1 /*
2 * Copyright (c) 1999, 2017, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 // no precompiled headers
26 #include "jvm.h"
27 #include "classfile/classLoader.hpp"
28 #include "classfile/systemDictionary.hpp"
29 #include "classfile/vmSymbols.hpp"
30 #include "code/icBuffer.hpp"
31 #include "code/vtableStubs.hpp"
32 #include "compiler/compileBroker.hpp"
33 #include "compiler/disassembler.hpp"
34 #include "interpreter/interpreter.hpp"
35 #include "logging/log.hpp"
36 #include "memory/allocation.inline.hpp"
37 #include "memory/filemap.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "os_linux.inline.hpp"
40 #include "os_share_linux.hpp"
41 #include "osContainer_linux.hpp"
42 #include "prims/jniFastGetField.hpp"
43 #include "prims/jvm_misc.hpp"
44 #include "runtime/arguments.hpp"
45 #include "runtime/atomic.hpp"
46 #include "runtime/extendedPC.hpp"
47 #include "runtime/globals.hpp"
48 #include "runtime/interfaceSupport.hpp"
49 #include "runtime/init.hpp"
50 #include "runtime/java.hpp"
51 #include "runtime/javaCalls.hpp"
52 #include "runtime/mutexLocker.hpp"
53 #include "runtime/objectMonitor.hpp"
54 #include "runtime/orderAccess.inline.hpp"
55 #include "runtime/osThread.hpp"
56 #include "runtime/perfMemory.hpp"
57 #include "runtime/sharedRuntime.hpp"
58 #include "runtime/statSampler.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "runtime/thread.inline.hpp"
61 #include "runtime/threadCritical.hpp"
62 #include "runtime/threadSMR.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/align.hpp"
69 #include "utilities/decoder.hpp"
70 #include "utilities/defaultStream.hpp"
71 #include "utilities/events.hpp"
72 #include "utilities/elfFile.hpp"
73 #include "utilities/growableArray.hpp"
74 #include "utilities/macros.hpp"
75 #include "utilities/vmError.hpp"
76
77 // put OS-includes here
78 # include <sys/types.h>
79 # include <sys/mman.h>
80 # include <sys/stat.h>
81 # include <sys/select.h>
82 # include <pthread.h>
83 # include <signal.h>
84 # include <errno.h>
85 # include <dlfcn.h>
86 # include <stdio.h>
87 # include <unistd.h>
88 # include <sys/resource.h>
89 # include <pthread.h>
90 # include <sys/stat.h>
91 # include <sys/time.h>
92 # include <sys/times.h>
93 # include <sys/utsname.h>
94 # include <sys/socket.h>
95 # include <sys/wait.h>
96 # include <pwd.h>
97 # include <poll.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 #define DAX_SHARED_BIT (1 << 8)
133 ////////////////////////////////////////////////////////////////////////////////
134 // global variables
135 julong os::Linux::_physical_memory = 0;
136
137 address os::Linux::_initial_thread_stack_bottom = NULL;
138 uintptr_t os::Linux::_initial_thread_stack_size = 0;
139
140 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
141 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
142 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
143 Mutex* os::Linux::_createThread_lock = NULL;
144 pthread_t os::Linux::_main_thread;
145 int os::Linux::_page_size = -1;
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
151 static jlong initial_time_count=0;
152
153 static int clock_tics_per_sec = 100;
154
155 // For diagnostics to print a message once. see run_periodic_checks
156 static sigset_t check_signal_done;
157 static bool check_signals = true;
158
159 // Signal number used to suspend/resume a thread
160
161 // do not use any signal number less than SIGSEGV, see 4355769
162 static int SR_signum = SIGUSR2;
163 sigset_t SR_sigset;
164
165 // utility functions
166
167 static int SR_initialize();
168
169 julong os::available_memory() {
170 return Linux::available_memory();
171 }
172
173 julong os::Linux::available_memory() {
174 // values in struct sysinfo are "unsigned long"
175 struct sysinfo si;
176 julong avail_mem;
177
178 if (OSContainer::is_containerized()) {
179 jlong mem_limit, mem_usage;
180 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
181 if ((mem_usage = OSContainer::memory_usage_in_bytes()) > 0) {
182 if (mem_limit > mem_usage) {
183 avail_mem = (julong)mem_limit - (julong)mem_usage;
184 } else {
185 avail_mem = 0;
186 }
187 log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
188 return avail_mem;
189 } else {
190 log_debug(os,container)("container memory usage call failed: " JLONG_FORMAT, mem_usage);
191 }
192 } else {
193 log_debug(os,container)("container memory unlimited or failed: " JLONG_FORMAT, mem_limit);
194 }
195 }
196
197 sysinfo(&si);
198 avail_mem = (julong)si.freeram * si.mem_unit;
199 log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
200 return avail_mem;
201 }
202
203 julong os::physical_memory() {
204 if (OSContainer::is_containerized()) {
205 jlong mem_limit;
206 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
207 log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
208 return (julong)mem_limit;
209 } else {
210 if (mem_limit == OSCONTAINER_ERROR) {
211 log_debug(os,container)("container memory limit call failed");
212 }
213 if (mem_limit == -1) {
214 log_debug(os,container)("container memory unlimited, using host value");
215 }
216 }
217 }
218
219 jlong phys_mem = Linux::physical_memory();
220 log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
221 return phys_mem;
222 }
223
224 // Return true if user is running as root.
225
226 bool os::have_special_privileges() {
227 static bool init = false;
228 static bool privileges = false;
229 if (!init) {
230 privileges = (getuid() != geteuid()) || (getgid() != getegid());
231 init = true;
232 }
233 return privileges;
234 }
235
236
237 #ifndef SYS_gettid
238 // i386: 224, ia64: 1105, amd64: 186, sparc 143
239 #ifdef __ia64__
240 #define SYS_gettid 1105
241 #else
242 #ifdef __i386__
243 #define SYS_gettid 224
244 #else
245 #ifdef __amd64__
246 #define SYS_gettid 186
247 #else
248 #ifdef __sparc__
249 #define SYS_gettid 143
250 #else
251 #error define gettid for the arch
252 #endif
253 #endif
254 #endif
255 #endif
256 #endif
257
258
259 // pid_t gettid()
260 //
261 // Returns the kernel thread id of the currently running thread. Kernel
262 // thread id is used to access /proc.
263 pid_t os::Linux::gettid() {
264 int rslt = syscall(SYS_gettid);
265 assert(rslt != -1, "must be."); // old linuxthreads implementation?
266 return (pid_t)rslt;
267 }
268
269 // Most versions of linux have a bug where the number of processors are
270 // determined by looking at the /proc file system. In a chroot environment,
271 // the system call returns 1. This causes the VM to act as if it is
272 // a single processor and elide locking (see is_MP() call).
273 static bool unsafe_chroot_detected = false;
274 static const char *unstable_chroot_error = "/proc file system not found.\n"
275 "Java may be unstable running multithreaded in a chroot "
276 "environment on Linux when /proc filesystem is not mounted.";
277
278 void os::Linux::initialize_system_info() {
279 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
280 if (processor_count() == 1) {
281 pid_t pid = os::Linux::gettid();
282 char fname[32];
283 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
284 FILE *fp = fopen(fname, "r");
285 if (fp == NULL) {
286 unsafe_chroot_detected = true;
287 } else {
288 fclose(fp);
289 }
290 }
291 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
292 assert(processor_count() > 0, "linux error");
293 }
294
295 void os::init_system_properties_values() {
296 // The next steps are taken in the product version:
297 //
298 // Obtain the JAVA_HOME value from the location of libjvm.so.
299 // This library should be located at:
300 // <JAVA_HOME>/lib/{client|server}/libjvm.so.
301 //
302 // If "/jre/lib/" appears at the right place in the path, then we
303 // assume libjvm.so is installed in a JDK and we use this path.
304 //
305 // Otherwise exit with message: "Could not create the Java virtual machine."
306 //
307 // The following extra steps are taken in the debugging version:
308 //
309 // If "/jre/lib/" does NOT appear at the right place in the path
310 // instead of exit check for $JAVA_HOME environment variable.
311 //
312 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
313 // then we append a fake suffix "hotspot/libjvm.so" to this path so
314 // it looks like libjvm.so is installed there
315 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
316 //
317 // Otherwise exit.
318 //
319 // Important note: if the location of libjvm.so changes this
320 // code needs to be changed accordingly.
321
322 // See ld(1):
323 // The linker uses the following search paths to locate required
324 // shared libraries:
325 // 1: ...
326 // ...
327 // 7: The default directories, normally /lib and /usr/lib.
328 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
329 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
330 #else
331 #define DEFAULT_LIBPATH "/lib:/usr/lib"
332 #endif
333
334 // Base path of extensions installed on the system.
335 #define SYS_EXT_DIR "/usr/java/packages"
336 #define EXTENSIONS_DIR "/lib/ext"
337
338 // Buffer that fits several sprintfs.
339 // Note that the space for the colon and the trailing null are provided
340 // by the nulls included by the sizeof operator.
341 const size_t bufsize =
342 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
343 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
344 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
345
346 // sysclasspath, java_home, dll_dir
347 {
348 char *pslash;
349 os::jvm_path(buf, bufsize);
350
351 // Found the full path to libjvm.so.
352 // Now cut the path to <java_home>/jre if we can.
353 pslash = strrchr(buf, '/');
354 if (pslash != NULL) {
355 *pslash = '\0'; // Get rid of /libjvm.so.
356 }
357 pslash = strrchr(buf, '/');
358 if (pslash != NULL) {
359 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
360 }
361 Arguments::set_dll_dir(buf);
362
363 if (pslash != NULL) {
364 pslash = strrchr(buf, '/');
365 if (pslash != NULL) {
366 *pslash = '\0'; // Get rid of /lib.
367 }
368 }
369 Arguments::set_java_home(buf);
370 set_boot_path('/', ':');
371 }
372
373 // Where to look for native libraries.
374 //
375 // Note: Due to a legacy implementation, most of the library path
376 // is set in the launcher. This was to accomodate linking restrictions
377 // on legacy Linux implementations (which are no longer supported).
378 // Eventually, all the library path setting will be done here.
379 //
380 // However, to prevent the proliferation of improperly built native
381 // libraries, the new path component /usr/java/packages is added here.
382 // Eventually, all the library path setting will be done here.
383 {
384 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
385 // should always exist (until the legacy problem cited above is
386 // addressed).
387 const char *v = ::getenv("LD_LIBRARY_PATH");
388 const char *v_colon = ":";
389 if (v == NULL) { v = ""; v_colon = ""; }
390 // That's +1 for the colon and +1 for the trailing '\0'.
391 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
392 strlen(v) + 1 +
393 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
394 mtInternal);
395 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
396 Arguments::set_library_path(ld_library_path);
397 FREE_C_HEAP_ARRAY(char, ld_library_path);
398 }
399
400 // Extensions directories.
401 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
402 Arguments::set_ext_dirs(buf);
403
404 FREE_C_HEAP_ARRAY(char, buf);
405
406 #undef DEFAULT_LIBPATH
407 #undef SYS_EXT_DIR
408 #undef EXTENSIONS_DIR
409 }
410
411 ////////////////////////////////////////////////////////////////////////////////
412 // breakpoint support
413
414 void os::breakpoint() {
415 BREAKPOINT;
416 }
417
418 extern "C" void breakpoint() {
419 // use debugger to set breakpoint here
420 }
421
422 ////////////////////////////////////////////////////////////////////////////////
423 // signal support
424
425 debug_only(static bool signal_sets_initialized = false);
426 static sigset_t unblocked_sigs, vm_sigs;
427
428 bool os::Linux::is_sig_ignored(int sig) {
429 struct sigaction oact;
430 sigaction(sig, (struct sigaction*)NULL, &oact);
431 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
432 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
433 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
434 return true;
435 } else {
436 return false;
437 }
438 }
439
440 void os::Linux::signal_sets_init() {
441 // Should also have an assertion stating we are still single-threaded.
442 assert(!signal_sets_initialized, "Already initialized");
443 // Fill in signals that are necessarily unblocked for all threads in
444 // the VM. Currently, we unblock the following signals:
445 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
446 // by -Xrs (=ReduceSignalUsage));
447 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
448 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
449 // the dispositions or masks wrt these signals.
450 // Programs embedding the VM that want to use the above signals for their
451 // own purposes must, at this time, use the "-Xrs" option to prevent
452 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
453 // (See bug 4345157, and other related bugs).
454 // In reality, though, unblocking these signals is really a nop, since
455 // these signals are not blocked by default.
456 sigemptyset(&unblocked_sigs);
457 sigaddset(&unblocked_sigs, SIGILL);
458 sigaddset(&unblocked_sigs, SIGSEGV);
459 sigaddset(&unblocked_sigs, SIGBUS);
460 sigaddset(&unblocked_sigs, SIGFPE);
461 #if defined(PPC64)
462 sigaddset(&unblocked_sigs, SIGTRAP);
463 #endif
464 sigaddset(&unblocked_sigs, SR_signum);
465
466 if (!ReduceSignalUsage) {
467 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
468 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
469 }
470 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
471 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
472 }
473 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
474 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
475 }
476 }
477 // Fill in signals that are blocked by all but the VM thread.
478 sigemptyset(&vm_sigs);
479 if (!ReduceSignalUsage) {
480 sigaddset(&vm_sigs, BREAK_SIGNAL);
481 }
482 debug_only(signal_sets_initialized = true);
483
484 }
485
486 // These are signals that are unblocked while a thread is running Java.
487 // (For some reason, they get blocked by default.)
488 sigset_t* os::Linux::unblocked_signals() {
489 assert(signal_sets_initialized, "Not initialized");
490 return &unblocked_sigs;
491 }
492
493 // These are the signals that are blocked while a (non-VM) thread is
494 // running Java. Only the VM thread handles these signals.
495 sigset_t* os::Linux::vm_signals() {
496 assert(signal_sets_initialized, "Not initialized");
497 return &vm_sigs;
498 }
499
500 void os::Linux::hotspot_sigmask(Thread* thread) {
501
502 //Save caller's signal mask before setting VM signal mask
503 sigset_t caller_sigmask;
504 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
505
506 OSThread* osthread = thread->osthread();
507 osthread->set_caller_sigmask(caller_sigmask);
508
509 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
510
511 if (!ReduceSignalUsage) {
512 if (thread->is_VM_thread()) {
513 // Only the VM thread handles BREAK_SIGNAL ...
514 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
515 } else {
516 // ... all other threads block BREAK_SIGNAL
517 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
518 }
519 }
520 }
521
522 //////////////////////////////////////////////////////////////////////////////
523 // detecting pthread library
524
525 void os::Linux::libpthread_init() {
526 // Save glibc and pthread version strings.
527 #if !defined(_CS_GNU_LIBC_VERSION) || \
528 !defined(_CS_GNU_LIBPTHREAD_VERSION)
529 #error "glibc too old (< 2.3.2)"
530 #endif
531
532 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
533 assert(n > 0, "cannot retrieve glibc version");
534 char *str = (char *)malloc(n, mtInternal);
535 confstr(_CS_GNU_LIBC_VERSION, str, n);
536 os::Linux::set_glibc_version(str);
537
538 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
539 assert(n > 0, "cannot retrieve pthread version");
540 str = (char *)malloc(n, mtInternal);
541 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
542 os::Linux::set_libpthread_version(str);
543 }
544
545 /////////////////////////////////////////////////////////////////////////////
546 // thread stack expansion
547
548 // os::Linux::manually_expand_stack() takes care of expanding the thread
549 // stack. Note that this is normally not needed: pthread stacks allocate
550 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
551 // committed. Therefore it is not necessary to expand the stack manually.
552 //
553 // Manually expanding the stack was historically needed on LinuxThreads
554 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
555 // it is kept to deal with very rare corner cases:
556 //
557 // For one, user may run the VM on an own implementation of threads
558 // whose stacks are - like the old LinuxThreads - implemented using
559 // mmap(MAP_GROWSDOWN).
560 //
561 // Also, this coding may be needed if the VM is running on the primordial
562 // thread. Normally we avoid running on the primordial thread; however,
563 // user may still invoke the VM on the primordial thread.
564 //
565 // The following historical comment describes the details about running
566 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
567
568
569 // Force Linux kernel to expand current thread stack. If "bottom" is close
570 // to the stack guard, caller should block all signals.
571 //
572 // MAP_GROWSDOWN:
573 // A special mmap() flag that is used to implement thread stacks. It tells
574 // kernel that the memory region should extend downwards when needed. This
575 // allows early versions of LinuxThreads to only mmap the first few pages
576 // when creating a new thread. Linux kernel will automatically expand thread
577 // stack as needed (on page faults).
578 //
579 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
580 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
581 // region, it's hard to tell if the fault is due to a legitimate stack
582 // access or because of reading/writing non-exist memory (e.g. buffer
583 // overrun). As a rule, if the fault happens below current stack pointer,
584 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
585 // application (see Linux kernel fault.c).
586 //
587 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
588 // stack overflow detection.
589 //
590 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
591 // not use MAP_GROWSDOWN.
592 //
593 // To get around the problem and allow stack banging on Linux, we need to
594 // manually expand thread stack after receiving the SIGSEGV.
595 //
596 // There are two ways to expand thread stack to address "bottom", we used
597 // both of them in JVM before 1.5:
598 // 1. adjust stack pointer first so that it is below "bottom", and then
599 // touch "bottom"
600 // 2. mmap() the page in question
601 //
602 // Now alternate signal stack is gone, it's harder to use 2. For instance,
603 // if current sp is already near the lower end of page 101, and we need to
604 // call mmap() to map page 100, it is possible that part of the mmap() frame
605 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
606 // That will destroy the mmap() frame and cause VM to crash.
607 //
608 // The following code works by adjusting sp first, then accessing the "bottom"
609 // page to force a page fault. Linux kernel will then automatically expand the
610 // stack mapping.
611 //
612 // _expand_stack_to() assumes its frame size is less than page size, which
613 // should always be true if the function is not inlined.
614
615 static void NOINLINE _expand_stack_to(address bottom) {
616 address sp;
617 size_t size;
618 volatile char *p;
619
620 // Adjust bottom to point to the largest address within the same page, it
621 // gives us a one-page buffer if alloca() allocates slightly more memory.
622 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
623 bottom += os::Linux::page_size() - 1;
624
625 // sp might be slightly above current stack pointer; if that's the case, we
626 // will alloca() a little more space than necessary, which is OK. Don't use
627 // os::current_stack_pointer(), as its result can be slightly below current
628 // stack pointer, causing us to not alloca enough to reach "bottom".
629 sp = (address)&sp;
630
631 if (sp > bottom) {
632 size = sp - bottom;
633 p = (volatile char *)alloca(size);
634 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
635 p[0] = '\0';
636 }
637 }
638
639 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
640 assert(t!=NULL, "just checking");
641 assert(t->osthread()->expanding_stack(), "expand should be set");
642 assert(t->stack_base() != NULL, "stack_base was not initialized");
643
644 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
645 sigset_t mask_all, old_sigset;
646 sigfillset(&mask_all);
647 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
648 _expand_stack_to(addr);
649 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
650 return true;
651 }
652 return false;
653 }
654
655 //////////////////////////////////////////////////////////////////////////////
656 // create new thread
657
658 // Thread start routine for all newly created threads
659 static void *thread_native_entry(Thread *thread) {
660 // Try to randomize the cache line index of hot stack frames.
661 // This helps when threads of the same stack traces evict each other's
662 // cache lines. The threads can be either from the same JVM instance, or
663 // from different JVM instances. The benefit is especially true for
664 // processors with hyperthreading technology.
665 static int counter = 0;
666 int pid = os::current_process_id();
667 alloca(((pid ^ counter++) & 7) * 128);
668
669 thread->initialize_thread_current();
670
671 OSThread* osthread = thread->osthread();
672 Monitor* sync = osthread->startThread_lock();
673
674 osthread->set_thread_id(os::current_thread_id());
675
676 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
677 os::current_thread_id(), (uintx) pthread_self());
678
679 if (UseNUMA) {
680 int lgrp_id = os::numa_get_group_id();
681 if (lgrp_id != -1) {
682 thread->set_lgrp_id(lgrp_id);
683 }
684 }
685 // initialize signal mask for this thread
686 os::Linux::hotspot_sigmask(thread);
687
688 // initialize floating point control register
689 os::Linux::init_thread_fpu_state();
690
691 // handshaking with parent thread
692 {
693 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
694
695 // notify parent thread
696 osthread->set_state(INITIALIZED);
697 sync->notify_all();
698
699 // wait until os::start_thread()
700 while (osthread->get_state() == INITIALIZED) {
701 sync->wait(Mutex::_no_safepoint_check_flag);
702 }
703 }
704
705 // call one more level start routine
706 thread->run();
707
708 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
709 os::current_thread_id(), (uintx) pthread_self());
710
711 // If a thread has not deleted itself ("delete this") as part of its
712 // termination sequence, we have to ensure thread-local-storage is
713 // cleared before we actually terminate. No threads should ever be
714 // deleted asynchronously with respect to their termination.
715 if (Thread::current_or_null_safe() != NULL) {
716 assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
717 thread->clear_thread_current();
718 }
719
720 return 0;
721 }
722
723 bool os::create_thread(Thread* thread, ThreadType thr_type,
724 size_t req_stack_size) {
725 assert(thread->osthread() == NULL, "caller responsible");
726
727 // Allocate the OSThread object
728 OSThread* osthread = new OSThread(NULL, NULL);
729 if (osthread == NULL) {
730 return false;
731 }
732
733 // set the correct thread state
734 osthread->set_thread_type(thr_type);
735
736 // Initial state is ALLOCATED but not INITIALIZED
737 osthread->set_state(ALLOCATED);
738
739 thread->set_osthread(osthread);
740
741 // init thread attributes
742 pthread_attr_t attr;
743 pthread_attr_init(&attr);
744 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
745
746 // Calculate stack size if it's not specified by caller.
747 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
748 // In the Linux NPTL pthread implementation the guard size mechanism
749 // is not implemented properly. The posix standard requires adding
750 // the size of the guard pages to the stack size, instead Linux
751 // takes the space out of 'stacksize'. Thus we adapt the requested
752 // stack_size by the size of the guard pages to mimick proper
753 // behaviour. However, be careful not to end up with a size
754 // of zero due to overflow. Don't add the guard page in that case.
755 size_t guard_size = os::Linux::default_guard_size(thr_type);
756 if (stack_size <= SIZE_MAX - guard_size) {
757 stack_size += guard_size;
758 }
759 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
760
761 int status = pthread_attr_setstacksize(&attr, stack_size);
762 assert_status(status == 0, status, "pthread_attr_setstacksize");
763
764 // Configure glibc guard page.
