_mutex);
5597 assert_status(status == 0, status, "mutex_lock");
5598 guarantee(_nParked == 0, "invariant");
5599 ++_nParked;
5600
5601 // Object.wait(timo) will return because of
5602 // (a) notification
5603 // (b) timeout
5604 // (c) thread.interrupt
5605 //
5606 // Thread.interrupt and object.notify{All} both call Event::set.
5607 // That is, we treat thread.interrupt as a special case of notification.
5608 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5609 // We assume all ETIME returns are valid.
5610 //
5611 // TODO: properly differentiate simultaneous notify+interrupt.
5612 // In that case, we should propagate the notify to another waiter.
5613
5614 while (_Event < 0) {
5615 status = pthread_cond_timedwait(_cond, _mutex, &abst);
5616 assert_status(status == 0 || status == EINTR ||
5617 status == ETIME || status == ETIMEDOUT,
5618 status, "cond_timedwait");
5619 if (!FilterSpuriousWakeups) break; // previous semantics
5620 if (status == ETIME || status == ETIMEDOUT) break;
5621 // We consume and ignore EINTR and spurious wakeups.
5622 }
5623 --_nParked;
5624 if (_Event >= 0) {
5625 ret = OS_OK;
5626 }
5627 _Event = 0;
5628 status = pthread_mutex_unlock(_mutex);
5629 assert_status(status == 0, status, "mutex_unlock");
5630 assert(_nParked == 0, "invariant");
5631 // Paranoia to ensure our locked and lock-free paths interact
5632 // correctly with each other.
5633 OrderAccess::fence();
5634 return ret;
5635 }
5636
5637 void os::PlatformEvent::unpark() {
5638 // Transitions for _Event:
5639 // 0 => 1 : just return
5640 // 1 => 1 : just return
5641 // -1 => either 0 or 1; must signal target thread
5642 // That is, we can safely transition _Event from -1 to either
5643 // 0 or 1.
5644 // See also: "Semaphores in Plan 9" by Mullender & Cox
5645 //
5646 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5647 // that it will take two back-to-back park() calls for the owning
5648 // thread to block. This has the benefit of forcing a spurious return
5649 // from the first park() call after an unpark() call which will help
5650 // shake out uses of park() and unpark() without condition variables.
5651
5652 if (Atomic::xchg(1, &_Event) >= 0) return;
5653
5654 // Wait for the thread associated with the event to vacate
5655 int status = pthread_mutex_lock(_mutex);
5656 assert_status(status == 0, status, "mutex_lock");
5657 int AnyWaiters = _nParked;
5658 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5659 status = pthread_mutex_unlock(_mutex);
5660 assert_status(status == 0, status, "mutex_unlock");
5661 if (AnyWaiters != 0) {
5662 // Note that we signal() *after* dropping the lock for "immortal" Events.
5663 // This is safe and avoids a common class of futile wakeups. In rare
5664 // circumstances this can cause a thread to return prematurely from
5665 // cond_{timed}wait() but the spurious wakeup is benign and the victim
5666 // will simply re-test the condition and re-park itself.
5667 // This provides particular benefit if the underlying platform does not
5668 // provide wait morphing.
5669 status = pthread_cond_signal(_cond);
5670 assert_status(status == 0, status, "cond_signal");
5671 }
5672 }
5673
5674
5675 // JSR166
5676 // -------------------------------------------------------
5677
5678 // The solaris and linux implementations of park/unpark are fairly
5679 // conservative for now, but can be improved. They currently use a
5680 // mutex/condvar pair, plus a a count.
5681 // Park decrements count if > 0, else does a condvar wait. Unpark
5682 // sets count to 1 and signals condvar. Only one thread ever waits
5683 // on the condvar. Contention seen when trying to park implies that someone
5684 // is unparking you, so don't wait. And spurious returns are fine, so there
5685 // is no need to track notifications.
5686
5687 // This code is common to linux and solaris and will be moved to a
5688 // common place in dolphin.
5689 //
5690 // The passed in time value is either a relative time in nanoseconds
5691 // or an absolute time in milliseconds. Either way it has to be unpacked
5692 // into suitable seconds and nanoseconds components and stored in the
5693 // given timespec structure.
5694 // Given time is a 64-bit value and the time_t used in the timespec is only
5695 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5696 // overflow if times way in the future are given. Further on Solaris versions
5697 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5698 // number of seconds, in abstime, is less than current_time + 100,000,000.
