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/timer.hpp"
63 #include "semaphore_posix.hpp"
64 #include "services/attachListener.hpp"
65 #include "services/memTracker.hpp"
66 #include "services/runtimeService.hpp"
67 #include "utilities/align.hpp"
68 #include "utilities/decoder.hpp"
69 #include "utilities/defaultStream.hpp"
70 #include "utilities/events.hpp"
71 #include "utilities/elfFile.hpp"
72 #include "utilities/growableArray.hpp"
73 #include "utilities/macros.hpp"
74 #include "utilities/vmError.hpp"
75
76 // put OS-includes here
77 # include <sys/types.h>
78 # include <sys/mman.h>
79 # include <sys/stat.h>
80 # include <sys/select.h>
81 # include <pthread.h>
82 # include <signal.h>
83 # include <errno.h>
84 # include <dlfcn.h>
85 # include <stdio.h>
86 # include <unistd.h>
87 # include <sys/resource.h>
88 # include <pthread.h>
89 # include <sys/stat.h>
90 # include <sys/time.h>
91 # include <sys/times.h>
92 # include <sys/utsname.h>
93 # include <sys/socket.h>
94 # include <sys/wait.h>
95 # include <pwd.h>
96 # include <poll.h>
97 # include <semaphore.h>
98 # include <fcntl.h>
99 # include <string.h>
100 # include <syscall.h>
101 # include <sys/sysinfo.h>
102 # include <gnu/libc-version.h>
103 # include <sys/ipc.h>
104 # include <sys/shm.h>
105 # include <link.h>
106 # include <stdint.h>
107 # include <inttypes.h>
108 # include <sys/ioctl.h>
109
110 #ifndef _GNU_SOURCE
111 #define _GNU_SOURCE
112 #include <sched.h>
113 #undef _GNU_SOURCE
114 #else
115 #include <sched.h>
116 #endif
117
118 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
119 // getrusage() is prepared to handle the associated failure.
120 #ifndef RUSAGE_THREAD
121 #define RUSAGE_THREAD (1) /* only the calling thread */
122 #endif
123
124 #define MAX_PATH (2 * K)
125
126 #define MAX_SECS 100000000
127
128 // for timer info max values which include all bits
129 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
130
131 #define LARGEPAGES_BIT (1 << 6)
132 ////////////////////////////////////////////////////////////////////////////////
133 // global variables
134 julong os::Linux::_physical_memory = 0;
135
136 address os::Linux::_initial_thread_stack_bottom = NULL;
137 uintptr_t os::Linux::_initial_thread_stack_size = 0;
138
139 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
140 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
141 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
142 Mutex* os::Linux::_createThread_lock = NULL;
143 pthread_t os::Linux::_main_thread;
144 int os::Linux::_page_size = -1;
145 bool os::Linux::_supports_fast_thread_cpu_time = false;
146 uint32_t os::Linux::_os_version = 0;
147 const char * os::Linux::_glibc_version = NULL;
148 const char * os::Linux::_libpthread_version = NULL;
149
150 static jlong initial_time_count=0;
151
152 static int clock_tics_per_sec = 100;
153
154 // For diagnostics to print a message once. see run_periodic_checks
155 static sigset_t check_signal_done;
156 static bool check_signals = true;
157
158 // Signal number used to suspend/resume a thread
159
160 // do not use any signal number less than SIGSEGV, see 4355769
161 static int SR_signum = SIGUSR2;
162 sigset_t SR_sigset;
163
164 // utility functions
165
166 static int SR_initialize();
167
168 julong os::available_memory() {
169 return Linux::available_memory();
170 }
171
172 julong os::Linux::available_memory() {
173 // values in struct sysinfo are "unsigned long"
174 struct sysinfo si;
175 julong avail_mem;
176
177 if (OSContainer::is_containerized()) {
178 jlong mem_limit, mem_usage;
179 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
180 if ((mem_usage = OSContainer::memory_usage_in_bytes()) > 0) {
181 if (mem_limit > mem_usage) {
182 avail_mem = (julong)mem_limit - (julong)mem_usage;
183 } else {
184 avail_mem = 0;
185 }
186 log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
187 return avail_mem;
188 } else {
189 log_debug(os,container)("container memory usage call failed: " JLONG_FORMAT, mem_usage);
190 }
191 } else {
192 log_debug(os,container)("container memory unlimited or failed: " JLONG_FORMAT, mem_limit);
193 }
194 }
195
196 sysinfo(&si);
197 avail_mem = (julong)si.freeram * si.mem_unit;
198 log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
199 return avail_mem;
200 }
201
202 julong os::physical_memory() {
203 if (OSContainer::is_containerized()) {
204 jlong mem_limit;
205 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
206 log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
207 return (julong)mem_limit;
208 } else {
209 if (mem_limit == OSCONTAINER_ERROR) {
210 log_debug(os,container)("container memory limit call failed");
211 }
212 if (mem_limit == -1) {
213 log_debug(os,container)("container memory unlimited, using host value");
214 }
215 }
216 }
217
218 jlong phys_mem = Linux::physical_memory();
219 log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
220 return phys_mem;
221 }
222
223 // Return true if user is running as root.
224
225 bool os::have_special_privileges() {
226 static bool init = false;
227 static bool privileges = false;
228 if (!init) {
229 privileges = (getuid() != geteuid()) || (getgid() != getegid());
230 init = true;
231 }
232 return privileges;
233 }
234
235
236 #ifndef SYS_gettid
237 // i386: 224, ia64: 1105, amd64: 186, sparc 143
238 #ifdef __ia64__
239 #define SYS_gettid 1105
240 #else
241 #ifdef __i386__
242 #define SYS_gettid 224
243 #else
244 #ifdef __amd64__
245 #define SYS_gettid 186
246 #else
247 #ifdef __sparc__
248 #define SYS_gettid 143
249 #else
250 #error define gettid for the arch
251 #endif
252 #endif
253 #endif
254 #endif
255 #endif
256
257
258 // pid_t gettid()
259 //
260 // Returns the kernel thread id of the currently running thread. Kernel
261 // thread id is used to access /proc.
262 pid_t os::Linux::gettid() {
263 int rslt = syscall(SYS_gettid);
264 assert(rslt != -1, "must be."); // old linuxthreads implementation?
265 return (pid_t)rslt;
266 }
267
268 // Most versions of linux have a bug where the number of processors are
269 // determined by looking at the /proc file system. In a chroot environment,
270 // the system call returns 1. This causes the VM to act as if it is
271 // a single processor and elide locking (see is_MP() call).
272 static bool unsafe_chroot_detected = false;
273 static const char *unstable_chroot_error = "/proc file system not found.\n"
274 "Java may be unstable running multithreaded in a chroot "
275 "environment on Linux when /proc filesystem is not mounted.";
276
277 void os::Linux::initialize_system_info() {
278 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
279 if (processor_count() == 1) {
280 pid_t pid = os::Linux::gettid();
281 char fname[32];
282 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
283 FILE *fp = fopen(fname, "r");
284 if (fp == NULL) {
285 unsafe_chroot_detected = true;
286 } else {
287 fclose(fp);
288 }
289 }
290 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
291 assert(processor_count() > 0, "linux error");
292 }
293
294 void os::init_system_properties_values() {
295 // The next steps are taken in the product version:
296 //
297 // Obtain the JAVA_HOME value from the location of libjvm.so.
298 // This library should be located at:
299 // <JAVA_HOME>/lib/{client|server}/libjvm.so.
300 //
301 // If "/jre/lib/" appears at the right place in the path, then we
302 // assume libjvm.so is installed in a JDK and we use this path.