765 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
766
767 ThreadState state;
768
769 {
770 pthread_t tid;
771 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
772
773 char buf[64];
774 if (ret == 0) {
775 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
776 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
777 } else {
778 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
779 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
780 }
781
782 pthread_attr_destroy(&attr);
783
784 if (ret != 0) {
785 // Need to clean up stuff we've allocated so far
786 thread->set_osthread(NULL);
787 delete osthread;
788 return false;
789 }
790
791 // Store pthread info into the OSThread
792 osthread->set_pthread_id(tid);
793
794 // Wait until child thread is either initialized or aborted
795 {
796 Monitor* sync_with_child = osthread->startThread_lock();
797 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
798 while ((state = osthread->get_state()) == ALLOCATED) {
799 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
800 }
801 }
802 }
803
804 // Aborted due to thread limit being reached
805 if (state == ZOMBIE) {
806 thread->set_osthread(NULL);
807 delete osthread;
808 return false;
809 }
810
811 // The thread is returned suspended (in state INITIALIZED),
812 // and is started higher up in the call chain
813 assert(state == INITIALIZED, "race condition");
814 return true;
815 }
816
817 /////////////////////////////////////////////////////////////////////////////
818 // attach existing thread
819
820 // bootstrap the main thread
821 bool os::create_main_thread(JavaThread* thread) {
822 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
823 return create_attached_thread(thread);
824 }
825
826 bool os::create_attached_thread(JavaThread* thread) {
827 #ifdef ASSERT
828 thread->verify_not_published();
829 #endif
830
831 // Allocate the OSThread object
832 OSThread* osthread = new OSThread(NULL, NULL);
833
834 if (osthread == NULL) {
835 return false;
836 }
837
838 // Store pthread info into the OSThread
839 osthread->set_thread_id(os::Linux::gettid());
840 osthread->set_pthread_id(::pthread_self());
841
842 // initialize floating point control register
843 os::Linux::init_thread_fpu_state();
844
845 // Initial thread state is RUNNABLE
846 osthread->set_state(RUNNABLE);
847
848 thread->set_osthread(osthread);
849
850 if (UseNUMA) {
851 int lgrp_id = os::numa_get_group_id();
852 if (lgrp_id != -1) {
853 thread->set_lgrp_id(lgrp_id);
854 }
855 }
856
857 if (os::is_primordial_thread()) {
858 // If current thread is primordial thread, its stack is mapped on demand,
859 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
860 // the entire stack region to avoid SEGV in stack banging.
861 // It is also useful to get around the heap-stack-gap problem on SuSE
862 // kernel (see 4821821 for details). We first expand stack to the top
863 // of yellow zone, then enable stack yellow zone (order is significant,
864 // enabling yellow zone first will crash JVM on SuSE Linux), so there
865 // is no gap between the last two virtual memory regions.
866
867 JavaThread *jt = (JavaThread *)thread;
868 address addr = jt->stack_reserved_zone_base();
869 assert(addr != NULL, "initialization problem?");
870 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
871
872 osthread->set_expanding_stack();
873 os::Linux::manually_expand_stack(jt, addr);
874 osthread->clear_expanding_stack();
875 }
876
877 // initialize signal mask for this thread
878 // and save the caller's signal mask
879 os::Linux::hotspot_sigmask(thread);
880
881 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
882 os::current_thread_id(), (uintx) pthread_self());
883
884 return true;
885 }
886
887 void os::pd_start_thread(Thread* thread) {
888 OSThread * osthread = thread->osthread();
889 assert(osthread->get_state() != INITIALIZED, "just checking");
890 Monitor* sync_with_child = osthread->startThread_lock();
891 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
892 sync_with_child->notify();
893 }
894
895 // Free Linux resources related to the OSThread
896 void os::free_thread(OSThread* osthread) {
897 assert(osthread != NULL, "osthread not set");
898
899 // We are told to free resources of the argument thread,
900 // but we can only really operate on the current thread.
901 assert(Thread::current()->osthread() == osthread,
902 "os::free_thread but not current thread");
903
904 #ifdef ASSERT
905 sigset_t current;
906 sigemptyset(¤t);
907 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
908 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
909 #endif
910
911 // Restore caller's signal mask
912 sigset_t sigmask = osthread->caller_sigmask();
913 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
914
915 delete osthread;
916 }
917
918 //////////////////////////////////////////////////////////////////////////////
919 // primordial thread
920
921 // Check if current thread is the primordial thread, similar to Solaris thr_main.
922 bool os::is_primordial_thread(void) {
923 char dummy;
924 // If called before init complete, thread stack bottom will be null.
925 // Can be called if fatal error occurs before initialization.
926 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
927 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
928 os::Linux::initial_thread_stack_size() != 0,
929 "os::init did not locate primordial thread's stack region");
930 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
931 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
932 os::Linux::initial_thread_stack_size()) {
933 return true;
934 } else {
935 return false;
936 }
937 }
938
939 // Find the virtual memory area that contains addr
940 static bool find_vma(address addr, address* vma_low, address* vma_high) {
941 FILE *fp = fopen("/proc/self/maps", "r");
942 if (fp) {
943 address low, high;
944 while (!feof(fp)) {
945 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
946 if (low <= addr && addr < high) {
947 if (vma_low) *vma_low = low;
948 if (vma_high) *vma_high = high;
949 fclose(fp);
950 return true;
951 }
952 }
953 for (;;) {
954 int ch = fgetc(fp);
955 if (ch == EOF || ch == (int)'\n') break;
956 }
957 }
958 fclose(fp);
959 }
960 return false;
961 }
962
963 // Locate primordial thread stack. This special handling of primordial thread stack
964 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
965 // bogus value for the primordial process thread. While the launcher has created
966 // the VM in a new thread since JDK 6, we still have to allow for the use of the
967 // JNI invocation API from a primordial thread.
968 void os::Linux::capture_initial_stack(size_t max_size) {
969
970 // max_size is either 0 (which means accept OS default for thread stacks) or
971 // a user-specified value known to be at least the minimum needed. If we
972 // are actually on the primordial thread we can make it appear that we have a
973 // smaller max_size stack by inserting the guard pages at that location. But we
974 // cannot do anything to emulate a larger stack than what has been provided by
975 // the OS or threading library. In fact if we try to use a stack greater than
976 // what is set by rlimit then we will crash the hosting process.
977
978 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
979 // If this is "unlimited" then it will be a huge value.
980 struct rlimit rlim;
981 getrlimit(RLIMIT_STACK, &rlim);
982 size_t stack_size = rlim.rlim_cur;
983
984 // 6308388: a bug in ld.so will relocate its own .data section to the
985 // lower end of primordial stack; reduce ulimit -s value a little bit
986 // so we won't install guard page on ld.so's data section.
987 // But ensure we don't underflow the stack size - allow 1 page spare
988 if (stack_size >= (size_t)(3 * page_size())) {
989 stack_size -= 2 * page_size();
990 }
991
992 // Try to figure out where the stack base (top) is. This is harder.
993 //
994 // When an application is started, glibc saves the initial stack pointer in
995 // a global variable "__libc_stack_end", which is then used by system
996 // libraries. __libc_stack_end should be pretty close to stack top. The
997 // variable is available since the very early days. However, because it is
998 // a private interface, it could disappear in the future.
999 //
1000 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1001 // to __libc_stack_end, it is very close to stack top, but isn't the real
1002 // stack top. Note that /proc may not exist if VM is running as a chroot
1003 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1004 // /proc/<pid>/stat could change in the future (though unlikely).
1005 //
1006 // We try __libc_stack_end first. If that doesn't work, look for
1007 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1008 // as a hint, which should work well in most cases.
1009
1010 uintptr_t stack_start;
1011
1012 // try __libc_stack_end first
1013 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1014 if (p && *p) {
1015 stack_start = *p;
1016 } else {
1017 // see if we can get the start_stack field from /proc/self/stat
1018 FILE *fp;
1019 int pid;
1020 char state;
1021 int ppid;
1022 int pgrp;
1023 int session;
1024 int nr;
1025 int tpgrp;
1026 unsigned long flags;
1027 unsigned long minflt;
1028 unsigned long cminflt;
1029 unsigned long majflt;
1030 unsigned long cmajflt;
1031 unsigned long utime;
1032 unsigned long stime;
1033 long cutime;
1034 long cstime;
1035 long prio;
1036 long nice;
1037 long junk;
1038 long it_real;
1039 uintptr_t start;
1040 uintptr_t vsize;
1041 intptr_t rss;
1042 uintptr_t rsslim;
1043 uintptr_t scodes;
1044 uintptr_t ecode;
1045 int i;
1046
1047 // Figure what the primordial thread stack base is. Code is inspired
1048 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1049 // followed by command name surrounded by parentheses, state, etc.
1050 char stat[2048];
1051 int statlen;
1052
1053 fp = fopen("/proc/self/stat", "r");
1054 if (fp) {
1055 statlen = fread(stat, 1, 2047, fp);
1056 stat[statlen] = '\0';
1057 fclose(fp);
1058
1059 // Skip pid and the command string. Note that we could be dealing with
1060 // weird command names, e.g. user could decide to rename java launcher
1061 // to "java 1.4.2 :)", then the stat file would look like
1062 // 1234 (java 1.4.2 :)) R ... ...
1063 // We don't really need to know the command string, just find the last
1064 // occurrence of ")" and then start parsing from there. See bug 4726580.
1065 char * s = strrchr(stat, ')');
1066
1067 i = 0;
1068 if (s) {
1069 // Skip blank chars
1070 do { s++; } while (s && isspace(*s));
1071
1072 #define _UFM UINTX_FORMAT
1073 #define _DFM INTX_FORMAT
1074
1075 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1076 // 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
1077 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,
1078 &state, // 3 %c
1079 &ppid, // 4 %d
1080 &pgrp, // 5 %d
1081 &session, // 6 %d
1082 &nr, // 7 %d
1083 &tpgrp, // 8 %d
1084 &flags, // 9 %lu
1085 &minflt, // 10 %lu
1086 &cminflt, // 11 %lu
1087 &majflt, // 12 %lu
1088 &cmajflt, // 13 %lu
1089 &utime, // 14 %lu
1090 &stime, // 15 %lu
1091 &cutime, // 16 %ld
1092 &cstime, // 17 %ld
1093 &prio, // 18 %ld
1094 &nice, // 19 %ld
1095 &junk, // 20 %ld
1096 &it_real, // 21 %ld
1097 &start, // 22 UINTX_FORMAT
1098 &vsize, // 23 UINTX_FORMAT
1099 &rss, // 24 INTX_FORMAT
1100 &rsslim, // 25 UINTX_FORMAT
1101 &scodes, // 26 UINTX_FORMAT
1102 &ecode, // 27 UINTX_FORMAT
1103 &stack_start); // 28 UINTX_FORMAT
1104 }
1105
1106 #undef _UFM
1107 #undef _DFM
1108
1109 if (i != 28 - 2) {
1110 assert(false, "Bad conversion from /proc/self/stat");
1111 // product mode - assume we are the primordial thread, good luck in the
1112 // embedded case.
1113 warning("Can't detect primordial thread stack location - bad conversion");
1114 stack_start = (uintptr_t) &rlim;
1115 }
1116 } else {
1117 // For some reason we can't open /proc/self/stat (for example, running on
1118 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1119 // most cases, so don't abort:
1120 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1121 stack_start = (uintptr_t) &rlim;
1122 }
1123 }
1124
1125 // Now we have a pointer (stack_start) very close to the stack top, the
1126 // next thing to do is to figure out the exact location of stack top. We
1127 // can find out the virtual memory area that contains stack_start by
1128 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1129 // and its upper limit is the real stack top. (again, this would fail if
1130 // running inside chroot, because /proc may not exist.)
1131
1132 uintptr_t stack_top;
1133 address low, high;
1134 if (find_vma((address)stack_start, &low, &high)) {
1135 // success, "high" is the true stack top. (ignore "low", because initial
1136 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1137 stack_top = (uintptr_t)high;
1138 } else {
1139 // failed, likely because /proc/self/maps does not exist
1140 warning("Can't detect primordial thread stack location - find_vma failed");
1141 // best effort: stack_start is normally within a few pages below the real
1142 // stack top, use it as stack top, and reduce stack size so we won't put
1143 // guard page outside stack.
1144 stack_top = stack_start;
1145 stack_size -= 16 * page_size();
1146 }
1147
1148 // stack_top could be partially down the page so align it
1149 stack_top = align_up(stack_top, page_size());
1150
1151 // Allowed stack value is minimum of max_size and what we derived from rlimit
1152 if (max_size > 0) {
1153 _initial_thread_stack_size = MIN2(max_size, stack_size);
1154 } else {
1155 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1156 // clamp it at 8MB as we do on Solaris
1157 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1158 }
1159 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1160 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1161
1162 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1163
1164 if (log_is_enabled(Info, os, thread)) {
1165 // See if we seem to be on primordial process thread
1166 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1167 uintptr_t(&rlim) < stack_top;
1168
1169 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1170 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1171 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1172 stack_top, intptr_t(_initial_thread_stack_bottom));
1173 }
1174 }
1175
1176 ////////////////////////////////////////////////////////////////////////////////
1177 // time support
1178
1179 // Time since start-up in seconds to a fine granularity.
1180 // Used by VMSelfDestructTimer and the MemProfiler.
1181 double os::elapsedTime() {
1182
1183 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1184 }
1185
1186 jlong os::elapsed_counter() {
1187 return javaTimeNanos() - initial_time_count;
1188 }
1189
1190 jlong os::elapsed_frequency() {
1191 return NANOSECS_PER_SEC; // nanosecond resolution
1192 }
1193
1194 bool os::supports_vtime() { return true; }
1195 bool os::enable_vtime() { return false; }
1196 bool os::vtime_enabled() { return false; }
1197
1198 double os::elapsedVTime() {
1199 struct rusage usage;
1200 int retval = getrusage(RUSAGE_THREAD, &usage);
1201 if (retval == 0) {
1202 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);
1203 } else {
1204 // better than nothing, but not much
1205 return elapsedTime();
1206 }
1207 }
1208
1209 jlong os::javaTimeMillis() {
1210 timeval time;
1211 int status = gettimeofday(&time, NULL);
1212 assert(status != -1, "linux error");
1213 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1214 }
1215
1216 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1217 timeval time;
1218 int status = gettimeofday(&time, NULL);
1219 assert(status != -1, "linux error");
1220 seconds = jlong(time.tv_sec);
1221 nanos = jlong(time.tv_usec) * 1000;
1222 }
1223
1224
1225 #ifndef CLOCK_MONOTONIC
1226 #define CLOCK_MONOTONIC (1)
1227 #endif
1228
1229 void os::Linux::clock_init() {
1230 // we do dlopen's in this particular order due to bug in linux
1231 // dynamical loader (see 6348968) leading to crash on exit
1232 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1233 if (handle == NULL) {
1234 handle = dlopen("librt.so", RTLD_LAZY);
1235 }
1236
1237 if (handle) {
1238 int (*clock_getres_func)(clockid_t, struct timespec*) =
1239 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1240 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1241 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1242 if (clock_getres_func && clock_gettime_func) {
1243 // See if monotonic clock is supported by the kernel. Note that some
1244 // early implementations simply return kernel jiffies (updated every
1245 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1246 // for nano time (though the monotonic property is still nice to have).
1247 // It's fixed in newer kernels, however clock_getres() still returns
1248 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1249 // resolution for now. Hopefully as people move to new kernels, this
1250 // won't be a problem.
1251 struct timespec res;
1252 struct timespec tp;
1253 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1254 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1255 // yes, monotonic clock is supported
1256 _clock_gettime = clock_gettime_func;
1257 return;
1258 } else {
1259 // close librt if there is no monotonic clock
1260 dlclose(handle);
1261 }
1262 }
1263 }
1264 warning("No monotonic clock was available - timed services may " \
1265 "be adversely affected if the time-of-day clock changes");
1266 }
1267
1268 #ifndef SYS_clock_getres
1269 #if defined(X86) || defined(PPC64) || defined(S390)
1270 #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261)
1271 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1272 #else
1273 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1274 #define sys_clock_getres(x,y) -1
1275 #endif
1276 #else
1277 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1278 #endif
1279
1280 void os::Linux::fast_thread_clock_init() {
1281 if (!UseLinuxPosixThreadCPUClocks) {
1282 return;
1283 }
1284 clockid_t clockid;
1285 struct timespec tp;
1286 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1287 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1288
1289 // Switch to using fast clocks for thread cpu time if
1290 // the sys_clock_getres() returns 0 error code.
1291 // Note, that some kernels may support the current thread
1292 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1293 // returned by the pthread_getcpuclockid().
1294 // If the fast Posix clocks are supported then the sys_clock_getres()
1295 // must return at least tp.tv_sec == 0 which means a resolution
1296 // better than 1 sec. This is extra check for reliability.
1297
1298 if (pthread_getcpuclockid_func &&
1299 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1300 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1301 _supports_fast_thread_cpu_time = true;
1302 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1303 }
1304 }
1305
1306 jlong os::javaTimeNanos() {
1307 if (os::supports_monotonic_clock()) {
1308 struct timespec tp;
1309 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1310 assert(status == 0, "gettime error");
1311 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1312 return result;
1313 } else {
1314 timeval time;
1315 int status = gettimeofday(&time, NULL);
1316 assert(status != -1, "linux error");
1317 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1318 return 1000 * usecs;
1319 }
1320 }
1321
1322 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1323 if (os::supports_monotonic_clock()) {
1324 info_ptr->max_value = ALL_64_BITS;
1325
1326 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1327 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1328 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1329 } else {
1330 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1331 info_ptr->max_value = ALL_64_BITS;
1332
1333 // gettimeofday is a real time clock so it skips
1334 info_ptr->may_skip_backward = true;
1335 info_ptr->may_skip_forward = true;
1336 }
1337
1338 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1339 }
1340
1341 // Return the real, user, and system times in seconds from an
1342 // arbitrary fixed point in the past.
1343 bool os::getTimesSecs(double* process_real_time,
1344 double* process_user_time,
1345 double* process_system_time) {
1346 struct tms ticks;
1347 clock_t real_ticks = times(&ticks);
1348
1349 if (real_ticks == (clock_t) (-1)) {
1350 return false;
1351 } else {
1352 double ticks_per_second = (double) clock_tics_per_sec;
1353 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1354 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1355 *process_real_time = ((double) real_ticks) / ticks_per_second;
1356
1357 return true;
1358 }
1359 }
1360
1361
1362 char * os::local_time_string(char *buf, size_t buflen) {
1363 struct tm t;
1364 time_t long_time;
1365 time(&long_time);
1366 localtime_r(&long_time, &t);
1367 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1368 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1369 t.tm_hour, t.tm_min, t.tm_sec);
1370 return buf;
1371 }
1372
1373 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1374 return localtime_r(clock, res);
1375 }
1376
1377 ////////////////////////////////////////////////////////////////////////////////
1378 // runtime exit support
1379
1380 // Note: os::shutdown() might be called very early during initialization, or
1381 // called from signal handler. Before adding something to os::shutdown(), make
1382 // sure it is async-safe and can handle partially initialized VM.
1383 void os::shutdown() {
1384
1385 // allow PerfMemory to attempt cleanup of any persistent resources
1386 perfMemory_exit();
1387
1388 // needs to remove object in file system
1389 AttachListener::abort();
1390
1391 // flush buffered output, finish log files
1392 ostream_abort();
1393
1394 // Check for abort hook
1395 abort_hook_t abort_hook = Arguments::abort_hook();
1396 if (abort_hook != NULL) {
1397 abort_hook();
1398 }
1399
1400 }
1401
1402 // Note: os::abort() might be called very early during initialization, or
1403 // called from signal handler. Before adding something to os::abort(), make
1404 // sure it is async-safe and can handle partially initialized VM.
1405 void os::abort(bool dump_core, void* siginfo, const void* context) {
1406 os::shutdown();
1407 if (dump_core) {
1408 #ifndef PRODUCT
1409 fdStream out(defaultStream::output_fd());
1410 out.print_raw("Current thread is ");
1411 char buf[16];
1412 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1413 out.print_raw_cr(buf);
1414 out.print_raw_cr("Dumping core ...");
1415 #endif
1416 ::abort(); // dump core
1417 }
1418
1419 ::exit(1);
1420 }
1421
1422 // Die immediately, no exit hook, no abort hook, no cleanup.
1423 void os::die() {
1424 ::abort();
1425 }
1426
1427
1428 // This method is a copy of JDK's sysGetLastErrorString
1429 // from src/solaris/hpi/src/system_md.c
1430
1431 size_t os::lasterror(char *buf, size_t len) {
1432 if (errno == 0) return 0;
1433
1434 const char *s = os::strerror(errno);
1435 size_t n = ::strlen(s);
1436 if (n >= len) {
1437 n = len - 1;
1438 }
1439 ::strncpy(buf, s, n);
1440 buf[n] = '\0';
1441 return n;
1442 }
1443
1444 // thread_id is kernel thread id (similar to Solaris LWP id)
1445 intx os::current_thread_id() { return os::Linux::gettid(); }
1446 int os::current_process_id() {
1447 return ::getpid();
1448 }
1449
1450 // DLL functions
1451
1452 const char* os::dll_file_extension() { return ".so"; }
1453
1454 // This must be hard coded because it's the system's temporary
1455 // directory not the java application's temp directory, ala java.io.tmpdir.
1456 const char* os::get_temp_directory() { return "/tmp"; }
1457
1458 static bool file_exists(const char* filename) {
1459 struct stat statbuf;
1460 if (filename == NULL || strlen(filename) == 0) {
1461 return false;
1462 }
1463 return os::stat(filename, &statbuf) == 0;
1464 }
1465
1466 // check if addr is inside libjvm.so
1467 bool os::address_is_in_vm(address addr) {
1468 static address libjvm_base_addr;
1469 Dl_info dlinfo;
1470
1471 if (libjvm_base_addr == NULL) {
1472 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1473 libjvm_base_addr = (address)dlinfo.dli_fbase;
1474 }
1475 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1476 }
1477
1478 if (dladdr((void *)addr, &dlinfo) != 0) {
1479 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1480 }
1481
1482 return false;
1483 }
1484
1485 bool os::dll_address_to_function_name(address addr, char *buf,
1486 int buflen, int *offset,
1487 bool demangle) {
1488 // buf is not optional, but offset is optional
1489 assert(buf != NULL, "sanity check");
1490
1491 Dl_info dlinfo;
1492
1493 if (dladdr((void*)addr, &dlinfo) != 0) {
1494 // see if we have a matching symbol
1495 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1496 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1497 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1498 }
1499 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1500 return true;
1501 }
1502 // no matching symbol so try for just file info
1503 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1504 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1505 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1506 return true;
1507 }
1508 }
1509 }
1510
1511 buf[0] = '\0';
1512 if (offset != NULL) *offset = -1;
1513 return false;
1514 }
1515
1516 struct _address_to_library_name {
1517 address addr; // input : memory address
1518 size_t buflen; // size of fname
1519 char* fname; // output: library name
1520 address base; // library base addr
1521 };
1522
1523 static int address_to_library_name_callback(struct dl_phdr_info *info,
1524 size_t size, void *data) {
1525 int i;
1526 bool found = false;
1527 address libbase = NULL;
1528 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1529
1530 // iterate through all loadable segments
1531 for (i = 0; i < info->dlpi_phnum; i++) {
1532 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1533 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1534 // base address of a library is the lowest address of its loaded
1535 // segments.