5699 // As it will be 28 years before "now + 100000000" will overflow we can
5700 // ignore overflow and just impose a hard-limit on seconds using the value
5701 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5702 // years from "now".
5703
5704 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5705 assert(time > 0, "convertTime");
5706 time_t max_secs = 0;
5707
5708 if (!os::supports_monotonic_clock() || isAbsolute) {
5709 struct timeval now;
5710 int status = gettimeofday(&now, NULL);
5711 assert(status == 0, "gettimeofday");
5712
5713 max_secs = now.tv_sec + MAX_SECS;
5714
5715 if (isAbsolute) {
5716 jlong secs = time / 1000;
5717 if (secs > max_secs) {
5718 absTime->tv_sec = max_secs;
5719 } else {
5720 absTime->tv_sec = secs;
5721 }
5722 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5723 } else {
5724 jlong secs = time / NANOSECS_PER_SEC;
5725 if (secs >= MAX_SECS) {
5726 absTime->tv_sec = max_secs;
5727 absTime->tv_nsec = 0;
5728 } else {
5729 absTime->tv_sec = now.tv_sec + secs;
5730 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5731 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5732 absTime->tv_nsec -= NANOSECS_PER_SEC;
5733 ++absTime->tv_sec; // note: this must be <= max_secs
5734 }
5735 }
5736 }
5737 } else {
5738 // must be relative using monotonic clock
5739 struct timespec now;
5740 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5741 assert_status(status == 0, status, "clock_gettime");
5742 max_secs = now.tv_sec + MAX_SECS;
5743 jlong secs = time / NANOSECS_PER_SEC;
5744 if (secs >= MAX_SECS) {
5745 absTime->tv_sec = max_secs;
5746 absTime->tv_nsec = 0;
5747 } else {
5748 absTime->tv_sec = now.tv_sec + secs;
5749 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5750 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5751 absTime->tv_nsec -= NANOSECS_PER_SEC;
5752 ++absTime->tv_sec; // note: this must be <= max_secs
5753 }
5754 }
5755 }
5756 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5757 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5758 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5759 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5760 }
5761
5762 void Parker::park(bool isAbsolute, jlong time) {
5763 // Ideally we'd do something useful while spinning, such
5764 // as calling unpackTime().
5765
5766 // Optional fast-path check:
5767 // Return immediately if a permit is available.
5768 // We depend on Atomic::xchg() having full barrier semantics
5769 // since we are doing a lock-free update to _counter.
5770 if (Atomic::xchg(0, &_counter) > 0) return;
5771
5772 Thread* thread = Thread::current();
5773 assert(thread->is_Java_thread(), "Must be JavaThread");
5774 JavaThread *jt = (JavaThread *)thread;
5775
5776 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5777 // Check interrupt before trying to wait
5778 if (Thread::is_interrupted(thread, false)) {
5779 return;
5780 }
5781
5782 // Next, demultiplex/decode time arguments
5783 timespec absTime;
5784 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5785 return;
5786 }
5787 if (time > 0) {
5788 unpackTime(&absTime, isAbsolute, time);
5789 }
5790
5791
5792 // Enter safepoint region
5793 // Beware of deadlocks such as 6317397.
5794 // The per-thread Parker:: mutex is a classic leaf-lock.
5795 // In particular a thread must never block on the Threads_lock while
5796 // holding the Parker:: mutex. If safepoints are pending both the
5797 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5798 ThreadBlockInVM tbivm(jt);
5799
5800 // Don't wait if cannot get lock since interference arises from
5801 // unblocking. Also. check interrupt before trying wait
5802 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5803 return;
5804 }
5805
5806 int status;
5807 if (_counter > 0) { // no wait needed
5808 _counter = 0;
5809 status = pthread_mutex_unlock(_mutex);
5810 assert_status(status == 0, status, "invariant");
5811 // Paranoia to ensure our locked and lock-free paths interact
5812 // correctly with each other and Java-level accesses.
5813 OrderAccess::fence();
5814 return;
5815 }
5816
5817 #ifdef ASSERT
5818 // Don't catch signals while blocked; let the running threads have the signals.
5819 // (This allows a debugger to break into the running thread.)