303 //
304 // Otherwise exit with message: "Could not create the Java virtual machine."
305 //
306 // The following extra steps are taken in the debugging version:
307 //
308 // If "/jre/lib/" does NOT appear at the right place in the path
309 // instead of exit check for $JAVA_HOME environment variable.
310 //
311 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
312 // then we append a fake suffix "hotspot/libjvm.so" to this path so
313 // it looks like libjvm.so is installed there
314 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
315 //
316 // Otherwise exit.
317 //
318 // Important note: if the location of libjvm.so changes this
319 // code needs to be changed accordingly.
320
321 // See ld(1):
322 // The linker uses the following search paths to locate required
323 // shared libraries:
324 // 1: ...
325 // ...
326 // 7: The default directories, normally /lib and /usr/lib.
327 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
328 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
329 #else
330 #define DEFAULT_LIBPATH "/lib:/usr/lib"
331 #endif
332
333 // Base path of extensions installed on the system.
334 #define SYS_EXT_DIR "/usr/java/packages"
335 #define EXTENSIONS_DIR "/lib/ext"
336
337 // Buffer that fits several sprintfs.
338 // Note that the space for the colon and the trailing null are provided
339 // by the nulls included by the sizeof operator.
340 const size_t bufsize =
341 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
342 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
343 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
344
345 // sysclasspath, java_home, dll_dir
346 {
347 char *pslash;
348 os::jvm_path(buf, bufsize);
349
350 // Found the full path to libjvm.so.
351 // Now cut the path to <java_home>/jre if we can.
352 pslash = strrchr(buf, '/');
353 if (pslash != NULL) {
354 *pslash = '\0'; // Get rid of /libjvm.so.
355 }
356 pslash = strrchr(buf, '/');
357 if (pslash != NULL) {
358 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
359 }
360 Arguments::set_dll_dir(buf);
361
362 if (pslash != NULL) {
363 pslash = strrchr(buf, '/');
364 if (pslash != NULL) {
365 *pslash = '\0'; // Get rid of /lib.
366 }
367 }
368 Arguments::set_java_home(buf);
369 set_boot_path('/', ':');
370 }
371
372 // Where to look for native libraries.
373 //
374 // Note: Due to a legacy implementation, most of the library path
375 // is set in the launcher. This was to accomodate linking restrictions
376 // on legacy Linux implementations (which are no longer supported).
377 // Eventually, all the library path setting will be done here.
378 //
379 // However, to prevent the proliferation of improperly built native
380 // libraries, the new path component /usr/java/packages is added here.
381 // Eventually, all the library path setting will be done here.
382 {
383 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
384 // should always exist (until the legacy problem cited above is
385 // addressed).
386 const char *v = ::getenv("LD_LIBRARY_PATH");
387 const char *v_colon = ":";
388 if (v == NULL) { v = ""; v_colon = ""; }
389 // That's +1 for the colon and +1 for the trailing '\0'.
390 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
391 strlen(v) + 1 +
392 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
393 mtInternal);
394 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
395 Arguments::set_library_path(ld_library_path);
396 FREE_C_HEAP_ARRAY(char, ld_library_path);
397 }
398
399 // Extensions directories.
400 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
401 Arguments::set_ext_dirs(buf);
402
403 FREE_C_HEAP_ARRAY(char, buf);
404
405 #undef DEFAULT_LIBPATH
406 #undef SYS_EXT_DIR
407 #undef EXTENSIONS_DIR
408 }
409
410 ////////////////////////////////////////////////////////////////////////////////
411 // breakpoint support
412
413 void os::breakpoint() {
414 BREAKPOINT;
415 }
416
417 extern "C" void breakpoint() {
418 // use debugger to set breakpoint here
419 }
420
421 ////////////////////////////////////////////////////////////////////////////////
422 // signal support
423
424 debug_only(static bool signal_sets_initialized = false);
425 static sigset_t unblocked_sigs, vm_sigs;
426
427 bool os::Linux::is_sig_ignored(int sig) {
428 struct sigaction oact;
429 sigaction(sig, (struct sigaction*)NULL, &oact);
430 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
431 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
432 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
433 return true;
434 } else {
435 return false;
436 }
437 }
438
439 void os::Linux::signal_sets_init() {
440 // Should also have an assertion stating we are still single-threaded.
441 assert(!signal_sets_initialized, "Already initialized");
442 // Fill in signals that are necessarily unblocked for all threads in
443 // the VM. Currently, we unblock the following signals:
444 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
445 // by -Xrs (=ReduceSignalUsage));
446 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
447 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
448 // the dispositions or masks wrt these signals.
449 // Programs embedding the VM that want to use the above signals for their
450 // own purposes must, at this time, use the "-Xrs" option to prevent
451 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
452 // (See bug 4345157, and other related bugs).
453 // In reality, though, unblocking these signals is really a nop, since
454 // these signals are not blocked by default.
455 sigemptyset(&unblocked_sigs);
456 sigaddset(&unblocked_sigs, SIGILL);
457 sigaddset(&unblocked_sigs, SIGSEGV);
458 sigaddset(&unblocked_sigs, SIGBUS);
459 sigaddset(&unblocked_sigs, SIGFPE);
460 #if defined(PPC64)
461 sigaddset(&unblocked_sigs, SIGTRAP);
462 #endif
463 sigaddset(&unblocked_sigs, SR_signum);
464
465 if (!ReduceSignalUsage) {
466 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
467 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
468 }
469 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
470 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
471 }
472 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
473 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
474 }
475 }
476 // Fill in signals that are blocked by all but the VM thread.
477 sigemptyset(&vm_sigs);
478 if (!ReduceSignalUsage) {
479 sigaddset(&vm_sigs, BREAK_SIGNAL);
480 }
481 debug_only(signal_sets_initialized = true);
482
483 }
484
485 // These are signals that are unblocked while a thread is running Java.
486 // (For some reason, they get blocked by default.)
487 sigset_t* os::Linux::unblocked_signals() {
488 assert(signal_sets_initialized, "Not initialized");
489 return &unblocked_sigs;
490 }
491
492 // These are the signals that are blocked while a (non-VM) thread is
493 // running Java. Only the VM thread handles these signals.
494 sigset_t* os::Linux::vm_signals() {
495 assert(signal_sets_initialized, "Not initialized");
496 return &vm_sigs;
497 }
498
499 void os::Linux::hotspot_sigmask(Thread* thread) {
500
501 //Save caller's signal mask before setting VM signal mask
502 sigset_t caller_sigmask;
503 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
504
505 OSThread* osthread = thread->osthread();
506 osthread->set_caller_sigmask(caller_sigmask);
507
508 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
509
510 if (!ReduceSignalUsage) {
511 if (thread->is_VM_thread()) {
512 // Only the VM thread handles BREAK_SIGNAL ...
513 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
514 } else {
515 // ... all other threads block BREAK_SIGNAL
516 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
517 }
518 }
519 }
520
521 //////////////////////////////////////////////////////////////////////////////
522 // detecting pthread library
523
524 void os::Linux::libpthread_init() {
525 // Save glibc and pthread version strings.
526 #if !defined(_CS_GNU_LIBC_VERSION) || \
527 !defined(_CS_GNU_LIBPTHREAD_VERSION)
528 #error "glibc too old (< 2.3.2)"
529 #endif
530
531 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
532 assert(n > 0, "cannot retrieve glibc version");
533 char *str = (char *)malloc(n, mtInternal);
534 confstr(_CS_GNU_LIBC_VERSION, str, n);
535 os::Linux::set_glibc_version(str);
536
537 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
538 assert(n > 0, "cannot retrieve pthread version");
539 str = (char *)malloc(n, mtInternal);
540 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
541 os::Linux::set_libpthread_version(str);
542 }
543
544 /////////////////////////////////////////////////////////////////////////////
545 // thread stack expansion
546
547 // os::Linux::manually_expand_stack() takes care of expanding the thread
548 // stack. Note that this is normally not needed: pthread stacks allocate
549 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
550 // committed. Therefore it is not necessary to expand the stack manually.