1536 if (libbase == NULL || libbase > segbase) {
1537 libbase = segbase;
1538 }
1539 // see if 'addr' is within current segment
1540 if (segbase <= d->addr &&
1541 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1542 found = true;
1543 }
1544 }
1545 }
1546
1547 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1548 // so dll_address_to_library_name() can fall through to use dladdr() which
1549 // can figure out executable name from argv[0].
1550 if (found && info->dlpi_name && info->dlpi_name[0]) {
1551 d->base = libbase;
1552 if (d->fname) {
1553 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1554 }
1555 return 1;
1556 }
1557 return 0;
1558 }
1559
1560 bool os::dll_address_to_library_name(address addr, char* buf,
1561 int buflen, int* offset) {
1562 // buf is not optional, but offset is optional
1563 assert(buf != NULL, "sanity check");
1564
1565 Dl_info dlinfo;
1566 struct _address_to_library_name data;
1567
1568 // There is a bug in old glibc dladdr() implementation that it could resolve
1569 // to wrong library name if the .so file has a base address != NULL. Here
1570 // we iterate through the program headers of all loaded libraries to find
1571 // out which library 'addr' really belongs to. This workaround can be
1572 // removed once the minimum requirement for glibc is moved to 2.3.x.
1573 data.addr = addr;
1574 data.fname = buf;
1575 data.buflen = buflen;
1576 data.base = NULL;
1577 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1578
1579 if (rslt) {
1580 // buf already contains library name
1581 if (offset) *offset = addr - data.base;
1582 return true;
1583 }
1584 if (dladdr((void*)addr, &dlinfo) != 0) {
1585 if (dlinfo.dli_fname != NULL) {
1586 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1587 }
1588 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1589 *offset = addr - (address)dlinfo.dli_fbase;
1590 }
1591 return true;
1592 }
1593
1594 buf[0] = '\0';
1595 if (offset) *offset = -1;
1596 return false;
1597 }
1598
1599 // Loads .dll/.so and
1600 // in case of error it checks if .dll/.so was built for the
1601 // same architecture as Hotspot is running on
1602
1603
1604 // Remember the stack's state. The Linux dynamic linker will change
1605 // the stack to 'executable' at most once, so we must safepoint only once.
1606 bool os::Linux::_stack_is_executable = false;
1607
1608 // VM operation that loads a library. This is necessary if stack protection
1609 // of the Java stacks can be lost during loading the library. If we
1610 // do not stop the Java threads, they can stack overflow before the stacks
1611 // are protected again.
1612 class VM_LinuxDllLoad: public VM_Operation {
1613 private:
1614 const char *_filename;
1615 char *_ebuf;
1616 int _ebuflen;
1617 void *_lib;
1618 public:
1619 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1620 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1621 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1622 void doit() {
1623 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1624 os::Linux::_stack_is_executable = true;
1625 }
1626 void* loaded_library() { return _lib; }
1627 };
1628
1629 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1630 void * result = NULL;
1631 bool load_attempted = false;
1632
1633 // Check whether the library to load might change execution rights
1634 // of the stack. If they are changed, the protection of the stack
1635 // guard pages will be lost. We need a safepoint to fix this.
1636 //
1637 // See Linux man page execstack(8) for more info.
1638 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1639 if (!ElfFile::specifies_noexecstack(filename)) {
1640 if (!is_init_completed()) {
1641 os::Linux::_stack_is_executable = true;
1642 // This is OK - No Java threads have been created yet, and hence no
1643 // stack guard pages to fix.
1644 //
1645 // This should happen only when you are building JDK7 using a very
1646 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1647 //
1648 // Dynamic loader will make all stacks executable after
1649 // this function returns, and will not do that again.
1650 #ifdef ASSERT
1651 ThreadsListHandle tlh;
1652 assert(tlh.length() == 0, "no Java threads should exist yet.");
1653 #endif
1654 } else {
1655 warning("You have loaded library %s which might have disabled stack guard. "
1656 "The VM will try to fix the stack guard now.\n"
1657 "It's highly recommended that you fix the library with "
1658 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1659 filename);
1660
1661 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1662 JavaThread *jt = JavaThread::current();
1663 if (jt->thread_state() != _thread_in_native) {
1664 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1665 // that requires ExecStack. Cannot enter safe point. Let's give up.
1666 warning("Unable to fix stack guard. Giving up.");
1667 } else {
1668 if (!LoadExecStackDllInVMThread) {
1669 // This is for the case where the DLL has an static
1670 // constructor function that executes JNI code. We cannot
1671 // load such DLLs in the VMThread.
1672 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1673 }
1674
1675 ThreadInVMfromNative tiv(jt);
1676 debug_only(VMNativeEntryWrapper vew;)
1677
1678 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1679 VMThread::execute(&op);
1680 if (LoadExecStackDllInVMThread) {
1681 result = op.loaded_library();
1682 }
1683 load_attempted = true;
1684 }
1685 }
1686 }
1687 }
1688
1689 if (!load_attempted) {
1690 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1691 }
1692
1693 if (result != NULL) {
1694 // Successful loading
1695 return result;
1696 }
1697
1698 Elf32_Ehdr elf_head;
1699 int diag_msg_max_length=ebuflen-strlen(ebuf);
1700 char* diag_msg_buf=ebuf+strlen(ebuf);
1701
1702 if (diag_msg_max_length==0) {
1703 // No more space in ebuf for additional diagnostics message
1704 return NULL;
1705 }
1706
1707
1708 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1709
1710 if (file_descriptor < 0) {
1711 // Can't open library, report dlerror() message
1712 return NULL;
1713 }
1714
1715 bool failed_to_read_elf_head=
1716 (sizeof(elf_head)!=
1717 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1718
1719 ::close(file_descriptor);
1720 if (failed_to_read_elf_head) {
1721 // file i/o error - report dlerror() msg
1722 return NULL;
1723 }
1724
1725 typedef struct {
1726 Elf32_Half code; // Actual value as defined in elf.h
1727 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1728 unsigned char elf_class; // 32 or 64 bit
1729 unsigned char endianess; // MSB or LSB
1730 char* name; // String representation
1731 } arch_t;
1732
1733 #ifndef EM_486
1734 #define EM_486 6 /* Intel 80486 */
1735 #endif
1736 #ifndef EM_AARCH64
1737 #define EM_AARCH64 183 /* ARM AARCH64 */
1738 #endif
1739
1740 static const arch_t arch_array[]={
1741 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1742 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1743 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1744 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1745 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1746 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1747 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1748 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1749 #if defined(VM_LITTLE_ENDIAN)
1750 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1751 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1752 #else
1753 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1754 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1755 #endif
1756 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1757 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1758 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1759 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1760 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1761 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1762 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1763 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1764 };
1765
1766 #if (defined IA32)
1767 static Elf32_Half running_arch_code=EM_386;
1768 #elif (defined AMD64)
1769 static Elf32_Half running_arch_code=EM_X86_64;
1770 #elif (defined IA64)
1771 static Elf32_Half running_arch_code=EM_IA_64;
1772 #elif (defined __sparc) && (defined _LP64)
1773 static Elf32_Half running_arch_code=EM_SPARCV9;
1774 #elif (defined __sparc) && (!defined _LP64)
1775 static Elf32_Half running_arch_code=EM_SPARC;
1776 #elif (defined __powerpc64__)
1777 static Elf32_Half running_arch_code=EM_PPC64;
1778 #elif (defined __powerpc__)
1779 static Elf32_Half running_arch_code=EM_PPC;
1780 #elif (defined AARCH64)
1781 static Elf32_Half running_arch_code=EM_AARCH64;
1782 #elif (defined ARM)
1783 static Elf32_Half running_arch_code=EM_ARM;
1784 #elif (defined S390)
1785 static Elf32_Half running_arch_code=EM_S390;
1786 #elif (defined ALPHA)
1787 static Elf32_Half running_arch_code=EM_ALPHA;
1788 #elif (defined MIPSEL)
1789 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1790 #elif (defined PARISC)
1791 static Elf32_Half running_arch_code=EM_PARISC;
1792 #elif (defined MIPS)
1793 static Elf32_Half running_arch_code=EM_MIPS;
1794 #elif (defined M68K)
1795 static Elf32_Half running_arch_code=EM_68K;
1796 #elif (defined SH)
1797 static Elf32_Half running_arch_code=EM_SH;
1798 #else
1799 #error Method os::dll_load requires that one of following is defined:\
1800 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1801 #endif
1802
1803 // Identify compatability class for VM's architecture and library's architecture
1804 // Obtain string descriptions for architectures
1805
1806 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1807 int running_arch_index=-1;
1808
1809 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1810 if (running_arch_code == arch_array[i].code) {
1811 running_arch_index = i;
1812 }
1813 if (lib_arch.code == arch_array[i].code) {
1814 lib_arch.compat_class = arch_array[i].compat_class;
1815 lib_arch.name = arch_array[i].name;
1816 }
1817 }
1818
1819 assert(running_arch_index != -1,
1820 "Didn't find running architecture code (running_arch_code) in arch_array");
1821 if (running_arch_index == -1) {
1822 // Even though running architecture detection failed
1823 // we may still continue with reporting dlerror() message
1824 return NULL;
1825 }
1826
1827 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1828 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1829 return NULL;
1830 }
1831
1832 #ifndef S390
1833 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1834 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1835 return NULL;
1836 }
1837 #endif // !S390
1838
1839 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1840 if (lib_arch.name!=NULL) {
1841 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1842 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1843 lib_arch.name, arch_array[running_arch_index].name);
1844 } else {
1845 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1846 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1847 lib_arch.code,
1848 arch_array[running_arch_index].name);
1849 }
1850 }
1851
1852 return NULL;
1853 }
1854
1855 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1856 int ebuflen) {
1857 void * result = ::dlopen(filename, RTLD_LAZY);
1858 if (result == NULL) {
1859 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1860 ebuf[ebuflen-1] = '\0';
1861 }
1862 return result;
1863 }
1864
1865 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1866 int ebuflen) {
1867 void * result = NULL;
1868 if (LoadExecStackDllInVMThread) {
1869 result = dlopen_helper(filename, ebuf, ebuflen);
1870 }
1871
1872 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1873 // library that requires an executable stack, or which does not have this
1874 // stack attribute set, dlopen changes the stack attribute to executable. The
1875 // read protection of the guard pages gets lost.
1876 //
1877 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1878 // may have been queued at the same time.
1879
1880 if (!_stack_is_executable) {
1881 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
1882 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1883 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1884 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1885 warning("Attempt to reguard stack yellow zone failed.");
1886 }
1887 }
1888 }
1889 }
1890
1891 return result;
1892 }
1893
1894 void* os::dll_lookup(void* handle, const char* name) {
1895 void* res = dlsym(handle, name);
1896 return res;
1897 }
1898
1899 void* os::get_default_process_handle() {
1900 return (void*)::dlopen(NULL, RTLD_LAZY);
1901 }
1902
1903 static bool _print_ascii_file(const char* filename, outputStream* st) {
1904 int fd = ::open(filename, O_RDONLY);
1905 if (fd == -1) {
1906 return false;
1907 }
1908
1909 char buf[33];
1910 int bytes;
1911 buf[32] = '\0';
1912 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1913 st->print_raw(buf, bytes);
1914 }
1915
1916 ::close(fd);
1917
1918 return true;
1919 }
1920
1921 void os::print_dll_info(outputStream *st) {
1922 st->print_cr("Dynamic libraries:");
1923
1924 char fname[32];
1925 pid_t pid = os::Linux::gettid();
1926
1927 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1928
1929 if (!_print_ascii_file(fname, st)) {
1930 st->print("Can not get library information for pid = %d\n", pid);
1931 }
1932 }
1933
1934 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1935 FILE *procmapsFile = NULL;
1936
1937 // Open the procfs maps file for the current process
1938 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1939 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1940 char line[PATH_MAX + 100];
1941
1942 // Read line by line from 'file'
1943 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1944 u8 base, top, offset, inode;
1945 char permissions[5];
1946 char device[6];
1947 char name[PATH_MAX + 1];
1948
1949 // Parse fields from line
1950 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1951 &base, &top, permissions, &offset, device, &inode, name);
1952
1953 // Filter by device id '00:00' so that we only get file system mapped files.
1954 if (strcmp(device, "00:00") != 0) {
1955
1956 // Call callback with the fields of interest
1957 if(callback(name, (address)base, (address)top, param)) {
1958 // Oops abort, callback aborted
1959 fclose(procmapsFile);
1960 return 1;
1961 }
1962 }
1963 }
1964 fclose(procmapsFile);
1965 }
1966 return 0;
1967 }
1968
1969 void os::print_os_info_brief(outputStream* st) {
1970 os::Linux::print_distro_info(st);
1971
1972 os::Posix::print_uname_info(st);
1973
1974 os::Linux::print_libversion_info(st);
1975
1976 }
1977
1978 void os::print_os_info(outputStream* st) {
1979 st->print("OS:");
1980
1981 os::Linux::print_distro_info(st);
1982
1983 os::Posix::print_uname_info(st);
1984
1985 // Print warning if unsafe chroot environment detected
1986 if (unsafe_chroot_detected) {
1987 st->print("WARNING!! ");
1988 st->print_cr("%s", unstable_chroot_error);
1989 }
1990
1991 os::Linux::print_libversion_info(st);
1992
1993 os::Posix::print_rlimit_info(st);
1994
1995 os::Posix::print_load_average(st);
1996
1997 os::Linux::print_full_memory_info(st);
1998
1999 os::Linux::print_container_info(st);
2000 }
2001
2002 // Try to identify popular distros.
2003 // Most Linux distributions have a /etc/XXX-release file, which contains
2004 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2005 // file that also contains the OS version string. Some have more than one
2006 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2007 // /etc/redhat-release.), so the order is important.
2008 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2009 // their own specific XXX-release file as well as a redhat-release file.
2010 // Because of this the XXX-release file needs to be searched for before the
2011 // redhat-release file.
2012 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2013 // search for redhat-release / SuSE-release needs to be before lsb-release.
2014 // Since the lsb-release file is the new standard it needs to be searched
2015 // before the older style release files.
2016 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2017 // next to last resort. The os-release file is a new standard that contains
2018 // distribution information and the system-release file seems to be an old
2019 // standard that has been replaced by the lsb-release and os-release files.
2020 // Searching for the debian_version file is the last resort. It contains
2021 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2022 // "Debian " is printed before the contents of the debian_version file.
2023
2024 const char* distro_files[] = {
2025 "/etc/oracle-release",
2026 "/etc/mandriva-release",
2027 "/etc/mandrake-release",
2028 "/etc/sun-release",
2029 "/etc/redhat-release",
2030 "/etc/SuSE-release",
2031 "/etc/lsb-release",
2032 "/etc/turbolinux-release",
2033 "/etc/gentoo-release",
2034 "/etc/ltib-release",
2035 "/etc/angstrom-version",
2036 "/etc/system-release",
2037 "/etc/os-release",
2038 NULL };
2039
2040 void os::Linux::print_distro_info(outputStream* st) {
2041 for (int i = 0;; i++) {
2042 const char* file = distro_files[i];
2043 if (file == NULL) {
2044 break; // done
2045 }
2046 // If file prints, we found it.
2047 if (_print_ascii_file(file, st)) {
2048 return;
2049 }
2050 }
2051
2052 if (file_exists("/etc/debian_version")) {
2053 st->print("Debian ");
2054 _print_ascii_file("/etc/debian_version", st);
2055 } else {
2056 st->print("Linux");
2057 }
2058 st->cr();
2059 }
2060
2061 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2062 char buf[256];
2063 while (fgets(buf, sizeof(buf), fp)) {
2064 // Edit out extra stuff in expected format
2065 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2066 char* ptr = strstr(buf, "\""); // the name is in quotes
2067 if (ptr != NULL) {
2068 ptr++; // go beyond first quote
2069 char* nl = strchr(ptr, '\"');
2070 if (nl != NULL) *nl = '\0';
2071 strncpy(distro, ptr, length);
2072 } else {
2073 ptr = strstr(buf, "=");
2074 ptr++; // go beyond equals then
2075 char* nl = strchr(ptr, '\n');
2076 if (nl != NULL) *nl = '\0';
2077 strncpy(distro, ptr, length);
2078 }
2079 return;
2080 } else if (get_first_line) {
2081 char* nl = strchr(buf, '\n');
2082 if (nl != NULL) *nl = '\0';
2083 strncpy(distro, buf, length);
2084 return;
2085 }
2086 }
2087 // print last line and close
2088 char* nl = strchr(buf, '\n');
2089 if (nl != NULL) *nl = '\0';
2090 strncpy(distro, buf, length);
2091 }
2092
2093 static void parse_os_info(char* distro, size_t length, const char* file) {
2094 FILE* fp = fopen(file, "r");
2095 if (fp != NULL) {
2096 // if suse format, print out first line
2097 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2098 parse_os_info_helper(fp, distro, length, get_first_line);
2099 fclose(fp);
2100 }
2101 }
2102
2103 void os::get_summary_os_info(char* buf, size_t buflen) {
2104 for (int i = 0;; i++) {
2105 const char* file = distro_files[i];
2106 if (file == NULL) {
2107 break; // ran out of distro_files
2108 }
2109 if (file_exists(file)) {
2110 parse_os_info(buf, buflen, file);
2111 return;
2112 }
2113 }
2114 // special case for debian
2115 if (file_exists("/etc/debian_version")) {
2116 strncpy(buf, "Debian ", buflen);
2117 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2118 } else {
2119 strncpy(buf, "Linux", buflen);
2120 }
2121 }
2122
2123 void os::Linux::print_libversion_info(outputStream* st) {
2124 // libc, pthread
2125 st->print("libc:");
2126 st->print("%s ", os::Linux::glibc_version());
2127 st->print("%s ", os::Linux::libpthread_version());
2128 st->cr();
2129 }
2130
2131 void os::Linux::print_full_memory_info(outputStream* st) {
2132 st->print("\n/proc/meminfo:\n");
2133 _print_ascii_file("/proc/meminfo", st);
2134 st->cr();
2135 }
2136
2137 void os::Linux::print_container_info(outputStream* st) {
2138 if (OSContainer::is_containerized()) {
2139 st->print("container (cgroup) information:\n");
2140
2141 char *p = OSContainer::container_type();
2142 if (p == NULL)
2143 st->print("container_type() failed\n");
2144 else {
2145 st->print("container_type: %s\n", p);
2146 }
2147
2148 p = OSContainer::cpu_cpuset_cpus();
2149 if (p == NULL)
2150 st->print("cpu_cpuset_cpus() failed\n");
2151 else {
2152 st->print("cpu_cpuset_cpus: %s\n", p);
2153 free(p);
2154 }
2155
2156 p = OSContainer::cpu_cpuset_memory_nodes();
2157 if (p < 0)
2158 st->print("cpu_memory_nodes() failed\n");
2159 else {
2160 st->print("cpu_memory_nodes: %s\n", p);
2161 free(p);
2162 }
2163
2164 int i = OSContainer::active_processor_count();
2165 if (i < 0)
2166 st->print("active_processor_count() failed\n");
2167 else
2168 st->print("active_processor_count: %d\n", i);
2169
2170 i = OSContainer::cpu_quota();
2171 st->print("cpu_quota: %d\n", i);
2172
2173 i = OSContainer::cpu_period();
2174 st->print("cpu_period: %d\n", i);
2175
2176 i = OSContainer::cpu_shares();
2177 st->print("cpu_shares: %d\n", i);
2178
2179 jlong j = OSContainer::memory_limit_in_bytes();
2180 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2181
2182 j = OSContainer::memory_and_swap_limit_in_bytes();
2183 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2184
2185 j = OSContainer::memory_soft_limit_in_bytes();
2186 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2187
2188 j = OSContainer::OSContainer::memory_usage_in_bytes();
2189 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2190
2191 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2192 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2193 st->cr();
2194 }
2195 }
2196
2197 void os::print_memory_info(outputStream* st) {
2198
2199 st->print("Memory:");
2200 st->print(" %dk page", os::vm_page_size()>>10);
2201
2202 // values in struct sysinfo are "unsigned long"
2203 struct sysinfo si;
2204 sysinfo(&si);
2205
2206 st->print(", physical " UINT64_FORMAT "k",
2207 os::physical_memory() >> 10);
2208 st->print("(" UINT64_FORMAT "k free)",
2209 os::available_memory() >> 10);
2210 st->print(", swap " UINT64_FORMAT "k",
2211 ((jlong)si.totalswap * si.mem_unit) >> 10);
2212 st->print("(" UINT64_FORMAT "k free)",
2213 ((jlong)si.freeswap * si.mem_unit) >> 10);
2214 st->cr();
2215 }
2216
2217 // Print the first "model name" line and the first "flags" line
2218 // that we find and nothing more. We assume "model name" comes
2219 // before "flags" so if we find a second "model name", then the
2220 // "flags" field is considered missing.
2221 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2222 #if defined(IA32) || defined(AMD64)
2223 // Other platforms have less repetitive cpuinfo files
2224 FILE *fp = fopen("/proc/cpuinfo", "r");
2225 if (fp) {
2226 while (!feof(fp)) {
2227 if (fgets(buf, buflen, fp)) {
2228 // Assume model name comes before flags
2229 bool model_name_printed = false;
2230 if (strstr(buf, "model name") != NULL) {
2231 if (!model_name_printed) {
2232 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2233 st->print_raw(buf);
2234 model_name_printed = true;
2235 } else {
2236 // model name printed but not flags? Odd, just return
2237 fclose(fp);
2238 return true;
2239 }
2240 }
2241 // print the flags line too
2242 if (strstr(buf, "flags") != NULL) {
2243 st->print_raw(buf);
2244 fclose(fp);
2245 return true;
2246 }
2247 }
2248 }
2249 fclose(fp);
2250 }
2251 #endif // x86 platforms
2252 return false;
2253 }
2254
2255 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2256 // Only print the model name if the platform provides this as a summary
2257 if (!print_model_name_and_flags(st, buf, buflen)) {
2258 st->print("\n/proc/cpuinfo:\n");
2259 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2260 st->print_cr(" <Not Available>");
2261 }
2262 }
2263 }
2264
2265 #if defined(AMD64) || defined(IA32) || defined(X32)
2266 const char* search_string = "model name";
2267 #elif defined(M68K)
2268 const char* search_string = "CPU";
2269 #elif defined(PPC64)
2270 const char* search_string = "cpu";
2271 #elif defined(S390)
2272 const char* search_string = "processor";
2273 #elif defined(SPARC)
2274 const char* search_string = "cpu";
2275 #else
2276 const char* search_string = "Processor";
2277 #endif
2278
2279 // Parses the cpuinfo file for string representing the model name.
