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[33];
1917 int bytes;
1918 buf[32] = '\0';
1919 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1920 st->print_raw(buf, bytes);
1921 }
1922
1923 ::close(fd);
1924
1925 return true;
1926 }
1927
1928 void os::print_dll_info(outputStream *st) {
1929 st->print_cr("Dynamic libraries:");
1930
1931 char fname[32];
1932 pid_t pid = os::Linux::gettid();
1933
1934 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1935
1936 if (!_print_ascii_file(fname, st)) {
1937 st->print("Can not get library information for pid = %d\n", pid);
1938 }
1939 }
1940
1941 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1942 FILE *procmapsFile = NULL;
1943
1944 // Open the procfs maps file for the current process
1945 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1946 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1947 char line[PATH_MAX + 100];
1948
1949 // Read line by line from 'file'
1950 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1951 u8 base, top, offset, inode;
1952 char permissions[5];
1953 char device[6];
1954 char name[PATH_MAX + 1];
1955
1956 // Parse fields from line
1957 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1958 &base, &top, permissions, &offset, device, &inode, name);
1959
1960 // Filter by device id '00:00' so that we only get file system mapped files.
1961 if (strcmp(device, "00:00") != 0) {
1962
1963 // Call callback with the fields of interest
1964 if(callback(name, (address)base, (address)top, param)) {
1965 // Oops abort, callback aborted
1966 fclose(procmapsFile);
1967 return 1;
1968 }
1969 }
1970 }
1971 fclose(procmapsFile);
1972 }
1973 return 0;
1974 }
1975
1976 void os::print_os_info_brief(outputStream* st) {
1977 os::Linux::print_distro_info(st);
1978
1979 os::Posix::print_uname_info(st);
1980
1981 os::Linux::print_libversion_info(st);
1982
1983 }
1984
1985 void os::print_os_info(outputStream* st) {
1986 st->print("OS:");
1987
1988 os::Linux::print_distro_info(st);
1989
1990 os::Posix::print_uname_info(st);
1991
1992 // Print warning if unsafe chroot environment detected
1993 if (unsafe_chroot_detected) {
1994 st->print("WARNING!! ");
1995 st->print_cr("%s", unstable_chroot_error);
1996 }
1997
1998 os::Linux::print_libversion_info(st);
1999
2000 os::Posix::print_rlimit_info(st);
2001
2002 os::Posix::print_load_average(st);
2003
2004 os::Linux::print_full_memory_info(st);
2005 }
2006
2007 // Try to identify popular distros.
2008 // Most Linux distributions have a /etc/XXX-release file, which contains
2009 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2010 // file that also contains the OS version string. Some have more than one
2011 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2012 // /etc/redhat-release.), so the order is important.
2013 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2014 // their own specific XXX-release file as well as a redhat-release file.
2015 // Because of this the XXX-release file needs to be searched for before the
2016 // redhat-release file.
2017 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2018 // search for redhat-release / SuSE-release needs to be before lsb-release.
2019 // Since the lsb-release file is the new standard it needs to be searched
2020 // before the older style release files.
2021 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2022 // next to last resort. The os-release file is a new standard that contains
2023 // distribution information and the system-release file seems to be an old
2024 // standard that has been replaced by the lsb-release and os-release files.
2025 // Searching for the debian_version file is the last resort. It contains
2026 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2027 // "Debian " is printed before the contents of the debian_version file.
2028
2029 const char* distro_files[] = {
2030 "/etc/oracle-release",
2031 "/etc/mandriva-release",
2032 "/etc/mandrake-release",
2033 "/etc/sun-release",
2034 "/etc/redhat-release",
2035 "/etc/SuSE-release",
2036 "/etc/lsb-release",
2037 "/etc/turbolinux-release",
2038 "/etc/gentoo-release",
2039 "/etc/ltib-release",
2040 "/etc/angstrom-version",
2041 "/etc/system-release",
2042 "/etc/os-release",
2043 NULL };
2044
2045 void os::Linux::print_distro_info(outputStream* st) {
2046 for (int i = 0;; i++) {
2047 const char* file = distro_files[i];
2048 if (file == NULL) {
2049 break; // done
2050 }
2051 // If file prints, we found it.
2052 if (_print_ascii_file(file, st)) {
2053 return;
2054 }
2055 }
2056
2057 if (file_exists("/etc/debian_version")) {
2058 st->print("Debian ");
2059 _print_ascii_file("/etc/debian_version", st);
2060 } else {
2061 st->print("Linux");
2062 }
2063 st->cr();
2064 }
2065
2066 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2067 char buf[256];
2068 while (fgets(buf, sizeof(buf), fp)) {
2069 // Edit out extra stuff in expected format
2070 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2071 char* ptr = strstr(buf, "\""); // the name is in quotes
2072 if (ptr != NULL) {
2073 ptr++; // go beyond first quote
2074 char* nl = strchr(ptr, '\"');
2075 if (nl != NULL) *nl = '\0';
2076 strncpy(distro, ptr, length);
2077 } else {
2078 ptr = strstr(buf, "=");
2079 ptr++; // go beyond equals then
2080 char* nl = strchr(ptr, '\n');
2081 if (nl != NULL) *nl = '\0';
2082 strncpy(distro, ptr, length);
2083 }
2084 return;
2085 } else if (get_first_line) {
2086 char* nl = strchr(buf, '\n');
2087 if (nl != NULL) *nl = '\0';
2088 strncpy(distro, buf, length);
2089 return;
2090 }
2091 }
2092 // print last line and close
2093 char* nl = strchr(buf, '\n');
2094 if (nl != NULL) *nl = '\0';
2095 strncpy(distro, buf, length);
2096 }
2097
2098 static void parse_os_info(char* distro, size_t length, const char* file) {
2099 FILE* fp = fopen(file, "r");
2100 if (fp != NULL) {
2101 // if suse format, print out first line
2102 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2103 parse_os_info_helper(fp, distro, length, get_first_line);
2104 fclose(fp);
2105 }
2106 }
2107
2108 void os::get_summary_os_info(char* buf, size_t buflen) {
2109 for (int i = 0;; i++) {
2110 const char* file = distro_files[i];
2111 if (file == NULL) {
2112 break; // ran out of distro_files
2113 }
2114 if (file_exists(file)) {
2115 parse_os_info(buf, buflen, file);
2116 return;
2117 }
2118 }
2119 // special case for debian
2120 if (file_exists("/etc/debian_version")) {
2121 strncpy(buf, "Debian ", buflen);
2122 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2123 } else {
2124 strncpy(buf, "Linux", buflen);
2125 }
2126 }
2127
2128 void os::Linux::print_libversion_info(outputStream* st) {
2129 // libc, pthread
2130 st->print("libc:");
2131 st->print("%s ", os::Linux::glibc_version());
2132 st->print("%s ", os::Linux::libpthread_version());
2133 st->cr();
2134 }
2135
2136 void os::Linux::print_full_memory_info(outputStream* st) {
2137 st->print("\n/proc/meminfo:\n");
2138 _print_ascii_file("/proc/meminfo", st);
2139 st->cr();
2140 }
2141
2142 void os::print_memory_info(outputStream* st) {
2143
2144 st->print("Memory:");
2145 st->print(" %dk page", os::vm_page_size()>>10);
2146
2147 // values in struct sysinfo are "unsigned long"
2148 struct sysinfo si;
2149 sysinfo(&si);
2150
2151 st->print(", physical " UINT64_FORMAT "k",
2152 os::physical_memory() >> 10);
2153 st->print("(" UINT64_FORMAT "k free)",
2154 os::available_memory() >> 10);
2155 st->print(", swap " UINT64_FORMAT "k",
2156 ((jlong)si.totalswap * si.mem_unit) >> 10);
2157 st->print("(" UINT64_FORMAT "k free)",
2158 ((jlong)si.freeswap * si.mem_unit) >> 10);
2159 st->cr();
2160 }
2161
2162 // Print the first "model name" line and the first "flags" line
2163 // that we find and nothing more. We assume "model name" comes
2164 // before "flags" so if we find a second "model name", then the
2165 // "flags" field is considered missing.
2166 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2167 #if defined(IA32) || defined(AMD64)
2168 // Other platforms have less repetitive cpuinfo files
2169 FILE *fp = fopen("/proc/cpuinfo", "r");
2170 if (fp) {
2171 while (!feof(fp)) {
2172 if (fgets(buf, buflen, fp)) {
2173 // Assume model name comes before flags
2174 bool model_name_printed = false;
2175 if (strstr(buf, "model name") != NULL) {
2176 if (!model_name_printed) {
2177 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2178 st->print_raw(buf);
2179 model_name_printed = true;
2180 } else {
2181 // model name printed but not flags? Odd, just return
2182 fclose(fp);
2183 return true;
2184 }
2185 }
2186 // print the flags line too
2187 if (strstr(buf, "flags") != NULL) {
2188 st->print_raw(buf);
2189 fclose(fp);
2190 return true;
2191 }
2192 }
2193 }
2194 fclose(fp);
2195 }
2196 #endif // x86 platforms
2197 return false;
2198 }
2199
2200 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2201 // Only print the model name if the platform provides this as a summary
2202 if (!print_model_name_and_flags(st, buf, buflen)) {
2203 st->print("\n/proc/cpuinfo:\n");
2204 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2205 st->print_cr(" <Not Available>");
2206 }
2207 }
2208 }
2209
2210 #if defined(AMD64) || defined(IA32) || defined(X32)
2211 const char* search_string = "model name";
2212 #elif defined(SPARC)
2213 const char* search_string = "cpu";
2214 #elif defined(PPC64)
2215 const char* search_string = "cpu";
2216 #else
2217 const char* search_string = "Processor";
2218 #endif
2219
2220 // Parses the cpuinfo file for string representing the model name.
2221 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2222 FILE* fp = fopen("/proc/cpuinfo", "r");
2223 if (fp != NULL) {
2224 while (!feof(fp)) {
2225 char buf[256];
2226 if (fgets(buf, sizeof(buf), fp)) {
2227 char* start = strstr(buf, search_string);
2228 if (start != NULL) {
2229 char *ptr = start + strlen(search_string);
2230 char *end = buf + strlen(buf);
2231 while (ptr != end) {
2232 // skip whitespace and colon for the rest of the name.
2233 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2234 break;
2235 }
2236 ptr++;
2237 }
2238 if (ptr != end) {
2239 // reasonable string, get rid of newline and keep the rest
2240 char* nl = strchr(buf, '\n');
2241 if (nl != NULL) *nl = '\0';
2242 strncpy(cpuinfo, ptr, length);
2243 fclose(fp);
2244 return;
2245 }
2246 }
2247 }
2248 }
2249 fclose(fp);
2250 }
2251 // cpuinfo not found or parsing failed, just print generic string. The entire
2252 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2253 #if defined(AMD64)
2254 strncpy(cpuinfo, "x86_64", length);
2255 #elif defined(IA32)
2256 strncpy(cpuinfo, "x86_32", length);
2257 #elif defined(IA64)
2258 strncpy(cpuinfo, "IA64", length);
2259 #elif defined(SPARC)
2260 strncpy(cpuinfo, "sparcv9", length);
2261 #elif defined(AARCH64)
2262 strncpy(cpuinfo, "AArch64", length);
2263 #elif defined(ARM)
2264 strncpy(cpuinfo, "ARM", length);
2265 #elif defined(PPC)
2266 strncpy(cpuinfo, "PPC64", length);
2267 #elif defined(ZERO_LIBARCH)
2268 strncpy(cpuinfo, ZERO_LIBARCH, length);
2269 #else
2270 strncpy(cpuinfo, "unknown", length);
2271 #endif
2272 }
2273
2274 static void print_signal_handler(outputStream* st, int sig,
2275 char* buf, size_t buflen);
2276
2277 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2278 st->print_cr("Signal Handlers:");
2279 print_signal_handler(st, SIGSEGV, buf, buflen);
2280 print_signal_handler(st, SIGBUS , buf, buflen);
2281 print_signal_handler(st, SIGFPE , buf, buflen);
2282 print_signal_handler(st, SIGPIPE, buf, buflen);
2283 print_signal_handler(st, SIGXFSZ, buf, buflen);
2284 print_signal_handler(st, SIGILL , buf, buflen);
2285 print_signal_handler(st, SR_signum, buf, buflen);
2286 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2287 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2288 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2289 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2290 #if defined(PPC64)
2291 print_signal_handler(st, SIGTRAP, buf, buflen);
2292 #endif
2293 }
2294
2295 static char saved_jvm_path[MAXPATHLEN] = {0};
2296
2297 // Find the full path to the current module, libjvm.so
2298 void os::jvm_path(char *buf, jint buflen) {
2299 // Error checking.
2300 if (buflen < MAXPATHLEN) {
2301 assert(false, "must use a large-enough buffer");
2302 buf[0] = '\0';
2303 return;
2304 }
2305 // Lazy resolve the path to current module.
2306 if (saved_jvm_path[0] != 0) {
2307 strcpy(buf, saved_jvm_path);
2308 return;
2309 }
2310
2311 char dli_fname[MAXPATHLEN];
2312 bool ret = dll_address_to_library_name(
2313 CAST_FROM_FN_PTR(address, os::jvm_path),
2314 dli_fname, sizeof(dli_fname), NULL);
2315 assert(ret, "cannot locate libjvm");
2316 char *rp = NULL;
2317 if (ret && dli_fname[0] != '\0') {
2318 rp = realpath(dli_fname, buf);
2319 }
2320 if (rp == NULL) {
2321 return;
2322 }
2323
2324 if (Arguments::sun_java_launcher_is_altjvm()) {
2325 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2326 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2327 // If "/jre/lib/" appears at the right place in the string, then
2328 // assume we are installed in a JDK and we're done. Otherwise, check
2329 // for a JAVA_HOME environment variable and fix up the path so it
2330 // looks like libjvm.so is installed there (append a fake suffix
2331 // hotspot/libjvm.so).
2332 const char *p = buf + strlen(buf) - 1;
2333 for (int count = 0; p > buf && count < 5; ++count) {
2334 for (--p; p > buf && *p != '/'; --p)
2335 /* empty */ ;
2336 }
2337
2338 if (strncmp(p, "/jre/lib/", 9) != 0) {
2339 // Look for JAVA_HOME in the environment.
2340 char* java_home_var = ::getenv("JAVA_HOME");
2341 if (java_home_var != NULL && java_home_var[0] != 0) {
2342 char* jrelib_p;
2343 int len;
2344
2345 // Check the current module name "libjvm.so".
2346 p = strrchr(buf, '/');
2347 if (p == NULL) {
2348 return;
2349 }
2350 assert(strstr(p, "/libjvm") == p, "invalid library name");
2351
2352 rp = realpath(java_home_var, buf);
2353 if (rp == NULL) {
2354 return;
2355 }
2356
2357 // determine if this is a legacy image or modules image
2358 // modules image doesn't have "jre" subdirectory
2359 len = strlen(buf);
2360 assert(len < buflen, "Ran out of buffer room");
2361 jrelib_p = buf + len;
2362 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2363 if (0 != access(buf, F_OK)) {
2364 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2365 }
2366
2367 if (0 == access(buf, F_OK)) {
2368 // Use current module name "libjvm.so"
2369 len = strlen(buf);
2370 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2371 } else {
2372 // Go back to path of .so
2373 rp = realpath(dli_fname, buf);
2374 if (rp == NULL) {
2375 return;
2376 }
2377 }
2378 }
2379 }
2380 }
2381
2382 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2383 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2384 }
2385
2386 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2387 // no prefix required, not even "_"
2388 }
2389
2390 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2391 // no suffix required
2392 }
2393
2394 ////////////////////////////////////////////////////////////////////////////////
2395 // sun.misc.Signal support
2396
2397 static volatile jint sigint_count = 0;
2398
2399 static void UserHandler(int sig, void *siginfo, void *context) {
2400 // 4511530 - sem_post is serialized and handled by the manager thread. When
2401 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2402 // don't want to flood the manager thread with sem_post requests.
2403 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2404 return;
2405 }
2406
2407 // Ctrl-C is pressed during error reporting, likely because the error
2408 // handler fails to abort. Let VM die immediately.