551 //
552 // Manually expanding the stack was historically needed on LinuxThreads
553 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
554 // it is kept to deal with very rare corner cases:
555 //
556 // For one, user may run the VM on an own implementation of threads
557 // whose stacks are - like the old LinuxThreads - implemented using
558 // mmap(MAP_GROWSDOWN).
559 //
560 // Also, this coding may be needed if the VM is running on the primordial
561 // thread. Normally we avoid running on the primordial thread; however,
562 // user may still invoke the VM on the primordial thread.
563 //
564 // The following historical comment describes the details about running
565 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
566
567
568 // Force Linux kernel to expand current thread stack. If "bottom" is close
569 // to the stack guard, caller should block all signals.
570 //
571 // MAP_GROWSDOWN:
572 // A special mmap() flag that is used to implement thread stacks. It tells
573 // kernel that the memory region should extend downwards when needed. This
574 // allows early versions of LinuxThreads to only mmap the first few pages
575 // when creating a new thread. Linux kernel will automatically expand thread
576 // stack as needed (on page faults).
577 //
578 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
579 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
580 // region, it's hard to tell if the fault is due to a legitimate stack
581 // access or because of reading/writing non-exist memory (e.g. buffer
582 // overrun). As a rule, if the fault happens below current stack pointer,
583 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
584 // application (see Linux kernel fault.c).
585 //
586 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
587 // stack overflow detection.
588 //
589 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
590 // not use MAP_GROWSDOWN.
591 //
592 // To get around the problem and allow stack banging on Linux, we need to
593 // manually expand thread stack after receiving the SIGSEGV.
594 //
595 // There are two ways to expand thread stack to address "bottom", we used
596 // both of them in JVM before 1.5:
597 // 1. adjust stack pointer first so that it is below "bottom", and then
598 // touch "bottom"
599 // 2. mmap() the page in question
600 //
601 // Now alternate signal stack is gone, it's harder to use 2. For instance,
602 // if current sp is already near the lower end of page 101, and we need to
603 // call mmap() to map page 100, it is possible that part of the mmap() frame
604 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
605 // That will destroy the mmap() frame and cause VM to crash.
606 //
607 // The following code works by adjusting sp first, then accessing the "bottom"
608 // page to force a page fault. Linux kernel will then automatically expand the
609 // stack mapping.
610 //
611 // _expand_stack_to() assumes its frame size is less than page size, which
612 // should always be true if the function is not inlined.
613
614 static void NOINLINE _expand_stack_to(address bottom) {
615 address sp;
616 size_t size;
617 volatile char *p;
618
619 // Adjust bottom to point to the largest address within the same page, it
620 // gives us a one-page buffer if alloca() allocates slightly more memory.
621 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
622 bottom += os::Linux::page_size() - 1;
623
624 // sp might be slightly above current stack pointer; if that's the case, we
625 // will alloca() a little more space than necessary, which is OK. Don't use
626 // os::current_stack_pointer(), as its result can be slightly below current
627 // stack pointer, causing us to not alloca enough to reach "bottom".
628 sp = (address)&sp;
629
630 if (sp > bottom) {
631 size = sp - bottom;
632 p = (volatile char *)alloca(size);
633 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
634 p[0] = '\0';
635 }
636 }
637
638 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
639 assert(t!=NULL, "just checking");
640 assert(t->osthread()->expanding_stack(), "expand should be set");
641 assert(t->stack_base() != NULL, "stack_base was not initialized");
642
643 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
644 sigset_t mask_all, old_sigset;
645 sigfillset(&mask_all);
646 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
647 _expand_stack_to(addr);
648 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
649 return true;
650 }
651 return false;
652 }
653
654 //////////////////////////////////////////////////////////////////////////////
655 // create new thread
656
657 // Thread start routine for all newly created threads
658 static void *thread_native_entry(Thread *thread) {
659 // Try to randomize the cache line index of hot stack frames.
660 // This helps when threads of the same stack traces evict each other's
661 // cache lines. The threads can be either from the same JVM instance, or
662 // from different JVM instances. The benefit is especially true for
663 // processors with hyperthreading technology.
664 static int counter = 0;
665 int pid = os::current_process_id();
666 alloca(((pid ^ counter++) & 7) * 128);
667
668 thread->initialize_thread_current();
669
670 OSThread* osthread = thread->osthread();
671 Monitor* sync = osthread->startThread_lock();
672
673 osthread->set_thread_id(os::current_thread_id());
674
675 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
676 os::current_thread_id(), (uintx) pthread_self());
677
678 if (UseNUMA) {
679 int lgrp_id = os::numa_get_group_id();
680 if (lgrp_id != -1) {
681 thread->set_lgrp_id(lgrp_id);
682 }
683 }
684 // initialize signal mask for this thread
685 os::Linux::hotspot_sigmask(thread);
686
687 // initialize floating point control register
688 os::Linux::init_thread_fpu_state();
689
690 // handshaking with parent thread
691 {
692 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
693
694 // notify parent thread
695 osthread->set_state(INITIALIZED);
696 sync->notify_all();
697
698 // wait until os::start_thread()
699 while (osthread->get_state() == INITIALIZED) {
700 sync->wait(Mutex::_no_safepoint_check_flag);
701 }
702 }
703
704 // call one more level start routine
705 thread->run();
706
707 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
708 os::current_thread_id(), (uintx) pthread_self());
709
710 // If a thread has not deleted itself ("delete this") as part of its
711 // termination sequence, we have to ensure thread-local-storage is
712 // cleared before we actually terminate. No threads should ever be
713 // deleted asynchronously with respect to their termination.
714 if (Thread::current_or_null_safe() != NULL) {
715 assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
716 thread->clear_thread_current();
717 }
718
719 return 0;
720 }
721
722 bool os::create_thread(Thread* thread, ThreadType thr_type,
723 size_t req_stack_size) {
724 assert(thread->osthread() == NULL, "caller responsible");
725
726 // Allocate the OSThread object
727 OSThread* osthread = new OSThread(NULL, NULL);
728 if (osthread == NULL) {
729 return false;
730 }
731
732 // set the correct thread state
733 osthread->set_thread_type(thr_type);
734
735 // Initial state is ALLOCATED but not INITIALIZED
736 osthread->set_state(ALLOCATED);
737
738 thread->set_osthread(osthread);
739
740 // init thread attributes
741 pthread_attr_t attr;
742 pthread_attr_init(&attr);
743 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
744
745 // Calculate stack size if it's not specified by caller.
746 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
747 // In the Linux NPTL pthread implementation the guard size mechanism
748 // is not implemented properly. The posix standard requires adding
749 // the size of the guard pages to the stack size, instead Linux
750 // takes the space out of 'stacksize'. Thus we adapt the requested
751 // stack_size by the size of the guard pages to mimick proper
752 // behaviour. However, be careful not to end up with a size
753 // of zero due to overflow. Don't add the guard page in that case.
754 size_t guard_size = os::Linux::default_guard_size(thr_type);
755 if (stack_size <= SIZE_MAX - guard_size) {
756 stack_size += guard_size;
757 }
758 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
759
760 int status = pthread_attr_setstacksize(&attr, stack_size);
761 assert_status(status == 0, status, "pthread_attr_setstacksize");
762
763 // Configure glibc guard page.