2280 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2281 FILE* fp = fopen("/proc/cpuinfo", "r");
2282 if (fp != NULL) {
2283 while (!feof(fp)) {
2284 char buf[256];
2285 if (fgets(buf, sizeof(buf), fp)) {
2286 char* start = strstr(buf, search_string);
2287 if (start != NULL) {
2288 char *ptr = start + strlen(search_string);
2289 char *end = buf + strlen(buf);
2290 while (ptr != end) {
2291 // skip whitespace and colon for the rest of the name.
2292 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2293 break;
2294 }
2295 ptr++;
2296 }
2297 if (ptr != end) {
2298 // reasonable string, get rid of newline and keep the rest
2299 char* nl = strchr(buf, '\n');
2300 if (nl != NULL) *nl = '\0';
2301 strncpy(cpuinfo, ptr, length);
2302 fclose(fp);
2303 return;
2304 }
2305 }
2306 }
2307 }
2308 fclose(fp);
2309 }
2310 // cpuinfo not found or parsing failed, just print generic string. The entire
2311 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2312 #if defined(AARCH64)
2313 strncpy(cpuinfo, "AArch64", length);
2314 #elif defined(AMD64)
2315 strncpy(cpuinfo, "x86_64", length);
2316 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2317 strncpy(cpuinfo, "ARM", length);
2318 #elif defined(IA32)
2319 strncpy(cpuinfo, "x86_32", length);
2320 #elif defined(IA64)
2321 strncpy(cpuinfo, "IA64", length);
2322 #elif defined(PPC)
2323 strncpy(cpuinfo, "PPC64", length);
2324 #elif defined(S390)
2325 strncpy(cpuinfo, "S390", length);
2326 #elif defined(SPARC)
2327 strncpy(cpuinfo, "sparcv9", length);
2328 #elif defined(ZERO_LIBARCH)
2329 strncpy(cpuinfo, ZERO_LIBARCH, length);
2330 #else
2331 strncpy(cpuinfo, "unknown", length);
2332 #endif
2333 }
2334
2335 static void print_signal_handler(outputStream* st, int sig,
2336 char* buf, size_t buflen);
2337
2338 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2339 st->print_cr("Signal Handlers:");
2340 print_signal_handler(st, SIGSEGV, buf, buflen);
2341 print_signal_handler(st, SIGBUS , buf, buflen);
2342 print_signal_handler(st, SIGFPE , buf, buflen);
2343 print_signal_handler(st, SIGPIPE, buf, buflen);
2344 print_signal_handler(st, SIGXFSZ, buf, buflen);
2345 print_signal_handler(st, SIGILL , buf, buflen);
2346 print_signal_handler(st, SR_signum, buf, buflen);
2347 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2348 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2349 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2350 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2351 #if defined(PPC64)
2352 print_signal_handler(st, SIGTRAP, buf, buflen);
2353 #endif
2354 }
2355
2356 static char saved_jvm_path[MAXPATHLEN] = {0};
2357
2358 // Find the full path to the current module, libjvm.so
2359 void os::jvm_path(char *buf, jint buflen) {
2360 // Error checking.
2361 if (buflen < MAXPATHLEN) {
2362 assert(false, "must use a large-enough buffer");
2363 buf[0] = '\0';
2364 return;
2365 }
2366 // Lazy resolve the path to current module.
2367 if (saved_jvm_path[0] != 0) {
2368 strcpy(buf, saved_jvm_path);
2369 return;
2370 }
2371
2372 char dli_fname[MAXPATHLEN];
2373 bool ret = dll_address_to_library_name(
2374 CAST_FROM_FN_PTR(address, os::jvm_path),
2375 dli_fname, sizeof(dli_fname), NULL);
2376 assert(ret, "cannot locate libjvm");
2377 char *rp = NULL;
2378 if (ret && dli_fname[0] != '\0') {
2379 rp = os::Posix::realpath(dli_fname, buf, buflen);
2380 }
2381 if (rp == NULL) {
2382 return;
2383 }
2384
2385 if (Arguments::sun_java_launcher_is_altjvm()) {
2386 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2387 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2388 // If "/jre/lib/" appears at the right place in the string, then
2389 // assume we are installed in a JDK and we're done. Otherwise, check
2390 // for a JAVA_HOME environment variable and fix up the path so it
2391 // looks like libjvm.so is installed there (append a fake suffix
2392 // hotspot/libjvm.so).
2393 const char *p = buf + strlen(buf) - 1;
2394 for (int count = 0; p > buf && count < 5; ++count) {
2395 for (--p; p > buf && *p != '/'; --p)
2396 /* empty */ ;
2397 }
2398
2399 if (strncmp(p, "/jre/lib/", 9) != 0) {
2400 // Look for JAVA_HOME in the environment.
2401 char* java_home_var = ::getenv("JAVA_HOME");
2402 if (java_home_var != NULL && java_home_var[0] != 0) {
2403 char* jrelib_p;
2404 int len;
2405
2406 // Check the current module name "libjvm.so".
2407 p = strrchr(buf, '/');
2408 if (p == NULL) {
2409 return;
2410 }
2411 assert(strstr(p, "/libjvm") == p, "invalid library name");
2412
2413 rp = os::Posix::realpath(java_home_var, buf, buflen);
2414 if (rp == NULL) {
2415 return;
2416 }
2417
2418 // determine if this is a legacy image or modules image
2419 // modules image doesn't have "jre" subdirectory
2420 len = strlen(buf);
2421 assert(len < buflen, "Ran out of buffer room");
2422 jrelib_p = buf + len;
2423 snprintf(jrelib_p, buflen-len, "/jre/lib");
2424 if (0 != access(buf, F_OK)) {
2425 snprintf(jrelib_p, buflen-len, "/lib");
2426 }
2427
2428 if (0 == access(buf, F_OK)) {
2429 // Use current module name "libjvm.so"
2430 len = strlen(buf);
2431 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2432 } else {
2433 // Go back to path of .so
2434 rp = os::Posix::realpath(dli_fname, buf, buflen);
2435 if (rp == NULL) {
2436 return;
2437 }
2438 }
2439 }
2440 }
2441 }
2442
2443 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2444 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2445 }
2446
2447 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2448 // no prefix required, not even "_"
2449 }
2450
2451 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2452 // no suffix required
2453 }
2454
2455 ////////////////////////////////////////////////////////////////////////////////
2456 // sun.misc.Signal support
2457
2458 static volatile jint sigint_count = 0;
2459
2460 static void UserHandler(int sig, void *siginfo, void *context) {
2461 // 4511530 - sem_post is serialized and handled by the manager thread. When
2462 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2463 // don't want to flood the manager thread with sem_post requests.
2464 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2465 return;
2466 }
2467
2468 // Ctrl-C is pressed during error reporting, likely because the error
2469 // handler fails to abort. Let VM die immediately.
2470 if (sig == SIGINT && VMError::is_error_reported()) {
2471 os::die();
2472 }
2473
2474 os::signal_notify(sig);
2475 }
2476
2477 void* os::user_handler() {
2478 return CAST_FROM_FN_PTR(void*, UserHandler);
2479 }
2480
2481 static struct timespec create_timespec(unsigned int sec, int nsec) {
2482 struct timespec ts;
2483 // Semaphore's are always associated with CLOCK_REALTIME
2484 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2485 // see unpackTime for discussion on overflow checking
2486 if (sec >= MAX_SECS) {
2487 ts.tv_sec += MAX_SECS;
2488 ts.tv_nsec = 0;
2489 } else {
2490 ts.tv_sec += sec;
2491 ts.tv_nsec += nsec;
2492 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2493 ts.tv_nsec -= NANOSECS_PER_SEC;
2494 ++ts.tv_sec; // note: this must be <= max_secs
2495 }
2496 }
2497
2498 return ts;
2499 }
2500
2501 extern "C" {
2502 typedef void (*sa_handler_t)(int);
2503 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2504 }
2505
2506 void* os::signal(int signal_number, void* handler) {
2507 struct sigaction sigAct, oldSigAct;
2508
2509 sigfillset(&(sigAct.sa_mask));
2510 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2511 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2512
2513 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2514 // -1 means registration failed
2515 return (void *)-1;
2516 }
2517
2518 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2519 }
2520
2521 void os::signal_raise(int signal_number) {
2522 ::raise(signal_number);
2523 }
2524
2525 // The following code is moved from os.cpp for making this
2526 // code platform specific, which it is by its very nature.
2527
2528 // Will be modified when max signal is changed to be dynamic
2529 int os::sigexitnum_pd() {
2530 return NSIG;
2531 }
2532
2533 // a counter for each possible signal value
2534 static volatile jint pending_signals[NSIG+1] = { 0 };
2535
2536 // Linux(POSIX) specific hand shaking semaphore.
2537 static Semaphore* sig_sem = NULL;
2538 static PosixSemaphore sr_semaphore;
2539
2540 void os::signal_init_pd() {
2541 // Initialize signal structures
2542 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2543
2544 // Initialize signal semaphore
2545 sig_sem = new Semaphore();
2546 }
2547
2548 void os::signal_notify(int sig) {
2549 if (sig_sem != NULL) {
2550 Atomic::inc(&pending_signals[sig]);
2551 sig_sem->signal();
2552 } else {
2553 // Signal thread is not created with ReduceSignalUsage and signal_init_pd
2554 // initialization isn't called.
2555 assert(ReduceSignalUsage, "signal semaphore should be created");
2556 }
2557 }
2558
2559 static int check_pending_signals(bool wait) {
2560 Atomic::store(0, &sigint_count);
2561 for (;;) {
2562 for (int i = 0; i < NSIG + 1; i++) {
2563 jint n = pending_signals[i];
2564 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2565 return i;
2566 }
2567 }
2568 if (!wait) {
2569 return -1;
2570 }
2571 JavaThread *thread = JavaThread::current();
2572 ThreadBlockInVM tbivm(thread);
2573
2574 bool threadIsSuspended;
2575 do {
2576 thread->set_suspend_equivalent();
2577 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2578 sig_sem->wait();
2579
2580 // were we externally suspended while we were waiting?
2581 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2582 if (threadIsSuspended) {
2583 // The semaphore has been incremented, but while we were waiting
2584 // another thread suspended us. We don't want to continue running
2585 // while suspended because that would surprise the thread that
2586 // suspended us.
2587 sig_sem->signal();
2588
2589 thread->java_suspend_self();
2590 }
2591 } while (threadIsSuspended);
2592 }
2593 }
2594
2595 int os::signal_wait() {
2596 return check_pending_signals(true);
2597 }
2598
2599 ////////////////////////////////////////////////////////////////////////////////
2600 // Virtual Memory
2601
2602 int os::vm_page_size() {
2603 // Seems redundant as all get out
2604 assert(os::Linux::page_size() != -1, "must call os::init");
2605 return os::Linux::page_size();
2606 }
2607
2608 // Solaris allocates memory by pages.
2609 int os::vm_allocation_granularity() {
2610 assert(os::Linux::page_size() != -1, "must call os::init");
2611 return os::Linux::page_size();
2612 }
2613
2614 // Rationale behind this function:
2615 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2616 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2617 // samples for JITted code. Here we create private executable mapping over the code cache
2618 // and then we can use standard (well, almost, as mapping can change) way to provide
2619 // info for the reporting script by storing timestamp and location of symbol
2620 void linux_wrap_code(char* base, size_t size) {
2621 static volatile jint cnt = 0;
2622
2623 if (!UseOprofile) {
2624 return;
2625 }
2626
2627 char buf[PATH_MAX+1];
2628 int num = Atomic::add(1, &cnt);
2629
2630 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2631 os::get_temp_directory(), os::current_process_id(), num);
2632 unlink(buf);
2633
2634 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2635
2636 if (fd != -1) {
2637 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2638 if (rv != (off_t)-1) {
2639 if (::write(fd, "", 1) == 1) {
2640 mmap(base, size,
2641 PROT_READ|PROT_WRITE|PROT_EXEC,
2642 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2643 }
2644 }
2645 ::close(fd);
2646 unlink(buf);
2647 }
2648 }
2649
2650 static bool recoverable_mmap_error(int err) {
2651 // See if the error is one we can let the caller handle. This
2652 // list of errno values comes from JBS-6843484. I can't find a
2653 // Linux man page that documents this specific set of errno
2654 // values so while this list currently matches Solaris, it may
2655 // change as we gain experience with this failure mode.
2656 switch (err) {
2657 case EBADF:
2658 case EINVAL:
2659 case ENOTSUP:
2660 // let the caller deal with these errors
2661 return true;
2662
2663 default:
2664 // Any remaining errors on this OS can cause our reserved mapping
2665 // to be lost. That can cause confusion where different data
2666 // structures think they have the same memory mapped. The worst
2667 // scenario is if both the VM and a library think they have the
2668 // same memory mapped.
2669 return false;
2670 }
2671 }
2672
2673 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2674 int err) {
2675 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2676 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2677 os::strerror(err), err);
2678 }
2679
2680 static void warn_fail_commit_memory(char* addr, size_t size,
2681 size_t alignment_hint, bool exec,
2682 int err) {
2683 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2684 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2685 alignment_hint, exec, os::strerror(err), err);
2686 }
2687
2688 // NOTE: Linux kernel does not really reserve the pages for us.
2689 // All it does is to check if there are enough free pages
2690 // left at the time of mmap(). This could be a potential
2691 // problem.
2692 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2693 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2694 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2695 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2696 if (res != (uintptr_t) MAP_FAILED) {
2697 if (UseNUMAInterleaving) {
2698 numa_make_global(addr, size);
2699 }
2700 return 0;
2701 }
2702
2703 int err = errno; // save errno from mmap() call above
2704
2705 if (!recoverable_mmap_error(err)) {
2706 warn_fail_commit_memory(addr, size, exec, err);
2707 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2708 }
2709
2710 return err;
2711 }
2712
2713 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2714 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2715 }
2716
2717 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2718 const char* mesg) {
2719 assert(mesg != NULL, "mesg must be specified");
2720 int err = os::Linux::commit_memory_impl(addr, size, exec);
2721 if (err != 0) {
2722 // the caller wants all commit errors to exit with the specified mesg:
2723 warn_fail_commit_memory(addr, size, exec, err);
2724 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2725 }
2726 }
2727
2728 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2729 #ifndef MAP_HUGETLB
2730 #define MAP_HUGETLB 0x40000
2731 #endif
2732
2733 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2734 #ifndef MADV_HUGEPAGE
2735 #define MADV_HUGEPAGE 14
2736 #endif
2737
2738 int os::Linux::commit_memory_impl(char* addr, size_t size,
2739 size_t alignment_hint, bool exec) {
2740 int err = os::Linux::commit_memory_impl(addr, size, exec);
2741 if (err == 0) {
2742 realign_memory(addr, size, alignment_hint);
2743 }
2744 return err;
2745 }
2746
2747 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2748 bool exec) {
2749 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2750 }
2751
2752 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2753 size_t alignment_hint, bool exec,
2754 const char* mesg) {
2755 assert(mesg != NULL, "mesg must be specified");
2756 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2757 if (err != 0) {
2758 // the caller wants all commit errors to exit with the specified mesg:
2759 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2760 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2761 }
2762 }
2763
2764 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2765 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2766 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2767 // be supported or the memory may already be backed by huge pages.
2768 ::madvise(addr, bytes, MADV_HUGEPAGE);
2769 }
2770 }
2771
2772 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2773 // This method works by doing an mmap over an existing mmaping and effectively discarding
2774 // the existing pages. However it won't work for SHM-based large pages that cannot be
2775 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2776 // small pages on top of the SHM segment. This method always works for small pages, so we
2777 // allow that in any case.
2778 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2779 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2780 }
2781 }
2782
2783 void os::numa_make_global(char *addr, size_t bytes) {
2784 Linux::numa_interleave_memory(addr, bytes);
2785 }
2786
2787 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2788 // bind policy to MPOL_PREFERRED for the current thread.
2789 #define USE_MPOL_PREFERRED 0
2790
2791 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2792 // To make NUMA and large pages more robust when both enabled, we need to ease
2793 // the requirements on where the memory should be allocated. MPOL_BIND is the
2794 // default policy and it will force memory to be allocated on the specified
2795 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2796 // the specified node, but will not force it. Using this policy will prevent
2797 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2798 // free large pages.
2799 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2800 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2801 }
2802
2803 bool os::numa_topology_changed() { return false; }
2804
2805 size_t os::numa_get_groups_num() {
2806 // Return just the number of nodes in which it's possible to allocate memory
2807 // (in numa terminology, configured nodes).
2808 return Linux::numa_num_configured_nodes();
2809 }
2810
2811 int os::numa_get_group_id() {
2812 int cpu_id = Linux::sched_getcpu();
2813 if (cpu_id != -1) {
2814 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2815 if (lgrp_id != -1) {
2816 return lgrp_id;
2817 }
2818 }
2819 return 0;
2820 }
2821
2822 int os::Linux::get_existing_num_nodes() {
2823 size_t node;
2824 size_t highest_node_number = Linux::numa_max_node();
2825 int num_nodes = 0;
2826
2827 // Get the total number of nodes in the system including nodes without memory.
2828 for (node = 0; node <= highest_node_number; node++) {
2829 if (isnode_in_existing_nodes(node)) {
2830 num_nodes++;
2831 }
2832 }
2833 return num_nodes;
2834 }
2835
2836 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2837 size_t highest_node_number = Linux::numa_max_node();
2838 size_t i = 0;
2839
2840 // Map all node ids in which is possible to allocate memory. Also nodes are
2841 // not always consecutively available, i.e. available from 0 to the highest
2842 // node number.
2843 for (size_t node = 0; node <= highest_node_number; node++) {
2844 if (Linux::isnode_in_configured_nodes(node)) {
2845 ids[i++] = node;
2846 }
2847 }
2848 return i;
2849 }
2850
2851 bool os::get_page_info(char *start, page_info* info) {
2852 return false;
2853 }
2854
2855 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2856 page_info* page_found) {
2857 return end;
2858 }
2859
2860
2861 int os::Linux::sched_getcpu_syscall(void) {
2862 unsigned int cpu = 0;
2863 int retval = -1;
2864
2865 #if defined(IA32)
2866 #ifndef SYS_getcpu
2867 #define SYS_getcpu 318
2868 #endif
2869 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2870 #elif defined(AMD64)
2871 // Unfortunately we have to bring all these macros here from vsyscall.h
2872 // to be able to compile on old linuxes.
2873 #define __NR_vgetcpu 2
2874 #define VSYSCALL_START (-10UL << 20)
2875 #define VSYSCALL_SIZE 1024
2876 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2877 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2878 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2879 retval = vgetcpu(&cpu, NULL, NULL);
2880 #endif
2881
2882 return (retval == -1) ? retval : cpu;
2883 }
2884
2885 void os::Linux::sched_getcpu_init() {
2886 // sched_getcpu() should be in libc.
2887 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2888 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2889
2890 // If it's not, try a direct syscall.
2891 if (sched_getcpu() == -1) {
2892 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2893 (void*)&sched_getcpu_syscall));
2894 }
2895 }
2896
2897 // Something to do with the numa-aware allocator needs these symbols
2898 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2899 extern "C" JNIEXPORT void numa_error(char *where) { }
2900
2901 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2902 // load symbol from base version instead.
2903 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2904 void *f = dlvsym(handle, name, "libnuma_1.1");
2905 if (f == NULL) {
2906 f = dlsym(handle, name);
2907 }
2908 return f;
2909 }
2910
2911 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2912 // Return NULL if the symbol is not defined in this particular version.
2913 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2914 return dlvsym(handle, name, "libnuma_1.2");
2915 }
2916
2917 bool os::Linux::libnuma_init() {
2918 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
2919 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2920 if (handle != NULL) {
2921 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2922 libnuma_dlsym(handle, "numa_node_to_cpus")));
2923 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2924 libnuma_dlsym(handle, "numa_max_node")));
2925 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2926 libnuma_dlsym(handle, "numa_num_configured_nodes")));
2927 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2928 libnuma_dlsym(handle, "numa_available")));
2929 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2930 libnuma_dlsym(handle, "numa_tonode_memory")));
2931 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2932 libnuma_dlsym(handle, "numa_interleave_memory")));
2933 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2934 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2935 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2936 libnuma_dlsym(handle, "numa_set_bind_policy")));
2937 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2938 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2939 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2940 libnuma_dlsym(handle, "numa_distance")));
2941
2942 if (numa_available() != -1) {
2943 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2944 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2945 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2946 // Create an index -> node mapping, since nodes are not always consecutive
2947 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2948 rebuild_nindex_to_node_map();
2949 // Create a cpu -> node mapping
2950 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2951 rebuild_cpu_to_node_map();
2952 return true;
2953 }
2954 }
2955 }
2956 return false;
2957 }
2958
2959 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2960 // Creating guard page is very expensive. Java thread has HotSpot
2961 // guard pages, only enable glibc guard page for non-Java threads.
2962 // (Remember: compiler thread is a Java thread, too!)
2963 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2964 }
2965
2966 void os::Linux::rebuild_nindex_to_node_map() {
2967 int highest_node_number = Linux::numa_max_node();
2968
2969 nindex_to_node()->clear();
2970 for (int node = 0; node <= highest_node_number; node++) {
2971 if (Linux::isnode_in_existing_nodes(node)) {
2972 nindex_to_node()->append(node);
2973 }
2974 }
2975 }
2976
2977 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2978 // The table is later used in get_node_by_cpu().
2979 void os::Linux::rebuild_cpu_to_node_map() {
2980 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2981 // in libnuma (possible values are starting from 16,
2982 // and continuing up with every other power of 2, but less
2983 // than the maximum number of CPUs supported by kernel), and
2984 // is a subject to change (in libnuma version 2 the requirements
2985 // are more reasonable) we'll just hardcode the number they use
2986 // in the library.
2987 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2988
2989 size_t cpu_num = processor_count();
2990 size_t cpu_map_size = NCPUS / BitsPerCLong;
2991 size_t cpu_map_valid_size =
2992 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2993
2994 cpu_to_node()->clear();
2995 cpu_to_node()->at_grow(cpu_num - 1);
2996
2997 size_t node_num = get_existing_num_nodes();
2998
2999 int distance = 0;
3000 int closest_distance = INT_MAX;
3001 int closest_node = 0;
3002 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
3003 for (size_t i = 0; i < node_num; i++) {
3004 // Check if node is configured (not a memory-less node). If it is not, find
3005 // the closest configured node.
3006 if (!isnode_in_configured_nodes(nindex_to_node()->at(i))) {
3007 closest_distance = INT_MAX;
3008 // Check distance from all remaining nodes in the system. Ignore distance
3009 // from itself and from another non-configured node.
3010 for (size_t m = 0; m < node_num; m++) {
3011 if (m != i && isnode_in_configured_nodes(nindex_to_node()->at(m))) {
3012 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3013 // If a closest node is found, update. There is always at least one
3014 // configured node in the system so there is always at least one node
3015 // close.
3016 if (distance != 0 && distance < closest_distance) {
3017 closest_distance = distance;
3018 closest_node = nindex_to_node()->at(m);
3019 }
3020 }
3021 }
3022 } else {
3023 // Current node is already a configured node.