2409 if (sig == SIGINT && is_error_reported()) {
2410 os::die();
2411 }
2412
2413 os::signal_notify(sig);
2414 }
2415
2416 void* os::user_handler() {
2417 return CAST_FROM_FN_PTR(void*, UserHandler);
2418 }
2419
2420 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2421 struct timespec ts;
2422 // Semaphore's are always associated with CLOCK_REALTIME
2423 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2424 // see unpackTime for discussion on overflow checking
2425 if (sec >= MAX_SECS) {
2426 ts.tv_sec += MAX_SECS;
2427 ts.tv_nsec = 0;
2428 } else {
2429 ts.tv_sec += sec;
2430 ts.tv_nsec += nsec;
2431 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2432 ts.tv_nsec -= NANOSECS_PER_SEC;
2433 ++ts.tv_sec; // note: this must be <= max_secs
2434 }
2435 }
2436
2437 return ts;
2438 }
2439
2440 extern "C" {
2441 typedef void (*sa_handler_t)(int);
2442 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2443 }
2444
2445 void* os::signal(int signal_number, void* handler) {
2446 struct sigaction sigAct, oldSigAct;
2447
2448 sigfillset(&(sigAct.sa_mask));
2449 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2450 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2451
2452 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2453 // -1 means registration failed
2454 return (void *)-1;
2455 }
2456
2457 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2458 }
2459
2460 void os::signal_raise(int signal_number) {
2461 ::raise(signal_number);
2462 }
2463
2464 // The following code is moved from os.cpp for making this
2465 // code platform specific, which it is by its very nature.
2466
2467 // Will be modified when max signal is changed to be dynamic
2468 int os::sigexitnum_pd() {
2469 return NSIG;
2470 }
2471
2472 // a counter for each possible signal value
2473 static volatile jint pending_signals[NSIG+1] = { 0 };
2474
2475 // Linux(POSIX) specific hand shaking semaphore.
2476 static sem_t sig_sem;
2477 static PosixSemaphore sr_semaphore;
2478
2479 void os::signal_init_pd() {
2480 // Initialize signal structures
2481 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2482
2483 // Initialize signal semaphore
2484 ::sem_init(&sig_sem, 0, 0);
2485 }
2486
2487 void os::signal_notify(int sig) {
2488 Atomic::inc(&pending_signals[sig]);
2489 ::sem_post(&sig_sem);
2490 }
2491
2492 static int check_pending_signals(bool wait) {
2493 Atomic::store(0, &sigint_count);
2494 for (;;) {
2495 for (int i = 0; i < NSIG + 1; i++) {
2496 jint n = pending_signals[i];
2497 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2498 return i;
2499 }
2500 }
2501 if (!wait) {
2502 return -1;
2503 }
2504 JavaThread *thread = JavaThread::current();
2505 ThreadBlockInVM tbivm(thread);
2506
2507 bool threadIsSuspended;
2508 do {
2509 thread->set_suspend_equivalent();
2510 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2511 ::sem_wait(&sig_sem);
2512
2513 // were we externally suspended while we were waiting?
2514 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2515 if (threadIsSuspended) {
2516 // The semaphore has been incremented, but while we were waiting
2517 // another thread suspended us. We don't want to continue running
2518 // while suspended because that would surprise the thread that
2519 // suspended us.
2520 ::sem_post(&sig_sem);
2521
2522 thread->java_suspend_self();
2523 }
2524 } while (threadIsSuspended);
2525 }
2526 }
2527
2528 int os::signal_lookup() {
2529 return check_pending_signals(false);
2530 }
2531
2532 int os::signal_wait() {
2533 return check_pending_signals(true);
2534 }
2535
2536 ////////////////////////////////////////////////////////////////////////////////
2537 // Virtual Memory
2538
2539 int os::vm_page_size() {
2540 // Seems redundant as all get out
2541 assert(os::Linux::page_size() != -1, "must call os::init");
2542 return os::Linux::page_size();
2543 }
2544
2545 // Solaris allocates memory by pages.
2546 int os::vm_allocation_granularity() {
2547 assert(os::Linux::page_size() != -1, "must call os::init");
2548 return os::Linux::page_size();
2549 }
2550
2551 // Rationale behind this function:
2552 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2553 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2554 // samples for JITted code. Here we create private executable mapping over the code cache
2555 // and then we can use standard (well, almost, as mapping can change) way to provide
2556 // info for the reporting script by storing timestamp and location of symbol
2557 void linux_wrap_code(char* base, size_t size) {
2558 static volatile jint cnt = 0;
2559
2560 if (!UseOprofile) {
2561 return;
2562 }
2563
2564 char buf[PATH_MAX+1];
2565 int num = Atomic::add(1, &cnt);
2566
2567 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2568 os::get_temp_directory(), os::current_process_id(), num);
2569 unlink(buf);
2570
2571 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2572
2573 if (fd != -1) {
2574 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2575 if (rv != (off_t)-1) {
2576 if (::write(fd, "", 1) == 1) {
2577 mmap(base, size,
2578 PROT_READ|PROT_WRITE|PROT_EXEC,
2579 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2580 }
2581 }
2582 ::close(fd);
2583 unlink(buf);
2584 }
2585 }
2586
2587 static bool recoverable_mmap_error(int err) {
2588 // See if the error is one we can let the caller handle. This
2589 // list of errno values comes from JBS-6843484. I can't find a
2590 // Linux man page that documents this specific set of errno
2591 // values so while this list currently matches Solaris, it may
2592 // change as we gain experience with this failure mode.
2593 switch (err) {
2594 case EBADF:
2595 case EINVAL:
2596 case ENOTSUP:
2597 // let the caller deal with these errors
2598 return true;
2599
2600 default:
2601 // Any remaining errors on this OS can cause our reserved mapping
2602 // to be lost. That can cause confusion where different data
2603 // structures think they have the same memory mapped. The worst
2604 // scenario is if both the VM and a library think they have the
2605 // same memory mapped.
2606 return false;
2607 }
2608 }
2609
2610 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2611 int err) {
2612 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2613 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2614 os::strerror(err), err);
2615 }
2616
2617 static void warn_fail_commit_memory(char* addr, size_t size,
2618 size_t alignment_hint, bool exec,
2619 int err) {
2620 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2621 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2622 alignment_hint, exec, os::strerror(err), err);
2623 }
2624
2625 // NOTE: Linux kernel does not really reserve the pages for us.
2626 // All it does is to check if there are enough free pages
2627 // left at the time of mmap(). This could be a potential
2628 // problem.
2629 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2630 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2631 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2632 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2633 if (res != (uintptr_t) MAP_FAILED) {
2634 if (UseNUMAInterleaving) {
2635 numa_make_global(addr, size);
2636 }
2637 return 0;
2638 }
2639
2640 int err = errno; // save errno from mmap() call above
2641
2642 if (!recoverable_mmap_error(err)) {
2643 warn_fail_commit_memory(addr, size, exec, err);
2644 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2645 }
2646
2647 return err;
2648 }
2649
2650 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2651 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2652 }
2653
2654 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2655 const char* mesg) {
2656 assert(mesg != NULL, "mesg must be specified");
2657 int err = os::Linux::commit_memory_impl(addr, size, exec);
2658 if (err != 0) {
2659 // the caller wants all commit errors to exit with the specified mesg:
2660 warn_fail_commit_memory(addr, size, exec, err);
2661 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2662 }
2663 }
2664
2665 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2666 #ifndef MAP_HUGETLB
2667 #define MAP_HUGETLB 0x40000
2668 #endif
2669
2670 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2671 #ifndef MADV_HUGEPAGE
2672 #define MADV_HUGEPAGE 14
2673 #endif
2674
2675 int os::Linux::commit_memory_impl(char* addr, size_t size,
2676 size_t alignment_hint, bool exec) {
2677 int err = os::Linux::commit_memory_impl(addr, size, exec);
2678 if (err == 0) {
2679 realign_memory(addr, size, alignment_hint);
2680 }
2681 return err;
2682 }
2683
2684 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2685 bool exec) {
2686 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2687 }
2688
2689 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2690 size_t alignment_hint, bool exec,
2691 const char* mesg) {
2692 assert(mesg != NULL, "mesg must be specified");
2693 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2694 if (err != 0) {
2695 // the caller wants all commit errors to exit with the specified mesg:
2696 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2697 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2698 }
2699 }
2700
2701 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2702 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2703 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2704 // be supported or the memory may already be backed by huge pages.
2705 ::madvise(addr, bytes, MADV_HUGEPAGE);
2706 }
2707 }
2708
2709 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2710 // This method works by doing an mmap over an existing mmaping and effectively discarding
2711 // the existing pages. However it won't work for SHM-based large pages that cannot be
2712 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2713 // small pages on top of the SHM segment. This method always works for small pages, so we
2714 // allow that in any case.
2715 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2716 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2717 }
2718 }
2719
2720 void os::numa_make_global(char *addr, size_t bytes) {
2721 Linux::numa_interleave_memory(addr, bytes);
2722 }
2723
2724 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2725 // bind policy to MPOL_PREFERRED for the current thread.
2726 #define USE_MPOL_PREFERRED 0
2727
2728 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2729 // To make NUMA and large pages more robust when both enabled, we need to ease
2730 // the requirements on where the memory should be allocated. MPOL_BIND is the
2731 // default policy and it will force memory to be allocated on the specified
2732 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2733 // the specified node, but will not force it. Using this policy will prevent
2734 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2735 // free large pages.
2736 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2737 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2738 }
2739
2740 bool os::numa_topology_changed() { return false; }
2741
2742 size_t os::numa_get_groups_num() {
2743 int max_node = Linux::numa_max_node();
2744 return max_node > 0 ? max_node + 1 : 1;
2745 }
2746
2747 int os::numa_get_group_id() {
2748 int cpu_id = Linux::sched_getcpu();
2749 if (cpu_id != -1) {
2750 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2751 if (lgrp_id != -1) {
2752 return lgrp_id;
2753 }
2754 }
2755 return 0;
2756 }
2757
2758 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2759 for (size_t i = 0; i < size; i++) {
2760 ids[i] = i;
2761 }
2762 return size;
2763 }
2764
2765 bool os::get_page_info(char *start, page_info* info) {
2766 return false;
2767 }
2768
2769 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2770 page_info* page_found) {
2771 return end;
2772 }
2773
2774
2775 int os::Linux::sched_getcpu_syscall(void) {
2776 unsigned int cpu = 0;
2777 int retval = -1;
2778
2779 #if defined(IA32)
2780 #ifndef SYS_getcpu
2781 #define SYS_getcpu 318
2782 #endif
2783 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2784 #elif defined(AMD64)
2785 // Unfortunately we have to bring all these macros here from vsyscall.h
2786 // to be able to compile on old linuxes.
2787 #define __NR_vgetcpu 2
2788 #define VSYSCALL_START (-10UL << 20)
2789 #define VSYSCALL_SIZE 1024
2790 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2791 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2792 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2793 retval = vgetcpu(&cpu, NULL, NULL);
2794 #endif
2795
2796 return (retval == -1) ? retval : cpu;
2797 }
2798
2799 // Something to do with the numa-aware allocator needs these symbols
2800 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2801 extern "C" JNIEXPORT void numa_error(char *where) { }
2802
2803
2804 // If we are running with libnuma version > 2, then we should
2805 // be trying to use symbols with versions 1.1
2806 // If we are running with earlier version, which did not have symbol versions,
2807 // we should use the base version.
2808 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2809 void *f = dlvsym(handle, name, "libnuma_1.1");
2810 if (f == NULL) {
2811 f = dlsym(handle, name);
2812 }
2813 return f;
2814 }
2815
2816 bool os::Linux::libnuma_init() {
2817 // sched_getcpu() should be in libc.
2818 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2819 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2820
2821 // If it's not, try a direct syscall.
2822 if (sched_getcpu() == -1) {
2823 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2824 (void*)&sched_getcpu_syscall));
2825 }
2826
2827 if (sched_getcpu() != -1) { // Does it work?
2828 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2829 if (handle != NULL) {
2830 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2831 libnuma_dlsym(handle, "numa_node_to_cpus")));
2832 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2833 libnuma_dlsym(handle, "numa_max_node")));
2834 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2835 libnuma_dlsym(handle, "numa_available")));
2836 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2837 libnuma_dlsym(handle, "numa_tonode_memory")));
2838 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2839 libnuma_dlsym(handle, "numa_interleave_memory")));
2840 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2841 libnuma_dlsym(handle, "numa_set_bind_policy")));
2842
2843
2844 if (numa_available() != -1) {
2845 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2846 // Create a cpu -> node mapping
2847 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2848 rebuild_cpu_to_node_map();
2849 return true;
2850 }
2851 }
2852 }
2853 return false;
2854 }
2855
2856 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2857 // The table is later used in get_node_by_cpu().
2858 void os::Linux::rebuild_cpu_to_node_map() {
2859 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2860 // in libnuma (possible values are starting from 16,
2861 // and continuing up with every other power of 2, but less
2862 // than the maximum number of CPUs supported by kernel), and
2863 // is a subject to change (in libnuma version 2 the requirements
2864 // are more reasonable) we'll just hardcode the number they use
2865 // in the library.
2866 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2867
2868 size_t cpu_num = os::active_processor_count();
2869 size_t cpu_map_size = NCPUS / BitsPerCLong;
2870 size_t cpu_map_valid_size =
2871 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2872
2873 cpu_to_node()->clear();
2874 cpu_to_node()->at_grow(cpu_num - 1);
2875 size_t node_num = numa_get_groups_num();
2876
2877 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2878 for (size_t i = 0; i < node_num; i++) {
2879 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2880 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2881 if (cpu_map[j] != 0) {
2882 for (size_t k = 0; k < BitsPerCLong; k++) {
2883 if (cpu_map[j] & (1UL << k)) {
2884 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2885 }
2886 }
2887 }
2888 }
2889 }
2890 }
2891 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2892 }
2893
2894 int os::Linux::get_node_by_cpu(int cpu_id) {
2895 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2896 return cpu_to_node()->at(cpu_id);
2897 }
2898 return -1;
2899 }
2900
2901 GrowableArray<int>* os::Linux::_cpu_to_node;
2902 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2903 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2904 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2905 os::Linux::numa_available_func_t os::Linux::_numa_available;
2906 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2907 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2908 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2909 unsigned long* os::Linux::_numa_all_nodes;
2910
2911 bool os::pd_uncommit_memory(char* addr, size_t size) {
2912 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2913 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2914 return res != (uintptr_t) MAP_FAILED;
2915 }
2916
2917 static address get_stack_commited_bottom(address bottom, size_t size) {
2918 address nbot = bottom;
2919 address ntop = bottom + size;
2920
2921 size_t page_sz = os::vm_page_size();
2922 unsigned pages = size / page_sz;
2923
2924 unsigned char vec[1];
2925 unsigned imin = 1, imax = pages + 1, imid;
2926 int mincore_return_value = 0;
2927
2928 assert(imin <= imax, "Unexpected page size");
2929
2930 while (imin < imax) {
2931 imid = (imax + imin) / 2;
2932 nbot = ntop - (imid * page_sz);
2933
2934 // Use a trick with mincore to check whether the page is mapped or not.
2935 // mincore sets vec to 1 if page resides in memory and to 0 if page
2936 // is swapped output but if page we are asking for is unmapped
2937 // it returns -1,ENOMEM
2938 mincore_return_value = mincore(nbot, page_sz, vec);
2939
2940 if (mincore_return_value == -1) {
2941 // Page is not mapped go up
2942 // to find first mapped page
2943 if (errno != EAGAIN) {
2944 assert(errno == ENOMEM, "Unexpected mincore errno");
2945 imax = imid;
2946 }
2947 } else {
2948 // Page is mapped go down
2949 // to find first not mapped page
2950 imin = imid + 1;
2951 }
2952 }
2953
2954 nbot = nbot + page_sz;
2955
2956 // Adjust stack bottom one page up if last checked page is not mapped
2957 if (mincore_return_value == -1) {
2958 nbot = nbot + page_sz;
2959 }
2960
2961 return nbot;
2962 }
2963
2964
2965 // Linux uses a growable mapping for the stack, and if the mapping for
2966 // the stack guard pages is not removed when we detach a thread the
2967 // stack cannot grow beyond the pages where the stack guard was
2968 // mapped. If at some point later in the process the stack expands to
2969 // that point, the Linux kernel cannot expand the stack any further
2970 // because the guard pages are in the way, and a segfault occurs.
2971 //
2972 // However, it's essential not to split the stack region by unmapping
2973 // a region (leaving a hole) that's already part of the stack mapping,
2974 // so if the stack mapping has already grown beyond the guard pages at
2975 // the time we create them, we have to truncate the stack mapping.
2976 // So, we need to know the extent of the stack mapping when
2977 // create_stack_guard_pages() is called.
2978
2979 // We only need this for stacks that are growable: at the time of
2980 // writing thread stacks don't use growable mappings (i.e. those
2981 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2982 // only applies to the main thread.