764 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
765
766 ThreadState state;
767
768 {
769 pthread_t tid;
770 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
771
772 char buf[64];
773 if (ret == 0) {
774 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
775 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
776 } else {
777 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
778 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
779 }
780
781 pthread_attr_destroy(&attr);
782
783 if (ret != 0) {
784 // Need to clean up stuff we've allocated so far
785 thread->set_osthread(NULL);
786 delete osthread;
787 return false;
788 }
789
790 // Store pthread info into the OSThread
791 osthread->set_pthread_id(tid);
792
793 // Wait until child thread is either initialized or aborted
794 {
795 Monitor* sync_with_child = osthread->startThread_lock();
796 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
797 while ((state = osthread->get_state()) == ALLOCATED) {
798 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
799 }
800 }
801 }
802
803 // Aborted due to thread limit being reached
804 if (state == ZOMBIE) {
805 thread->set_osthread(NULL);
806 delete osthread;
807 return false;
808 }
809
810 // The thread is returned suspended (in state INITIALIZED),
811 // and is started higher up in the call chain
812 assert(state == INITIALIZED, "race condition");
813 return true;
814 }
815
816 /////////////////////////////////////////////////////////////////////////////
817 // attach existing thread
818
819 // bootstrap the main thread
820 bool os::create_main_thread(JavaThread* thread) {
821 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
822 return create_attached_thread(thread);
823 }
824
825 bool os::create_attached_thread(JavaThread* thread) {
826 #ifdef ASSERT
827 thread->verify_not_published();
828 #endif
829
830 // Allocate the OSThread object
831 OSThread* osthread = new OSThread(NULL, NULL);
832
833 if (osthread == NULL) {
834 return false;
835 }
836
837 // Store pthread info into the OSThread
838 osthread->set_thread_id(os::Linux::gettid());
839 osthread->set_pthread_id(::pthread_self());
840
841 // initialize floating point control register
842 os::Linux::init_thread_fpu_state();
843
844 // Initial thread state is RUNNABLE
845 osthread->set_state(RUNNABLE);
846
847 thread->set_osthread(osthread);
848
849 if (UseNUMA) {
850 int lgrp_id = os::numa_get_group_id();
851 if (lgrp_id != -1) {
852 thread->set_lgrp_id(lgrp_id);
853 }
854 }
855
856 if (os::Linux::is_initial_thread()) {
857 // If current thread is initial thread, its stack is mapped on demand,
858 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
859 // the entire stack region to avoid SEGV in stack banging.
860 // It is also useful to get around the heap-stack-gap problem on SuSE
861 // kernel (see 4821821 for details). We first expand stack to the top
862 // of yellow zone, then enable stack yellow zone (order is significant,
863 // enabling yellow zone first will crash JVM on SuSE Linux), so there
864 // is no gap between the last two virtual memory regions.
865
866 JavaThread *jt = (JavaThread *)thread;
867 address addr = jt->stack_reserved_zone_base();
868 assert(addr != NULL, "initialization problem?");
869 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
870
871 osthread->set_expanding_stack();
872 os::Linux::manually_expand_stack(jt, addr);
873 osthread->clear_expanding_stack();
874 }
875
876 // initialize signal mask for this thread
877 // and save the caller's signal mask
878 os::Linux::hotspot_sigmask(thread);
879
880 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
881 os::current_thread_id(), (uintx) pthread_self());
882
883 return true;
884 }
885
886 void os::pd_start_thread(Thread* thread) {
887 OSThread * osthread = thread->osthread();
888 assert(osthread->get_state() != INITIALIZED, "just checking");
889 Monitor* sync_with_child = osthread->startThread_lock();
890 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
891 sync_with_child->notify();
892 }
893
894 // Free Linux resources related to the OSThread
895 void os::free_thread(OSThread* osthread) {
896 assert(osthread != NULL, "osthread not set");
897
898 // We are told to free resources of the argument thread,
899 // but we can only really operate on the current thread.
900 assert(Thread::current()->osthread() == osthread,
901 "os::free_thread but not current thread");
902
903 #ifdef ASSERT
904 sigset_t current;
905 sigemptyset(¤t);
906 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
907 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
908 #endif
909
910 // Restore caller's signal mask
911 sigset_t sigmask = osthread->caller_sigmask();
912 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
913
914 delete osthread;
915 }
916
917 //////////////////////////////////////////////////////////////////////////////
918 // initial thread
919
920 // Check if current thread is the initial thread, similar to Solaris thr_main.
921 bool os::Linux::is_initial_thread(void) {
922 char dummy;
923 // If called before init complete, thread stack bottom will be null.
924 // Can be called if fatal error occurs before initialization.
925 if (initial_thread_stack_bottom() == NULL) return false;
926 assert(initial_thread_stack_bottom() != NULL &&
927 initial_thread_stack_size() != 0,
928 "os::init did not locate initial thread's stack region");
929 if ((address)&dummy >= initial_thread_stack_bottom() &&
930 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
931 return true;
932 } else {
933 return false;
934 }
935 }
936
937 // Find the virtual memory area that contains addr
938 static bool find_vma(address addr, address* vma_low, address* vma_high) {
939 FILE *fp = fopen("/proc/self/maps", "r");
940 if (fp) {
941 address low, high;
942 while (!feof(fp)) {
943 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
944 if (low <= addr && addr < high) {
945 if (vma_low) *vma_low = low;
946 if (vma_high) *vma_high = high;
947 fclose(fp);
948 return true;
949 }
950 }
951 for (;;) {
952 int ch = fgetc(fp);
953 if (ch == EOF || ch == (int)'\n') break;
954 }
955 }
956 fclose(fp);
957 }
958 return false;
959 }
960
961 // Locate initial thread stack. This special handling of initial thread stack
962 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
963 // bogus value for the primordial process thread. While the launcher has created
964 // the VM in a new thread since JDK 6, we still have to allow for the use of the
965 // JNI invocation API from a primordial thread.
966 void os::Linux::capture_initial_stack(size_t max_size) {
967
968 // max_size is either 0 (which means accept OS default for thread stacks) or
969 // a user-specified value known to be at least the minimum needed. If we
970 // are actually on the primordial thread we can make it appear that we have a
971 // smaller max_size stack by inserting the guard pages at that location. But we
972 // cannot do anything to emulate a larger stack than what has been provided by
973 // the OS or threading library. In fact if we try to use a stack greater than
974 // what is set by rlimit then we will crash the hosting process.
975
976 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
977 // If this is "unlimited" then it will be a huge value.
978 struct rlimit rlim;
979 getrlimit(RLIMIT_STACK, &rlim);
980 size_t stack_size = rlim.rlim_cur;
981
982 // 6308388: a bug in ld.so will relocate its own .data section to the
983 // lower end of primordial stack; reduce ulimit -s value a little bit
984 // so we won't install guard page on ld.so's data section.
985 stack_size -= 2 * page_size();
986
987 // Try to figure out where the stack base (top) is. This is harder.
988 //
989 // When an application is started, glibc saves the initial stack pointer in
990 // a global variable "__libc_stack_end", which is then used by system
991 // libraries. __libc_stack_end should be pretty close to stack top. The
992 // variable is available since the very early days. However, because it is
993 // a private interface, it could disappear in the future.
994 //
995 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
996 // to __libc_stack_end, it is very close to stack top, but isn't the real
997 // stack top. Note that /proc may not exist if VM is running as a chroot
998 // program, so reading /proc/<pid>/stat could fail. Also the contents of
999 // /proc/<pid>/stat could change in the future (though unlikely).