5820 sigset_t oldsigs;
5821 sigemptyset(&oldsigs);
5822 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5823 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5824 #endif
5825
5826 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5827 jt->set_suspend_equivalent();
5828 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5829
5830 assert(_cur_index == -1, "invariant");
5831 if (time == 0) {
5832 _cur_index = REL_INDEX; // arbitrary choice when not timed
5833 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5834 } else {
5835 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5836 status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5837 }
5838 _cur_index = -1;
5839 assert_status(status == 0 || status == EINTR ||
5840 status == ETIME || status == ETIMEDOUT,
5841 status, "cond_timedwait");
5842
5843 #ifdef ASSERT
5844 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5845 #endif
5846
5847 _counter = 0;
5848 status = pthread_mutex_unlock(_mutex);
5849 assert_status(status == 0, status, "invariant");
5850 // Paranoia to ensure our locked and lock-free paths interact
5851 // correctly with each other and Java-level accesses.
5852 OrderAccess::fence();
5853
5854 // If externally suspended while waiting, re-suspend
5855 if (jt->handle_special_suspend_equivalent_condition()) {
5856 jt->java_suspend_self();
5857 }
5858 }
5859
5860 void Parker::unpark() {
5861 int status = pthread_mutex_lock(_mutex);
5862 assert_status(status == 0, status, "invariant");
5863 const int s = _counter;
5864 _counter = 1;
5865 // must capture correct index before unlocking
5866 int index = _cur_index;
5867 status = pthread_mutex_unlock(_mutex);
5868 assert_status(status == 0, status, "invariant");
5869 if (s < 1 && index != -1) {
5870 // thread is definitely parked
5871 status = pthread_cond_signal(&_cond[index]);
5872 assert_status(status == 0, status, "invariant");
5873 }
5874 }
5875
5876
5877 extern char** environ;
5878
5879 // Run the specified command in a separate process. Return its exit value,
5880 // or -1 on failure (e.g. can't fork a new process).
5881 // Unlike system(), this function can be called from signal handler. It
5882 // doesn't block SIGINT et al.
5883 int os::fork_and_exec(char* cmd) {
5884 const char * argv[4] = {"sh", "-c", cmd, NULL};
5885
5886 pid_t pid = fork();
5887
5888 if (pid < 0) {
5889 // fork failed
5890 return -1;
5891
5892 } else if (pid == 0) {
5893 // child process
5894
5895 execve("/bin/sh", (char* const*)argv, environ);
5896
5897 // execve failed
5898 _exit(-1);
5899
5900 } else {
5901 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5902 // care about the actual exit code, for now.
5903
5904 int status;
5905
5906 // Wait for the child process to exit. This returns immediately if
5907 // the child has already exited. */
5908 while (waitpid(pid, &status, 0) < 0) {
5909 switch (errno) {
5910 case ECHILD: return 0;
5911 case EINTR: break;
5912 default: return -1;
5913 }
5914 }
5915
5916 if (WIFEXITED(status)) {
5917 // The child exited normally; get its exit code.
5918 return WEXITSTATUS(status);
5919 } else if (WIFSIGNALED(status)) {
5920 // The child exited because of a signal
5921 // The best value to return is 0x80 + signal number,
5922 // because that is what all Unix shells do, and because
5923 // it allows callers to distinguish between process exit and
5924 // process death by signal.
5925 return 0x80 + WTERMSIG(status);
5926 } else {
5927 // Unknown exit code; pass it through
5928 return status;
5929 }
5930 }
5931 }
5932
5933 // is_headless_jre()
5934 //
5935 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5936 // in order to report if we are running in a headless jre
5937 //
5938 // Since JDK8 xawt/libmawt.so was moved into the same directory
5939 // as libawt.so, and renamed libawt_xawt.so
5940 //
5941 bool os::is_headless_jre() {
5942 struct stat statbuf;
5943 char buf[MAXPATHLEN];
5944 char libmawtpath[MAXPATHLEN];
5945 const char *xawtstr = "/xawt/libmawt.so";
5946 const char *new_xawtstr = "/libawt_xawt.so";
5947 char *p;
5948
5949 // Get path to libjvm.so
5950 os::jvm_path(buf, sizeof(buf));
5951
5952 // Get rid of libjvm.so
5953 p = strrchr(buf, '/');
5954 if (p == NULL) {
5955 return false;
5956 } else {
5957 *p = '\0';
5958 }
5959
5960 // Get rid of client or server
5961 p = strrchr(buf, '/');
5962 if (p == NULL) {
5963 return false;
5964 } else {
5965 *p = '\0';
5966 }
5967
5968 // check xawt/libmawt.so
5969 strcpy(libmawtpath, buf);
5970 strcat(libmawtpath, xawtstr);
5971 if (::stat(libmawtpath, &statbuf) == 0) return false;
5972
5973 // check libawt_xawt.so
5974 strcpy(libmawtpath, buf);
5975 strcat(libmawtpath, new_xawtstr);
5976 if (::stat(libmawtpath, &statbuf) == 0) return false;
5977
5978 return true;
5979 }
5980
5981 // Get the default path to the core file
5982 // Returns the length of the string
5983 int os::get_core_path(char* buffer, size_t bufferSize) {
5984 /*
5985 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5986 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5987 */
5988 const int core_pattern_len = 129;