2983
2984 // If the (growable) stack mapping already extends beyond the point
2985 // where we're going to put our guard pages, truncate the mapping at
2986 // that point by munmap()ping it. This ensures that when we later
2987 // munmap() the guard pages we don't leave a hole in the stack
2988 // mapping. This only affects the main/initial thread
2989
2990 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2991 if (os::Linux::is_initial_thread()) {
2992 // As we manually grow stack up to bottom inside create_attached_thread(),
2993 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
2994 // we don't need to do anything special.
2995 // Check it first, before calling heavy function.
2996 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
2997 unsigned char vec[1];
2998
2999 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3000 // Fallback to slow path on all errors, including EAGAIN
3001 stack_extent = (uintptr_t) get_stack_commited_bottom(
3002 os::Linux::initial_thread_stack_bottom(),
3003 (size_t)addr - stack_extent);
3004 }
3005
3006 if (stack_extent < (uintptr_t)addr) {
3007 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3008 }
3009 }
3010
3011 return os::commit_memory(addr, size, !ExecMem);
3012 }
3013
3014 // If this is a growable mapping, remove the guard pages entirely by
3015 // munmap()ping them. If not, just call uncommit_memory(). This only
3016 // affects the main/initial thread, but guard against future OS changes
3017 // It's safe to always unmap guard pages for initial thread because we
3018 // always place it right after end of the mapped region
3019
3020 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3021 uintptr_t stack_extent, stack_base;
3022
3023 if (os::Linux::is_initial_thread()) {
3024 return ::munmap(addr, size) == 0;
3025 }
3026
3027 return os::uncommit_memory(addr, size);
3028 }
3029
3030 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3031 // at 'requested_addr'. If there are existing memory mappings at the same
3032 // location, however, they will be overwritten. If 'fixed' is false,
3033 // 'requested_addr' is only treated as a hint, the return value may or
3034 // may not start from the requested address. Unlike Linux mmap(), this
3035 // function returns NULL to indicate failure.
3036 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3037 char * addr;
3038 int flags;
3039
3040 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3041 if (fixed) {
3042 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3043 flags |= MAP_FIXED;
3044 }
3045
3046 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3047 // touch an uncommitted page. Otherwise, the read/write might
3048 // succeed if we have enough swap space to back the physical page.
3049 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3050 flags, -1, 0);
3051
3052 return addr == MAP_FAILED ? NULL : addr;
3053 }
3054
3055 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3056 // (req_addr != NULL) or with a given alignment.
3057 // - bytes shall be a multiple of alignment.
3058 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3059 // - alignment sets the alignment at which memory shall be allocated.
3060 // It must be a multiple of allocation granularity.
3061 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3062 // req_addr or NULL.
3063 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3064
3065 size_t extra_size = bytes;
3066 if (req_addr == NULL && alignment > 0) {
3067 extra_size += alignment;
3068 }
3069
3070 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3071 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3072 -1, 0);
3073 if (start == MAP_FAILED) {
3074 start = NULL;
3075 } else {
3076 if (req_addr != NULL) {
3077 if (start != req_addr) {
3078 ::munmap(start, extra_size);
3079 start = NULL;
3080 }
3081 } else {
3082 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3083 char* const end_aligned = start_aligned + bytes;
3084 char* const end = start + extra_size;
3085 if (start_aligned > start) {
3086 ::munmap(start, start_aligned - start);
3087 }
3088 if (end_aligned < end) {
3089 ::munmap(end_aligned, end - end_aligned);
3090 }
3091 start = start_aligned;
3092 }
3093 }
3094 return start;
3095 }
3096
3097 static int anon_munmap(char * addr, size_t size) {
3098 return ::munmap(addr, size) == 0;
3099 }
3100
3101 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3102 size_t alignment_hint) {
3103 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3104 }
3105
3106 bool os::pd_release_memory(char* addr, size_t size) {
3107 return anon_munmap(addr, size);
3108 }
3109
3110 static bool linux_mprotect(char* addr, size_t size, int prot) {
3111 // Linux wants the mprotect address argument to be page aligned.
3112 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3113
3114 // According to SUSv3, mprotect() should only be used with mappings
3115 // established by mmap(), and mmap() always maps whole pages. Unaligned
3116 // 'addr' likely indicates problem in the VM (e.g. trying to change
3117 // protection of malloc'ed or statically allocated memory). Check the
3118 // caller if you hit this assert.
3119 assert(addr == bottom, "sanity check");
3120
3121 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3122 return ::mprotect(bottom, size, prot) == 0;
3123 }
3124
3125 // Set protections specified
3126 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3127 bool is_committed) {
3128 unsigned int p = 0;
3129 switch (prot) {
3130 case MEM_PROT_NONE: p = PROT_NONE; break;
3131 case MEM_PROT_READ: p = PROT_READ; break;
3132 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3133 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3134 default:
3135 ShouldNotReachHere();
3136 }
3137 // is_committed is unused.
3138 return linux_mprotect(addr, bytes, p);
3139 }
3140
3141 bool os::guard_memory(char* addr, size_t size) {
3142 return linux_mprotect(addr, size, PROT_NONE);
3143 }
3144
3145 bool os::unguard_memory(char* addr, size_t size) {
3146 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3147 }
3148
3149 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3150 size_t page_size) {
3151 bool result = false;
3152 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3153 MAP_ANONYMOUS|MAP_PRIVATE,
3154 -1, 0);
3155 if (p != MAP_FAILED) {
3156 void *aligned_p = align_ptr_up(p, page_size);
3157
3158 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3159
3160 munmap(p, page_size * 2);
3161 }
3162
3163 if (warn && !result) {
3164 warning("TransparentHugePages is not supported by the operating system.");
3165 }
3166
3167 return result;
3168 }
3169
3170 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3171 bool result = false;
3172 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3173 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3174 -1, 0);
3175
3176 if (p != MAP_FAILED) {
3177 // We don't know if this really is a huge page or not.
3178 FILE *fp = fopen("/proc/self/maps", "r");
3179 if (fp) {
3180 while (!feof(fp)) {
3181 char chars[257];
3182 long x = 0;
3183 if (fgets(chars, sizeof(chars), fp)) {
3184 if (sscanf(chars, "%lx-%*x", &x) == 1
3185 && x == (long)p) {
3186 if (strstr (chars, "hugepage")) {
3187 result = true;
3188 break;
3189 }
3190 }
3191 }
3192 }
3193 fclose(fp);
3194 }
3195 munmap(p, page_size);
3196 }
3197
3198 if (warn && !result) {
3199 warning("HugeTLBFS is not supported by the operating system.");
3200 }
3201
3202 return result;
3203 }
3204
3205 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3206 //
3207 // From the coredump_filter documentation:
3208 //
3209 // - (bit 0) anonymous private memory
3210 // - (bit 1) anonymous shared memory
3211 // - (bit 2) file-backed private memory
3212 // - (bit 3) file-backed shared memory
3213 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3214 // effective only if the bit 2 is cleared)
3215 // - (bit 5) hugetlb private memory
3216 // - (bit 6) hugetlb shared memory
3217 //
3218 static void set_coredump_filter(void) {
3219 FILE *f;
3220 long cdm;
3221
3222 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3223 return;
3224 }
3225
3226 if (fscanf(f, "%lx", &cdm) != 1) {
3227 fclose(f);
3228 return;
3229 }
3230
3231 rewind(f);
3232
3233 if ((cdm & LARGEPAGES_BIT) == 0) {
3234 cdm |= LARGEPAGES_BIT;
3235 fprintf(f, "%#lx", cdm);
3236 }
3237
3238 fclose(f);
3239 }
3240
3241 // Large page support
3242
3243 static size_t _large_page_size = 0;
3244
3245 size_t os::Linux::find_large_page_size() {
3246 size_t large_page_size = 0;
3247
3248 // large_page_size on Linux is used to round up heap size. x86 uses either
3249 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3250 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3251 // page as large as 256M.
3252 //
3253 // Here we try to figure out page size by parsing /proc/meminfo and looking
3254 // for a line with the following format:
3255 // Hugepagesize: 2048 kB
3256 //
3257 // If we can't determine the value (e.g. /proc is not mounted, or the text
3258 // format has been changed), we'll use the largest page size supported by
3259 // the processor.
3260
3261 #ifndef ZERO
3262 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3263 ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3264 #endif // ZERO
3265
3266 FILE *fp = fopen("/proc/meminfo", "r");
3267 if (fp) {
3268 while (!feof(fp)) {
3269 int x = 0;
3270 char buf[16];
3271 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3272 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3273 large_page_size = x * K;
3274 break;
3275 }
3276 } else {
3277 // skip to next line
3278 for (;;) {
3279 int ch = fgetc(fp);
3280 if (ch == EOF || ch == (int)'\n') break;
3281 }
3282 }
3283 }
3284 fclose(fp);
3285 }
3286
3287 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3288 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3289 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3290 proper_unit_for_byte_size(large_page_size));
3291 }
3292
3293 return large_page_size;
3294 }
3295
3296 size_t os::Linux::setup_large_page_size() {
3297 _large_page_size = Linux::find_large_page_size();
3298 const size_t default_page_size = (size_t)Linux::page_size();
3299 if (_large_page_size > default_page_size) {
3300 _page_sizes[0] = _large_page_size;
3301 _page_sizes[1] = default_page_size;
3302 _page_sizes[2] = 0;
3303 }
3304
3305 return _large_page_size;
3306 }
3307
3308 bool os::Linux::setup_large_page_type(size_t page_size) {
3309 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3310 FLAG_IS_DEFAULT(UseSHM) &&
3311 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3312
3313 // The type of large pages has not been specified by the user.
3314
3315 // Try UseHugeTLBFS and then UseSHM.
3316 UseHugeTLBFS = UseSHM = true;
3317
3318 // Don't try UseTransparentHugePages since there are known
3319 // performance issues with it turned on. This might change in the future.
3320 UseTransparentHugePages = false;
3321 }
3322
3323 if (UseTransparentHugePages) {
3324 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3325 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3326 UseHugeTLBFS = false;
3327 UseSHM = false;
3328 return true;
3329 }
3330 UseTransparentHugePages = false;
3331 }
3332
3333 if (UseHugeTLBFS) {
3334 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3335 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3336 UseSHM = false;
3337 return true;
3338 }
3339 UseHugeTLBFS = false;
3340 }
3341
3342 return UseSHM;
3343 }
3344
3345 void os::large_page_init() {
3346 if (!UseLargePages &&
3347 !UseTransparentHugePages &&
3348 !UseHugeTLBFS &&
3349 !UseSHM) {
3350 // Not using large pages.
3351 return;
3352 }
3353
3354 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3355 // The user explicitly turned off large pages.
3356 // Ignore the rest of the large pages flags.
3357 UseTransparentHugePages = false;
3358 UseHugeTLBFS = false;
3359 UseSHM = false;
3360 return;
3361 }
3362
3363 size_t large_page_size = Linux::setup_large_page_size();
3364 UseLargePages = Linux::setup_large_page_type(large_page_size);
3365
3366 set_coredump_filter();
3367 }
3368
3369 #ifndef SHM_HUGETLB
3370 #define SHM_HUGETLB 04000
3371 #endif
3372
3373 #define shm_warning_format(format, ...) \
3374 do { \
3375 if (UseLargePages && \
3376 (!FLAG_IS_DEFAULT(UseLargePages) || \
3377 !FLAG_IS_DEFAULT(UseSHM) || \
3378 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3379 warning(format, __VA_ARGS__); \
3380 } \
3381 } while (0)
3382
3383 #define shm_warning(str) shm_warning_format("%s", str)
3384
3385 #define shm_warning_with_errno(str) \
3386 do { \
3387 int err = errno; \
3388 shm_warning_format(str " (error = %d)", err); \
3389 } while (0)
3390
3391 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3392 assert(is_size_aligned(bytes, alignment), "Must be divisible by the alignment");
3393
3394 if (!is_size_aligned(alignment, SHMLBA)) {
3395 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3396 return NULL;
3397 }
3398
3399 // To ensure that we get 'alignment' aligned memory from shmat,
3400 // we pre-reserve aligned virtual memory and then attach to that.
3401
3402 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3403 if (pre_reserved_addr == NULL) {
3404 // Couldn't pre-reserve aligned memory.
3405 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3406 return NULL;
3407 }
3408
3409 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3410 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3411
3412 if ((intptr_t)addr == -1) {
3413 int err = errno;
3414 shm_warning_with_errno("Failed to attach shared memory.");
3415
3416 assert(err != EACCES, "Unexpected error");
3417 assert(err != EIDRM, "Unexpected error");
3418 assert(err != EINVAL, "Unexpected error");
3419
3420 // Since we don't know if the kernel unmapped the pre-reserved memory area
3421 // we can't unmap it, since that would potentially unmap memory that was
3422 // mapped from other threads.
3423 return NULL;
3424 }
3425
3426 return addr;
3427 }
3428
3429 static char* shmat_at_address(int shmid, char* req_addr) {
3430 if (!is_ptr_aligned(req_addr, SHMLBA)) {
3431 assert(false, "Requested address needs to be SHMLBA aligned");
3432 return NULL;
3433 }
3434
3435 char* addr = (char*)shmat(shmid, req_addr, 0);
3436
3437 if ((intptr_t)addr == -1) {
3438 shm_warning_with_errno("Failed to attach shared memory.");
3439 return NULL;
3440 }
3441
3442 return addr;
3443 }
3444
3445 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3446 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3447 if (req_addr != NULL) {
3448 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3449 assert(is_ptr_aligned(req_addr, alignment), "Must be divisible by given alignment");
3450 return shmat_at_address(shmid, req_addr);
3451 }
3452
3453 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3454 // return large page size aligned memory addresses when req_addr == NULL.
3455 // However, if the alignment is larger than the large page size, we have
3456 // to manually ensure that the memory returned is 'alignment' aligned.
3457 if (alignment > os::large_page_size()) {
3458 assert(is_size_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3459 return shmat_with_alignment(shmid, bytes, alignment);
3460 } else {
3461 return shmat_at_address(shmid, NULL);
3462 }
3463 }
3464
3465 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3466 char* req_addr, bool exec) {
3467 // "exec" is passed in but not used. Creating the shared image for
3468 // the code cache doesn't have an SHM_X executable permission to check.
3469 assert(UseLargePages && UseSHM, "only for SHM large pages");
3470 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3471 assert(is_ptr_aligned(req_addr, alignment), "Unaligned address");
3472
3473 if (!is_size_aligned(bytes, os::large_page_size())) {
3474 return NULL; // Fallback to small pages.
3475 }
3476
3477 // Create a large shared memory region to attach to based on size.
3478 // Currently, size is the total size of the heap.
3479 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3480 if (shmid == -1) {
3481 // Possible reasons for shmget failure:
3482 // 1. shmmax is too small for Java heap.
3483 // > check shmmax value: cat /proc/sys/kernel/shmmax
3484 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3485 // 2. not enough large page memory.
3486 // > check available large pages: cat /proc/meminfo
3487 // > increase amount of large pages:
3488 // echo new_value > /proc/sys/vm/nr_hugepages
3489 // Note 1: different Linux may use different name for this property,
3490 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3491 // Note 2: it's possible there's enough physical memory available but
3492 // they are so fragmented after a long run that they can't
3493 // coalesce into large pages. Try to reserve large pages when
3494 // the system is still "fresh".
3495 shm_warning_with_errno("Failed to reserve shared memory.");
3496 return NULL;
3497 }
3498
3499 // Attach to the region.
3500 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3501
3502 // Remove shmid. If shmat() is successful, the actual shared memory segment
3503 // will be deleted when it's detached by shmdt() or when the process
3504 // terminates. If shmat() is not successful this will remove the shared
3505 // segment immediately.
3506 shmctl(shmid, IPC_RMID, NULL);
3507
3508 return addr;
3509 }
3510
3511 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3512 int error) {
3513 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3514
3515 bool warn_on_failure = UseLargePages &&
3516 (!FLAG_IS_DEFAULT(UseLargePages) ||
3517 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3518 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3519
3520 if (warn_on_failure) {
3521 char msg[128];
3522 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3523 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3524 warning("%s", msg);
3525 }
3526 }
3527
3528 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3529 char* req_addr,
3530 bool exec) {
3531 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3532 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3533 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3534
3535 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3536 char* addr = (char*)::mmap(req_addr, bytes, prot,
3537 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3538 -1, 0);
3539
3540 if (addr == MAP_FAILED) {
3541 warn_on_large_pages_failure(req_addr, bytes, errno);
3542 return NULL;
3543 }
3544
3545 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3546
3547 return addr;
3548 }
3549
3550 // Reserve memory using mmap(MAP_HUGETLB).
3551 // - bytes shall be a multiple of alignment.
3552 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3553 // - alignment sets the alignment at which memory shall be allocated.