1000 //
1001 // We try __libc_stack_end first. If that doesn't work, look for
1002 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1003 // as a hint, which should work well in most cases.
1004
1005 uintptr_t stack_start;
1006
1007 // try __libc_stack_end first
1008 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1009 if (p && *p) {
1010 stack_start = *p;
1011 } else {
1012 // see if we can get the start_stack field from /proc/self/stat
1013 FILE *fp;
1014 int pid;
1015 char state;
1016 int ppid;
1017 int pgrp;
1018 int session;
1019 int nr;
1020 int tpgrp;
1021 unsigned long flags;
1022 unsigned long minflt;
1023 unsigned long cminflt;
1024 unsigned long majflt;
1025 unsigned long cmajflt;
1026 unsigned long utime;
1027 unsigned long stime;
1028 long cutime;
1029 long cstime;
1030 long prio;
1031 long nice;
1032 long junk;
1033 long it_real;
1034 uintptr_t start;
1035 uintptr_t vsize;
1036 intptr_t rss;
1037 uintptr_t rsslim;
1038 uintptr_t scodes;
1039 uintptr_t ecode;
1040 int i;
1041
1042 // Figure what the primordial thread stack base is. Code is inspired
1043 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1044 // followed by command name surrounded by parentheses, state, etc.
1045 char stat[2048];
1046 int statlen;
1047
1048 fp = fopen("/proc/self/stat", "r");
1049 if (fp) {
1050 statlen = fread(stat, 1, 2047, fp);
1051 stat[statlen] = '\0';
1052 fclose(fp);
1053
1054 // Skip pid and the command string. Note that we could be dealing with
1055 // weird command names, e.g. user could decide to rename java launcher
1056 // to "java 1.4.2 :)", then the stat file would look like
1057 // 1234 (java 1.4.2 :)) R ... ...
1058 // We don't really need to know the command string, just find the last
1059 // occurrence of ")" and then start parsing from there. See bug 4726580.
1060 char * s = strrchr(stat, ')');
1061
1062 i = 0;
1063 if (s) {
1064 // Skip blank chars
1065 do { s++; } while (s && isspace(*s));
1066
1067 #define _UFM UINTX_FORMAT
1068 #define _DFM INTX_FORMAT
1069
1070 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1071 // 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
1072 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,
1073 &state, // 3 %c
1074 &ppid, // 4 %d
1075 &pgrp, // 5 %d
1076 &session, // 6 %d
1077 &nr, // 7 %d
1078 &tpgrp, // 8 %d
1079 &flags, // 9 %lu
1080 &minflt, // 10 %lu
1081 &cminflt, // 11 %lu
1082 &majflt, // 12 %lu
1083 &cmajflt, // 13 %lu
1084 &utime, // 14 %lu
1085 &stime, // 15 %lu
1086 &cutime, // 16 %ld
1087 &cstime, // 17 %ld
1088 &prio, // 18 %ld
1089 &nice, // 19 %ld
1090 &junk, // 20 %ld
1091 &it_real, // 21 %ld
1092 &start, // 22 UINTX_FORMAT
1093 &vsize, // 23 UINTX_FORMAT
1094 &rss, // 24 INTX_FORMAT
1095 &rsslim, // 25 UINTX_FORMAT
1096 &scodes, // 26 UINTX_FORMAT
1097 &ecode, // 27 UINTX_FORMAT
1098 &stack_start); // 28 UINTX_FORMAT
1099 }
1100
1101 #undef _UFM
1102 #undef _DFM
1103
1104 if (i != 28 - 2) {
1105 assert(false, "Bad conversion from /proc/self/stat");
1106 // product mode - assume we are the initial thread, good luck in the
1107 // embedded case.
1108 warning("Can't detect initial thread stack location - bad conversion");
1109 stack_start = (uintptr_t) &rlim;
1110 }
1111 } else {
1112 // For some reason we can't open /proc/self/stat (for example, running on
1113 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1114 // most cases, so don't abort:
1115 warning("Can't detect initial thread stack location - no /proc/self/stat");
1116 stack_start = (uintptr_t) &rlim;
1117 }
1118 }
1119
1120 // Now we have a pointer (stack_start) very close to the stack top, the
1121 // next thing to do is to figure out the exact location of stack top. We
1122 // can find out the virtual memory area that contains stack_start by
1123 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1124 // and its upper limit is the real stack top. (again, this would fail if
1125 // running inside chroot, because /proc may not exist.)
1126
1127 uintptr_t stack_top;
1128 address low, high;
1129 if (find_vma((address)stack_start, &low, &high)) {
1130 // success, "high" is the true stack top. (ignore "low", because initial
1131 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1132 stack_top = (uintptr_t)high;
1133 } else {
1134 // failed, likely because /proc/self/maps does not exist
1135 warning("Can't detect initial thread stack location - find_vma failed");
1136 // best effort: stack_start is normally within a few pages below the real
1137 // stack top, use it as stack top, and reduce stack size so we won't put
1138 // guard page outside stack.
1139 stack_top = stack_start;
1140 stack_size -= 16 * page_size();
1141 }
1142
1143 // stack_top could be partially down the page so align it
1144 stack_top = align_up(stack_top, page_size());
1145
1146 // Allowed stack value is minimum of max_size and what we derived from rlimit
1147 if (max_size > 0) {
1148 _initial_thread_stack_size = MIN2(max_size, stack_size);
1149 } else {
1150 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1151 // clamp it at 8MB as we do on Solaris
1152 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1153 }
1154 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1155 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1156
1157 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1158
1159 if (log_is_enabled(Info, os, thread)) {
1160 // See if we seem to be on primordial process thread
1161 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1162 uintptr_t(&rlim) < stack_top;
1163
1164 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1165 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1166 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1167 stack_top, intptr_t(_initial_thread_stack_bottom));
1168 }
1169 }
1170
1171 ////////////////////////////////////////////////////////////////////////////////
1172 // time support
1173
1174 // Time since start-up in seconds to a fine granularity.
1175 // Used by VMSelfDestructTimer and the MemProfiler.
1176 double os::elapsedTime() {
1177
1178 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1179 }
1180
1181 jlong os::elapsed_counter() {
1182 return javaTimeNanos() - initial_time_count;
1183 }
1184
1185 jlong os::elapsed_frequency() {
1186 return NANOSECS_PER_SEC; // nanosecond resolution
1187 }
1188
1189 bool os::supports_vtime() { return true; }
1190 bool os::enable_vtime() { return false; }
1191 bool os::vtime_enabled() { return false; }
1192
1193 double os::elapsedVTime() {
1194 struct rusage usage;
1195 int retval = getrusage(RUSAGE_THREAD, &usage);
1196 if (retval == 0) {
1197 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);
1198 } else {
1199 // better than nothing, but not much
1200 return elapsedTime();
1201 }
1202 }
1203
1204 jlong os::javaTimeMillis() {
1205 timeval time;
1206 int status = gettimeofday(&time, NULL);
1207 assert(status != -1, "linux error");
1208 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1209 }
1210
1211 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1212 timeval time;
1213 int status = gettimeofday(&time, NULL);
1214 assert(status != -1, "linux error");
1215 seconds = jlong(time.tv_sec);
1216 nanos = jlong(time.tv_usec) * 1000;
1217 }
1218
1219
1220 #ifndef CLOCK_MONOTONIC
1221 #define CLOCK_MONOTONIC (1)
1222 #endif
1223
1224 void os::Linux::clock_init() {
1225 // we do dlopen's in this particular order due to bug in linux
1226 // dynamical loader (see 6348968) leading to crash on exit
1227 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1228 if (handle == NULL) {
1229 handle = dlopen("librt.so", RTLD_LAZY);
1230 }
1231
1232 if (handle) {
1233 int (*clock_getres_func)(clockid_t, struct timespec*) =
1234 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1235 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1236 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1237 if (clock_getres_func && clock_gettime_func) {
1238 // See if monotonic clock is supported by the kernel. Note that some
1239 // early implementations simply return kernel jiffies (updated every
1240 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1241 // for nano time (though the monotonic property is still nice to have).