5989 char core_pattern[core_pattern_len] = {0};
5990
5991 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5992 if (core_pattern_file == -1) {
5993 return -1;
5994 }
5995
5996 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5997 ::close(core_pattern_file);
5998 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5999 return -1;
6000 }
6001 if (core_pattern[ret-1] == '\n') {
6002 core_pattern[ret-1] = '\0';
6003 } else {
6004 core_pattern[ret] = '\0';
6005 }
6006
6007 char *pid_pos = strstr(core_pattern, "%p");
6008 int written;
6009
6010 if (core_pattern[0] == '/') {
6011 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
6012 } else {
6013 char cwd[PATH_MAX];
6014
6015 const char* p = get_current_directory(cwd, PATH_MAX);
6016 if (p == NULL) {
6017 return -1;
6018 }
6019
6020 if (core_pattern[0] == '|') {
6021 written = jio_snprintf(buffer, bufferSize,
6022 "\"%s\" (or dumping to %s/core.%d)",
6023 &core_pattern[1], p, current_process_id());
6024 } else {
6025 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6026 }
6027 }
6028
6029 if (written < 0) {
6030 return -1;
6031 }
6032
6033 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6034 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6035
6036 if (core_uses_pid_file != -1) {
6037 char core_uses_pid = 0;
6038 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6039 ::close(core_uses_pid_file);
6040
6041 if (core_uses_pid == '1') {
6042 jio_snprintf(buffer + written, bufferSize - written,
6043 ".%d", current_process_id());
6044 }
6045 }
6046 }
6047
6048 return strlen(buffer);
6049 }
6050
6051 bool os::start_debugging(char *buf, int buflen) {
6052 int len = (int)strlen(buf);
6053 char *p = &buf[len];
6054
6055 jio_snprintf(p, buflen-len,
6056 "\n\n"
6057 "Do you want to debug the problem?\n\n"
6058 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
6059 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
6060 "Otherwise, press RETURN to abort...",
6061 os::current_process_id(), os::current_process_id(),
6062 os::current_thread_id(), os::current_thread_id());
6063
6064 bool yes = os::message_box("Unexpected Error", buf);
6065
6066 if (yes) {
6067 // yes, user asked VM to launch debugger
6068 jio_snprintf(buf, sizeof(buf), "gdb /proc/%d/exe %d",
6069 os::current_process_id(), os::current_process_id());
6070
6071 os::fork_and_exec(buf);
6072 yes = false;
6073 }
6074 return yes;
6075 }
6076
6077 static inline struct timespec get_mtime(const char* filename) {
6078 struct stat st;
6079 int ret = os::stat(filename, &st);
6080 assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
6081 return st.st_mtim;
6082 }
6083
6084 int os::compare_file_modified_times(const char* file1, const char* file2) {
6085 struct timespec filetime1 = get_mtime(file1);
6086 struct timespec filetime2 = get_mtime(file2);
6087 int diff = filetime1.tv_sec - filetime2.tv_sec;
6088 if (diff == 0) {
6089 return filetime1.tv_nsec - filetime2.tv_nsec;
6090 }
6091 return diff;
6092 }
6093
6094 /////////////// Unit tests ///////////////
6095
6096 #ifndef PRODUCT
6097
6098 #define test_log(...) \
6099 do { \
6100 if (VerboseInternalVMTests) { \
6101 tty->print_cr(__VA_ARGS__); \
6102 tty->flush(); \
6103 } \
6104 } while (false)
6105
6106 class TestReserveMemorySpecial : AllStatic {
6107 public:
6108 static void small_page_write(void* addr, size_t size) {
6109 size_t page_size = os::vm_page_size();
6110
6111 char* end = (char*)addr + size;
6112 for (char* p = (char*)addr; p < end; p += page_size) {
6113 *p = 1;
6114 }
6115 }
6116
6117 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6118 if (!UseHugeTLBFS) {
6119 return;
6120 }
6121
6122 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6123
6124 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6125
6126 if (addr != NULL) {
6127 small_page_write(addr, size);
6128
6129 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6130 }
6131 }
6132
6133 static void test_reserve_memory_special_huge_tlbfs_only() {
6134 if (!UseHugeTLBFS) {
6135 return;
6136 }
6137
6138 size_t lp = os::large_page_size();
6139
6140 for (size_t size = lp; size <= lp * 10; size += lp) {
6141 test_reserve_memory_special_huge_tlbfs_only(size);
6142 }
6143 }
6144
6145 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6146 size_t lp = os::large_page_size();
6147 size_t ag = os::vm_allocation_granularity();
6148
6149 // sizes to test
6150 const size_t sizes[] = {
6151 lp, lp + ag, lp + lp / 2, lp * 2,
6152 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6153 lp * 10, lp * 10 + lp / 2
6154 };
6155 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6156
6157 // For each size/alignment combination, we test three scenarios:
6158 // 1) with req_addr == NULL
6159 // 2) with a non-null req_addr at which we expect to successfully allocate
6160 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6161 // expect the allocation to either fail or to ignore req_addr
6162
6163 // Pre-allocate two areas; they shall be as large as the largest allocation
6164 // and aligned to the largest alignment we will be testing.