3024 closest_node = nindex_to_node()->at(i);
3025 }
3026
3027 // Get cpus from the original node and map them to the closest node. If node
3028 // is a configured node (not a memory-less node), then original node and
3029 // closest node are the same.
3030 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3031 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3032 if (cpu_map[j] != 0) {
3033 for (size_t k = 0; k < BitsPerCLong; k++) {
3034 if (cpu_map[j] & (1UL << k)) {
3035 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3036 }
3037 }
3038 }
3039 }
3040 }
3041 }
3042 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3043 }
3044
3045 int os::Linux::get_node_by_cpu(int cpu_id) {
3046 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3047 return cpu_to_node()->at(cpu_id);
3048 }
3049 return -1;
3050 }
3051
3052 GrowableArray<int>* os::Linux::_cpu_to_node;
3053 GrowableArray<int>* os::Linux::_nindex_to_node;
3054 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3055 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3056 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3057 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3058 os::Linux::numa_available_func_t os::Linux::_numa_available;
3059 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3060 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3061 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3062 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3063 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3064 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3065 unsigned long* os::Linux::_numa_all_nodes;
3066 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3067 struct bitmask* os::Linux::_numa_nodes_ptr;
3068
3069 bool os::pd_uncommit_memory(char* addr, size_t size) {
3070 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3071 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3072 return res != (uintptr_t) MAP_FAILED;
3073 }
3074
3075 static address get_stack_commited_bottom(address bottom, size_t size) {
3076 address nbot = bottom;
3077 address ntop = bottom + size;
3078
3079 size_t page_sz = os::vm_page_size();
3080 unsigned pages = size / page_sz;
3081
3082 unsigned char vec[1];
3083 unsigned imin = 1, imax = pages + 1, imid;
3084 int mincore_return_value = 0;
3085
3086 assert(imin <= imax, "Unexpected page size");
3087
3088 while (imin < imax) {
3089 imid = (imax + imin) / 2;
3090 nbot = ntop - (imid * page_sz);
3091
3092 // Use a trick with mincore to check whether the page is mapped or not.
3093 // mincore sets vec to 1 if page resides in memory and to 0 if page
3094 // is swapped output but if page we are asking for is unmapped
3095 // it returns -1,ENOMEM
3096 mincore_return_value = mincore(nbot, page_sz, vec);
3097
3098 if (mincore_return_value == -1) {
3099 // Page is not mapped go up
3100 // to find first mapped page
3101 if (errno != EAGAIN) {
3102 assert(errno == ENOMEM, "Unexpected mincore errno");
3103 imax = imid;
3104 }
3105 } else {
3106 // Page is mapped go down
3107 // to find first not mapped page
3108 imin = imid + 1;
3109 }
3110 }
3111
3112 nbot = nbot + page_sz;
3113
3114 // Adjust stack bottom one page up if last checked page is not mapped
3115 if (mincore_return_value == -1) {
3116 nbot = nbot + page_sz;
3117 }
3118
3119 return nbot;
3120 }
3121
3122
3123 // Linux uses a growable mapping for the stack, and if the mapping for
3124 // the stack guard pages is not removed when we detach a thread the
3125 // stack cannot grow beyond the pages where the stack guard was
3126 // mapped. If at some point later in the process the stack expands to
3127 // that point, the Linux kernel cannot expand the stack any further
3128 // because the guard pages are in the way, and a segfault occurs.
3129 //
3130 // However, it's essential not to split the stack region by unmapping
3131 // a region (leaving a hole) that's already part of the stack mapping,
3132 // so if the stack mapping has already grown beyond the guard pages at
3133 // the time we create them, we have to truncate the stack mapping.
3134 // So, we need to know the extent of the stack mapping when
3135 // create_stack_guard_pages() is called.
3136
3137 // We only need this for stacks that are growable: at the time of
3138 // writing thread stacks don't use growable mappings (i.e. those
3139 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3140 // only applies to the main thread.
3141
3142 // If the (growable) stack mapping already extends beyond the point
3143 // where we're going to put our guard pages, truncate the mapping at
3144 // that point by munmap()ping it. This ensures that when we later
3145 // munmap() the guard pages we don't leave a hole in the stack
3146 // mapping. This only affects the main/primordial thread
3147
3148 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3149 if (os::is_primordial_thread()) {
3150 // As we manually grow stack up to bottom inside create_attached_thread(),
3151 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3152 // we don't need to do anything special.
3153 // Check it first, before calling heavy function.
3154 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3155 unsigned char vec[1];
3156
3157 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3158 // Fallback to slow path on all errors, including EAGAIN
3159 stack_extent = (uintptr_t) get_stack_commited_bottom(
3160 os::Linux::initial_thread_stack_bottom(),
3161 (size_t)addr - stack_extent);
3162 }
3163
3164 if (stack_extent < (uintptr_t)addr) {
3165 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3166 }
3167 }
3168
3169 return os::commit_memory(addr, size, !ExecMem);
3170 }
3171
3172 // If this is a growable mapping, remove the guard pages entirely by
3173 // munmap()ping them. If not, just call uncommit_memory(). This only
3174 // affects the main/primordial thread, but guard against future OS changes.
3175 // It's safe to always unmap guard pages for primordial thread because we
3176 // always place it right after end of the mapped region.
3177
3178 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3179 uintptr_t stack_extent, stack_base;
3180
3181 if (os::is_primordial_thread()) {
3182 return ::munmap(addr, size) == 0;
3183 }
3184
3185 return os::uncommit_memory(addr, size);
3186 }
3187
3188 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3189 // at 'requested_addr'. If there are existing memory mappings at the same
3190 // location, however, they will be overwritten. If 'fixed' is false,
3191 // 'requested_addr' is only treated as a hint, the return value may or
3192 // may not start from the requested address. Unlike Linux mmap(), this
3193 // function returns NULL to indicate failure.
3194 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3195 char * addr;
3196 int flags;
3197
3198 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3199 if (fixed) {
3200 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3201 flags |= MAP_FIXED;
3202 }
3203
3204 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3205 // touch an uncommitted page. Otherwise, the read/write might
3206 // succeed if we have enough swap space to back the physical page.
3207 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3208 flags, -1, 0);
3209
3210 return addr == MAP_FAILED ? NULL : addr;
3211 }
3212
3213 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3214 // (req_addr != NULL) or with a given alignment.
3215 // - bytes shall be a multiple of alignment.
3216 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3217 // - alignment sets the alignment at which memory shall be allocated.
3218 // It must be a multiple of allocation granularity.
3219 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3220 // req_addr or NULL.
3221 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3222
3223 size_t extra_size = bytes;
3224 if (req_addr == NULL && alignment > 0) {
3225 extra_size += alignment;
3226 }
3227
3228 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3229 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3230 -1, 0);
3231 if (start == MAP_FAILED) {
3232 start = NULL;
3233 } else {
3234 if (req_addr != NULL) {
3235 if (start != req_addr) {
3236 ::munmap(start, extra_size);
3237 start = NULL;
3238 }
3239 } else {
3240 char* const start_aligned = align_up(start, alignment);
3241 char* const end_aligned = start_aligned + bytes;
3242 char* const end = start + extra_size;
3243 if (start_aligned > start) {
3244 ::munmap(start, start_aligned - start);
3245 }
3246 if (end_aligned < end) {
3247 ::munmap(end_aligned, end - end_aligned);
3248 }
3249 start = start_aligned;
3250 }
3251 }
3252 return start;
3253 }
3254
3255 static int anon_munmap(char * addr, size_t size) {
3256 return ::munmap(addr, size) == 0;
3257 }
3258
3259 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3260 size_t alignment_hint) {
3261 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3262 }
3263
3264 bool os::pd_release_memory(char* addr, size_t size) {
3265 return anon_munmap(addr, size);
3266 }
3267
3268 static bool linux_mprotect(char* addr, size_t size, int prot) {
3269 // Linux wants the mprotect address argument to be page aligned.
3270 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3271
3272 // According to SUSv3, mprotect() should only be used with mappings
3273 // established by mmap(), and mmap() always maps whole pages. Unaligned
3274 // 'addr' likely indicates problem in the VM (e.g. trying to change
3275 // protection of malloc'ed or statically allocated memory). Check the
3276 // caller if you hit this assert.
3277 assert(addr == bottom, "sanity check");
3278
3279 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3280 return ::mprotect(bottom, size, prot) == 0;
3281 }
3282
3283 // Set protections specified
3284 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3285 bool is_committed) {
3286 unsigned int p = 0;
3287 switch (prot) {
3288 case MEM_PROT_NONE: p = PROT_NONE; break;
3289 case MEM_PROT_READ: p = PROT_READ; break;
3290 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3291 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3292 default:
3293 ShouldNotReachHere();
3294 }
3295 // is_committed is unused.
3296 return linux_mprotect(addr, bytes, p);
3297 }
3298
3299 bool os::guard_memory(char* addr, size_t size) {
3300 return linux_mprotect(addr, size, PROT_NONE);
3301 }
3302
3303 bool os::unguard_memory(char* addr, size_t size) {
3304 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3305 }
3306
3307 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3308 size_t page_size) {
3309 bool result = false;
3310 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3311 MAP_ANONYMOUS|MAP_PRIVATE,
3312 -1, 0);
3313 if (p != MAP_FAILED) {
3314 void *aligned_p = align_up(p, page_size);
3315
3316 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3317
3318 munmap(p, page_size * 2);
3319 }
3320
3321 if (warn && !result) {
3322 warning("TransparentHugePages is not supported by the operating system.");
3323 }
3324
3325 return result;
3326 }
3327
3328 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3329 bool result = false;
3330 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3331 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3332 -1, 0);
3333
3334 if (p != MAP_FAILED) {
3335 // We don't know if this really is a huge page or not.
3336 FILE *fp = fopen("/proc/self/maps", "r");
3337 if (fp) {
3338 while (!feof(fp)) {
3339 char chars[257];
3340 long x = 0;
3341 if (fgets(chars, sizeof(chars), fp)) {
3342 if (sscanf(chars, "%lx-%*x", &x) == 1
3343 && x == (long)p) {
3344 if (strstr (chars, "hugepage")) {
3345 result = true;
3346 break;
3347 }
3348 }
3349 }
3350 }
3351 fclose(fp);
3352 }
3353 munmap(p, page_size);
3354 }
3355
3356 if (warn && !result) {
3357 warning("HugeTLBFS is not supported by the operating system.");
3358 }
3359
3360 return result;
3361 }
3362
3363 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3364 //
3365 // From the coredump_filter documentation:
3366 //
3367 // - (bit 0) anonymous private memory
3368 // - (bit 1) anonymous shared memory
3369 // - (bit 2) file-backed private memory
3370 // - (bit 3) file-backed shared memory
3371 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3372 // effective only if the bit 2 is cleared)
3373 // - (bit 5) hugetlb private memory
3374 // - (bit 6) hugetlb shared memory
3375 // - (bit 7) dax private memory
3376 // - (bit 8) dax shared memory
3377 //
3378 static void set_coredump_filter(bool largepages, bool dax_shared) {
3379 FILE *f;
3380 long cdm;
3381 bool filter_changed = false;
3382
3383 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3384 return;
3385 }
3386
3387 if (fscanf(f, "%lx", &cdm) != 1) {
3388 fclose(f);
3389 return;
3390 }
3391
3392 rewind(f);
3393
3394 if (largepages && (cdm & LARGEPAGES_BIT) == 0) {
3395 cdm |= LARGEPAGES_BIT;
3396 filter_changed = true;
3397 }
3398 if (dax_shared && (cdm & DAX_SHARED_BIT) == 0) {
3399 cdm |= DAX_SHARED_BIT;
3400 filter_changed = true;
3401 }
3402 if (filter_changed) {
3403 fprintf(f, "%#lx", cdm);
3404 }
3405
3406 fclose(f);
3407 }
3408
3409 // Large page support
3410
3411 static size_t _large_page_size = 0;
3412
3413 size_t os::Linux::find_large_page_size() {
3414 size_t large_page_size = 0;
3415
3416 // large_page_size on Linux is used to round up heap size. x86 uses either
3417 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3418 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3419 // page as large as 256M.
3420 //
3421 // Here we try to figure out page size by parsing /proc/meminfo and looking
3422 // for a line with the following format:
3423 // Hugepagesize: 2048 kB
3424 //
3425 // If we can't determine the value (e.g. /proc is not mounted, or the text
3426 // format has been changed), we'll use the largest page size supported by
3427 // the processor.
3428
3429 #ifndef ZERO
3430 large_page_size =
3431 AARCH64_ONLY(2 * M)
3432 AMD64_ONLY(2 * M)
3433 ARM32_ONLY(2 * M)
3434 IA32_ONLY(4 * M)
3435 IA64_ONLY(256 * M)
3436 PPC_ONLY(4 * M)
3437 S390_ONLY(1 * M)
3438 SPARC_ONLY(4 * M);
3439 #endif // ZERO
3440
3441 FILE *fp = fopen("/proc/meminfo", "r");
3442 if (fp) {
3443 while (!feof(fp)) {
3444 int x = 0;
3445 char buf[16];
3446 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3447 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3448 large_page_size = x * K;
3449 break;
3450 }
3451 } else {
3452 // skip to next line
3453 for (;;) {
3454 int ch = fgetc(fp);
3455 if (ch == EOF || ch == (int)'\n') break;
3456 }
3457 }
3458 }
3459 fclose(fp);
3460 }
3461
3462 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3463 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3464 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3465 proper_unit_for_byte_size(large_page_size));
3466 }
3467
3468 return large_page_size;
3469 }
3470
3471 size_t os::Linux::setup_large_page_size() {
3472 _large_page_size = Linux::find_large_page_size();
3473 const size_t default_page_size = (size_t)Linux::page_size();
3474 if (_large_page_size > default_page_size) {
3475 _page_sizes[0] = _large_page_size;
3476 _page_sizes[1] = default_page_size;
3477 _page_sizes[2] = 0;
3478 }
3479
3480 return _large_page_size;
3481 }
3482
3483 bool os::Linux::setup_large_page_type(size_t page_size) {
3484 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3485 FLAG_IS_DEFAULT(UseSHM) &&
3486 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3487
3488 // The type of large pages has not been specified by the user.
3489
3490 // Try UseHugeTLBFS and then UseSHM.
3491 UseHugeTLBFS = UseSHM = true;
3492
3493 // Don't try UseTransparentHugePages since there are known
3494 // performance issues with it turned on. This might change in the future.
3495 UseTransparentHugePages = false;
3496 }
3497
3498 if (UseTransparentHugePages) {
3499 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3500 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3501 UseHugeTLBFS = false;
3502 UseSHM = false;
3503 return true;
3504 }
3505 UseTransparentHugePages = false;
3506 }
3507
3508 if (UseHugeTLBFS) {
3509 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3510 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3511 UseSHM = false;
3512 return true;
3513 }
3514 UseHugeTLBFS = false;
3515 }
3516
3517 return UseSHM;
3518 }
3519
3520 void os::large_page_init() {
3521 if (!UseLargePages &&
3522 !UseTransparentHugePages &&
3523 !UseHugeTLBFS &&
3524 !UseSHM) {
3525 // Not using large pages.
3526 return;
3527 }
3528
3529 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3530 // The user explicitly turned off large pages.
3531 // Ignore the rest of the large pages flags.
3532 UseTransparentHugePages = false;
3533 UseHugeTLBFS = false;
3534 UseSHM = false;
3535 return;
3536 }
3537
3538 size_t large_page_size = Linux::setup_large_page_size();
3539 UseLargePages = Linux::setup_large_page_type(large_page_size);
3540
3541 set_coredump_filter(true /*largepages*/, false /*dax_shared*/);
3542 }
3543
3544 #ifndef SHM_HUGETLB
3545 #define SHM_HUGETLB 04000
3546 #endif
3547
3548 #define shm_warning_format(format, ...) \
3549 do { \
3550 if (UseLargePages && \
3551 (!FLAG_IS_DEFAULT(UseLargePages) || \
3552 !FLAG_IS_DEFAULT(UseSHM) || \
3553 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3554 warning(format, __VA_ARGS__); \
3555 } \
3556 } while (0)
3557
3558 #define shm_warning(str) shm_warning_format("%s", str)
3559
3560 #define shm_warning_with_errno(str) \
3561 do { \
3562 int err = errno; \
3563 shm_warning_format(str " (error = %d)", err); \
3564 } while (0)
3565
3566 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3567 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3568
3569 if (!is_aligned(alignment, SHMLBA)) {
3570 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3571 return NULL;
3572 }
3573
3574 // To ensure that we get 'alignment' aligned memory from shmat,
3575 // we pre-reserve aligned virtual memory and then attach to that.
3576
3577 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3578 if (pre_reserved_addr == NULL) {
3579 // Couldn't pre-reserve aligned memory.
3580 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3581 return NULL;
3582 }
3583
3584 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3585 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3586
3587 if ((intptr_t)addr == -1) {
3588 int err = errno;
3589 shm_warning_with_errno("Failed to attach shared memory.");
3590
3591 assert(err != EACCES, "Unexpected error");
3592 assert(err != EIDRM, "Unexpected error");
3593 assert(err != EINVAL, "Unexpected error");
3594
3595 // Since we don't know if the kernel unmapped the pre-reserved memory area
3596 // we can't unmap it, since that would potentially unmap memory that was
3597 // mapped from other threads.
3598 return NULL;
3599 }
3600
3601 return addr;
3602 }
3603
3604 static char* shmat_at_address(int shmid, char* req_addr) {
3605 if (!is_aligned(req_addr, SHMLBA)) {
3606 assert(false, "Requested address needs to be SHMLBA aligned");
3607 return NULL;
3608 }
3609
3610 char* addr = (char*)shmat(shmid, req_addr, 0);
3611
3612 if ((intptr_t)addr == -1) {
3613 shm_warning_with_errno("Failed to attach shared memory.");
3614 return NULL;
3615 }
3616
3617 return addr;
3618 }
3619
3620 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3621 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3622 if (req_addr != NULL) {
3623 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3624 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3625 return shmat_at_address(shmid, req_addr);
3626 }
3627
3628 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3629 // return large page size aligned memory addresses when req_addr == NULL.
3630 // However, if the alignment is larger than the large page size, we have
3631 // to manually ensure that the memory returned is 'alignment' aligned.
3632 if (alignment > os::large_page_size()) {
3633 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3634 return shmat_with_alignment(shmid, bytes, alignment);
3635 } else {
3636 return shmat_at_address(shmid, NULL);
3637 }
3638 }
3639
3640 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3641 char* req_addr, bool exec) {
3642 // "exec" is passed in but not used. Creating the shared image for
3643 // the code cache doesn't have an SHM_X executable permission to check.
3644 assert(UseLargePages && UseSHM, "only for SHM large pages");
3645 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3646 assert(is_aligned(req_addr, alignment), "Unaligned address");
3647
3648 if (!is_aligned(bytes, os::large_page_size())) {
3649 return NULL; // Fallback to small pages.
3650 }
3651
3652 // Create a large shared memory region to attach to based on size.
3653 // Currently, size is the total size of the heap.
3654 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3655 if (shmid == -1) {
3656 // Possible reasons for shmget failure:
3657 // 1. shmmax is too small for Java heap.
3658 // > check shmmax value: cat /proc/sys/kernel/shmmax
3659 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3660 // 2. not enough large page memory.
3661 // > check available large pages: cat /proc/meminfo
3662 // > increase amount of large pages:
3663 // echo new_value > /proc/sys/vm/nr_hugepages
3664 // Note 1: different Linux may use different name for this property,
3665 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3666 // Note 2: it's possible there's enough physical memory available but
3667 // they are so fragmented after a long run that they can't
3668 // coalesce into large pages. Try to reserve large pages when
3669 // the system is still "fresh".
3670 shm_warning_with_errno("Failed to reserve shared memory.");
3671 return NULL;
3672 }
3673
3674 // Attach to the region.
3675 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3676
3677 // Remove shmid. If shmat() is successful, the actual shared memory segment
3678 // will be deleted when it's detached by shmdt() or when the process
3679 // terminates. If shmat() is not successful this will remove the shared
3680 // segment immediately.
3681 shmctl(shmid, IPC_RMID, NULL);
3682
3683 return addr;
3684 }
3685
3686 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3687 int error) {
3688 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3689
3690 bool warn_on_failure = UseLargePages &&
3691 (!FLAG_IS_DEFAULT(UseLargePages) ||
3692 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3693 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3694
3695 if (warn_on_failure) {
3696 char msg[128];
3697 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3698 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3699 warning("%s", msg);
3700 }
3701 }
3702
3703 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3704 char* req_addr,
3705 bool exec) {
3706 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3707 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3708 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3709
3710 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3711 char* addr = (char*)::mmap(req_addr, bytes, prot,
3712 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3713 -1, 0);
3714
3715 if (addr == MAP_FAILED) {
3716 warn_on_large_pages_failure(req_addr, bytes, errno);
3717 return NULL;
3718 }
3719
3720 assert(is_aligned(addr, os::large_page_size()), "Must be");
3721
3722 return addr;
3723 }
3724
3725 // Reserve memory using mmap(MAP_HUGETLB).
3726 // - bytes shall be a multiple of alignment.
3727 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3728 // - alignment sets the alignment at which memory shall be allocated.
3729 // It must be a multiple of allocation granularity.
3730 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3731 // req_addr or NULL.
3732 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3733 size_t alignment,
3734 char* req_addr,
3735 bool exec) {
3736 size_t large_page_size = os::large_page_size();
3737 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3738
3739 assert(is_aligned(req_addr, alignment), "Must be");
3740 assert(is_aligned(bytes, alignment), "Must be");
3741
3742 // First reserve - but not commit - the address range in small pages.
3743 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3744
3745 if (start == NULL) {
3746 return NULL;
3747 }
3748
3749 assert(is_aligned(start, alignment), "Must be");
3750
3751 char* end = start + bytes;
3752
3753 // Find the regions of the allocated chunk that can be promoted to large pages.
3754 char* lp_start = align_up(start, large_page_size);
3755 char* lp_end = align_down(end, large_page_size);
3756
3757 size_t lp_bytes = lp_end - lp_start;
3758
3759 assert(is_aligned(lp_bytes, large_page_size), "Must be");
3760
3761 if (lp_bytes == 0) {
3762 // The mapped region doesn't even span the start and the end of a large page.
3763 // Fall back to allocate a non-special area.
3764 ::munmap(start, end - start);
3765 return NULL;
3766 }
3767
3768 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3769
3770 void* result;
3771
3772 // Commit small-paged leading area.
3773 if (start != lp_start) {
3774 result = ::mmap(start, lp_start - start, prot,
3775 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3776 -1, 0);
3777 if (result == MAP_FAILED) {
3778 ::munmap(lp_start, end - lp_start);
3779 return NULL;
3780 }
3781 }
3782
3783 // Commit large-paged area.