3554 // It must be a multiple of allocation granularity.
3555 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3556 // req_addr or NULL.
3557 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3558 size_t alignment,
3559 char* req_addr,
3560 bool exec) {
3561 size_t large_page_size = os::large_page_size();
3562 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3563
3564 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3565 assert(is_size_aligned(bytes, alignment), "Must be");
3566
3567 // First reserve - but not commit - the address range in small pages.
3568 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3569
3570 if (start == NULL) {
3571 return NULL;
3572 }
3573
3574 assert(is_ptr_aligned(start, alignment), "Must be");
3575
3576 char* end = start + bytes;
3577
3578 // Find the regions of the allocated chunk that can be promoted to large pages.
3579 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3580 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3581
3582 size_t lp_bytes = lp_end - lp_start;
3583
3584 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3585
3586 if (lp_bytes == 0) {
3587 // The mapped region doesn't even span the start and the end of a large page.
3588 // Fall back to allocate a non-special area.
3589 ::munmap(start, end - start);
3590 return NULL;
3591 }
3592
3593 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3594
3595 void* result;
3596
3597 // Commit small-paged leading area.
3598 if (start != lp_start) {
3599 result = ::mmap(start, lp_start - start, prot,
3600 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3601 -1, 0);
3602 if (result == MAP_FAILED) {
3603 ::munmap(lp_start, end - lp_start);
3604 return NULL;
3605 }
3606 }
3607
3608 // Commit large-paged area.
3609 result = ::mmap(lp_start, lp_bytes, prot,
3610 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3611 -1, 0);
3612 if (result == MAP_FAILED) {
3613 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3614 // If the mmap above fails, the large pages region will be unmapped and we
3615 // have regions before and after with small pages. Release these regions.
3616 //
3617 // | mapped | unmapped | mapped |
3618 // ^ ^ ^ ^
3619 // start lp_start lp_end end
3620 //
3621 ::munmap(start, lp_start - start);
3622 ::munmap(lp_end, end - lp_end);
3623 return NULL;
3624 }
3625
3626 // Commit small-paged trailing area.
3627 if (lp_end != end) {
3628 result = ::mmap(lp_end, end - lp_end, prot,
3629 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3630 -1, 0);
3631 if (result == MAP_FAILED) {
3632 ::munmap(start, lp_end - start);
3633 return NULL;
3634 }
3635 }
3636
3637 return start;
3638 }
3639
3640 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3641 size_t alignment,
3642 char* req_addr,
3643 bool exec) {
3644 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3645 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3646 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3647 assert(is_power_of_2(os::large_page_size()), "Must be");
3648 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3649
3650 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3651 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3652 } else {
3653 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3654 }
3655 }
3656
3657 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3658 char* req_addr, bool exec) {
3659 assert(UseLargePages, "only for large pages");
3660
3661 char* addr;
3662 if (UseSHM) {
3663 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3664 } else {
3665 assert(UseHugeTLBFS, "must be");
3666 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3667 }
3668
3669 if (addr != NULL) {
3670 if (UseNUMAInterleaving) {
3671 numa_make_global(addr, bytes);
3672 }
3673
3674 // The memory is committed
3675 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3676 }
3677
3678 return addr;
3679 }
3680
3681 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3682 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3683 return shmdt(base) == 0;
3684 }
3685
3686 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3687 return pd_release_memory(base, bytes);
3688 }
3689
3690 bool os::release_memory_special(char* base, size_t bytes) {
3691 bool res;
3692 if (MemTracker::tracking_level() > NMT_minimal) {
3693 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3694 res = os::Linux::release_memory_special_impl(base, bytes);
3695 if (res) {
3696 tkr.record((address)base, bytes);
3697 }
3698
3699 } else {
3700 res = os::Linux::release_memory_special_impl(base, bytes);
3701 }
3702 return res;
3703 }
3704
3705 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3706 assert(UseLargePages, "only for large pages");
3707 bool res;
3708
3709 if (UseSHM) {
3710 res = os::Linux::release_memory_special_shm(base, bytes);
3711 } else {
3712 assert(UseHugeTLBFS, "must be");
3713 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3714 }
3715 return res;
3716 }
3717
3718 size_t os::large_page_size() {
3719 return _large_page_size;
3720 }
3721
3722 // With SysV SHM the entire memory region must be allocated as shared
3723 // memory.
3724 // HugeTLBFS allows application to commit large page memory on demand.
3725 // However, when committing memory with HugeTLBFS fails, the region
3726 // that was supposed to be committed will lose the old reservation
3727 // and allow other threads to steal that memory region. Because of this
3728 // behavior we can't commit HugeTLBFS memory.
3729 bool os::can_commit_large_page_memory() {
3730 return UseTransparentHugePages;
3731 }
3732
3733 bool os::can_execute_large_page_memory() {
3734 return UseTransparentHugePages || UseHugeTLBFS;
3735 }
3736
3737 // Reserve memory at an arbitrary address, only if that area is
3738 // available (and not reserved for something else).
3739
3740 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3741 const int max_tries = 10;
3742 char* base[max_tries];
3743 size_t size[max_tries];
3744 const size_t gap = 0x000000;
3745
3746 // Assert only that the size is a multiple of the page size, since
3747 // that's all that mmap requires, and since that's all we really know
3748 // about at this low abstraction level. If we need higher alignment,
3749 // we can either pass an alignment to this method or verify alignment
3750 // in one of the methods further up the call chain. See bug 5044738.
3751 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3752
3753 // Repeatedly allocate blocks until the block is allocated at the
3754 // right spot.
3755
3756 // Linux mmap allows caller to pass an address as hint; give it a try first,
3757 // if kernel honors the hint then we can return immediately.
3758 char * addr = anon_mmap(requested_addr, bytes, false);
3759 if (addr == requested_addr) {
3760 return requested_addr;
3761 }
3762
3763 if (addr != NULL) {
3764 // mmap() is successful but it fails to reserve at the requested address
3765 anon_munmap(addr, bytes);
3766 }
3767
3768 int i;
3769 for (i = 0; i < max_tries; ++i) {
3770 base[i] = reserve_memory(bytes);
3771
3772 if (base[i] != NULL) {
3773 // Is this the block we wanted?
3774 if (base[i] == requested_addr) {
3775 size[i] = bytes;
3776 break;
3777 }
3778
3779 // Does this overlap the block we wanted? Give back the overlapped
3780 // parts and try again.
3781
3782 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3783 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3784 unmap_memory(base[i], top_overlap);
3785 base[i] += top_overlap;
3786 size[i] = bytes - top_overlap;
3787 } else {
3788 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3789 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3790 unmap_memory(requested_addr, bottom_overlap);
3791 size[i] = bytes - bottom_overlap;
3792 } else {
3793 size[i] = bytes;
3794 }
3795 }
3796 }
3797 }
3798
3799 // Give back the unused reserved pieces.
3800
3801 for (int j = 0; j < i; ++j) {
3802 if (base[j] != NULL) {
3803 unmap_memory(base[j], size[j]);
3804 }
3805 }
3806
3807 if (i < max_tries) {
3808 return requested_addr;
3809 } else {
3810 return NULL;
3811 }
3812 }
3813
3814 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3815 return ::read(fd, buf, nBytes);
3816 }
3817
3818 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3819 return ::pread(fd, buf, nBytes, offset);
3820 }
3821
3822 // Short sleep, direct OS call.
3823 //
3824 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3825 // sched_yield(2) will actually give up the CPU:
3826 //
3827 // * Alone on this pariticular CPU, keeps running.
3828 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3829 // (pre 2.6.39).
3830 //
3831 // So calling this with 0 is an alternative.
3832 //
3833 void os::naked_short_sleep(jlong ms) {
3834 struct timespec req;
3835
3836 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3837 req.tv_sec = 0;
3838 if (ms > 0) {
3839 req.tv_nsec = (ms % 1000) * 1000000;
3840 } else {
3841 req.tv_nsec = 1;
3842 }
3843
3844 nanosleep(&req, NULL);
3845
3846 return;
3847 }
3848
3849 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3850 void os::infinite_sleep() {
3851 while (true) { // sleep forever ...
3852 ::sleep(100); // ... 100 seconds at a time
3853 }
3854 }
3855
3856 // Used to convert frequent JVM_Yield() to nops
3857 bool os::dont_yield() {
3858 return DontYieldALot;
3859 }
3860
3861 void os::naked_yield() {
3862 sched_yield();
3863 }
3864
3865 ////////////////////////////////////////////////////////////////////////////////
3866 // thread priority support
3867
3868 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3869 // only supports dynamic priority, static priority must be zero. For real-time
3870 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3871 // However, for large multi-threaded applications, SCHED_RR is not only slower
3872 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3873 // of 5 runs - Sep 2005).
3874 //
3875 // The following code actually changes the niceness of kernel-thread/LWP. It
3876 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3877 // not the entire user process, and user level threads are 1:1 mapped to kernel
3878 // threads. It has always been the case, but could change in the future. For
3879 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3880 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3881
3882 int os::java_to_os_priority[CriticalPriority + 1] = {
3883 19, // 0 Entry should never be used
3884
3885 4, // 1 MinPriority
3886 3, // 2
3887 2, // 3
3888
3889 1, // 4
3890 0, // 5 NormPriority
3891 -1, // 6
3892
3893 -2, // 7
3894 -3, // 8
3895 -4, // 9 NearMaxPriority
3896
3897 -5, // 10 MaxPriority
3898
3899 -5 // 11 CriticalPriority
3900 };
3901
3902 static int prio_init() {
3903 if (ThreadPriorityPolicy == 1) {
3904 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3905 // if effective uid is not root. Perhaps, a more elegant way of doing
3906 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3907 if (geteuid() != 0) {
3908 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3909 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3910 }
3911 ThreadPriorityPolicy = 0;
3912 }
3913 }
3914 if (UseCriticalJavaThreadPriority) {
3915 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3916 }
3917 return 0;
3918 }
3919
3920 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3921 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3922
3923 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3924 return (ret == 0) ? OS_OK : OS_ERR;
3925 }
3926
3927 OSReturn os::get_native_priority(const Thread* const thread,
3928 int *priority_ptr) {
3929 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3930 *priority_ptr = java_to_os_priority[NormPriority];
3931 return OS_OK;
3932 }
3933
3934 errno = 0;
3935 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3936 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3937 }
3938
3939 // Hint to the underlying OS that a task switch would not be good.
3940 // Void return because it's a hint and can fail.
3941 void os::hint_no_preempt() {}
3942
3943 ////////////////////////////////////////////////////////////////////////////////
3944 // suspend/resume support
3945
3946 // the low-level signal-based suspend/resume support is a remnant from the
3947 // old VM-suspension that used to be for java-suspension, safepoints etc,
3948 // within hotspot. Now there is a single use-case for this:
3949 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3950 // that runs in the watcher thread.
3951 // The remaining code is greatly simplified from the more general suspension
3952 // code that used to be used.
3953 //
3954 // The protocol is quite simple:
3955 // - suspend:
3956 // - sends a signal to the target thread
3957 // - polls the suspend state of the osthread using a yield loop
3958 // - target thread signal handler (SR_handler) sets suspend state
3959 // and blocks in sigsuspend until continued
3960 // - resume:
3961 // - sets target osthread state to continue
3962 // - sends signal to end the sigsuspend loop in the SR_handler
3963 //
3964 // Note that the SR_lock plays no role in this suspend/resume protocol.
3965
3966 static void resume_clear_context(OSThread *osthread) {
3967 osthread->set_ucontext(NULL);
3968 osthread->set_siginfo(NULL);
3969 }
3970
3971 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3972 ucontext_t* context) {
3973 osthread->set_ucontext(context);
3974 osthread->set_siginfo(siginfo);
3975 }
3976
3977 // Handler function invoked when a thread's execution is suspended or
3978 // resumed. We have to be careful that only async-safe functions are
3979 // called here (Note: most pthread functions are not async safe and
3980 // should be avoided.)
3981 //
3982 // Note: sigwait() is a more natural fit than sigsuspend() from an
3983 // interface point of view, but sigwait() prevents the signal hander
3984 // from being run. libpthread would get very confused by not having
3985 // its signal handlers run and prevents sigwait()'s use with the
3986 // mutex granting granting signal.
3987 //
3988 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3989 //
3990 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3991 // Save and restore errno to avoid confusing native code with EINTR
3992 // after sigsuspend.
3993 int old_errno = errno;
3994
3995 Thread* thread = Thread::current_or_null_safe();
3996 assert(thread != NULL, "Missing current thread in SR_handler");
3997 OSThread* osthread = thread->osthread();
3998 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3999
4000 os::SuspendResume::State current = osthread->sr.state();
4001 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4002 suspend_save_context(osthread, siginfo, context);
4003
4004 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4005 os::SuspendResume::State state = osthread->sr.suspended();
4006 if (state == os::SuspendResume::SR_SUSPENDED) {
4007 sigset_t suspend_set; // signals for sigsuspend()
4008 sigemptyset(&suspend_set);
4009 // get current set of blocked signals and unblock resume signal
4010 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4011 sigdelset(&suspend_set, SR_signum);
4012
4013 sr_semaphore.signal();
4014 // wait here until we are resumed
4015 while (1) {
4016 sigsuspend(&suspend_set);
4017
4018 os::SuspendResume::State result = osthread->sr.running();
4019 if (result == os::SuspendResume::SR_RUNNING) {
4020 sr_semaphore.signal();
4021 break;
4022 }
4023 }
4024
4025 } else if (state == os::SuspendResume::SR_RUNNING) {
4026 // request was cancelled, continue
4027 } else {
4028 ShouldNotReachHere();
4029 }
4030
4031 resume_clear_context(osthread);
4032 } else if (current == os::SuspendResume::SR_RUNNING) {
4033 // request was cancelled, continue
4034 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4035 // ignore
4036 } else {
4037 // ignore
4038 }
4039
4040 errno = old_errno;
4041 }
4042
4043 static int SR_initialize() {
4044 struct sigaction act;
4045 char *s;
4046
4047 // Get signal number to use for suspend/resume
4048 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4049 int sig = ::strtol(s, 0, 10);
4050 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4051 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4052 SR_signum = sig;
4053 } else {
4054 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4055 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4056 }
4057 }
4058
4059 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4060 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4061
4062 sigemptyset(&SR_sigset);
4063 sigaddset(&SR_sigset, SR_signum);
4064
4065 // Set up signal handler for suspend/resume
4066 act.sa_flags = SA_RESTART|SA_SIGINFO;
4067 act.sa_handler = (void (*)(int)) SR_handler;
4068
4069 // SR_signum is blocked by default.
4070 // 4528190 - We also need to block pthread restart signal (32 on all
4071 // supported Linux platforms). Note that LinuxThreads need to block
4072 // this signal for all threads to work properly. So we don't have
4073 // to use hard-coded signal number when setting up the mask.
4074 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4075
4076 if (sigaction(SR_signum, &act, 0) == -1) {
4077 return -1;
4078 }
4079
4080 // Save signal flag
4081 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4082 return 0;
4083 }
4084
4085 static int sr_notify(OSThread* osthread) {
4086 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4087 assert_status(status == 0, status, "pthread_kill");
4088 return status;
4089 }
4090
4091 // "Randomly" selected value for how long we want to spin
4092 // before bailing out on suspending a thread, also how often
4093 // we send a signal to a thread we want to resume
4094 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4095 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4096
4097 // returns true on success and false on error - really an error is fatal
4098 // but this seems the normal response to library errors
4099 static bool do_suspend(OSThread* osthread) {
4100 assert(osthread->sr.is_running(), "thread should be running");
4101 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4102
4103 // mark as suspended and send signal
4104 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4105 // failed to switch, state wasn't running?
4106 ShouldNotReachHere();
4107 return false;
4108 }
4109
4110 if (sr_notify(osthread) != 0) {
4111 ShouldNotReachHere();
4112 }
4113
4114 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4115 while (true) {
4116 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4117 break;
4118 } else {
4119 // timeout
4120 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4121 if (cancelled == os::SuspendResume::SR_RUNNING) {
4122 return false;
4123 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4124 // make sure that we consume the signal on the semaphore as well
4125 sr_semaphore.wait();
4126 break;
4127 } else {
4128 ShouldNotReachHere();
4129 return false;
4130 }
4131 }
4132 }
4133
4134 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4135 return true;
4136 }
4137
4138 static void do_resume(OSThread* osthread) {
4139 assert(osthread->sr.is_suspended(), "thread should be suspended");
4140 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4141
4142 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4143 // failed to switch to WAKEUP_REQUEST
4144 ShouldNotReachHere();
4145 return;
4146 }
4147
4148 while (true) {
4149 if (sr_notify(osthread) == 0) {
4150 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4151 if (osthread->sr.is_running()) {
4152 return;
4153 }
4154 }
4155 } else {
4156 ShouldNotReachHere();
4157 }
4158 }
4159
4160 guarantee(osthread->sr.is_running(), "Must be running!");
4161 }
4162
4163 ///////////////////////////////////////////////////////////////////////////////////
4164 // signal handling (except suspend/resume)
4165
4166 // This routine may be used by user applications as a "hook" to catch signals.