6165 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6166 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6167 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6168 -1, 0);
6169 assert(mapping1 != MAP_FAILED, "should work");
6170
6171 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6172 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6173 -1, 0);
6174 assert(mapping2 != MAP_FAILED, "should work");
6175
6176 // Unmap the first mapping, but leave the second mapping intact: the first
6177 // mapping will serve as a value for a "good" req_addr (case 2). The second
6178 // mapping, still intact, as "bad" req_addr (case 3).
6179 ::munmap(mapping1, mapping_size);
6180
6181 // Case 1
6182 test_log("%s, req_addr NULL:", __FUNCTION__);
6183 test_log("size align result");
6184
6185 for (int i = 0; i < num_sizes; i++) {
6186 const size_t size = sizes[i];
6187 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6188 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6189 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6190 size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6191 if (p != NULL) {
6192 assert(is_ptr_aligned(p, alignment), "must be");
6193 small_page_write(p, size);
6194 os::Linux::release_memory_special_huge_tlbfs(p, size);
6195 }
6196 }
6197 }
6198
6199 // Case 2
6200 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6201 test_log("size align req_addr result");
6202
6203 for (int i = 0; i < num_sizes; i++) {
6204 const size_t size = sizes[i];
6205 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6206 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6207 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6208 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6209 size, alignment, p2i(req_addr), p2i(p),
6210 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6211 if (p != NULL) {
6212 assert(p == req_addr, "must be");
6213 small_page_write(p, size);
6214 os::Linux::release_memory_special_huge_tlbfs(p, size);
6215 }
6216 }
6217 }
6218
6219 // Case 3
6220 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6221 test_log("size align req_addr result");
6222
6223 for (int i = 0; i < num_sizes; i++) {
6224 const size_t size = sizes[i];
6225 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6226 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6227 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6228 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6229 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6230 // as the area around req_addr contains already existing mappings, the API should always
6231 // return NULL (as per contract, it cannot return another address)
6232 assert(p == NULL, "must be");
6233 }
6234 }
6235
6236 ::munmap(mapping2, mapping_size);
6237
6238 }
6239
6240 static void test_reserve_memory_special_huge_tlbfs() {
6241 if (!UseHugeTLBFS) {
6242 return;
6243 }
6244
6245 test_reserve_memory_special_huge_tlbfs_only();
6246 test_reserve_memory_special_huge_tlbfs_mixed();
6247 }
6248
6249 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6250 if (!UseSHM) {
6251 return;
6252 }
6253
6254 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6255
6256 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6257
6258 if (addr != NULL) {
6259 assert(is_ptr_aligned(addr, alignment), "Check");
6260 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6261
6262 small_page_write(addr, size);
6263
6264 os::Linux::release_memory_special_shm(addr, size);
6265 }
6266 }
6267
6268 static void test_reserve_memory_special_shm() {
6269 size_t lp = os::large_page_size();
6270 size_t ag = os::vm_allocation_granularity();
6271
6272 for (size_t size = ag; size < lp * 3; size += ag) {
6273 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6274 test_reserve_memory_special_shm(size, alignment);
6275 }
6276 }
6277 }
6278
6279 static void test() {
6280 test_reserve_memory_special_huge_tlbfs();
6281 test_reserve_memory_special_shm();
6282 }
6283 };
6284
6285 void TestReserveMemorySpecial_test() {
6286 TestReserveMemorySpecial::test();
6287 }
6288
6289 #endif