3784 result = ::mmap(lp_start, lp_bytes, prot,
3785 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3786 -1, 0);
3787 if (result == MAP_FAILED) {
3788 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3789 // If the mmap above fails, the large pages region will be unmapped and we
3790 // have regions before and after with small pages. Release these regions.
3791 //
3792 // | mapped | unmapped | mapped |
3793 // ^ ^ ^ ^
3794 // start lp_start lp_end end
3795 //
3796 ::munmap(start, lp_start - start);
3797 ::munmap(lp_end, end - lp_end);
3798 return NULL;
3799 }
3800
3801 // Commit small-paged trailing area.
3802 if (lp_end != end) {
3803 result = ::mmap(lp_end, end - lp_end, prot,
3804 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3805 -1, 0);
3806 if (result == MAP_FAILED) {
3807 ::munmap(start, lp_end - start);
3808 return NULL;
3809 }
3810 }
3811
3812 return start;
3813 }
3814
3815 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3816 size_t alignment,
3817 char* req_addr,
3818 bool exec) {
3819 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3820 assert(is_aligned(req_addr, alignment), "Must be");
3821 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3822 assert(is_power_of_2(os::large_page_size()), "Must be");
3823 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3824
3825 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3826 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3827 } else {
3828 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3829 }
3830 }
3831
3832 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3833 char* req_addr, bool exec) {
3834 assert(UseLargePages, "only for large pages");
3835
3836 char* addr;
3837 if (UseSHM) {
3838 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3839 } else {
3840 assert(UseHugeTLBFS, "must be");
3841 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3842 }
3843
3844 if (addr != NULL) {
3845 if (UseNUMAInterleaving) {
3846 numa_make_global(addr, bytes);
3847 }
3848
3849 // The memory is committed
3850 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3851 }
3852
3853 return addr;
3854 }
3855
3856 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3857 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3858 return shmdt(base) == 0;
3859 }
3860
3861 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3862 return pd_release_memory(base, bytes);
3863 }
3864
3865 bool os::release_memory_special(char* base, size_t bytes) {
3866 bool res;
3867 if (MemTracker::tracking_level() > NMT_minimal) {
3868 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3869 res = os::Linux::release_memory_special_impl(base, bytes);
3870 if (res) {
3871 tkr.record((address)base, bytes);
3872 }
3873
3874 } else {
3875 res = os::Linux::release_memory_special_impl(base, bytes);
3876 }
3877 return res;
3878 }
3879
3880 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3881 assert(UseLargePages, "only for large pages");
3882 bool res;
3883
3884 if (UseSHM) {
3885 res = os::Linux::release_memory_special_shm(base, bytes);
3886 } else {
3887 assert(UseHugeTLBFS, "must be");
3888 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3889 }
3890 return res;
3891 }
3892
3893 size_t os::large_page_size() {
3894 return _large_page_size;
3895 }
3896
3897 // With SysV SHM the entire memory region must be allocated as shared
3898 // memory.
3899 // HugeTLBFS allows application to commit large page memory on demand.
3900 // However, when committing memory with HugeTLBFS fails, the region
3901 // that was supposed to be committed will lose the old reservation
3902 // and allow other threads to steal that memory region. Because of this
3903 // behavior we can't commit HugeTLBFS memory.
3904 bool os::can_commit_large_page_memory() {
3905 return UseTransparentHugePages;
3906 }
3907
3908 bool os::can_execute_large_page_memory() {
3909 return UseTransparentHugePages || UseHugeTLBFS;
3910 }
3911
3912 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
3913 assert(file_desc >= 0, "file_desc is not valid");
3914 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
3915 if (result != NULL) {
3916 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
3917 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
3918 }
3919 }
3920 return result;
3921 }
3922
3923 // Reserve memory at an arbitrary address, only if that area is
3924 // available (and not reserved for something else).
3925
3926 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3927 const int max_tries = 10;
3928 char* base[max_tries];
3929 size_t size[max_tries];
3930 const size_t gap = 0x000000;
3931
3932 // Assert only that the size is a multiple of the page size, since
3933 // that's all that mmap requires, and since that's all we really know
3934 // about at this low abstraction level. If we need higher alignment,
3935 // we can either pass an alignment to this method or verify alignment
3936 // in one of the methods further up the call chain. See bug 5044738.
3937 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3938
3939 // Repeatedly allocate blocks until the block is allocated at the
3940 // right spot.
3941
3942 // Linux mmap allows caller to pass an address as hint; give it a try first,
3943 // if kernel honors the hint then we can return immediately.
3944 char * addr = anon_mmap(requested_addr, bytes, false);
3945 if (addr == requested_addr) {
3946 return requested_addr;
3947 }
3948
3949 if (addr != NULL) {
3950 // mmap() is successful but it fails to reserve at the requested address
3951 anon_munmap(addr, bytes);
3952 }
3953
3954 int i;
3955 for (i = 0; i < max_tries; ++i) {
3956 base[i] = reserve_memory(bytes);
3957
3958 if (base[i] != NULL) {
3959 // Is this the block we wanted?
3960 if (base[i] == requested_addr) {
3961 size[i] = bytes;
3962 break;
3963 }
3964
3965 // Does this overlap the block we wanted? Give back the overlapped
3966 // parts and try again.
3967
3968 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3969 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3970 unmap_memory(base[i], top_overlap);
3971 base[i] += top_overlap;
3972 size[i] = bytes - top_overlap;
3973 } else {
3974 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3975 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3976 unmap_memory(requested_addr, bottom_overlap);
3977 size[i] = bytes - bottom_overlap;
3978 } else {
3979 size[i] = bytes;
3980 }
3981 }
3982 }
3983 }
3984
3985 // Give back the unused reserved pieces.
3986
3987 for (int j = 0; j < i; ++j) {
3988 if (base[j] != NULL) {
3989 unmap_memory(base[j], size[j]);
3990 }
3991 }
3992
3993 if (i < max_tries) {
3994 return requested_addr;
3995 } else {
3996 return NULL;
3997 }
3998 }
3999
4000 size_t os::read(int fd, void *buf, unsigned int nBytes) {
4001 return ::read(fd, buf, nBytes);
4002 }
4003
4004 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
4005 return ::pread(fd, buf, nBytes, offset);
4006 }
4007
4008 // Short sleep, direct OS call.
4009 //
4010 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
4011 // sched_yield(2) will actually give up the CPU:
4012 //
4013 // * Alone on this pariticular CPU, keeps running.
4014 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
4015 // (pre 2.6.39).
4016 //
4017 // So calling this with 0 is an alternative.
4018 //
4019 void os::naked_short_sleep(jlong ms) {
4020 struct timespec req;
4021
4022 assert(ms < 1000, "Un-interruptable sleep, short time use only");
4023 req.tv_sec = 0;
4024 if (ms > 0) {
4025 req.tv_nsec = (ms % 1000) * 1000000;
4026 } else {
4027 req.tv_nsec = 1;
4028 }
4029
4030 nanosleep(&req, NULL);
4031
4032 return;
4033 }
4034
4035 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4036 void os::infinite_sleep() {
4037 while (true) { // sleep forever ...
4038 ::sleep(100); // ... 100 seconds at a time
4039 }
4040 }
4041
4042 // Used to convert frequent JVM_Yield() to nops
4043 bool os::dont_yield() {
4044 return DontYieldALot;
4045 }
4046
4047 void os::naked_yield() {
4048 sched_yield();
4049 }
4050
4051 ////////////////////////////////////////////////////////////////////////////////
4052 // thread priority support
4053
4054 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4055 // only supports dynamic priority, static priority must be zero. For real-time
4056 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4057 // However, for large multi-threaded applications, SCHED_RR is not only slower
4058 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4059 // of 5 runs - Sep 2005).
4060 //
4061 // The following code actually changes the niceness of kernel-thread/LWP. It
4062 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4063 // not the entire user process, and user level threads are 1:1 mapped to kernel
4064 // threads. It has always been the case, but could change in the future. For
4065 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4066 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4067
4068 int os::java_to_os_priority[CriticalPriority + 1] = {
4069 19, // 0 Entry should never be used
4070
4071 4, // 1 MinPriority
4072 3, // 2
4073 2, // 3
4074
4075 1, // 4
4076 0, // 5 NormPriority
4077 -1, // 6
4078
4079 -2, // 7
4080 -3, // 8
4081 -4, // 9 NearMaxPriority
4082
4083 -5, // 10 MaxPriority
4084
4085 -5 // 11 CriticalPriority
4086 };
4087
4088 static int prio_init() {
4089 if (ThreadPriorityPolicy == 1) {
4090 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4091 // if effective uid is not root. Perhaps, a more elegant way of doing
4092 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4093 if (geteuid() != 0) {
4094 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4095 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4096 }
4097 ThreadPriorityPolicy = 0;
4098 }
4099 }
4100 if (UseCriticalJavaThreadPriority) {
4101 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4102 }
4103 return 0;
4104 }
4105
4106 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4107 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4108
4109 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4110 return (ret == 0) ? OS_OK : OS_ERR;
4111 }
4112
4113 OSReturn os::get_native_priority(const Thread* const thread,
4114 int *priority_ptr) {
4115 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4116 *priority_ptr = java_to_os_priority[NormPriority];
4117 return OS_OK;
4118 }
4119
4120 errno = 0;
4121 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4122 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4123 }
4124
4125 // Hint to the underlying OS that a task switch would not be good.
4126 // Void return because it's a hint and can fail.
4127 void os::hint_no_preempt() {}
4128
4129 ////////////////////////////////////////////////////////////////////////////////
4130 // suspend/resume support
4131
4132 // The low-level signal-based suspend/resume support is a remnant from the
4133 // old VM-suspension that used to be for java-suspension, safepoints etc,
4134 // within hotspot. Currently used by JFR's OSThreadSampler
4135 //
4136 // The remaining code is greatly simplified from the more general suspension
4137 // code that used to be used.
4138 //
4139 // The protocol is quite simple:
4140 // - suspend:
4141 // - sends a signal to the target thread
4142 // - polls the suspend state of the osthread using a yield loop
4143 // - target thread signal handler (SR_handler) sets suspend state
4144 // and blocks in sigsuspend until continued
4145 // - resume:
4146 // - sets target osthread state to continue
4147 // - sends signal to end the sigsuspend loop in the SR_handler
4148 //
4149 // Note that the SR_lock plays no role in this suspend/resume protocol,
4150 // but is checked for NULL in SR_handler as a thread termination indicator.
4151 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4152 //
4153 // Note that resume_clear_context() and suspend_save_context() are needed
4154 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4155 // which in part is used by:
4156 // - Forte Analyzer: AsyncGetCallTrace()
4157 // - StackBanging: get_frame_at_stack_banging_point()
4158
4159 static void resume_clear_context(OSThread *osthread) {
4160 osthread->set_ucontext(NULL);
4161 osthread->set_siginfo(NULL);
4162 }
4163
4164 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4165 ucontext_t* context) {
4166 osthread->set_ucontext(context);
4167 osthread->set_siginfo(siginfo);
4168 }
4169
4170 // Handler function invoked when a thread's execution is suspended or
4171 // resumed. We have to be careful that only async-safe functions are
4172 // called here (Note: most pthread functions are not async safe and
4173 // should be avoided.)
4174 //
4175 // Note: sigwait() is a more natural fit than sigsuspend() from an
4176 // interface point of view, but sigwait() prevents the signal hander
4177 // from being run. libpthread would get very confused by not having
4178 // its signal handlers run and prevents sigwait()'s use with the
4179 // mutex granting granting signal.
4180 //
4181 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4182 //
4183 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4184 // Save and restore errno to avoid confusing native code with EINTR
4185 // after sigsuspend.
4186 int old_errno = errno;
4187
4188 Thread* thread = Thread::current_or_null_safe();
4189 assert(thread != NULL, "Missing current thread in SR_handler");
4190
4191 // On some systems we have seen signal delivery get "stuck" until the signal
4192 // mask is changed as part of thread termination. Check that the current thread
4193 // has not already terminated (via SR_lock()) - else the following assertion
4194 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4195 // destructor has completed.
4196
4197 if (thread->SR_lock() == NULL) {
4198 return;
4199 }
4200
4201 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4202
4203 OSThread* osthread = thread->osthread();
4204
4205 os::SuspendResume::State current = osthread->sr.state();
4206 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4207 suspend_save_context(osthread, siginfo, context);
4208
4209 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4210 os::SuspendResume::State state = osthread->sr.suspended();
4211 if (state == os::SuspendResume::SR_SUSPENDED) {
4212 sigset_t suspend_set; // signals for sigsuspend()
4213 sigemptyset(&suspend_set);
4214 // get current set of blocked signals and unblock resume signal
4215 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4216 sigdelset(&suspend_set, SR_signum);
4217
4218 sr_semaphore.signal();
4219 // wait here until we are resumed
4220 while (1) {
4221 sigsuspend(&suspend_set);
4222
4223 os::SuspendResume::State result = osthread->sr.running();
4224 if (result == os::SuspendResume::SR_RUNNING) {
4225 sr_semaphore.signal();
4226 break;
4227 }
4228 }
4229
4230 } else if (state == os::SuspendResume::SR_RUNNING) {
4231 // request was cancelled, continue
4232 } else {
4233 ShouldNotReachHere();
4234 }
4235
4236 resume_clear_context(osthread);
4237 } else if (current == os::SuspendResume::SR_RUNNING) {
4238 // request was cancelled, continue
4239 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4240 // ignore
4241 } else {
4242 // ignore
4243 }
4244
4245 errno = old_errno;
4246 }
4247
4248 static int SR_initialize() {
4249 struct sigaction act;
4250 char *s;
4251
4252 // Get signal number to use for suspend/resume
4253 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4254 int sig = ::strtol(s, 0, 10);
4255 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4256 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4257 SR_signum = sig;
4258 } else {
4259 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4260 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4261 }
4262 }
4263
4264 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4265 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4266
4267 sigemptyset(&SR_sigset);
4268 sigaddset(&SR_sigset, SR_signum);
4269
4270 // Set up signal handler for suspend/resume
4271 act.sa_flags = SA_RESTART|SA_SIGINFO;
4272 act.sa_handler = (void (*)(int)) SR_handler;
4273
4274 // SR_signum is blocked by default.
4275 // 4528190 - We also need to block pthread restart signal (32 on all
4276 // supported Linux platforms). Note that LinuxThreads need to block
4277 // this signal for all threads to work properly. So we don't have
4278 // to use hard-coded signal number when setting up the mask.
4279 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4280
4281 if (sigaction(SR_signum, &act, 0) == -1) {
4282 return -1;
4283 }
4284
4285 // Save signal flag
4286 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4287 return 0;
4288 }
4289
4290 static int sr_notify(OSThread* osthread) {
4291 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4292 assert_status(status == 0, status, "pthread_kill");
4293 return status;
4294 }
4295
4296 // "Randomly" selected value for how long we want to spin
4297 // before bailing out on suspending a thread, also how often
4298 // we send a signal to a thread we want to resume
4299 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4300 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4301
4302 // returns true on success and false on error - really an error is fatal
4303 // but this seems the normal response to library errors
4304 static bool do_suspend(OSThread* osthread) {
4305 assert(osthread->sr.is_running(), "thread should be running");
4306 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4307
4308 // mark as suspended and send signal
4309 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4310 // failed to switch, state wasn't running?
4311 ShouldNotReachHere();
4312 return false;
4313 }
4314
4315 if (sr_notify(osthread) != 0) {
4316 ShouldNotReachHere();
4317 }
4318
4319 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4320 while (true) {
4321 if (sr_semaphore.timedwait(create_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4322 break;
4323 } else {
4324 // timeout
4325 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4326 if (cancelled == os::SuspendResume::SR_RUNNING) {
4327 return false;
4328 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4329 // make sure that we consume the signal on the semaphore as well
4330 sr_semaphore.wait();
4331 break;
4332 } else {
4333 ShouldNotReachHere();
4334 return false;
4335 }
4336 }
4337 }
4338
4339 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4340 return true;
4341 }
4342
4343 static void do_resume(OSThread* osthread) {
4344 assert(osthread->sr.is_suspended(), "thread should be suspended");
4345 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4346
4347 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4348 // failed to switch to WAKEUP_REQUEST
4349 ShouldNotReachHere();
4350 return;
4351 }
4352
4353 while (true) {
4354 if (sr_notify(osthread) == 0) {
4355 if (sr_semaphore.timedwait(create_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4356 if (osthread->sr.is_running()) {
4357 return;
4358 }
4359 }
4360 } else {
4361 ShouldNotReachHere();
4362 }
4363 }
4364
4365 guarantee(osthread->sr.is_running(), "Must be running!");
4366 }
4367
4368 ///////////////////////////////////////////////////////////////////////////////////
4369 // signal handling (except suspend/resume)
4370
4371 // This routine may be used by user applications as a "hook" to catch signals.
4372 // The user-defined signal handler must pass unrecognized signals to this
4373 // routine, and if it returns true (non-zero), then the signal handler must
4374 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4375 // routine will never retun false (zero), but instead will execute a VM panic
4376 // routine kill the process.
4377 //
4378 // If this routine returns false, it is OK to call it again. This allows
4379 // the user-defined signal handler to perform checks either before or after
4380 // the VM performs its own checks. Naturally, the user code would be making
4381 // a serious error if it tried to handle an exception (such as a null check
4382 // or breakpoint) that the VM was generating for its own correct operation.
4383 //
4384 // This routine may recognize any of the following kinds of signals:
4385 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4386 // It should be consulted by handlers for any of those signals.
4387 //
4388 // The caller of this routine must pass in the three arguments supplied
4389 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4390 // field of the structure passed to sigaction(). This routine assumes that
4391 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4392 //
4393 // Note that the VM will print warnings if it detects conflicting signal
4394 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4395 //
4396 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4397 siginfo_t* siginfo,
4398 void* ucontext,
4399 int abort_if_unrecognized);
4400
4401 void signalHandler(int sig, siginfo_t* info, void* uc) {
4402 assert(info != NULL && uc != NULL, "it must be old kernel");
4403 int orig_errno = errno; // Preserve errno value over signal handler.
4404 JVM_handle_linux_signal(sig, info, uc, true);
4405 errno = orig_errno;
4406 }
4407
4408
4409 // This boolean allows users to forward their own non-matching signals
4410 // to JVM_handle_linux_signal, harmlessly.
4411 bool os::Linux::signal_handlers_are_installed = false;
4412
4413 // For signal-chaining
4414 struct sigaction sigact[NSIG];
4415 uint64_t sigs = 0;
4416 #if (64 < NSIG-1)
4417 #error "Not all signals can be encoded in sigs. Adapt its type!"
4418 #endif
4419 bool os::Linux::libjsig_is_loaded = false;
4420 typedef struct sigaction *(*get_signal_t)(int);
4421 get_signal_t os::Linux::get_signal_action = NULL;
4422
4423 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4424 struct sigaction *actp = NULL;
4425
4426 if (libjsig_is_loaded) {
4427 // Retrieve the old signal handler from libjsig
4428 actp = (*get_signal_action)(sig);
4429 }
4430 if (actp == NULL) {
4431 // Retrieve the preinstalled signal handler from jvm
4432 actp = get_preinstalled_handler(sig);
4433 }
4434
4435 return actp;
4436 }
4437
4438 static bool call_chained_handler(struct sigaction *actp, int sig,
4439 siginfo_t *siginfo, void *context) {
4440 // Call the old signal handler
4441 if (actp->sa_handler == SIG_DFL) {
4442 // It's more reasonable to let jvm treat it as an unexpected exception
4443 // instead of taking the default action.
4444 return false;
4445 } else if (actp->sa_handler != SIG_IGN) {
4446 if ((actp->sa_flags & SA_NODEFER) == 0) {
4447 // automaticlly block the signal
4448 sigaddset(&(actp->sa_mask), sig);
4449 }
4450
4451 sa_handler_t hand = NULL;
4452 sa_sigaction_t sa = NULL;
4453 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4454 // retrieve the chained handler
4455 if (siginfo_flag_set) {
4456 sa = actp->sa_sigaction;
4457 } else {
4458 hand = actp->sa_handler;
4459 }
4460
4461 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4462 actp->sa_handler = SIG_DFL;
4463 }
4464
4465 // try to honor the signal mask
4466 sigset_t oset;
4467 sigemptyset(&oset);
4468 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4469
4470 // call into the chained handler
4471 if (siginfo_flag_set) {
4472 (*sa)(sig, siginfo, context);
4473 } else {
4474 (*hand)(sig);
4475 }
4476
4477 // restore the signal mask
4478 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4479 }
4480 // Tell jvm's signal handler the signal is taken care of.
4481 return true;
4482 }
4483
4484 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4485 bool chained = false;
4486 // signal-chaining
4487 if (UseSignalChaining) {
4488 struct sigaction *actp = get_chained_signal_action(sig);
4489 if (actp != NULL) {
4490 chained = call_chained_handler(actp, sig, siginfo, context);
4491 }
4492 }
4493 return chained;
4494 }
4495
4496 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4497 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4498 return &sigact[sig];
4499 }
4500 return NULL;
4501 }
4502
4503 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4504 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4505 sigact[sig] = oldAct;
4506 sigs |= (uint64_t)1 << (sig-1);
4507 }
4508
4509 // for diagnostic
4510 int sigflags[NSIG];
4511
4512 int os::Linux::get_our_sigflags(int sig) {
4513 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4514 return sigflags[sig];
4515 }
4516
4517 void os::Linux::set_our_sigflags(int sig, int flags) {
4518 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4519 if (sig > 0 && sig < NSIG) {
4520 sigflags[sig] = flags;
4521 }
4522 }
4523
4524 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4525 // Check for overwrite.
4526 struct sigaction oldAct;
4527 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4528
4529 void* oldhand = oldAct.sa_sigaction
4530 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4531 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4532 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4533 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4534 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4535 if (AllowUserSignalHandlers || !set_installed) {
4536 // Do not overwrite; user takes responsibility to forward to us.
4537 return;
4538 } else if (UseSignalChaining) {
4539 // save the old handler in jvm
4540 save_preinstalled_handler(sig, oldAct);
4541 // libjsig also interposes the sigaction() call below and saves the
4542 // old sigaction on it own.
4543 } else {
4544 fatal("Encountered unexpected pre-existing sigaction handler "
4545 "%#lx for signal %d.", (long)oldhand, sig);
4546 }
4547 }
4548
4549 struct sigaction sigAct;
4550 sigfillset(&(sigAct.sa_mask));
4551 sigAct.sa_handler = SIG_DFL;
4552 if (!set_installed) {
4553 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4554 } else {
4555 sigAct.sa_sigaction = signalHandler;
4556 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4557 }
4558 // Save flags, which are set by ours
4559 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4560 sigflags[sig] = sigAct.sa_flags;
4561
4562 int ret = sigaction(sig, &sigAct, &oldAct);
4563 assert(ret == 0, "check");
4564
4565 void* oldhand2 = oldAct.sa_sigaction
4566 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4567 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4568 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4569 }
4570
4571 // install signal handlers for signals that HotSpot needs to
4572 // handle in order to support Java-level exception handling.