4167 // The user-defined signal handler must pass unrecognized signals to this
4168 // routine, and if it returns true (non-zero), then the signal handler must
4169 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4170 // routine will never retun false (zero), but instead will execute a VM panic
4171 // routine kill the process.
4172 //
4173 // If this routine returns false, it is OK to call it again. This allows
4174 // the user-defined signal handler to perform checks either before or after
4175 // the VM performs its own checks. Naturally, the user code would be making
4176 // a serious error if it tried to handle an exception (such as a null check
4177 // or breakpoint) that the VM was generating for its own correct operation.
4178 //
4179 // This routine may recognize any of the following kinds of signals:
4180 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4181 // It should be consulted by handlers for any of those signals.
4182 //
4183 // The caller of this routine must pass in the three arguments supplied
4184 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4185 // field of the structure passed to sigaction(). This routine assumes that
4186 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4187 //
4188 // Note that the VM will print warnings if it detects conflicting signal
4189 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4190 //
4191 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4192 siginfo_t* siginfo,
4193 void* ucontext,
4194 int abort_if_unrecognized);
4195
4196 void signalHandler(int sig, siginfo_t* info, void* uc) {
4197 assert(info != NULL && uc != NULL, "it must be old kernel");
4198 int orig_errno = errno; // Preserve errno value over signal handler.
4199 JVM_handle_linux_signal(sig, info, uc, true);
4200 errno = orig_errno;
4201 }
4202
4203
4204 // This boolean allows users to forward their own non-matching signals
4205 // to JVM_handle_linux_signal, harmlessly.
4206 bool os::Linux::signal_handlers_are_installed = false;
4207
4208 // For signal-chaining
4209 struct sigaction sigact[NSIG];
4210 uint64_t sigs = 0;
4211 #if (64 < NSIG-1)
4212 #error "Not all signals can be encoded in sigs. Adapt its type!"
4213 #endif
4214 bool os::Linux::libjsig_is_loaded = false;
4215 typedef struct sigaction *(*get_signal_t)(int);
4216 get_signal_t os::Linux::get_signal_action = NULL;
4217
4218 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4219 struct sigaction *actp = NULL;
4220
4221 if (libjsig_is_loaded) {
4222 // Retrieve the old signal handler from libjsig
4223 actp = (*get_signal_action)(sig);
4224 }
4225 if (actp == NULL) {
4226 // Retrieve the preinstalled signal handler from jvm
4227 actp = get_preinstalled_handler(sig);
4228 }
4229
4230 return actp;
4231 }
4232
4233 static bool call_chained_handler(struct sigaction *actp, int sig,
4234 siginfo_t *siginfo, void *context) {
4235 // Call the old signal handler
4236 if (actp->sa_handler == SIG_DFL) {
4237 // It's more reasonable to let jvm treat it as an unexpected exception
4238 // instead of taking the default action.
4239 return false;
4240 } else if (actp->sa_handler != SIG_IGN) {
4241 if ((actp->sa_flags & SA_NODEFER) == 0) {
4242 // automaticlly block the signal
4243 sigaddset(&(actp->sa_mask), sig);
4244 }
4245
4246 sa_handler_t hand = NULL;
4247 sa_sigaction_t sa = NULL;
4248 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4249 // retrieve the chained handler
4250 if (siginfo_flag_set) {
4251 sa = actp->sa_sigaction;
4252 } else {
4253 hand = actp->sa_handler;
4254 }
4255
4256 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4257 actp->sa_handler = SIG_DFL;
4258 }
4259
4260 // try to honor the signal mask
4261 sigset_t oset;
4262 sigemptyset(&oset);
4263 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4264
4265 // call into the chained handler
4266 if (siginfo_flag_set) {
4267 (*sa)(sig, siginfo, context);
4268 } else {
4269 (*hand)(sig);
4270 }
4271
4272 // restore the signal mask
4273 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4274 }
4275 // Tell jvm's signal handler the signal is taken care of.
4276 return true;
4277 }
4278
4279 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4280 bool chained = false;
4281 // signal-chaining
4282 if (UseSignalChaining) {
4283 struct sigaction *actp = get_chained_signal_action(sig);
4284 if (actp != NULL) {
4285 chained = call_chained_handler(actp, sig, siginfo, context);
4286 }
4287 }
4288 return chained;
4289 }
4290
4291 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4292 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4293 return &sigact[sig];
4294 }
4295 return NULL;
4296 }
4297
4298 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4299 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4300 sigact[sig] = oldAct;
4301 sigs |= (uint64_t)1 << (sig-1);
4302 }
4303
4304 // for diagnostic
4305 int sigflags[NSIG];
4306
4307 int os::Linux::get_our_sigflags(int sig) {
4308 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4309 return sigflags[sig];
4310 }
4311
4312 void os::Linux::set_our_sigflags(int sig, int flags) {
4313 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4314 if (sig > 0 && sig < NSIG) {
4315 sigflags[sig] = flags;
4316 }
4317 }
4318
4319 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4320 // Check for overwrite.
4321 struct sigaction oldAct;
4322 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4323
4324 void* oldhand = oldAct.sa_sigaction
4325 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4326 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4327 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4328 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4329 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4330 if (AllowUserSignalHandlers || !set_installed) {
4331 // Do not overwrite; user takes responsibility to forward to us.
4332 return;
4333 } else if (UseSignalChaining) {
4334 // save the old handler in jvm
4335 save_preinstalled_handler(sig, oldAct);
4336 // libjsig also interposes the sigaction() call below and saves the
4337 // old sigaction on it own.
4338 } else {
4339 fatal("Encountered unexpected pre-existing sigaction handler "
4340 "%#lx for signal %d.", (long)oldhand, sig);
4341 }
4342 }
4343
4344 struct sigaction sigAct;
4345 sigfillset(&(sigAct.sa_mask));
4346 sigAct.sa_handler = SIG_DFL;
4347 if (!set_installed) {
4348 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4349 } else {
4350 sigAct.sa_sigaction = signalHandler;
4351 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4352 }
4353 // Save flags, which are set by ours
4354 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4355 sigflags[sig] = sigAct.sa_flags;
4356
4357 int ret = sigaction(sig, &sigAct, &oldAct);
4358 assert(ret == 0, "check");
4359
4360 void* oldhand2 = oldAct.sa_sigaction
4361 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4362 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4363 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4364 }
4365
4366 // install signal handlers for signals that HotSpot needs to
4367 // handle in order to support Java-level exception handling.
4368
4369 void os::Linux::install_signal_handlers() {
4370 if (!signal_handlers_are_installed) {
4371 signal_handlers_are_installed = true;
4372
4373 // signal-chaining
4374 typedef void (*signal_setting_t)();
4375 signal_setting_t begin_signal_setting = NULL;
4376 signal_setting_t end_signal_setting = NULL;
4377 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4378 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4379 if (begin_signal_setting != NULL) {
4380 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4381 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4382 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4383 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4384 libjsig_is_loaded = true;
4385 assert(UseSignalChaining, "should enable signal-chaining");
4386 }
4387 if (libjsig_is_loaded) {
4388 // Tell libjsig jvm is setting signal handlers
4389 (*begin_signal_setting)();
4390 }
4391
4392 set_signal_handler(SIGSEGV, true);
4393 set_signal_handler(SIGPIPE, true);
4394 set_signal_handler(SIGBUS, true);
4395 set_signal_handler(SIGILL, true);
4396 set_signal_handler(SIGFPE, true);
4397 #if defined(PPC64)
4398 set_signal_handler(SIGTRAP, true);
4399 #endif
4400 set_signal_handler(SIGXFSZ, true);
4401
4402 if (libjsig_is_loaded) {
4403 // Tell libjsig jvm finishes setting signal handlers
4404 (*end_signal_setting)();
4405 }
4406
4407 // We don't activate signal checker if libjsig is in place, we trust ourselves
4408 // and if UserSignalHandler is installed all bets are off.
4409 // Log that signal checking is off only if -verbose:jni is specified.
4410 if (CheckJNICalls) {
4411 if (libjsig_is_loaded) {
4412 if (PrintJNIResolving) {
4413 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4414 }
4415 check_signals = false;
4416 }
4417 if (AllowUserSignalHandlers) {
4418 if (PrintJNIResolving) {
4419 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4420 }
4421 check_signals = false;
4422 }
4423 }
4424 }
4425 }
4426
4427 // This is the fastest way to get thread cpu time on Linux.
4428 // Returns cpu time (user+sys) for any thread, not only for current.
4429 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4430 // It might work on 2.6.10+ with a special kernel/glibc patch.
4431 // For reference, please, see IEEE Std 1003.1-2004:
4432 // http://www.unix.org/single_unix_specification
4433
4434 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4435 struct timespec tp;
4436 int rc = os::Linux::clock_gettime(clockid, &tp);
4437 assert(rc == 0, "clock_gettime is expected to return 0 code");
4438
4439 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4440 }
4441
4442 void os::Linux::initialize_os_info() {
4443 assert(_os_version == 0, "OS info already initialized");
4444
4445 struct utsname _uname;
4446
4447 uint32_t major;
4448 uint32_t minor;
4449 uint32_t fix;
4450
4451 int rc;
4452
4453 // Kernel version is unknown if
4454 // verification below fails.
4455 _os_version = 0x01000000;
4456
4457 rc = uname(&_uname);
4458 if (rc != -1) {
4459
4460 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4461 if (rc == 3) {
4462
4463 if (major < 256 && minor < 256 && fix < 256) {
4464 // Kernel version format is as expected,
4465 // set it overriding unknown state.
4466 _os_version = (major << 16) |
4467 (minor << 8 ) |
4468 (fix << 0 ) ;
4469 }
4470 }
4471 }
4472 }
4473
4474 uint32_t os::Linux::os_version() {
4475 assert(_os_version != 0, "not initialized");
4476 return _os_version & 0x00FFFFFF;
4477 }
4478
4479 bool os::Linux::os_version_is_known() {
4480 assert(_os_version != 0, "not initialized");
4481 return _os_version & 0x01000000 ? false : true;
4482 }
4483
4484 /////
4485 // glibc on Linux platform uses non-documented flag
4486 // to indicate, that some special sort of signal
4487 // trampoline is used.
4488 // We will never set this flag, and we should
4489 // ignore this flag in our diagnostic
4490 #ifdef SIGNIFICANT_SIGNAL_MASK
4491 #undef SIGNIFICANT_SIGNAL_MASK
4492 #endif
4493 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4494
4495 static const char* get_signal_handler_name(address handler,
4496 char* buf, int buflen) {
4497 int offset = 0;
4498 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4499 if (found) {
4500 // skip directory names
4501 const char *p1, *p2;
4502 p1 = buf;
4503 size_t len = strlen(os::file_separator());
4504 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4505 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4506 } else {
4507 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4508 }
4509 return buf;
4510 }
4511
4512 static void print_signal_handler(outputStream* st, int sig,
4513 char* buf, size_t buflen) {
4514 struct sigaction sa;
4515
4516 sigaction(sig, NULL, &sa);
4517
4518 // See comment for SIGNIFICANT_SIGNAL_MASK define
4519 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4520
4521 st->print("%s: ", os::exception_name(sig, buf, buflen));
4522
4523 address handler = (sa.sa_flags & SA_SIGINFO)
4524 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4525 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4526
4527 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4528 st->print("SIG_DFL");
4529 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4530 st->print("SIG_IGN");
4531 } else {
4532 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4533 }
4534
4535 st->print(", sa_mask[0]=");
4536 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4537
4538 address rh = VMError::get_resetted_sighandler(sig);
4539 // May be, handler was resetted by VMError?
4540 if (rh != NULL) {
4541 handler = rh;
4542 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4543 }
4544
4545 st->print(", sa_flags=");
4546 os::Posix::print_sa_flags(st, sa.sa_flags);
4547
4548 // Check: is it our handler?
4549 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4550 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4551 // It is our signal handler
4552 // check for flags, reset system-used one!
4553 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4554 st->print(
4555 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4556 os::Linux::get_our_sigflags(sig));
4557 }
4558 }
4559 st->cr();
4560 }
4561
4562
4563 #define DO_SIGNAL_CHECK(sig) \
4564 do { \
4565 if (!sigismember(&check_signal_done, sig)) { \
4566 os::Linux::check_signal_handler(sig); \
4567 } \
4568 } while (0)
4569
4570 // This method is a periodic task to check for misbehaving JNI applications
4571 // under CheckJNI, we can add any periodic checks here
4572
4573 void os::run_periodic_checks() {
4574 if (check_signals == false) return;
4575
4576 // SEGV and BUS if overridden could potentially prevent
4577 // generation of hs*.log in the event of a crash, debugging
4578 // such a case can be very challenging, so we absolutely
4579 // check the following for a good measure:
4580 DO_SIGNAL_CHECK(SIGSEGV);
4581 DO_SIGNAL_CHECK(SIGILL);
4582 DO_SIGNAL_CHECK(SIGFPE);
4583 DO_SIGNAL_CHECK(SIGBUS);
4584 DO_SIGNAL_CHECK(SIGPIPE);
4585 DO_SIGNAL_CHECK(SIGXFSZ);
4586 #if defined(PPC64)
4587 DO_SIGNAL_CHECK(SIGTRAP);
4588 #endif
4589
4590 // ReduceSignalUsage allows the user to override these handlers
4591 // see comments at the very top and jvm_solaris.h
4592 if (!ReduceSignalUsage) {
4593 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4594 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4595 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4596 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4597 }
4598
4599 DO_SIGNAL_CHECK(SR_signum);
4600 }
4601
4602 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4603
4604 static os_sigaction_t os_sigaction = NULL;
4605
4606 void os::Linux::check_signal_handler(int sig) {
4607 char buf[O_BUFLEN];
4608 address jvmHandler = NULL;
4609
4610
4611 struct sigaction act;
4612 if (os_sigaction == NULL) {
4613 // only trust the default sigaction, in case it has been interposed
4614 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4615 if (os_sigaction == NULL) return;
4616 }
4617
4618 os_sigaction(sig, (struct sigaction*)NULL, &act);
4619
4620
4621 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4622
4623 address thisHandler = (act.sa_flags & SA_SIGINFO)
4624 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4625 : CAST_FROM_FN_PTR(address, act.sa_handler);
4626
4627
4628 switch (sig) {
4629 case SIGSEGV:
4630 case SIGBUS:
4631 case SIGFPE:
4632 case SIGPIPE:
4633 case SIGILL:
4634 case SIGXFSZ:
4635 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4636 break;
4637
4638 case SHUTDOWN1_SIGNAL:
4639 case SHUTDOWN2_SIGNAL:
4640 case SHUTDOWN3_SIGNAL:
4641 case BREAK_SIGNAL:
4642 jvmHandler = (address)user_handler();
4643 break;
4644
4645 default:
4646 if (sig == SR_signum) {
4647 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4648 } else {
4649 return;
4650 }
4651 break;
4652 }
4653
4654 if (thisHandler != jvmHandler) {
4655 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4656 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4657 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4658 // No need to check this sig any longer
4659 sigaddset(&check_signal_done, sig);
4660 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4661 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4662 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4663 exception_name(sig, buf, O_BUFLEN));
4664 }
4665 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4666 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4667 tty->print("expected:");
4668 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4669 tty->cr();
4670 tty->print(" found:");
4671 os::Posix::print_sa_flags(tty, act.sa_flags);
4672 tty->cr();
4673 // No need to check this sig any longer
4674 sigaddset(&check_signal_done, sig);
4675 }
4676
4677 // Dump all the signal
4678 if (sigismember(&check_signal_done, sig)) {
4679 print_signal_handlers(tty, buf, O_BUFLEN);
4680 }
4681 }
4682
4683 extern void report_error(char* file_name, int line_no, char* title,
4684 char* format, ...);
4685
4686 // this is called _before_ the most of global arguments have been parsed
4687 void os::init(void) {
4688 char dummy; // used to get a guess on initial stack address
4689 // first_hrtime = gethrtime();
4690
4691 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4692
4693 init_random(1234567);
4694
4695 ThreadCritical::initialize();
4696
4697 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4698 if (Linux::page_size() == -1) {
4699 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4700 os::strerror(errno));
4701 }
4702 init_page_sizes((size_t) Linux::page_size());
4703
4704 Linux::initialize_system_info();
4705
4706 Linux::initialize_os_info();
4707
4708 // main_thread points to the aboriginal thread
4709 Linux::_main_thread = pthread_self();
4710
4711 Linux::clock_init();
4712 initial_time_count = javaTimeNanos();
4713
4714 // pthread_condattr initialization for monotonic clock
4715 int status;
4716 pthread_condattr_t* _condattr = os::Linux::condAttr();
4717 if ((status = pthread_condattr_init(_condattr)) != 0) {
4718 fatal("pthread_condattr_init: %s", os::strerror(status));
4719 }
4720 // Only set the clock if CLOCK_MONOTONIC is available
4721 if (os::supports_monotonic_clock()) {
4722 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4723 if (status == EINVAL) {
4724 warning("Unable to use monotonic clock with relative timed-waits" \
4725 " - changes to the time-of-day clock may have adverse affects");
4726 } else {
4727 fatal("pthread_condattr_setclock: %s", os::strerror(status));
4728 }
4729 }
4730 }
4731 // else it defaults to CLOCK_REALTIME
4732
4733 // retrieve entry point for pthread_setname_np
4734 Linux::_pthread_setname_np =
4735 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4736
4737 }
4738
4739 // To install functions for atexit system call
4740 extern "C" {
4741 static void perfMemory_exit_helper() {
4742 perfMemory_exit();
4743 }
4744 }
4745
4746 // this is called _after_ the global arguments have been parsed
4747 jint os::init_2(void) {
4748 Linux::fast_thread_clock_init();
4749
4750 // Allocate a single page and mark it as readable for safepoint polling
4751 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4752 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4753
4754 os::set_polling_page(polling_page);
4755 log_info(os)("SafePoint Polling address: " INTPTR_FORMAT, p2i(polling_page));
4756
4757 if (!UseMembar) {
4758 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4759 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4760 os::set_memory_serialize_page(mem_serialize_page);
4761 log_info(os)("Memory Serialize Page address: " INTPTR_FORMAT, p2i(mem_serialize_page));
4762 }
4763
4764 // initialize suspend/resume support - must do this before signal_sets_init()
4765 if (SR_initialize() != 0) {
4766 perror("SR_initialize failed");
4767 return JNI_ERR;
4768 }
4769
4770 Linux::signal_sets_init();
4771 Linux::install_signal_handlers();
4772
4773 // Check minimum allowable stack size for thread creation and to initialize
4774 // the java system classes, including StackOverflowError - depends on page
4775 // size. Add a page for compiler2 recursion in main thread.