1242 // It's fixed in newer kernels, however clock_getres() still returns
1243 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1244 // resolution for now. Hopefully as people move to new kernels, this
1245 // won't be a problem.
1246 struct timespec res;
1247 struct timespec tp;
1248 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1249 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1250 // yes, monotonic clock is supported
1251 _clock_gettime = clock_gettime_func;
1252 return;
1253 } else {
1254 // close librt if there is no monotonic clock
1255 dlclose(handle);
1256 }
1257 }
1258 }
1259 warning("No monotonic clock was available - timed services may " \
1260 "be adversely affected if the time-of-day clock changes");
1261 }
1262
1263 #ifndef SYS_clock_getres
1264 #if defined(X86) || defined(PPC64) || defined(S390)
1265 #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261)
1266 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1267 #else
1268 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1269 #define sys_clock_getres(x,y) -1
1270 #endif
1271 #else
1272 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1273 #endif
1274
1275 void os::Linux::fast_thread_clock_init() {
1276 if (!UseLinuxPosixThreadCPUClocks) {
1277 return;
1278 }
1279 clockid_t clockid;
1280 struct timespec tp;
1281 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1282 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1283
1284 // Switch to using fast clocks for thread cpu time if
1285 // the sys_clock_getres() returns 0 error code.
1286 // Note, that some kernels may support the current thread
1287 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1288 // returned by the pthread_getcpuclockid().
1289 // If the fast Posix clocks are supported then the sys_clock_getres()
1290 // must return at least tp.tv_sec == 0 which means a resolution
1291 // better than 1 sec. This is extra check for reliability.
1292
1293 if (pthread_getcpuclockid_func &&
1294 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1295 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1296 _supports_fast_thread_cpu_time = true;
1297 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1298 }
1299 }
1300
1301 jlong os::javaTimeNanos() {
1302 if (os::supports_monotonic_clock()) {
1303 struct timespec tp;
1304 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1305 assert(status == 0, "gettime error");
1306 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1307 return result;
1308 } else {
1309 timeval time;
1310 int status = gettimeofday(&time, NULL);
1311 assert(status != -1, "linux error");
1312 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1313 return 1000 * usecs;
1314 }
1315 }
1316
1317 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1318 if (os::supports_monotonic_clock()) {
1319 info_ptr->max_value = ALL_64_BITS;
1320
1321 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1322 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1323 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1324 } else {
1325 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1326 info_ptr->max_value = ALL_64_BITS;
1327
1328 // gettimeofday is a real time clock so it skips
1329 info_ptr->may_skip_backward = true;
1330 info_ptr->may_skip_forward = true;
1331 }
1332
1333 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1334 }
1335
1336 // Return the real, user, and system times in seconds from an
1337 // arbitrary fixed point in the past.
1338 bool os::getTimesSecs(double* process_real_time,
1339 double* process_user_time,
1340 double* process_system_time) {
1341 struct tms ticks;
1342 clock_t real_ticks = times(&ticks);
1343
1344 if (real_ticks == (clock_t) (-1)) {
1345 return false;
1346 } else {
1347 double ticks_per_second = (double) clock_tics_per_sec;
1348 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1349 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1350 *process_real_time = ((double) real_ticks) / ticks_per_second;
1351
1352 return true;
1353 }
1354 }
1355
1356
1357 char * os::local_time_string(char *buf, size_t buflen) {
1358 struct tm t;
1359 time_t long_time;
1360 time(&long_time);
1361 localtime_r(&long_time, &t);
1362 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1363 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1364 t.tm_hour, t.tm_min, t.tm_sec);
1365 return buf;
1366 }
1367
1368 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1369 return localtime_r(clock, res);
1370 }
1371
1372 ////////////////////////////////////////////////////////////////////////////////
1373 // runtime exit support
1374
1375 // Note: os::shutdown() might be called very early during initialization, or
1376 // called from signal handler. Before adding something to os::shutdown(), make
1377 // sure it is async-safe and can handle partially initialized VM.
1378 void os::shutdown() {
1379
1380 // allow PerfMemory to attempt cleanup of any persistent resources
1381 perfMemory_exit();
1382
1383 // needs to remove object in file system
1384 AttachListener::abort();
1385
1386 // flush buffered output, finish log files
1387 ostream_abort();
1388
1389 // Check for abort hook
1390 abort_hook_t abort_hook = Arguments::abort_hook();
1391 if (abort_hook != NULL) {
1392 abort_hook();
1393 }
1394
1395 }
1396
1397 // Note: os::abort() might be called very early during initialization, or
1398 // called from signal handler. Before adding something to os::abort(), make
1399 // sure it is async-safe and can handle partially initialized VM.
1400 void os::abort(bool dump_core, void* siginfo, const void* context) {
1401 os::shutdown();
1402 if (dump_core) {
1403 #ifndef PRODUCT
1404 fdStream out(defaultStream::output_fd());
1405 out.print_raw("Current thread is ");
1406 char buf[16];
1407 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1408 out.print_raw_cr(buf);
1409 out.print_raw_cr("Dumping core ...");
1410 #endif
1411 ::abort(); // dump core
1412 }
1413
1414 ::exit(1);
1415 }
1416
1417 // Die immediately, no exit hook, no abort hook, no cleanup.
1418 void os::die() {
1419 ::abort();
1420 }
1421
1422
1423 // This method is a copy of JDK's sysGetLastErrorString
1424 // from src/solaris/hpi/src/system_md.c
1425
1426 size_t os::lasterror(char *buf, size_t len) {
1427 if (errno == 0) return 0;
1428
1429 const char *s = os::strerror(errno);
1430 size_t n = ::strlen(s);
1431 if (n >= len) {
1432 n = len - 1;
1433 }
1434 ::strncpy(buf, s, n);
1435 buf[n] = '\0';
1436 return n;
1437 }
1438
1439 // thread_id is kernel thread id (similar to Solaris LWP id)
1440 intx os::current_thread_id() { return os::Linux::gettid(); }
1441 int os::current_process_id() {
1442 return ::getpid();
1443 }
1444
1445 // DLL functions
1446
1447 const char* os::dll_file_extension() { return ".so"; }
1448
1449 // This must be hard coded because it's the system's temporary
1450 // directory not the java application's temp directory, ala java.io.tmpdir.