4573
4574 void os::Linux::install_signal_handlers() {
4575 if (!signal_handlers_are_installed) {
4576 signal_handlers_are_installed = true;
4577
4578 // signal-chaining
4579 typedef void (*signal_setting_t)();
4580 signal_setting_t begin_signal_setting = NULL;
4581 signal_setting_t end_signal_setting = NULL;
4582 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4583 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4584 if (begin_signal_setting != NULL) {
4585 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4586 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4587 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4588 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4589 libjsig_is_loaded = true;
4590 assert(UseSignalChaining, "should enable signal-chaining");
4591 }
4592 if (libjsig_is_loaded) {
4593 // Tell libjsig jvm is setting signal handlers
4594 (*begin_signal_setting)();
4595 }
4596
4597 set_signal_handler(SIGSEGV, true);
4598 set_signal_handler(SIGPIPE, true);
4599 set_signal_handler(SIGBUS, true);
4600 set_signal_handler(SIGILL, true);
4601 set_signal_handler(SIGFPE, true);
4602 #if defined(PPC64)
4603 set_signal_handler(SIGTRAP, true);
4604 #endif
4605 set_signal_handler(SIGXFSZ, true);
4606
4607 if (libjsig_is_loaded) {
4608 // Tell libjsig jvm finishes setting signal handlers
4609 (*end_signal_setting)();
4610 }
4611
4612 // We don't activate signal checker if libjsig is in place, we trust ourselves
4613 // and if UserSignalHandler is installed all bets are off.
4614 // Log that signal checking is off only if -verbose:jni is specified.
4615 if (CheckJNICalls) {
4616 if (libjsig_is_loaded) {
4617 if (PrintJNIResolving) {
4618 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4619 }
4620 check_signals = false;
4621 }
4622 if (AllowUserSignalHandlers) {
4623 if (PrintJNIResolving) {
4624 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4625 }
4626 check_signals = false;
4627 }
4628 }
4629 }
4630 }
4631
4632 // This is the fastest way to get thread cpu time on Linux.
4633 // Returns cpu time (user+sys) for any thread, not only for current.
4634 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4635 // It might work on 2.6.10+ with a special kernel/glibc patch.
4636 // For reference, please, see IEEE Std 1003.1-2004:
4637 // http://www.unix.org/single_unix_specification
4638
4639 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4640 struct timespec tp;
4641 int rc = os::Linux::clock_gettime(clockid, &tp);
4642 assert(rc == 0, "clock_gettime is expected to return 0 code");
4643
4644 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4645 }
4646
4647 void os::Linux::initialize_os_info() {
4648 assert(_os_version == 0, "OS info already initialized");
4649
4650 struct utsname _uname;
4651
4652 uint32_t major;
4653 uint32_t minor;
4654 uint32_t fix;
4655
4656 int rc;
4657
4658 // Kernel version is unknown if
4659 // verification below fails.
4660 _os_version = 0x01000000;
4661
4662 rc = uname(&_uname);
4663 if (rc != -1) {
4664
4665 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4666 if (rc == 3) {
4667
4668 if (major < 256 && minor < 256 && fix < 256) {
4669 // Kernel version format is as expected,
4670 // set it overriding unknown state.
4671 _os_version = (major << 16) |
4672 (minor << 8 ) |
4673 (fix << 0 ) ;
4674 }
4675 }
4676 }
4677 }
4678
4679 uint32_t os::Linux::os_version() {
4680 assert(_os_version != 0, "not initialized");
4681 return _os_version & 0x00FFFFFF;
4682 }
4683
4684 bool os::Linux::os_version_is_known() {
4685 assert(_os_version != 0, "not initialized");
4686 return _os_version & 0x01000000 ? false : true;
4687 }
4688
4689 /////
4690 // glibc on Linux platform uses non-documented flag
4691 // to indicate, that some special sort of signal
4692 // trampoline is used.
4693 // We will never set this flag, and we should
4694 // ignore this flag in our diagnostic
4695 #ifdef SIGNIFICANT_SIGNAL_MASK
4696 #undef SIGNIFICANT_SIGNAL_MASK
4697 #endif
4698 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4699
4700 static const char* get_signal_handler_name(address handler,
4701 char* buf, int buflen) {
4702 int offset = 0;
4703 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4704 if (found) {
4705 // skip directory names
4706 const char *p1, *p2;
4707 p1 = buf;
4708 size_t len = strlen(os::file_separator());
4709 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4710 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4711 } else {
4712 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4713 }
4714 return buf;
4715 }
4716
4717 static void print_signal_handler(outputStream* st, int sig,
4718 char* buf, size_t buflen) {
4719 struct sigaction sa;
4720
4721 sigaction(sig, NULL, &sa);
4722
4723 // See comment for SIGNIFICANT_SIGNAL_MASK define
4724 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4725
4726 st->print("%s: ", os::exception_name(sig, buf, buflen));
4727
4728 address handler = (sa.sa_flags & SA_SIGINFO)
4729 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4730 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4731
4732 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4733 st->print("SIG_DFL");
4734 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4735 st->print("SIG_IGN");
4736 } else {
4737 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4738 }
4739
4740 st->print(", sa_mask[0]=");
4741 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4742
4743 address rh = VMError::get_resetted_sighandler(sig);
4744 // May be, handler was resetted by VMError?
4745 if (rh != NULL) {
4746 handler = rh;
4747 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4748 }
4749
4750 st->print(", sa_flags=");
4751 os::Posix::print_sa_flags(st, sa.sa_flags);
4752
4753 // Check: is it our handler?
4754 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4755 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4756 // It is our signal handler
4757 // check for flags, reset system-used one!
4758 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4759 st->print(
4760 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4761 os::Linux::get_our_sigflags(sig));
4762 }
4763 }
4764 st->cr();
4765 }
4766
4767
4768 #define DO_SIGNAL_CHECK(sig) \
4769 do { \
4770 if (!sigismember(&check_signal_done, sig)) { \
4771 os::Linux::check_signal_handler(sig); \
4772 } \
4773 } while (0)
4774
4775 // This method is a periodic task to check for misbehaving JNI applications
4776 // under CheckJNI, we can add any periodic checks here
4777
4778 void os::run_periodic_checks() {
4779 if (check_signals == false) return;
4780
4781 // SEGV and BUS if overridden could potentially prevent
4782 // generation of hs*.log in the event of a crash, debugging
4783 // such a case can be very challenging, so we absolutely
4784 // check the following for a good measure:
4785 DO_SIGNAL_CHECK(SIGSEGV);
4786 DO_SIGNAL_CHECK(SIGILL);
4787 DO_SIGNAL_CHECK(SIGFPE);
4788 DO_SIGNAL_CHECK(SIGBUS);
4789 DO_SIGNAL_CHECK(SIGPIPE);
4790 DO_SIGNAL_CHECK(SIGXFSZ);
4791 #if defined(PPC64)
4792 DO_SIGNAL_CHECK(SIGTRAP);
4793 #endif
4794
4795 // ReduceSignalUsage allows the user to override these handlers
4796 // see comments at the very top and jvm_md.h
4797 if (!ReduceSignalUsage) {
4798 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4799 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4800 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4801 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4802 }
4803
4804 DO_SIGNAL_CHECK(SR_signum);
4805 }
4806
4807 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4808
4809 static os_sigaction_t os_sigaction = NULL;
4810
4811 void os::Linux::check_signal_handler(int sig) {
4812 char buf[O_BUFLEN];
4813 address jvmHandler = NULL;
4814
4815
4816 struct sigaction act;
4817 if (os_sigaction == NULL) {
4818 // only trust the default sigaction, in case it has been interposed
4819 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4820 if (os_sigaction == NULL) return;
4821 }
4822
4823 os_sigaction(sig, (struct sigaction*)NULL, &act);
4824
4825
4826 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4827
4828 address thisHandler = (act.sa_flags & SA_SIGINFO)
4829 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4830 : CAST_FROM_FN_PTR(address, act.sa_handler);
4831
4832
4833 switch (sig) {
4834 case SIGSEGV:
4835 case SIGBUS:
4836 case SIGFPE:
4837 case SIGPIPE:
4838 case SIGILL:
4839 case SIGXFSZ:
4840 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4841 break;
4842
4843 case SHUTDOWN1_SIGNAL:
4844 case SHUTDOWN2_SIGNAL:
4845 case SHUTDOWN3_SIGNAL:
4846 case BREAK_SIGNAL:
4847 jvmHandler = (address)user_handler();
4848 break;
4849
4850 default:
4851 if (sig == SR_signum) {
4852 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4853 } else {
4854 return;
4855 }
4856 break;
4857 }
4858
4859 if (thisHandler != jvmHandler) {
4860 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4861 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4862 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4863 // No need to check this sig any longer
4864 sigaddset(&check_signal_done, sig);
4865 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4866 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4867 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4868 exception_name(sig, buf, O_BUFLEN));
4869 }
4870 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4871 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4872 tty->print("expected:");
4873 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4874 tty->cr();
4875 tty->print(" found:");
4876 os::Posix::print_sa_flags(tty, act.sa_flags);
4877 tty->cr();
4878 // No need to check this sig any longer
4879 sigaddset(&check_signal_done, sig);
4880 }
4881
4882 // Dump all the signal
4883 if (sigismember(&check_signal_done, sig)) {
4884 print_signal_handlers(tty, buf, O_BUFLEN);
4885 }
4886 }
4887
4888 extern void report_error(char* file_name, int line_no, char* title,
4889 char* format, ...);
4890
4891 // this is called _before_ most of the global arguments have been parsed
4892 void os::init(void) {
4893 char dummy; // used to get a guess on initial stack address
4894
4895 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4896
4897 init_random(1234567);
4898
4899 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4900 if (Linux::page_size() == -1) {
4901 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4902 os::strerror(errno));
4903 }
4904 init_page_sizes((size_t) Linux::page_size());
4905
4906 Linux::initialize_system_info();
4907
4908 Linux::initialize_os_info();
4909
4910 // _main_thread points to the thread that created/loaded the JVM.
4911 Linux::_main_thread = pthread_self();
4912
4913 Linux::clock_init();
4914 initial_time_count = javaTimeNanos();
4915
4916 // retrieve entry point for pthread_setname_np
4917 Linux::_pthread_setname_np =
4918 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4919
4920 os::Posix::init();
4921 }
4922
4923 // To install functions for atexit system call
4924 extern "C" {
4925 static void perfMemory_exit_helper() {
4926 perfMemory_exit();
4927 }
4928 }
4929
4930 void os::pd_init_container_support() {
4931 OSContainer::init();
4932 }
4933
4934 // this is called _after_ the global arguments have been parsed
4935 jint os::init_2(void) {
4936
4937 os::Posix::init_2();
4938
4939 Linux::fast_thread_clock_init();
4940
4941 // initialize suspend/resume support - must do this before signal_sets_init()
4942 if (SR_initialize() != 0) {
4943 perror("SR_initialize failed");
4944 return JNI_ERR;
4945 }
4946
4947 Linux::signal_sets_init();
4948 Linux::install_signal_handlers();
4949
4950 // Check and sets minimum stack sizes against command line options
4951 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4952 return JNI_ERR;
4953 }
4954 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4955
4956 #if defined(IA32)
4957 workaround_expand_exec_shield_cs_limit();
4958 #endif
4959
4960 Linux::libpthread_init();
4961 Linux::sched_getcpu_init();
4962 log_info(os)("HotSpot is running with %s, %s",
4963 Linux::glibc_version(), Linux::libpthread_version());
4964
4965 if (UseNUMA) {
4966 if (!Linux::libnuma_init()) {
4967 UseNUMA = false;
4968 } else {
4969 if ((Linux::numa_max_node() < 1)) {
4970 // There's only one node(they start from 0), disable NUMA.
4971 UseNUMA = false;
4972 }
4973 }
4974
4975 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4976 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4977 // we can make the adaptive lgrp chunk resizing work. If the user specified both
4978 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
4979 // and disable adaptive resizing.
4980 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4981 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
4982 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4983 UseAdaptiveSizePolicy = false;
4984 UseAdaptiveNUMAChunkSizing = false;
4985 }
4986 }
4987
4988 if (!UseNUMA && ForceNUMA) {
4989 UseNUMA = true;
4990 }
4991 }
4992
4993 if (MaxFDLimit) {
4994 // set the number of file descriptors to max. print out error
4995 // if getrlimit/setrlimit fails but continue regardless.
4996 struct rlimit nbr_files;
4997 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4998 if (status != 0) {
4999 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5000 } else {
5001 nbr_files.rlim_cur = nbr_files.rlim_max;
5002 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5003 if (status != 0) {
5004 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5005 }
5006 }
5007 }
5008
5009 // Initialize lock used to serialize thread creation (see os::create_thread)
5010 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5011
5012 // at-exit methods are called in the reverse order of their registration.
5013 // atexit functions are called on return from main or as a result of a
5014 // call to exit(3C). There can be only 32 of these functions registered
5015 // and atexit() does not set errno.
5016
5017 if (PerfAllowAtExitRegistration) {
5018 // only register atexit functions if PerfAllowAtExitRegistration is set.
5019 // atexit functions can be delayed until process exit time, which
5020 // can be problematic for embedded VM situations. Embedded VMs should
5021 // call DestroyJavaVM() to assure that VM resources are released.
5022
5023 // note: perfMemory_exit_helper atexit function may be removed in
5024 // the future if the appropriate cleanup code can be added to the
5025 // VM_Exit VMOperation's doit method.
5026 if (atexit(perfMemory_exit_helper) != 0) {
5027 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5028 }
5029 }
5030
5031 // initialize thread priority policy
5032 prio_init();
5033
5034 if (!FLAG_IS_DEFAULT(AllocateHeapAt)) {
5035 set_coredump_filter(false /*largepages*/, true /*dax_shared*/);
5036 }
5037 return JNI_OK;
5038 }
5039
5040 // Mark the polling page as unreadable
5041 void os::make_polling_page_unreadable(void) {
5042 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5043 fatal("Could not disable polling page");
5044 }
5045 }
5046
5047 // Mark the polling page as readable
5048 void os::make_polling_page_readable(void) {
5049 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5050 fatal("Could not enable polling page");
5051 }
5052 }
5053
5054 // older glibc versions don't have this macro (which expands to
5055 // an optimized bit-counting function) so we have to roll our own
5056 #ifndef CPU_COUNT
5057
5058 static int _cpu_count(const cpu_set_t* cpus) {
5059 int count = 0;
5060 // only look up to the number of configured processors
5061 for (int i = 0; i < os::processor_count(); i++) {
5062 if (CPU_ISSET(i, cpus)) {
5063 count++;
5064 }
5065 }
5066 return count;
5067 }
5068
5069 #define CPU_COUNT(cpus) _cpu_count(cpus)
5070
5071 #endif // CPU_COUNT
5072
5073 // Get the current number of available processors for this process.
5074 // This value can change at any time during a process's lifetime.
5075 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5076 // If it appears there may be more than 1024 processors then we do a
5077 // dynamic check - see 6515172 for details.
5078 // If anything goes wrong we fallback to returning the number of online
5079 // processors - which can be greater than the number available to the process.
5080 int os::Linux::active_processor_count() {
5081 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5082 cpu_set_t* cpus_p = &cpus;
5083 int cpus_size = sizeof(cpu_set_t);
5084
5085 int configured_cpus = os::processor_count(); // upper bound on available cpus
5086 int cpu_count = 0;
5087
5088 // old build platforms may not support dynamic cpu sets
5089 #ifdef CPU_ALLOC
5090
5091 // To enable easy testing of the dynamic path on different platforms we
5092 // introduce a diagnostic flag: UseCpuAllocPath
5093 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5094 // kernel may use a mask bigger than cpu_set_t
5095 log_trace(os)("active_processor_count: using dynamic path %s"
5096 "- configured processors: %d",
5097 UseCpuAllocPath ? "(forced) " : "",
5098 configured_cpus);
5099 cpus_p = CPU_ALLOC(configured_cpus);
5100 if (cpus_p != NULL) {
5101 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5102 // zero it just to be safe
5103 CPU_ZERO_S(cpus_size, cpus_p);
5104 }
5105 else {
5106 // failed to allocate so fallback to online cpus
5107 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5108 log_trace(os)("active_processor_count: "
5109 "CPU_ALLOC failed (%s) - using "
5110 "online processor count: %d",
5111 os::strerror(errno), online_cpus);
5112 return online_cpus;
5113 }
5114 }
5115 else {
5116 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5117 configured_cpus);
5118 }
5119 #else // CPU_ALLOC
5120 // these stubs won't be executed
5121 #define CPU_COUNT_S(size, cpus) -1
5122 #define CPU_FREE(cpus)
5123
5124 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5125 configured_cpus);
5126 #endif // CPU_ALLOC
5127
5128 // pid 0 means the current thread - which we have to assume represents the process
5129 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5130 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5131 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5132 }
5133 else {
5134 cpu_count = CPU_COUNT(cpus_p);
5135 }
5136 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5137 }
5138 else {
5139 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5140 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5141 "which may exceed available processors", os::strerror(errno), cpu_count);
5142 }
5143
5144 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5145 CPU_FREE(cpus_p);
5146 }
5147
5148 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5149 return cpu_count;
5150 }
5151
5152 // Determine the active processor count from one of
5153 // three different sources:
5154 //
5155 // 1. User option -XX:ActiveProcessorCount
5156 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5157 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5158 //
5159 // Option 1, if specified, will always override.
5160 // If the cgroup subsystem is active and configured, we
5161 // will return the min of the cgroup and option 2 results.
5162 // This is required since tools, such as numactl, that
5163 // alter cpu affinity do not update cgroup subsystem
5164 // cpuset configuration files.
5165 int os::active_processor_count() {
5166 // User has overridden the number of active processors
5167 if (ActiveProcessorCount > 0) {
5168 log_trace(os)("active_processor_count: "
5169 "active processor count set by user : %d",
5170 ActiveProcessorCount);
5171 return ActiveProcessorCount;
5172 }
5173
5174 int active_cpus;
5175 if (OSContainer::is_containerized()) {
5176 active_cpus = OSContainer::active_processor_count();
5177 log_trace(os)("active_processor_count: determined by OSContainer: %d",
5178 active_cpus);
5179 } else {
5180 active_cpus = os::Linux::active_processor_count();
5181 }
5182
5183 return active_cpus;
5184 }
5185
5186 void os::set_native_thread_name(const char *name) {
5187 if (Linux::_pthread_setname_np) {
5188 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5189 snprintf(buf, sizeof(buf), "%s", name);
5190 buf[sizeof(buf) - 1] = '\0';
5191 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5192 // ERANGE should not happen; all other errors should just be ignored.
5193 assert(rc != ERANGE, "pthread_setname_np failed");
5194 }
5195 }
5196
5197 bool os::distribute_processes(uint length, uint* distribution) {
5198 // Not yet implemented.
5199 return false;
5200 }
5201
5202 bool os::bind_to_processor(uint processor_id) {
5203 // Not yet implemented.
5204 return false;
5205 }
5206
5207 ///
5208
5209 void os::SuspendedThreadTask::internal_do_task() {
5210 if (do_suspend(_thread->osthread())) {
5211 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5212 do_task(context);
5213 do_resume(_thread->osthread());
5214 }
5215 }
5216
5217 ////////////////////////////////////////////////////////////////////////////////
5218 // debug support
5219
5220 bool os::find(address addr, outputStream* st) {
5221 Dl_info dlinfo;
5222 memset(&dlinfo, 0, sizeof(dlinfo));
5223 if (dladdr(addr, &dlinfo) != 0) {
5224 st->print(PTR_FORMAT ": ", p2i(addr));
5225 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5226 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5227 p2i(addr) - p2i(dlinfo.dli_saddr));
5228 } else if (dlinfo.dli_fbase != NULL) {
5229 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5230 } else {
5231 st->print("<absolute address>");
5232 }
5233 if (dlinfo.dli_fname != NULL) {
5234 st->print(" in %s", dlinfo.dli_fname);
5235 }
5236 if (dlinfo.dli_fbase != NULL) {
5237 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5238 }
5239 st->cr();
5240
5241 if (Verbose) {
5242 // decode some bytes around the PC
5243 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5244 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5245 address lowest = (address) dlinfo.dli_sname;
5246 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5247 if (begin < lowest) begin = lowest;
5248 Dl_info dlinfo2;
5249 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5250 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5251 end = (address) dlinfo2.dli_saddr;
5252 }
5253 Disassembler::decode(begin, end, st);
5254 }
5255 return true;
5256 }
5257 return false;
5258 }
5259
5260 ////////////////////////////////////////////////////////////////////////////////
5261 // misc
5262
5263 // This does not do anything on Linux. This is basically a hook for being
5264 // able to use structured exception handling (thread-local exception filters)
5265 // on, e.g., Win32.
5266 void
5267 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5268 JavaCallArguments* args, Thread* thread) {
5269 f(value, method, args, thread);
5270 }
5271
5272 void os::print_statistics() {
5273 }
5274
5275 bool os::message_box(const char* title, const char* message) {
5276 int i;
5277 fdStream err(defaultStream::error_fd());
5278 for (i = 0; i < 78; i++) err.print_raw("=");
5279 err.cr();
5280 err.print_raw_cr(title);
5281 for (i = 0; i < 78; i++) err.print_raw("-");
5282 err.cr();
5283 err.print_raw_cr(message);
5284 for (i = 0; i < 78; i++) err.print_raw("=");
5285 err.cr();
5286
5287 char buf[16];
5288 // Prevent process from exiting upon "read error" without consuming all CPU
5289 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5290
5291 return buf[0] == 'y' || buf[0] == 'Y';
5292 }
5293
5294 int os::stat(const char *path, struct stat *sbuf) {
5295 char pathbuf[MAX_PATH];
5296 if (strlen(path) > MAX_PATH - 1) {
5297 errno = ENAMETOOLONG;
5298 return -1;
5299 }
5300 os::native_path(strcpy(pathbuf, path));
5301 return ::stat(pathbuf, sbuf);
5302 }
5303
5304 // Is a (classpath) directory empty?
5305 bool os::dir_is_empty(const char* path) {
5306 DIR *dir = NULL;
5307 struct dirent *ptr;
5308
5309 dir = opendir(path);
5310 if (dir == NULL) return true;
5311
5312 // Scan the directory
5313 bool result = true;
5314 char buf[sizeof(struct dirent) + MAX_PATH];
5315 while (result && (ptr = ::readdir(dir)) != NULL) {
5316 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5317 result = false;
5318 }
5319 }
5320 closedir(dir);
5321 return result;
5322 }
5323
5324 // This code originates from JDK's sysOpen and open64_w
5325 // from src/solaris/hpi/src/system_md.c
5326
5327 int os::open(const char *path, int oflag, int mode) {
5328 if (strlen(path) > MAX_PATH - 1) {
5329 errno = ENAMETOOLONG;
5330 return -1;
5331 }
5332
5333 // All file descriptors that are opened in the Java process and not
5334 // specifically destined for a subprocess should have the close-on-exec
5335 // flag set. If we don't set it, then careless 3rd party native code
5336 // might fork and exec without closing all appropriate file descriptors
5337 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5338 // turn might:
5339 //
5340 // - cause end-of-file to fail to be detected on some file
5341 // descriptors, resulting in mysterious hangs, or
5342 //
5343 // - might cause an fopen in the subprocess to fail on a system
5344 // suffering from bug 1085341.