4776 // Add in 2*BytesPerWord times page size to account for VM stack during
4777 // class initialization depending on 32 or 64 bit VM.
4778 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4779 JavaThread::stack_guard_zone_size() +
4780 JavaThread::stack_shadow_zone_size() +
4781 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4782
4783 size_t threadStackSizeInBytes = ThreadStackSize * K;
4784 if (threadStackSizeInBytes != 0 &&
4785 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4786 tty->print_cr("\nThe stack size specified is too small, "
4787 "Specify at least " SIZE_FORMAT "k",
4788 os::Linux::min_stack_allowed/ K);
4789 return JNI_ERR;
4790 }
4791
4792 // Make the stack size a multiple of the page size so that
4793 // the yellow/red zones can be guarded.
4794 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4795 vm_page_size()));
4796
4797 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4798
4799 #if defined(IA32)
4800 workaround_expand_exec_shield_cs_limit();
4801 #endif
4802
4803 Linux::libpthread_init();
4804 log_info(os)("HotSpot is running with %s, %s",
4805 Linux::glibc_version(), Linux::libpthread_version());
4806
4807 if (UseNUMA) {
4808 if (!Linux::libnuma_init()) {
4809 UseNUMA = false;
4810 } else {
4811 if ((Linux::numa_max_node() < 1)) {
4812 // There's only one node(they start from 0), disable NUMA.
4813 UseNUMA = false;
4814 }
4815 }
4816 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4817 // we can make the adaptive lgrp chunk resizing work. If the user specified
4818 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4819 // disable adaptive resizing.
4820 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4821 if (FLAG_IS_DEFAULT(UseNUMA)) {
4822 UseNUMA = false;
4823 } else {
4824 if (FLAG_IS_DEFAULT(UseLargePages) &&
4825 FLAG_IS_DEFAULT(UseSHM) &&
4826 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4827 UseLargePages = false;
4828 } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4829 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4830 UseAdaptiveSizePolicy = false;
4831 UseAdaptiveNUMAChunkSizing = false;
4832 }
4833 }
4834 }
4835 if (!UseNUMA && ForceNUMA) {
4836 UseNUMA = true;
4837 }
4838 }
4839
4840 if (MaxFDLimit) {
4841 // set the number of file descriptors to max. print out error
4842 // if getrlimit/setrlimit fails but continue regardless.
4843 struct rlimit nbr_files;
4844 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4845 if (status != 0) {
4846 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4847 } else {
4848 nbr_files.rlim_cur = nbr_files.rlim_max;
4849 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4850 if (status != 0) {
4851 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4852 }
4853 }
4854 }
4855
4856 // Initialize lock used to serialize thread creation (see os::create_thread)
4857 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4858
4859 // at-exit methods are called in the reverse order of their registration.
4860 // atexit functions are called on return from main or as a result of a
4861 // call to exit(3C). There can be only 32 of these functions registered
4862 // and atexit() does not set errno.
4863
4864 if (PerfAllowAtExitRegistration) {
4865 // only register atexit functions if PerfAllowAtExitRegistration is set.
4866 // atexit functions can be delayed until process exit time, which
4867 // can be problematic for embedded VM situations. Embedded VMs should
4868 // call DestroyJavaVM() to assure that VM resources are released.
4869
4870 // note: perfMemory_exit_helper atexit function may be removed in
4871 // the future if the appropriate cleanup code can be added to the
4872 // VM_Exit VMOperation's doit method.
4873 if (atexit(perfMemory_exit_helper) != 0) {
4874 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4875 }
4876 }
4877
4878 // initialize thread priority policy
4879 prio_init();
4880
4881 return JNI_OK;
4882 }
4883
4884 // Mark the polling page as unreadable
4885 void os::make_polling_page_unreadable(void) {
4886 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4887 fatal("Could not disable polling page");
4888 }
4889 }
4890
4891 // Mark the polling page as readable
4892 void os::make_polling_page_readable(void) {
4893 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4894 fatal("Could not enable polling page");
4895 }
4896 }
4897
4898 // older glibc versions don't have this macro (which expands to
4899 // an optimized bit-counting function) so we have to roll our own
4900 #ifndef CPU_COUNT
4901
4902 static int _cpu_count(const cpu_set_t* cpus) {
4903 int count = 0;
4904 // only look up to the number of configured processors
4905 for (int i = 0; i < os::processor_count(); i++) {
4906 if (CPU_ISSET(i, cpus)) {
4907 count++;
4908 }
4909 }
4910 return count;
4911 }
4912
4913 #define CPU_COUNT(cpus) _cpu_count(cpus)
4914
4915 #endif // CPU_COUNT
4916
4917 // Get the current number of available processors for this process.
4918 // This value can change at any time during a process's lifetime.
4919 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
4920 // If it appears there may be more than 1024 processors then we do a
4921 // dynamic check - see 6515172 for details.
4922 // If anything goes wrong we fallback to returning the number of online
4923 // processors - which can be greater than the number available to the process.
4924 int os::active_processor_count() {
4925 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
4926 cpu_set_t* cpus_p = &cpus;
4927 int cpus_size = sizeof(cpu_set_t);
4928
4929 int configured_cpus = processor_count(); // upper bound on available cpus
4930 int cpu_count = 0;
4931
4932 // old build platforms may not support dynamic cpu sets
4933 #ifdef CPU_ALLOC
4934
4935 // To enable easy testing of the dynamic path on different platforms we
4936 // introduce a diagnostic flag: UseCpuAllocPath
4937 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
4938 // kernel may use a mask bigger than cpu_set_t
4939 log_trace(os)("active_processor_count: using dynamic path %s"
4940 "- configured processors: %d",
4941 UseCpuAllocPath ? "(forced) " : "",
4942 configured_cpus);
4943 cpus_p = CPU_ALLOC(configured_cpus);
4944 if (cpus_p != NULL) {
4945 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
4946 // zero it just to be safe
4947 CPU_ZERO_S(cpus_size, cpus_p);
4948 }
4949 else {
4950 // failed to allocate so fallback to online cpus
4951 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4952 log_trace(os)("active_processor_count: "
4953 "CPU_ALLOC failed (%s) - using "
4954 "online processor count: %d",
4955 os::strerror(errno), online_cpus);
4956 return online_cpus;
4957 }
4958 }
4959 else {
4960 log_trace(os)("active_processor_count: using static path - configured processors: %d",
4961 configured_cpus);
4962 }
4963 #else // CPU_ALLOC
4964 // these stubs won't be executed
4965 #define CPU_COUNT_S(size, cpus) -1
4966 #define CPU_FREE(cpus)
4967
4968 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
4969 configured_cpus);
4970 #endif // CPU_ALLOC
4971
4972 // pid 0 means the current thread - which we have to assume represents the process
4973 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
4974 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
4975 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
4976 }
4977 else {
4978 cpu_count = CPU_COUNT(cpus_p);
4979 }
4980 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
4981 }
4982 else {
4983 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
4984 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
4985 "which may exceed available processors", os::strerror(errno), cpu_count);
4986 }
4987
4988 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
4989 CPU_FREE(cpus_p);
4990 }
4991
4992 assert(cpu_count > 0 && cpu_count <= processor_count(), "sanity check");
4993 return cpu_count;
4994 }
4995
4996 void os::set_native_thread_name(const char *name) {
4997 if (Linux::_pthread_setname_np) {
4998 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4999 snprintf(buf, sizeof(buf), "%s", name);
5000 buf[sizeof(buf) - 1] = '\0';
5001 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5002 // ERANGE should not happen; all other errors should just be ignored.
5003 assert(rc != ERANGE, "pthread_setname_np failed");
5004 }
5005 }
5006
5007 bool os::distribute_processes(uint length, uint* distribution) {
5008 // Not yet implemented.
5009 return false;
5010 }
5011
5012 bool os::bind_to_processor(uint processor_id) {
5013 // Not yet implemented.
5014 return false;
5015 }
5016
5017 ///
5018
5019 void os::SuspendedThreadTask::internal_do_task() {
5020 if (do_suspend(_thread->osthread())) {
5021 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5022 do_task(context);
5023 do_resume(_thread->osthread());
5024 }
5025 }
5026
5027 class PcFetcher : public os::SuspendedThreadTask {
5028 public:
5029 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5030 ExtendedPC result();
5031 protected:
5032 void do_task(const os::SuspendedThreadTaskContext& context);
5033 private:
5034 ExtendedPC _epc;
5035 };
5036
5037 ExtendedPC PcFetcher::result() {
5038 guarantee(is_done(), "task is not done yet.");
5039 return _epc;
5040 }
5041
5042 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5043 Thread* thread = context.thread();
5044 OSThread* osthread = thread->osthread();
5045 if (osthread->ucontext() != NULL) {
5046 _epc = os::Linux::ucontext_get_pc((const ucontext_t *) context.ucontext());
5047 } else {
5048 // NULL context is unexpected, double-check this is the VMThread
5049 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5050 }
5051 }
5052
5053 // Suspends the target using the signal mechanism and then grabs the PC before
5054 // resuming the target. Used by the flat-profiler only
5055 ExtendedPC os::get_thread_pc(Thread* thread) {
5056 // Make sure that it is called by the watcher for the VMThread
5057 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5058 assert(thread->is_VM_thread(), "Can only be called for VMThread");
5059
5060 PcFetcher fetcher(thread);
5061 fetcher.run();
5062 return fetcher.result();
5063 }
5064
5065 ////////////////////////////////////////////////////////////////////////////////
5066 // debug support
5067
5068 bool os::find(address addr, outputStream* st) {
5069 Dl_info dlinfo;
5070 memset(&dlinfo, 0, sizeof(dlinfo));
5071 if (dladdr(addr, &dlinfo) != 0) {
5072 st->print(PTR_FORMAT ": ", p2i(addr));
5073 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5074 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5075 p2i(addr) - p2i(dlinfo.dli_saddr));
5076 } else if (dlinfo.dli_fbase != NULL) {
5077 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5078 } else {
5079 st->print("<absolute address>");
5080 }
5081 if (dlinfo.dli_fname != NULL) {
5082 st->print(" in %s", dlinfo.dli_fname);
5083 }
5084 if (dlinfo.dli_fbase != NULL) {
5085 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5086 }
5087 st->cr();
5088
5089 if (Verbose) {
5090 // decode some bytes around the PC
5091 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5092 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5093 address lowest = (address) dlinfo.dli_sname;
5094 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5095 if (begin < lowest) begin = lowest;
5096 Dl_info dlinfo2;
5097 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5098 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5099 end = (address) dlinfo2.dli_saddr;
5100 }
5101 Disassembler::decode(begin, end, st);
5102 }
5103 return true;
5104 }
5105 return false;
5106 }
5107
5108 ////////////////////////////////////////////////////////////////////////////////
5109 // misc
5110
5111 // This does not do anything on Linux. This is basically a hook for being
5112 // able to use structured exception handling (thread-local exception filters)
5113 // on, e.g., Win32.
5114 void
5115 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5116 JavaCallArguments* args, Thread* thread) {
5117 f(value, method, args, thread);
5118 }
5119
5120 void os::print_statistics() {
5121 }
5122
5123 bool os::message_box(const char* title, const char* message) {
5124 int i;
5125 fdStream err(defaultStream::error_fd());
5126 for (i = 0; i < 78; i++) err.print_raw("=");
5127 err.cr();
5128 err.print_raw_cr(title);
5129 for (i = 0; i < 78; i++) err.print_raw("-");
5130 err.cr();
5131 err.print_raw_cr(message);
5132 for (i = 0; i < 78; i++) err.print_raw("=");
5133 err.cr();
5134
5135 char buf[16];
5136 // Prevent process from exiting upon "read error" without consuming all CPU
5137 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5138
5139 return buf[0] == 'y' || buf[0] == 'Y';
5140 }
5141
5142 int os::stat(const char *path, struct stat *sbuf) {
5143 char pathbuf[MAX_PATH];
5144 if (strlen(path) > MAX_PATH - 1) {
5145 errno = ENAMETOOLONG;
5146 return -1;
5147 }
5148 os::native_path(strcpy(pathbuf, path));
5149 return ::stat(pathbuf, sbuf);
5150 }
5151
5152 bool os::check_heap(bool force) {
5153 return true;
5154 }
5155
5156 // Is a (classpath) directory empty?
5157 bool os::dir_is_empty(const char* path) {
5158 DIR *dir = NULL;
5159 struct dirent *ptr;
5160
5161 dir = opendir(path);
5162 if (dir == NULL) return true;
5163
5164 // Scan the directory
5165 bool result = true;
5166 char buf[sizeof(struct dirent) + MAX_PATH];
5167 while (result && (ptr = ::readdir(dir)) != NULL) {
5168 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5169 result = false;
5170 }
5171 }
5172 closedir(dir);
5173 return result;
5174 }
5175
5176 // This code originates from JDK's sysOpen and open64_w
5177 // from src/solaris/hpi/src/system_md.c
5178
5179 int os::open(const char *path, int oflag, int mode) {
5180 if (strlen(path) > MAX_PATH - 1) {
5181 errno = ENAMETOOLONG;
5182 return -1;
5183 }
5184
5185 // All file descriptors that are opened in the Java process and not
5186 // specifically destined for a subprocess should have the close-on-exec
5187 // flag set. If we don't set it, then careless 3rd party native code
5188 // might fork and exec without closing all appropriate file descriptors
5189 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5190 // turn might:
5191 //
5192 // - cause end-of-file to fail to be detected on some file
5193 // descriptors, resulting in mysterious hangs, or
5194 //
5195 // - might cause an fopen in the subprocess to fail on a system
5196 // suffering from bug 1085341.
5197 //
5198 // (Yes, the default setting of the close-on-exec flag is a Unix
5199 // design flaw)
5200 //
5201 // See:
5202 // 1085341: 32-bit stdio routines should support file descriptors >255
5203 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5204 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5205 //
5206 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5207 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5208 // because it saves a system call and removes a small window where the flag
5209 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5210 // and we fall back to using FD_CLOEXEC (see below).