1451 const char* os::get_temp_directory() { return "/tmp"; }
1452
1453 static bool file_exists(const char* filename) {
1454 struct stat statbuf;
1455 if (filename == NULL || strlen(filename) == 0) {
1456 return false;
1457 }
1458 return os::stat(filename, &statbuf) == 0;
1459 }
1460
1461 // check if addr is inside libjvm.so
1462 bool os::address_is_in_vm(address addr) {
1463 static address libjvm_base_addr;
1464 Dl_info dlinfo;
1465
1466 if (libjvm_base_addr == NULL) {
1467 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1468 libjvm_base_addr = (address)dlinfo.dli_fbase;
1469 }
1470 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1471 }
1472
1473 if (dladdr((void *)addr, &dlinfo) != 0) {
1474 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1475 }
1476
1477 return false;
1478 }
1479
1480 bool os::dll_address_to_function_name(address addr, char *buf,
1481 int buflen, int *offset,
1482 bool demangle) {
1483 // buf is not optional, but offset is optional
1484 assert(buf != NULL, "sanity check");
1485
1486 Dl_info dlinfo;
1487
1488 if (dladdr((void*)addr, &dlinfo) != 0) {
1489 // see if we have a matching symbol
1490 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1491 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1492 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1493 }
1494 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1495 return true;
1496 }
1497 // no matching symbol so try for just file info
1498 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1499 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1500 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1501 return true;
1502 }
1503 }
1504 }
1505
1506 buf[0] = '\0';
1507 if (offset != NULL) *offset = -1;
1508 return false;
1509 }
1510
1511 struct _address_to_library_name {
1512 address addr; // input : memory address
1513 size_t buflen; // size of fname
1514 char* fname; // output: library name
1515 address base; // library base addr
1516 };
1517
1518 static int address_to_library_name_callback(struct dl_phdr_info *info,
1519 size_t size, void *data) {
1520 int i;
1521 bool found = false;
1522 address libbase = NULL;
1523 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1524
1525 // iterate through all loadable segments
1526 for (i = 0; i < info->dlpi_phnum; i++) {
1527 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1528 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1529 // base address of a library is the lowest address of its loaded
1530 // segments.
1531 if (libbase == NULL || libbase > segbase) {
1532 libbase = segbase;
1533 }
1534 // see if 'addr' is within current segment
1535 if (segbase <= d->addr &&
1536 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1537 found = true;
1538 }
1539 }
1540 }
1541
1542 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1543 // so dll_address_to_library_name() can fall through to use dladdr() which
1544 // can figure out executable name from argv[0].
1545 if (found && info->dlpi_name && info->dlpi_name[0]) {
1546 d->base = libbase;
1547 if (d->fname) {
1548 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1549 }
1550 return 1;
1551 }
1552 return 0;
1553 }
1554
1555 bool os::dll_address_to_library_name(address addr, char* buf,
1556 int buflen, int* offset) {
1557 // buf is not optional, but offset is optional
1558 assert(buf != NULL, "sanity check");
1559
1560 Dl_info dlinfo;
1561 struct _address_to_library_name data;
1562
1563 // There is a bug in old glibc dladdr() implementation that it could resolve
1564 // to wrong library name if the .so file has a base address != NULL. Here
1565 // we iterate through the program headers of all loaded libraries to find
1566 // out which library 'addr' really belongs to. This workaround can be
1567 // removed once the minimum requirement for glibc is moved to 2.3.x.
1568 data.addr = addr;
1569 data.fname = buf;
1570 data.buflen = buflen;
1571 data.base = NULL;
1572 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1573
1574 if (rslt) {
1575 // buf already contains library name
1576 if (offset) *offset = addr - data.base;
1577 return true;
1578 }
1579 if (dladdr((void*)addr, &dlinfo) != 0) {
1580 if (dlinfo.dli_fname != NULL) {
1581 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1582 }
1583 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1584 *offset = addr - (address)dlinfo.dli_fbase;
1585 }
1586 return true;
1587 }
1588
1589 buf[0] = '\0';
1590 if (offset) *offset = -1;
1591 return false;
1592 }
1593
1594 // Loads .dll/.so and
1595 // in case of error it checks if .dll/.so was built for the
1596 // same architecture as Hotspot is running on
1597
1598
1599 // Remember the stack's state. The Linux dynamic linker will change
1600 // the stack to 'executable' at most once, so we must safepoint only once.
1601 bool os::Linux::_stack_is_executable = false;
1602
1603 // VM operation that loads a library. This is necessary if stack protection
1604 // of the Java stacks can be lost during loading the library. If we
1605 // do not stop the Java threads, they can stack overflow before the stacks
1606 // are protected again.
1607 class VM_LinuxDllLoad: public VM_Operation {
1608 private:
1609 const char *_filename;
1610 char *_ebuf;
1611 int _ebuflen;
1612 void *_lib;
1613 public:
1614 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1615 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1616 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1617 void doit() {
1618 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1619 os::Linux::_stack_is_executable = true;
1620 }
1621 void* loaded_library() { return _lib; }
1622 };
1623
1624 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1625 void * result = NULL;
1626 bool load_attempted = false;
1627
1628 // Check whether the library to load might change execution rights
1629 // of the stack. If they are changed, the protection of the stack
1630 // guard pages will be lost. We need a safepoint to fix this.
1631 //
1632 // See Linux man page execstack(8) for more info.
1633 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1634 if (!ElfFile::specifies_noexecstack(filename)) {
1635 if (!is_init_completed()) {
1636 os::Linux::_stack_is_executable = true;
1637 // This is OK - No Java threads have been created yet, and hence no
1638 // stack guard pages to fix.
1639 //
1640 // This should happen only when you are building JDK7 using a very
1641 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1642 //
1643 // Dynamic loader will make all stacks executable after
1644 // this function returns, and will not do that again.
1645 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1646 } else {
1647 warning("You have loaded library %s which might have disabled stack guard. "
1648 "The VM will try to fix the stack guard now.\n"
1649 "It's highly recommended that you fix the library with "
1650 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1651 filename);
1652
1653 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1654 JavaThread *jt = JavaThread::current();
1655 if (jt->thread_state() != _thread_in_native) {
1656 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1657 // that requires ExecStack. Cannot enter safe point. Let's give up.
1658 warning("Unable to fix stack guard. Giving up.");
1659 } else {
1660 if (!LoadExecStackDllInVMThread) {
1661 // This is for the case where the DLL has an static
1662 // constructor function that executes JNI code. We cannot
1663 // load such DLLs in the VMThread.