5345 //
5346 // (Yes, the default setting of the close-on-exec flag is a Unix
5347 // design flaw)
5348 //
5349 // See:
5350 // 1085341: 32-bit stdio routines should support file descriptors >255
5351 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5352 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5353 //
5354 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5355 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5356 // because it saves a system call and removes a small window where the flag
5357 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5358 // and we fall back to using FD_CLOEXEC (see below).
5359 #ifdef O_CLOEXEC
5360 oflag |= O_CLOEXEC;
5361 #endif
5362
5363 int fd = ::open64(path, oflag, mode);
5364 if (fd == -1) return -1;
5365
5366 //If the open succeeded, the file might still be a directory
5367 {
5368 struct stat64 buf64;
5369 int ret = ::fstat64(fd, &buf64);
5370 int st_mode = buf64.st_mode;
5371
5372 if (ret != -1) {
5373 if ((st_mode & S_IFMT) == S_IFDIR) {
5374 errno = EISDIR;
5375 ::close(fd);
5376 return -1;
5377 }
5378 } else {
5379 ::close(fd);
5380 return -1;
5381 }
5382 }
5383
5384 #ifdef FD_CLOEXEC
5385 // Validate that the use of the O_CLOEXEC flag on open above worked.
5386 // With recent kernels, we will perform this check exactly once.
5387 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5388 if (!O_CLOEXEC_is_known_to_work) {
5389 int flags = ::fcntl(fd, F_GETFD);
5390 if (flags != -1) {
5391 if ((flags & FD_CLOEXEC) != 0)
5392 O_CLOEXEC_is_known_to_work = 1;
5393 else
5394 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5395 }
5396 }
5397 #endif
5398
5399 return fd;
5400 }
5401
5402
5403 // create binary file, rewriting existing file if required
5404 int os::create_binary_file(const char* path, bool rewrite_existing) {
5405 int oflags = O_WRONLY | O_CREAT;
5406 if (!rewrite_existing) {
5407 oflags |= O_EXCL;
5408 }
5409 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5410 }
5411
5412 // return current position of file pointer
5413 jlong os::current_file_offset(int fd) {
5414 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5415 }
5416
5417 // move file pointer to the specified offset
5418 jlong os::seek_to_file_offset(int fd, jlong offset) {
5419 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5420 }
5421
5422 // This code originates from JDK's sysAvailable
5423 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5424
5425 int os::available(int fd, jlong *bytes) {
5426 jlong cur, end;
5427 int mode;
5428 struct stat64 buf64;
5429
5430 if (::fstat64(fd, &buf64) >= 0) {
5431 mode = buf64.st_mode;
5432 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5433 int n;
5434 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5435 *bytes = n;
5436 return 1;
5437 }
5438 }
5439 }
5440 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5441 return 0;
5442 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5443 return 0;
5444 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5445 return 0;
5446 }
5447 *bytes = end - cur;
5448 return 1;
5449 }
5450
5451 // Map a block of memory.
5452 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5453 char *addr, size_t bytes, bool read_only,
5454 bool allow_exec) {
5455 int prot;
5456 int flags = MAP_PRIVATE;
5457
5458 if (read_only) {
5459 prot = PROT_READ;
5460 } else {
5461 prot = PROT_READ | PROT_WRITE;
5462 }
5463
5464 if (allow_exec) {
5465 prot |= PROT_EXEC;
5466 }
5467
5468 if (addr != NULL) {
5469 flags |= MAP_FIXED;
5470 }
5471
5472 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5473 fd, file_offset);
5474 if (mapped_address == MAP_FAILED) {
5475 return NULL;
5476 }
5477 return mapped_address;
5478 }
5479
5480
5481 // Remap a block of memory.
5482 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5483 char *addr, size_t bytes, bool read_only,
5484 bool allow_exec) {
5485 // same as map_memory() on this OS
5486 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5487 allow_exec);
5488 }
5489
5490
5491 // Unmap a block of memory.
5492 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5493 return munmap(addr, bytes) == 0;
5494 }
5495
5496 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5497
5498 static clockid_t thread_cpu_clockid(Thread* thread) {
5499 pthread_t tid = thread->osthread()->pthread_id();
5500 clockid_t clockid;
5501
5502 // Get thread clockid
5503 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5504 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5505 return clockid;
5506 }
5507
5508 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5509 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5510 // of a thread.
5511 //
5512 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5513 // the fast estimate available on the platform.
5514
5515 jlong os::current_thread_cpu_time() {
5516 if (os::Linux::supports_fast_thread_cpu_time()) {
5517 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5518 } else {
5519 // return user + sys since the cost is the same
5520 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5521 }
5522 }
5523
5524 jlong os::thread_cpu_time(Thread* thread) {
5525 // consistent with what current_thread_cpu_time() returns
5526 if (os::Linux::supports_fast_thread_cpu_time()) {
5527 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5528 } else {
5529 return slow_thread_cpu_time(thread, true /* user + sys */);
5530 }
5531 }
5532
5533 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5534 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5535 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5536 } else {
5537 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5538 }
5539 }
5540
5541 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5542 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5543 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5544 } else {
5545 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5546 }
5547 }
5548
5549 // -1 on error.
5550 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5551 pid_t tid = thread->osthread()->thread_id();
5552 char *s;
5553 char stat[2048];
5554 int statlen;
5555 char proc_name[64];
5556 int count;
5557 long sys_time, user_time;
5558 char cdummy;
5559 int idummy;
5560 long ldummy;
5561 FILE *fp;
5562
5563 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5564 fp = fopen(proc_name, "r");
5565 if (fp == NULL) return -1;
5566 statlen = fread(stat, 1, 2047, fp);
5567 stat[statlen] = '\0';
5568 fclose(fp);
5569
5570 // Skip pid and the command string. Note that we could be dealing with
5571 // weird command names, e.g. user could decide to rename java launcher
5572 // to "java 1.4.2 :)", then the stat file would look like
5573 // 1234 (java 1.4.2 :)) R ... ...
5574 // We don't really need to know the command string, just find the last
5575 // occurrence of ")" and then start parsing from there. See bug 4726580.
5576 s = strrchr(stat, ')');
5577 if (s == NULL) return -1;
5578
5579 // Skip blank chars
5580 do { s++; } while (s && isspace(*s));
5581
5582 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5583 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5584 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5585 &user_time, &sys_time);
5586 if (count != 13) return -1;
5587 if (user_sys_cpu_time) {
5588 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5589 } else {
5590 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5591 }
5592 }
5593
5594 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5595 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5596 info_ptr->may_skip_backward = false; // elapsed time not wall time
5597 info_ptr->may_skip_forward = false; // elapsed time not wall time
5598 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5599 }
5600
5601 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5602 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5603 info_ptr->may_skip_backward = false; // elapsed time not wall time
5604 info_ptr->may_skip_forward = false; // elapsed time not wall time
5605 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5606 }
5607
5608 bool os::is_thread_cpu_time_supported() {
5609 return true;
5610 }
5611
5612 // System loadavg support. Returns -1 if load average cannot be obtained.
5613 // Linux doesn't yet have a (official) notion of processor sets,
5614 // so just return the system wide load average.
5615 int os::loadavg(double loadavg[], int nelem) {
5616 return ::getloadavg(loadavg, nelem);
5617 }
5618
5619 void os::pause() {
5620 char filename[MAX_PATH];
5621 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5622 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5623 } else {
5624 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5625 }
5626
5627 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5628 if (fd != -1) {
5629 struct stat buf;
5630 ::close(fd);
5631 while (::stat(filename, &buf) == 0) {
5632 (void)::poll(NULL, 0, 100);
5633 }
5634 } else {
5635 jio_fprintf(stderr,
5636 "Could not open pause file '%s', continuing immediately.\n", filename);
5637 }
5638 }
5639
5640 extern char** environ;
5641
5642 // Run the specified command in a separate process. Return its exit value,
5643 // or -1 on failure (e.g. can't fork a new process).
5644 // Unlike system(), this function can be called from signal handler. It
5645 // doesn't block SIGINT et al.
5646 int os::fork_and_exec(char* cmd) {
5647 const char * argv[4] = {"sh", "-c", cmd, NULL};
5648
5649 pid_t pid = fork();
5650
5651 if (pid < 0) {
5652 // fork failed
5653 return -1;
5654
5655 } else if (pid == 0) {
5656 // child process
5657
5658 execve("/bin/sh", (char* const*)argv, environ);
5659
5660 // execve failed
5661 _exit(-1);
5662
5663 } else {
5664 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5665 // care about the actual exit code, for now.
5666
5667 int status;
5668
5669 // Wait for the child process to exit. This returns immediately if
5670 // the child has already exited. */
5671 while (waitpid(pid, &status, 0) < 0) {
5672 switch (errno) {
5673 case ECHILD: return 0;
5674 case EINTR: break;
5675 default: return -1;
5676 }
5677 }
5678
5679 if (WIFEXITED(status)) {
5680 // The child exited normally; get its exit code.
5681 return WEXITSTATUS(status);
5682 } else if (WIFSIGNALED(status)) {
5683 // The child exited because of a signal
5684 // The best value to return is 0x80 + signal number,
5685 // because that is what all Unix shells do, and because
5686 // it allows callers to distinguish between process exit and
5687 // process death by signal.
5688 return 0x80 + WTERMSIG(status);
5689 } else {
5690 // Unknown exit code; pass it through
5691 return status;
5692 }
5693 }
5694 }
5695
5696 // is_headless_jre()
5697 //
5698 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5699 // in order to report if we are running in a headless jre
5700 //
5701 // Since JDK8 xawt/libmawt.so was moved into the same directory
5702 // as libawt.so, and renamed libawt_xawt.so
5703 //
5704 bool os::is_headless_jre() {
5705 struct stat statbuf;
5706 char buf[MAXPATHLEN];
5707 char libmawtpath[MAXPATHLEN];
5708 const char *xawtstr = "/xawt/libmawt.so";
5709 const char *new_xawtstr = "/libawt_xawt.so";
5710 char *p;
5711
5712 // Get path to libjvm.so
5713 os::jvm_path(buf, sizeof(buf));
5714
5715 // Get rid of libjvm.so
5716 p = strrchr(buf, '/');
5717 if (p == NULL) {
5718 return false;
5719 } else {
5720 *p = '\0';
5721 }
5722
5723 // Get rid of client or server
5724 p = strrchr(buf, '/');
5725 if (p == NULL) {
5726 return false;
5727 } else {
5728 *p = '\0';
5729 }
5730
5731 // check xawt/libmawt.so
5732 strcpy(libmawtpath, buf);
5733 strcat(libmawtpath, xawtstr);
5734 if (::stat(libmawtpath, &statbuf) == 0) return false;
5735
5736 // check libawt_xawt.so
5737 strcpy(libmawtpath, buf);
5738 strcat(libmawtpath, new_xawtstr);
5739 if (::stat(libmawtpath, &statbuf) == 0) return false;
5740
5741 return true;
5742 }
5743
5744 // Get the default path to the core file
5745 // Returns the length of the string
5746 int os::get_core_path(char* buffer, size_t bufferSize) {
5747 /*
5748 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5749 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5750 */
5751 const int core_pattern_len = 129;
5752 char core_pattern[core_pattern_len] = {0};
5753
5754 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5755 if (core_pattern_file == -1) {
5756 return -1;
5757 }
5758
5759 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5760 ::close(core_pattern_file);
5761 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5762 return -1;
5763 }
5764 if (core_pattern[ret-1] == '\n') {
5765 core_pattern[ret-1] = '\0';
5766 } else {
5767 core_pattern[ret] = '\0';
5768 }
5769
5770 char *pid_pos = strstr(core_pattern, "%p");
5771 int written;
5772
5773 if (core_pattern[0] == '/') {
5774 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5775 } else {
5776 char cwd[PATH_MAX];
5777
5778 const char* p = get_current_directory(cwd, PATH_MAX);
5779 if (p == NULL) {
5780 return -1;
5781 }
5782
5783 if (core_pattern[0] == '|') {
5784 written = jio_snprintf(buffer, bufferSize,
5785 "\"%s\" (or dumping to %s/core.%d)",
5786 &core_pattern[1], p, current_process_id());
5787 } else {
5788 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5789 }
5790 }
5791
5792 if (written < 0) {
5793 return -1;
5794 }
5795
5796 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5797 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5798
5799 if (core_uses_pid_file != -1) {
5800 char core_uses_pid = 0;
5801 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5802 ::close(core_uses_pid_file);
5803
5804 if (core_uses_pid == '1') {
5805 jio_snprintf(buffer + written, bufferSize - written,
5806 ".%d", current_process_id());
5807 }
5808 }
5809 }
5810
5811 return strlen(buffer);
5812 }
5813
5814 bool os::start_debugging(char *buf, int buflen) {
5815 int len = (int)strlen(buf);
5816 char *p = &buf[len];
5817
5818 jio_snprintf(p, buflen-len,
5819 "\n\n"
5820 "Do you want to debug the problem?\n\n"
5821 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5822 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5823 "Otherwise, press RETURN to abort...",
5824 os::current_process_id(), os::current_process_id(),
5825 os::current_thread_id(), os::current_thread_id());
5826
5827 bool yes = os::message_box("Unexpected Error", buf);
5828
5829 if (yes) {
5830 // yes, user asked VM to launch debugger
5831 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5832 os::current_process_id(), os::current_process_id());
5833
5834 os::fork_and_exec(buf);
5835 yes = false;
5836 }
5837 return yes;
5838 }
5839
5840
5841 // Java/Compiler thread:
5842 //
5843 // Low memory addresses
5844 // P0 +------------------------+
5845 // | |\ Java thread created by VM does not have glibc
5846 // | glibc guard page | - guard page, attached Java thread usually has
5847 // | |/ 1 glibc guard page.
5848 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5849 // | |\
5850 // | HotSpot Guard Pages | - red, yellow and reserved pages
5851 // | |/
5852 // +------------------------+ JavaThread::stack_reserved_zone_base()
5853 // | |\
5854 // | Normal Stack | -
5855 // | |/
5856 // P2 +------------------------+ Thread::stack_base()
5857 //
5858 // Non-Java thread:
5859 //
5860 // Low memory addresses
5861 // P0 +------------------------+
5862 // | |\
5863 // | glibc guard page | - usually 1 page
5864 // | |/
5865 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5866 // | |\
5867 // | Normal Stack | -
5868 // | |/
5869 // P2 +------------------------+ Thread::stack_base()
5870 //
5871 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
5872 // returned from pthread_attr_getstack().
5873 // ** Due to NPTL implementation error, linux takes the glibc guard page out
5874 // of the stack size given in pthread_attr. We work around this for
5875 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
5876 //
5877 #ifndef ZERO
5878 static void current_stack_region(address * bottom, size_t * size) {
5879 if (os::is_primordial_thread()) {
5880 // primordial thread needs special handling because pthread_getattr_np()
5881 // may return bogus value.
5882 *bottom = os::Linux::initial_thread_stack_bottom();
5883 *size = os::Linux::initial_thread_stack_size();
5884 } else {
5885 pthread_attr_t attr;
5886
5887 int rslt = pthread_getattr_np(pthread_self(), &attr);
5888
5889 // JVM needs to know exact stack location, abort if it fails
5890 if (rslt != 0) {
5891 if (rslt == ENOMEM) {
5892 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
5893 } else {
5894 fatal("pthread_getattr_np failed with error = %d", rslt);
5895 }
5896 }
5897
5898 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
5899 fatal("Cannot locate current stack attributes!");
5900 }
5901
5902 // Work around NPTL stack guard error.
5903 size_t guard_size = 0;
5904 rslt = pthread_attr_getguardsize(&attr, &guard_size);
5905 if (rslt != 0) {
5906 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
5907 }
5908 *bottom += guard_size;
5909 *size -= guard_size;
5910
5911 pthread_attr_destroy(&attr);
5912
5913 }
5914 assert(os::current_stack_pointer() >= *bottom &&
5915 os::current_stack_pointer() < *bottom + *size, "just checking");
5916 }
5917
5918 address os::current_stack_base() {
5919 address bottom;
5920 size_t size;
5921 current_stack_region(&bottom, &size);
5922 return (bottom + size);
5923 }
5924
5925 size_t os::current_stack_size() {
5926 // This stack size includes the usable stack and HotSpot guard pages
5927 // (for the threads that have Hotspot guard pages).
5928 address bottom;
5929 size_t size;
5930 current_stack_region(&bottom, &size);
5931 return size;
5932 }
5933 #endif
5934
5935 static inline struct timespec get_mtime(const char* filename) {
5936 struct stat st;
5937 int ret = os::stat(filename, &st);
5938 assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
5939 return st.st_mtim;
5940 }
5941
5942 int os::compare_file_modified_times(const char* file1, const char* file2) {
5943 struct timespec filetime1 = get_mtime(file1);
5944 struct timespec filetime2 = get_mtime(file2);
5945 int diff = filetime1.tv_sec - filetime2.tv_sec;
5946 if (diff == 0) {
5947 return filetime1.tv_nsec - filetime2.tv_nsec;
5948 }
5949 return diff;
5950 }
5951
5952 /////////////// Unit tests ///////////////
5953
5954 #ifndef PRODUCT
5955
5956 #define test_log(...) \
5957 do { \
5958 if (VerboseInternalVMTests) { \
5959 tty->print_cr(__VA_ARGS__); \
5960 tty->flush(); \
5961 } \
5962 } while (false)
5963
5964 class TestReserveMemorySpecial : AllStatic {
5965 public:
5966 static void small_page_write(void* addr, size_t size) {
5967 size_t page_size = os::vm_page_size();
5968
5969 char* end = (char*)addr + size;
5970 for (char* p = (char*)addr; p < end; p += page_size) {
5971 *p = 1;
5972 }
5973 }
5974
5975 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
5976 if (!UseHugeTLBFS) {
5977 return;
5978 }
5979
5980 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
5981
5982 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
5983
5984 if (addr != NULL) {
5985 small_page_write(addr, size);
5986
5987 os::Linux::release_memory_special_huge_tlbfs(addr, size);
5988 }
5989 }
5990
5991 static void test_reserve_memory_special_huge_tlbfs_only() {
5992 if (!UseHugeTLBFS) {
5993 return;
5994 }
5995
5996 size_t lp = os::large_page_size();
5997
5998 for (size_t size = lp; size <= lp * 10; size += lp) {
5999 test_reserve_memory_special_huge_tlbfs_only(size);
6000 }
6001 }
6002
6003 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6004 size_t lp = os::large_page_size();
6005 size_t ag = os::vm_allocation_granularity();
6006
6007 // sizes to test
6008 const size_t sizes[] = {
6009 lp, lp + ag, lp + lp / 2, lp * 2,
6010 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6011 lp * 10, lp * 10 + lp / 2
6012 };
6013 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6014
6015 // For each size/alignment combination, we test three scenarios:
6016 // 1) with req_addr == NULL
6017 // 2) with a non-null req_addr at which we expect to successfully allocate
6018 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6019 // expect the allocation to either fail or to ignore req_addr
6020
6021 // Pre-allocate two areas; they shall be as large as the largest allocation
6022 // and aligned to the largest alignment we will be testing.
6023 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6024 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6025 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6026 -1, 0);
6027 assert(mapping1 != MAP_FAILED, "should work");
6028
6029 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6030 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6031 -1, 0);
6032 assert(mapping2 != MAP_FAILED, "should work");
6033
6034 // Unmap the first mapping, but leave the second mapping intact: the first
6035 // mapping will serve as a value for a "good" req_addr (case 2). The second
6036 // mapping, still intact, as "bad" req_addr (case 3).
6037 ::munmap(mapping1, mapping_size);
6038
6039 // Case 1
6040 test_log("%s, req_addr NULL:", __FUNCTION__);
6041 test_log("size align result");
6042
6043 for (int i = 0; i < num_sizes; i++) {
6044 const size_t size = sizes[i];
6045 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6046 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6047 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6048 size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6049 if (p != NULL) {
6050 assert(is_aligned(p, alignment), "must be");
6051 small_page_write(p, size);
6052 os::Linux::release_memory_special_huge_tlbfs(p, size);
6053 }
6054 }
6055 }
6056
6057 // Case 2
6058 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6059 test_log("size align req_addr result");
6060
6061 for (int i = 0; i < num_sizes; i++) {
6062 const size_t size = sizes[i];
6063 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6064 char* const req_addr = align_up(mapping1, alignment);
6065 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6066 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6067 size, alignment, p2i(req_addr), p2i(p),
6068 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6069 if (p != NULL) {
6070 assert(p == req_addr, "must be");
6071 small_page_write(p, size);
6072 os::Linux::release_memory_special_huge_tlbfs(p, size);
6073 }
6074 }
6075 }
6076
6077 // Case 3
6078 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6079 test_log("size align req_addr result");
6080
6081 for (int i = 0; i < num_sizes; i++) {
6082 const size_t size = sizes[i];
6083 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6084 char* const req_addr = align_up(mapping2, alignment);
6085 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6086 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6087 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6088 // as the area around req_addr contains already existing mappings, the API should always
6089 // return NULL (as per contract, it cannot return another address)
6090 assert(p == NULL, "must be");
6091 }
6092 }
6093
6094 ::munmap(mapping2, mapping_size);
6095
6096 }
6097
6098 static void test_reserve_memory_special_huge_tlbfs() {
6099 if (!UseHugeTLBFS) {
6100 return;
6101 }
6102
6103 test_reserve_memory_special_huge_tlbfs_only();
6104 test_reserve_memory_special_huge_tlbfs_mixed();
6105 }
6106
6107 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6108 if (!UseSHM) {
6109 return;
6110 }
6111
6112 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6113
6114 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6115
6116 if (addr != NULL) {
6117 assert(is_aligned(addr, alignment), "Check");
6118 assert(is_aligned(addr, os::large_page_size()), "Check");
6119
6120 small_page_write(addr, size);
6121
6122 os::Linux::release_memory_special_shm(addr, size);
6123 }
6124 }
6125
6126 static void test_reserve_memory_special_shm() {
6127 size_t lp = os::large_page_size();
6128 size_t ag = os::vm_allocation_granularity();
6129
6130 for (size_t size = ag; size < lp * 3; size += ag) {
6131 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6132 test_reserve_memory_special_shm(size, alignment);
6133 }
6134 }
6135 }
6136
6137 static void test() {
6138 test_reserve_memory_special_huge_tlbfs();
6139 test_reserve_memory_special_shm();
6140 }
6141 };
6142
6143 void TestReserveMemorySpecial_test() {
6144 TestReserveMemorySpecial::test();
6145 }
6146
6147 #endif