5211 #ifdef O_CLOEXEC
5212 oflag |= O_CLOEXEC;
5213 #endif
5214
5215 int fd = ::open64(path, oflag, mode);
5216 if (fd == -1) return -1;
5217
5218 //If the open succeeded, the file might still be a directory
5219 {
5220 struct stat64 buf64;
5221 int ret = ::fstat64(fd, &buf64);
5222 int st_mode = buf64.st_mode;
5223
5224 if (ret != -1) {
5225 if ((st_mode & S_IFMT) == S_IFDIR) {
5226 errno = EISDIR;
5227 ::close(fd);
5228 return -1;
5229 }
5230 } else {
5231 ::close(fd);
5232 return -1;
5233 }
5234 }
5235
5236 #ifdef FD_CLOEXEC
5237 // Validate that the use of the O_CLOEXEC flag on open above worked.
5238 // With recent kernels, we will perform this check exactly once.
5239 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5240 if (!O_CLOEXEC_is_known_to_work) {
5241 int flags = ::fcntl(fd, F_GETFD);
5242 if (flags != -1) {
5243 if ((flags & FD_CLOEXEC) != 0)
5244 O_CLOEXEC_is_known_to_work = 1;
5245 else
5246 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5247 }
5248 }
5249 #endif
5250
5251 return fd;
5252 }
5253
5254
5255 // create binary file, rewriting existing file if required
5256 int os::create_binary_file(const char* path, bool rewrite_existing) {
5257 int oflags = O_WRONLY | O_CREAT;
5258 if (!rewrite_existing) {
5259 oflags |= O_EXCL;
5260 }
5261 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5262 }
5263
5264 // return current position of file pointer
5265 jlong os::current_file_offset(int fd) {
5266 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5267 }
5268
5269 // move file pointer to the specified offset
5270 jlong os::seek_to_file_offset(int fd, jlong offset) {
5271 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5272 }
5273
5274 // This code originates from JDK's sysAvailable
5275 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5276
5277 int os::available(int fd, jlong *bytes) {
5278 jlong cur, end;
5279 int mode;
5280 struct stat64 buf64;
5281
5282 if (::fstat64(fd, &buf64) >= 0) {
5283 mode = buf64.st_mode;
5284 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5285 int n;
5286 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5287 *bytes = n;
5288 return 1;
5289 }
5290 }
5291 }
5292 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5293 return 0;
5294 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5295 return 0;
5296 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5297 return 0;
5298 }
5299 *bytes = end - cur;
5300 return 1;
5301 }
5302
5303 // Map a block of memory.
5304 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5305 char *addr, size_t bytes, bool read_only,
5306 bool allow_exec) {
5307 int prot;
5308 int flags = MAP_PRIVATE;
5309
5310 if (read_only) {
5311 prot = PROT_READ;
5312 } else {
5313 prot = PROT_READ | PROT_WRITE;
5314 }
5315
5316 if (allow_exec) {
5317 prot |= PROT_EXEC;
5318 }
5319
5320 if (addr != NULL) {
5321 flags |= MAP_FIXED;
5322 }
5323
5324 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5325 fd, file_offset);
5326 if (mapped_address == MAP_FAILED) {
5327 return NULL;
5328 }
5329 return mapped_address;
5330 }
5331
5332
5333 // Remap a block of memory.
5334 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5335 char *addr, size_t bytes, bool read_only,
5336 bool allow_exec) {
5337 // same as map_memory() on this OS
5338 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5339 allow_exec);
5340 }
5341
5342
5343 // Unmap a block of memory.
5344 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5345 return munmap(addr, bytes) == 0;
5346 }
5347
5348 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5349
5350 static clockid_t thread_cpu_clockid(Thread* thread) {
5351 pthread_t tid = thread->osthread()->pthread_id();
5352 clockid_t clockid;
5353
5354 // Get thread clockid
5355 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5356 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5357 return clockid;
5358 }
5359
5360 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5361 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5362 // of a thread.
5363 //
5364 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5365 // the fast estimate available on the platform.
5366
5367 jlong os::current_thread_cpu_time() {
5368 if (os::Linux::supports_fast_thread_cpu_time()) {
5369 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5370 } else {
5371 // return user + sys since the cost is the same
5372 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5373 }
5374 }
5375
5376 jlong os::thread_cpu_time(Thread* thread) {
5377 // consistent with what current_thread_cpu_time() returns
5378 if (os::Linux::supports_fast_thread_cpu_time()) {
5379 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5380 } else {
5381 return slow_thread_cpu_time(thread, true /* user + sys */);
5382 }
5383 }
5384
5385 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5386 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5387 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5388 } else {
5389 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5390 }
5391 }
5392
5393 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5394 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5395 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5396 } else {
5397 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5398 }
5399 }
5400
5401 // -1 on error.
5402 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5403 pid_t tid = thread->osthread()->thread_id();
5404 char *s;
5405 char stat[2048];
5406 int statlen;
5407 char proc_name[64];
5408 int count;
5409 long sys_time, user_time;
5410 char cdummy;
5411 int idummy;
5412 long ldummy;
5413 FILE *fp;
5414
5415 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5416 fp = fopen(proc_name, "r");
5417 if (fp == NULL) return -1;
5418 statlen = fread(stat, 1, 2047, fp);
5419 stat[statlen] = '\0';
5420 fclose(fp);
5421
5422 // Skip pid and the command string. Note that we could be dealing with
5423 // weird command names, e.g. user could decide to rename java launcher
5424 // to "java 1.4.2 :)", then the stat file would look like
5425 // 1234 (java 1.4.2 :)) R ... ...
5426 // We don't really need to know the command string, just find the last
5427 // occurrence of ")" and then start parsing from there. See bug 4726580.
5428 s = strrchr(stat, ')');
5429 if (s == NULL) return -1;
5430
5431 // Skip blank chars
5432 do { s++; } while (s && isspace(*s));
5433
5434 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5435 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5436 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5437 &user_time, &sys_time);
5438 if (count != 13) return -1;
5439 if (user_sys_cpu_time) {
5440 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5441 } else {
5442 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5443 }
5444 }
5445
5446 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5447 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5448 info_ptr->may_skip_backward = false; // elapsed time not wall time
5449 info_ptr->may_skip_forward = false; // elapsed time not wall time
5450 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5451 }
5452
5453 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5454 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5455 info_ptr->may_skip_backward = false; // elapsed time not wall time
5456 info_ptr->may_skip_forward = false; // elapsed time not wall time
5457 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5458 }
5459
5460 bool os::is_thread_cpu_time_supported() {
5461 return true;
5462 }
5463
5464 // System loadavg support. Returns -1 if load average cannot be obtained.
5465 // Linux doesn't yet have a (official) notion of processor sets,
5466 // so just return the system wide load average.
5467 int os::loadavg(double loadavg[], int nelem) {
5468 return ::getloadavg(loadavg, nelem);
5469 }
5470
5471 void os::pause() {
5472 char filename[MAX_PATH];
5473 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5474 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5475 } else {
5476 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5477 }
5478
5479 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5480 if (fd != -1) {
5481 struct stat buf;
5482 ::close(fd);
5483 while (::stat(filename, &buf) == 0) {
5484 (void)::poll(NULL, 0, 100);
5485 }
5486 } else {
5487 jio_fprintf(stderr,
5488 "Could not open pause file '%s', continuing immediately.\n", filename);
5489 }
5490 }
5491
5492
5493 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5494 // comment paragraphs are worth repeating here:
5495 //
5496 // Assumption:
5497 // Only one parker can exist on an event, which is why we allocate
5498 // them per-thread. Multiple unparkers can coexist.
5499 //
5500 // _Event serves as a restricted-range semaphore.
5501 // -1 : thread is blocked, i.e. there is a waiter
5502 // 0 : neutral: thread is running or ready,
5503 // could have been signaled after a wait started
5504 // 1 : signaled - thread is running or ready
5505 //
5506
5507 // utility to compute the abstime argument to timedwait:
5508 // millis is the relative timeout time
5509 // abstime will be the absolute timeout time
5510 // TODO: replace compute_abstime() with unpackTime()
5511
5512 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5513 if (millis < 0) millis = 0;
5514
5515 jlong seconds = millis / 1000;
5516 millis %= 1000;
5517 if (seconds > 50000000) { // see man cond_timedwait(3T)
5518 seconds = 50000000;
5519 }
5520
5521 if (os::supports_monotonic_clock()) {
5522 struct timespec now;
5523 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5524 assert_status(status == 0, status, "clock_gettime");
5525 abstime->tv_sec = now.tv_sec + seconds;
5526 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5527 if (nanos >= NANOSECS_PER_SEC) {
5528 abstime->tv_sec += 1;
5529 nanos -= NANOSECS_PER_SEC;
5530 }
5531 abstime->tv_nsec = nanos;
5532 } else {
5533 struct timeval now;
5534 int status = gettimeofday(&now, NULL);
5535 assert(status == 0, "gettimeofday");
5536 abstime->tv_sec = now.tv_sec + seconds;
5537 long usec = now.tv_usec + millis * 1000;
5538 if (usec >= 1000000) {
5539 abstime->tv_sec += 1;
5540 usec -= 1000000;
5541 }
5542 abstime->tv_nsec = usec * 1000;
5543 }
5544 return abstime;
5545 }
5546
5547 void os::PlatformEvent::park() { // AKA "down()"
5548 // Transitions for _Event:
5549 // -1 => -1 : illegal
5550 // 1 => 0 : pass - return immediately
5551 // 0 => -1 : block; then set _Event to 0 before returning
5552
5553 // Invariant: Only the thread associated with the Event/PlatformEvent
5554 // may call park().
5555 // TODO: assert that _Assoc != NULL or _Assoc == Self
5556 assert(_nParked == 0, "invariant");
5557
5558 int v;
5559 for (;;) {
5560 v = _Event;
5561 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5562 }
5563 guarantee(v >= 0, "invariant");
5564 if (v == 0) {
5565 // Do this the hard way by blocking ...
5566 int status = pthread_mutex_lock(_mutex);
5567 assert_status(status == 0, status, "mutex_lock");
5568 guarantee(_nParked == 0, "invariant");
5569 ++_nParked;
5570 while (_Event < 0) {
5571 status = pthread_cond_wait(_cond, _mutex);
5572 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5573 // Treat this the same as if the wait was interrupted
5574 if (status == ETIME) { status = EINTR; }
5575 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5576 }
5577 --_nParked;
5578
5579 _Event = 0;
5580 status = pthread_mutex_unlock(_mutex);
5581 assert_status(status == 0, status, "mutex_unlock");
5582 // Paranoia to ensure our locked and lock-free paths interact
5583 // correctly with each other.
5584 OrderAccess::fence();
5585 }
5586 guarantee(_Event >= 0, "invariant");
5587 }
5588
5589 int os::PlatformEvent::park(jlong millis) {
5590 // Transitions for _Event:
5591 // -1 => -1 : illegal
5592 // 1 => 0 : pass - return immediately
5593 // 0 => -1 : block; then set _Event to 0 before returning
5594
5595 guarantee(_nParked == 0, "invariant");
5596
5597 int v;
5598 for (;;) {
5599 v = _Event;
5600 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5601 }
5602 guarantee(v >= 0, "invariant");
5603 if (v != 0) return OS_OK;
5604
5605 // We do this the hard way, by blocking the thread.
5606 // Consider enforcing a minimum timeout value.
5607 struct timespec abst;
5608 compute_abstime(&abst, millis);
5609
5610 int ret = OS_TIMEOUT;
5611 int status = pthread_mutex_lock(_mutex);
5612 assert_status(status == 0, status, "mutex_lock");
5613 guarantee(_nParked == 0, "invariant");
5614 ++_nParked;
5615
5616 // Object.wait(timo) will return because of
5617 // (a) notification
5618 // (b) timeout
5619 // (c) thread.interrupt
5620 //
5621 // Thread.interrupt and object.notify{All} both call Event::set.
5622 // That is, we treat thread.interrupt as a special case of notification.
5623 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5624 // We assume all ETIME returns are valid.
5625 //
5626 // TODO: properly differentiate simultaneous notify+interrupt.
5627 // In that case, we should propagate the notify to another waiter.
5628
5629 while (_Event < 0) {
5630 status = pthread_cond_timedwait(_cond, _mutex, &abst);
5631 assert_status(status == 0 || status == EINTR ||
5632 status == ETIME || status == ETIMEDOUT,
5633 status, "cond_timedwait");
5634 if (!FilterSpuriousWakeups) break; // previous semantics
5635 if (status == ETIME || status == ETIMEDOUT) break;
5636 // We consume and ignore EINTR and spurious wakeups.
5637 }
5638 --_nParked;
5639 if (_Event >= 0) {
5640 ret = OS_OK;
5641 }
5642 _Event = 0;
5643 status = pthread_mutex_unlock(_mutex);
5644 assert_status(status == 0, status, "mutex_unlock");
5645 assert(_nParked == 0, "invariant");
5646 // Paranoia to ensure our locked and lock-free paths interact
5647 // correctly with each other.
5648 OrderAccess::fence();
5649 return ret;
5650 }
5651
5652 void os::PlatformEvent::unpark() {
5653 // Transitions for _Event:
5654 // 0 => 1 : just return
5655 // 1 => 1 : just return
5656 // -1 => either 0 or 1; must signal target thread
5657 // That is, we can safely transition _Event from -1 to either
5658 // 0 or 1.
5659 // See also: "Semaphores in Plan 9" by Mullender & Cox
5660 //
5661 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5662 // that it will take two back-to-back park() calls for the owning
5663 // thread to block. This has the benefit of forcing a spurious return
5664 // from the first park() call after an unpark() call which will help
5665 // shake out uses of park() and unpark() without condition variables.
5666
5667 if (Atomic::xchg(1, &_Event) >= 0) return;
5668
5669 // Wait for the thread associated with the event to vacate
5670 int status = pthread_mutex_lock(_mutex);
5671 assert_status(status == 0, status, "mutex_lock");
5672 int AnyWaiters = _nParked;
5673 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5674 status = pthread_mutex_unlock(_mutex);
5675 assert_status(status == 0, status, "mutex_unlock");
5676 if (AnyWaiters != 0) {
5677 // Note that we signal() *after* dropping the lock for "immortal" Events.
5678 // This is safe and avoids a common class of futile wakeups. In rare
5679 // circumstances this can cause a thread to return prematurely from
5680 // cond_{timed}wait() but the spurious wakeup is benign and the victim
5681 // will simply re-test the condition and re-park itself.
5682 // This provides particular benefit if the underlying platform does not
5683 // provide wait morphing.
5684 status = pthread_cond_signal(_cond);
5685 assert_status(status == 0, status, "cond_signal");
5686 }
5687 }
5688
5689
5690 // JSR166
5691 // -------------------------------------------------------
5692
5693 // The solaris and linux implementations of park/unpark are fairly
5694 // conservative for now, but can be improved. They currently use a
5695 // mutex/condvar pair, plus a a count.
5696 // Park decrements count if > 0, else does a condvar wait. Unpark
5697 // sets count to 1 and signals condvar. Only one thread ever waits
5698 // on the condvar. Contention seen when trying to park implies that someone
5699 // is unparking you, so don't wait. And spurious returns are fine, so there
5700 // is no need to track notifications.
5701
5702 // This code is common to linux and solaris and will be moved to a
5703 // common place in dolphin.
5704 //
5705 // The passed in time value is either a relative time in nanoseconds
5706 // or an absolute time in milliseconds. Either way it has to be unpacked
5707 // into suitable seconds and nanoseconds components and stored in the
5708 // given timespec structure.
5709 // Given time is a 64-bit value and the time_t used in the timespec is only
5710 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5711 // overflow if times way in the future are given. Further on Solaris versions
5712 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5713 // number of seconds, in abstime, is less than current_time + 100,000,000.
5714 // As it will be 28 years before "now + 100000000" will overflow we can
5715 // ignore overflow and just impose a hard-limit on seconds using the value
5716 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5717 // years from "now".
5718
5719 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5720 assert(time > 0, "convertTime");
5721 time_t max_secs = 0;
5722
5723 if (!os::supports_monotonic_clock() || isAbsolute) {
5724 struct timeval now;
5725 int status = gettimeofday(&now, NULL);
5726 assert(status == 0, "gettimeofday");
5727
5728 max_secs = now.tv_sec + MAX_SECS;
5729
5730 if (isAbsolute) {
5731 jlong secs = time / 1000;
5732 if (secs > max_secs) {
5733 absTime->tv_sec = max_secs;
5734 } else {
5735 absTime->tv_sec = secs;
5736 }
5737 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5738 } else {
5739 jlong secs = time / NANOSECS_PER_SEC;
5740 if (secs >= MAX_SECS) {
5741 absTime->tv_sec = max_secs;
5742 absTime->tv_nsec = 0;
5743 } else {
5744 absTime->tv_sec = now.tv_sec + secs;
5745 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5746 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5747 absTime->tv_nsec -= NANOSECS_PER_SEC;
5748 ++absTime->tv_sec; // note: this must be <= max_secs
5749 }
5750 }
5751 }
5752 } else {
5753 // must be relative using monotonic clock
5754 struct timespec now;
5755 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5756 assert_status(status == 0, status, "clock_gettime");
5757 max_secs = now.tv_sec + MAX_SECS;
5758 jlong secs = time / NANOSECS_PER_SEC;
5759 if (secs >= MAX_SECS) {
5760 absTime->tv_sec = max_secs;
5761 absTime->tv_nsec = 0;
5762 } else {
5763 absTime->tv_sec = now.tv_sec + secs;
5764 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5765 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5766 absTime->tv_nsec -= NANOSECS_PER_SEC;
5767 ++absTime->tv_sec; // note: this must be <= max_secs
5768 }
5769 }
5770 }
5771 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5772 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5773 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5774 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5775 }
5776
5777 void Parker::park(bool isAbsolute, jlong time) {
5778 // Ideally we'd do something useful while spinning, such
5779 // as calling unpackTime().
5780
5781 // Optional fast-path check:
5782 // Return immediately if a permit is available.
5783 // We depend on Atomic::xchg() having full barrier semantics
5784 // since we are doing a lock-free update to _counter.
5785 if (Atomic::xchg(0, &_counter) > 0) return;
5786
5787 Thread* thread = Thread::current();
5788 assert(thread->is_Java_thread(), "Must be JavaThread");
5789 JavaThread *jt = (JavaThread *)thread;
5790
5791 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5792 // Check interrupt before trying to wait
5793 if (Thread::is_interrupted(thread, false)) {
5794 return;
5795 }
5796
5797 // Next, demultiplex/decode time arguments
5798 timespec absTime;
5799 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5800 return;
5801 }
5802 if (time > 0) {
5803 unpackTime(&absTime, isAbsolute, time);
5804 }
5805
5806
5807 // Enter safepoint region
5808 // Beware of deadlocks such as 6317397.
5809 // The per-thread Parker:: mutex is a classic leaf-lock.
5810 // In particular a thread must never block on the Threads_lock while
5811 // holding the Parker:: mutex. If safepoints are pending both the
5812 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5813 ThreadBlockInVM tbivm(jt);
5814
5815 // Don't wait if cannot get lock since interference arises from
5816 // unblocking. Also. check interrupt before trying wait
5817 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5818 return;
5819 }
5820
5821 int status;
5822 if (_counter > 0) { // no wait needed
5823 _counter = 0;
5824 status = pthread_mutex_unlock(_mutex);
5825 assert_status(status == 0, status, "invariant");
5826 // Paranoia to ensure our locked and lock-free paths interact
5827 // correctly with each other and Java-level accesses.
5828 OrderAccess::fence();
5829 return;
5830 }
5831
5832 #ifdef ASSERT
5833 // Don't catch signals while blocked; let the running threads have the signals.
5834 // (This allows a debugger to break into the running thread.)
5835 sigset_t oldsigs;
5836 sigemptyset(&oldsigs);
5837 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5838 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5839 #endif
5840
5841 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5842 jt->set_suspend_equivalent();
5843 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5844
5845 assert(_cur_index == -1, "invariant");
5846 if (time == 0) {
5847 _cur_index = REL_INDEX; // arbitrary choice when not timed
5848 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5849 } else {
5850 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5851 status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5852 }
5853 _cur_index = -1;
5854 assert_status(status == 0 || status == EINTR ||
5855 status == ETIME || status == ETIMEDOUT,
5856 status, "cond_timedwait");
5857
5858 #ifdef ASSERT
5859 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5860 #endif
5861
5862 _counter = 0;
5863 status = pthread_mutex_unlock(_mutex);
5864 assert_status(status == 0, status, "invariant");
5865 // Paranoia to ensure our locked and lock-free paths interact
5866 // correctly with each other and Java-level accesses.
5867 OrderAccess::fence();
5868
5869 // If externally suspended while waiting, re-suspend
5870 if (jt->handle_special_suspend_equivalent_condition()) {
5871 jt->java_suspend_self();
5872 }
5873 }
5874
5875 void Parker::unpark() {
5876 int status = pthread_mutex_lock(_mutex);
5877 assert_status(status == 0, status, "invariant");
5878 const int s = _counter;
5879 _counter = 1;
5880 // must capture correct index before unlocking
5881 int index = _cur_index;
5882 status = pthread_mutex_unlock(_mutex);
5883 assert_status(status == 0, status, "invariant");
5884 if (s < 1 && index != -1) {
5885 // thread is definitely parked
5886 status = pthread_cond_signal(&_cond[index]);
5887 assert_status(status == 0, status, "invariant");
5888 }
5889 }
5890
5891
5892 extern char** environ;
5893
5894 // Run the specified command in a separate process. Return its exit value,
5895 // or -1 on failure (e.g. can't fork a new process).
5896 // Unlike system(), this function can be called from signal handler. It
5897 // doesn't block SIGINT et al.
5898 int os::fork_and_exec(char* cmd) {
5899 const char * argv[4] = {"sh", "-c", cmd, NULL};
5900
5901 pid_t pid = fork();
5902
5903 if (pid < 0) {
5904 // fork failed
5905 return -1;
5906
5907 } else if (pid == 0) {
5908 // child process
5909
5910 execve("/bin/sh", (char* const*)argv, environ);
5911
5912 // execve failed
5913 _exit(-1);
5914
5915 } else {
5916 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5917 // care about the actual exit code, for now.
5918
5919 int status;
5920
5921 // Wait for the child process to exit. This returns immediately if
5922 // the child has already exited. */
5923 while (waitpid(pid, &status, 0) < 0) {
5924 switch (errno) {
5925 case ECHILD: return 0;
5926 case EINTR: break;
5927 default: return -1;
5928 }
5929 }
5930
5931 if (WIFEXITED(status)) {
5932 // The child exited normally; get its exit code.
5933 return WEXITSTATUS(status);
5934 } else if (WIFSIGNALED(status)) {
5935 // The child exited because of a signal
5936 // The best value to return is 0x80 + signal number,
5937 // because that is what all Unix shells do, and because
5938 // it allows callers to distinguish between process exit and
5939 // process death by signal.
5940 return 0x80 + WTERMSIG(status);
5941 } else {
5942 // Unknown exit code; pass it through
5943 return status;
5944 }
5945 }
5946 }
5947
5948 // is_headless_jre()
5949 //
5950 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5951 // in order to report if we are running in a headless jre
5952 //
5953 // Since JDK8 xawt/libmawt.so was moved into the same directory
5954 // as libawt.so, and renamed libawt_xawt.so
5955 //
5956 bool os::is_headless_jre() {
5957 struct stat statbuf;
5958 char buf[MAXPATHLEN];
5959 char libmawtpath[MAXPATHLEN];
5960 const char *xawtstr = "/xawt/libmawt.so";
5961 const char *new_xawtstr = "/libawt_xawt.so";
5962 char *p;
5963
5964 // Get path to libjvm.so
5965 os::jvm_path(buf, sizeof(buf));
5966
5967 // Get rid of libjvm.so
5968 p = strrchr(buf, '/');
5969 if (p == NULL) {
5970 return false;
5971 } else {
5972 *p = '\0';
5973 }
5974
5975 // Get rid of client or server
5976 p = strrchr(buf, '/');
5977 if (p == NULL) {
5978 return false;
5979 } else {
5980 *p = '\0';
5981 }
5982
5983 // check xawt/libmawt.so
5984 strcpy(libmawtpath, buf);
5985 strcat(libmawtpath, xawtstr);
5986 if (::stat(libmawtpath, &statbuf) == 0) return false;
5987
5988 // check libawt_xawt.so
5989 strcpy(libmawtpath, buf);
5990 strcat(libmawtpath, new_xawtstr);
5991 if (::stat(libmawtpath, &statbuf) == 0) return false;
5992
5993 return true;
5994 }
5995
5996 // Get the default path to the core file
5997 // Returns the length of the string
5998 int os::get_core_path(char* buffer, size_t bufferSize) {
5999 /*
6000 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
6001 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
6002 */
6003 const int core_pattern_len = 129;
6004 char core_pattern[core_pattern_len] = {0};
6005
6006 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
6007 if (core_pattern_file == -1) {
6008 return -1;
6009 }
6010
6011 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
6012 ::close(core_pattern_file);
6013 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
6014 return -1;
6015 }
6016 if (core_pattern[ret-1] == '\n') {
6017 core_pattern[ret-1] = '\0';
6018 } else {
6019 core_pattern[ret] = '\0';
6020 }
6021
6022 char *pid_pos = strstr(core_pattern, "%p");
6023 int written;
6024
6025 if (core_pattern[0] == '/') {
6026 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
6027 } else {
6028 char cwd[PATH_MAX];
6029
6030 const char* p = get_current_directory(cwd, PATH_MAX);
6031 if (p == NULL) {
6032 return -1;
6033 }
6034
6035 if (core_pattern[0] == '|') {
6036 written = jio_snprintf(buffer, bufferSize,
6037 "\"%s\" (or dumping to %s/core.%d)",
6038 &core_pattern[1], p, current_process_id());
6039 } else {
6040 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6041 }
6042 }
6043
6044 if (written < 0) {
6045 return -1;
6046 }
6047
6048 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6049 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6050
6051 if (core_uses_pid_file != -1) {
6052 char core_uses_pid = 0;
6053 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6054 ::close(core_uses_pid_file);
6055
6056 if (core_uses_pid == '1') {
6057 jio_snprintf(buffer + written, bufferSize - written,
6058 ".%d", current_process_id());
6059 }
6060 }
6061 }
6062
6063 return strlen(buffer);
6064 }
6065
6066 bool os::start_debugging(char *buf, int buflen) {
6067 int len = (int)strlen(buf);
6068 char *p = &buf[len];
6069
6070 jio_snprintf(p, buflen-len,
6071 "\n\n"
6072 "Do you want to debug the problem?\n\n"
6073 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
6074 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
6075 "Otherwise, press RETURN to abort...",
6076 os::current_process_id(), os::current_process_id(),
6077 os::current_thread_id(), os::current_thread_id());
6078
6079 bool yes = os::message_box("Unexpected Error", buf);
6080
6081 if (yes) {
6082 // yes, user asked VM to launch debugger
6083 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
6084 os::current_process_id(), os::current_process_id());
6085
6086 os::fork_and_exec(buf);
6087 yes = false;
6088 }
6089 return yes;
6090 }
6091
6092 static inline struct timespec get_mtime(const char* filename) {
6093 struct stat st;
6094 int ret = os::stat(filename, &st);
6095 assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
6096 return st.st_mtim;
6097 }
6098
6099 int os::compare_file_modified_times(const char* file1, const char* file2) {
6100 struct timespec filetime1 = get_mtime(file1);
6101 struct timespec filetime2 = get_mtime(file2);
6102 int diff = filetime1.tv_sec - filetime2.tv_sec;
6103 if (diff == 0) {
6104 return filetime1.tv_nsec - filetime2.tv_nsec;
6105 }
6106 return diff;
6107 }
6108
6109 /////////////// Unit tests ///////////////
6110
6111 #ifndef PRODUCT
6112
6113 #define test_log(...) \
6114 do { \
6115 if (VerboseInternalVMTests) { \
6116 tty->print_cr(__VA_ARGS__); \
6117 tty->flush(); \
6118 } \
6119 } while (false)
6120
6121 class TestReserveMemorySpecial : AllStatic {
6122 public:
6123 static void small_page_write(void* addr, size_t size) {
6124 size_t page_size = os::vm_page_size();
6125
6126 char* end = (char*)addr + size;
6127 for (char* p = (char*)addr; p < end; p += page_size) {
6128 *p = 1;
6129 }
6130 }
6131
6132 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6133 if (!UseHugeTLBFS) {
6134 return;
6135 }
6136
6137 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6138
6139 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6140
6141 if (addr != NULL) {
6142 small_page_write(addr, size);
6143
6144 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6145 }
6146 }
6147
6148 static void test_reserve_memory_special_huge_tlbfs_only() {
6149 if (!UseHugeTLBFS) {
6150 return;
6151 }
6152
6153 size_t lp = os::large_page_size();
6154
6155 for (size_t size = lp; size <= lp * 10; size += lp) {
6156 test_reserve_memory_special_huge_tlbfs_only(size);
6157 }
6158 }
6159
6160 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6161 size_t lp = os::large_page_size();
6162 size_t ag = os::vm_allocation_granularity();
6163
6164 // sizes to test
6165 const size_t sizes[] = {
6166 lp, lp + ag, lp + lp / 2, lp * 2,
6167 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6168 lp * 10, lp * 10 + lp / 2
6169 };
6170 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6171
6172 // For each size/alignment combination, we test three scenarios:
6173 // 1) with req_addr == NULL
6174 // 2) with a non-null req_addr at which we expect to successfully allocate
6175 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6176 // expect the allocation to either fail or to ignore req_addr
6177
6178 // Pre-allocate two areas; they shall be as large as the largest allocation
6179 // and aligned to the largest alignment we will be testing.
6180 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6181 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6182 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6183 -1, 0);
6184 assert(mapping1 != MAP_FAILED, "should work");
6185
6186 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6187 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6188 -1, 0);
6189 assert(mapping2 != MAP_FAILED, "should work");
6190
6191 // Unmap the first mapping, but leave the second mapping intact: the first
6192 // mapping will serve as a value for a "good" req_addr (case 2). The second
6193 // mapping, still intact, as "bad" req_addr (case 3).
6194 ::munmap(mapping1, mapping_size);
6195
6196 // Case 1
6197 test_log("%s, req_addr NULL:", __FUNCTION__);
6198 test_log("size align result");
6199
6200 for (int i = 0; i < num_sizes; i++) {
6201 const size_t size = sizes[i];
6202 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6203 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6204 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6205 size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6206 if (p != NULL) {
6207 assert(is_ptr_aligned(p, alignment), "must be");
6208 small_page_write(p, size);
6209 os::Linux::release_memory_special_huge_tlbfs(p, size);
6210 }
6211 }
6212 }
6213
6214 // Case 2
6215 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6216 test_log("size align req_addr result");
6217
6218 for (int i = 0; i < num_sizes; i++) {
6219 const size_t size = sizes[i];
6220 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6221 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6222 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6223 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6224 size, alignment, p2i(req_addr), p2i(p),
6225 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6226 if (p != NULL) {
6227 assert(p == req_addr, "must be");
6228 small_page_write(p, size);
6229 os::Linux::release_memory_special_huge_tlbfs(p, size);
6230 }
6231 }
6232 }
6233
6234 // Case 3
6235 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6236 test_log("size align req_addr result");
6237
6238 for (int i = 0; i < num_sizes; i++) {
6239 const size_t size = sizes[i];
6240 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6241 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6242 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6243 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6244 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6245 // as the area around req_addr contains already existing mappings, the API should always
6246 // return NULL (as per contract, it cannot return another address)
6247 assert(p == NULL, "must be");
6248 }
6249 }
6250
6251 ::munmap(mapping2, mapping_size);
6252
6253 }
6254
6255 static void test_reserve_memory_special_huge_tlbfs() {
6256 if (!UseHugeTLBFS) {
6257 return;
6258 }
6259
6260 test_reserve_memory_special_huge_tlbfs_only();
6261 test_reserve_memory_special_huge_tlbfs_mixed();
6262 }
6263
6264 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6265 if (!UseSHM) {
6266 return;
6267 }
6268
6269 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6270
6271 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6272
6273 if (addr != NULL) {
6274 assert(is_ptr_aligned(addr, alignment), "Check");
6275 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6276
6277 small_page_write(addr, size);
6278
6279 os::Linux::release_memory_special_shm(addr, size);
6280 }
6281 }
6282
6283 static void test_reserve_memory_special_shm() {
6284 size_t lp = os::large_page_size();
6285 size_t ag = os::vm_allocation_granularity();
6286
6287 for (size_t size = ag; size < lp * 3; size += ag) {
6288 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6289 test_reserve_memory_special_shm(size, alignment);
6290 }
6291 }
6292 }
6293
6294 static void test() {
6295 test_reserve_memory_special_huge_tlbfs();
6296 test_reserve_memory_special_shm();
6297 }
6298 };
6299
6300 void TestReserveMemorySpecial_test() {
6301 TestReserveMemorySpecial::test();
6302 }
6303
6304 #endif