1664 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1665 }
1666
1667 ThreadInVMfromNative tiv(jt);
1668 debug_only(VMNativeEntryWrapper vew;)
1669
1670 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1671 VMThread::execute(&op);
1672 if (LoadExecStackDllInVMThread) {
1673 result = op.loaded_library();
1674 }
1675 load_attempted = true;
1676 }
1677 }
1678 }
1679 }
1680
1681 if (!load_attempted) {
1682 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1683 }
1684
1685 if (result != NULL) {
1686 // Successful loading
1687 return result;
1688 }
1689
1690 Elf32_Ehdr elf_head;
1691 int diag_msg_max_length=ebuflen-strlen(ebuf);
1692 char* diag_msg_buf=ebuf+strlen(ebuf);
1693
1694 if (diag_msg_max_length==0) {
1695 // No more space in ebuf for additional diagnostics message
1696 return NULL;
1697 }
1698
1699
1700 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1701
1702 if (file_descriptor < 0) {
1703 // Can't open library, report dlerror() message
1704 return NULL;
1705 }
1706
1707 bool failed_to_read_elf_head=
1708 (sizeof(elf_head)!=
1709 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1710
1711 ::close(file_descriptor);
1712 if (failed_to_read_elf_head) {
1713 // file i/o error - report dlerror() msg
1714 return NULL;
1715 }
1716
1717 typedef struct {
1718 Elf32_Half code; // Actual value as defined in elf.h
1719 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1720 unsigned char elf_class; // 32 or 64 bit
1721 unsigned char endianess; // MSB or LSB
1722 char* name; // String representation
1723 } arch_t;
1724
1725 #ifndef EM_486
1726 #define EM_486 6 /* Intel 80486 */
1727 #endif
1728 #ifndef EM_AARCH64
1729 #define EM_AARCH64 183 /* ARM AARCH64 */
1730 #endif
1731
1732 static const arch_t arch_array[]={
1733 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1734 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1735 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1736 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1737 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1738 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1739 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1740 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1741 #if defined(VM_LITTLE_ENDIAN)
1742 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1743 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1744 #else
1745 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1746 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1747 #endif
1748 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1749 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1750 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1751 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1752 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1753 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1754 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1755 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1756 };
1757
1758 #if (defined IA32)
1759 static Elf32_Half running_arch_code=EM_386;
1760 #elif (defined AMD64)
1761 static Elf32_Half running_arch_code=EM_X86_64;
1762 #elif (defined IA64)
1763 static Elf32_Half running_arch_code=EM_IA_64;
1764 #elif (defined __sparc) && (defined _LP64)
1765 static Elf32_Half running_arch_code=EM_SPARCV9;
1766 #elif (defined __sparc) && (!defined _LP64)
1767 static Elf32_Half running_arch_code=EM_SPARC;
1768 #elif (defined __powerpc64__)
1769 static Elf32_Half running_arch_code=EM_PPC64;
1770 #elif (defined __powerpc__)
1771 static Elf32_Half running_arch_code=EM_PPC;
1772 #elif (defined AARCH64)
1773 static Elf32_Half running_arch_code=EM_AARCH64;
1774 #elif (defined ARM)
1775 static Elf32_Half running_arch_code=EM_ARM;
1776 #elif (defined S390)
1777 static Elf32_Half running_arch_code=EM_S390;
1778 #elif (defined ALPHA)
1779 static Elf32_Half running_arch_code=EM_ALPHA;
1780 #elif (defined MIPSEL)
1781 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1782 #elif (defined PARISC)
1783 static Elf32_Half running_arch_code=EM_PARISC;
1784 #elif (defined MIPS)
1785 static Elf32_Half running_arch_code=EM_MIPS;
1786 #elif (defined M68K)
1787 static Elf32_Half running_arch_code=EM_68K;
1788 #elif (defined SH)
1789 static Elf32_Half running_arch_code=EM_SH;
1790 #else
1791 #error Method os::dll_load requires that one of following is defined:\
1792 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1793 #endif
1794
1795 // Identify compatability class for VM's architecture and library's architecture
1796 // Obtain string descriptions for architectures
1797
1798 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1799 int running_arch_index=-1;
1800
1801 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1802 if (running_arch_code == arch_array[i].code) {
1803 running_arch_index = i;
1804 }
1805 if (lib_arch.code == arch_array[i].code) {
1806 lib_arch.compat_class = arch_array[i].compat_class;
1807 lib_arch.name = arch_array[i].name;
1808 }
1809 }
1810
1811 assert(running_arch_index != -1,
1812 "Didn't find running architecture code (running_arch_code) in arch_array");
1813 if (running_arch_index == -1) {
1814 // Even though running architecture detection failed
1815 // we may still continue with reporting dlerror() message
1816 return NULL;
1817 }
1818
1819 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1820 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1821 return NULL;
1822 }
1823
1824 #ifndef S390
1825 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1826 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1827 return NULL;
1828 }
1829 #endif // !S390
1830
1831 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1832 if (lib_arch.name!=NULL) {
1833 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1834 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1835 lib_arch.name, arch_array[running_arch_index].name);
1836 } else {
1837 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1838 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1839 lib_arch.code,
1840 arch_array[running_arch_index].name);
1841 }
1842 }
1843
1844 return NULL;
1845 }
1846
1847 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1848 int ebuflen) {
1849 void * result = ::dlopen(filename, RTLD_LAZY);
1850 if (result == NULL) {
1851 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1852 ebuf[ebuflen-1] = '\0';
1853 }
1854 return result;
1855 }
1856
1857 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1858 int ebuflen) {
1859 void * result = NULL;
1860 if (LoadExecStackDllInVMThread) {
1861 result = dlopen_helper(filename, ebuf, ebuflen);
1862 }
1863
1864 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1865 // library that requires an executable stack, or which does not have this
1866 // stack attribute set, dlopen changes the stack attribute to executable. The
1867 // read protection of the guard pages gets lost.
1868 //
1869 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1870 // may have been queued at the same time.
1871
1872 if (!_stack_is_executable) {
1873 JavaThread *jt = Threads::first();
1874
1875 while (jt) {
1876 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1877 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1878 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1879 warning("Attempt to reguard stack yellow zone failed.");
1880 }
1881 }
1882 jt = jt->next();
1883 }
1884 }
1885
1886 return result;
1887 }
1888
1889 void* os::dll_lookup(void* handle, const char* name) {
1890 void* res = dlsym(handle, name);
1891 return res;
1892 }
1893
1894 void* os::get_default_process_handle() {
1895 return (void*)::dlopen(NULL, RTLD_LAZY);
1896 }
1897
1898 static bool _print_ascii_file(const char* filename, outputStream* st) {
1899 int fd = ::open(filename, O_RDONLY);
1900 if (fd == -1) {
1901 return false;
1902 }
1903
1904 char buf[33];
1905 int bytes;
1906 buf[32] = '\0';
1907 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1908 st->print_raw(buf, bytes);
1909 }
1910
1911 ::close(fd);
1912
1913 return true;
1914 }
1915
1916 void os::print_dll_info(outputStream *st) {
1917 st->print_cr("Dynamic libraries:");
1918
1919 char fname[32];
1920 pid_t pid = os::Linux::gettid();
1921
1922 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1923
1924 if (!_print_ascii_file(fname, st)) {
1925 st->print("Can not get library information for pid = %d\n", pid);
1926 }
1927 }
1928
1929 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1930 FILE *procmapsFile = NULL;
1931
1932 // Open the procfs maps file for the current process
1933 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1934 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1935 char line[PATH_MAX + 100];
1936
1937 // Read line by line from 'file'
1938 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1939 u8 base, top, offset, inode;
1940 char permissions[5];
1941 char device[6];
1942 char name[PATH_MAX + 1];
1943
1944 // Parse fields from line
1945 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1946 &base, &top, permissions, &offset, device, &inode, name);
1947
1948 // Filter by device id '00:00' so that we only get file system mapped files.
1949 if (strcmp(device, "00:00") != 0) {
1950
1951 // Call callback with the fields of interest
1952 if(callback(name, (address)base, (address)top, param)) {
1953 // Oops abort, callback aborted
1954 fclose(procmapsFile);
1955 return 1;
1956 }
1957 }
1958 }
1959 fclose(procmapsFile);
1960 }
1961 return 0;
1962 }
1963
1964 void os::print_os_info_brief(outputStream* st) {
1965 os::Linux::print_distro_info(st);
1966
1967 os::Posix::print_uname_info(st);
1968
1969 os::Linux::print_libversion_info(st);
1970
1971 }
1972
1973 void os::print_os_info(outputStream* st) {
1974 st->print("OS:");
1975
1976 os::Linux::print_distro_info(st);
1977
1978 os::Posix::print_uname_info(st);
1979
1980 // Print warning if unsafe chroot environment detected
1981 if (unsafe_chroot_detected) {
1982 st->print("WARNING!! ");
1983 st->print_cr("%s", unstable_chroot_error);
1984 }
1985
1986 os::Linux::print_libversion_info(st);
1987
1988 os::Posix::print_rlimit_info(st);
1989
1990 os::Posix::print_load_average(st);
1991
1992 os::Linux::print_full_memory_info(st);
1993
1994 os::Linux::print_container_info(st);
1995 }
1996
1997 // Try to identify popular distros.
1998 // Most Linux distributions have a /etc/XXX-release file, which contains
1999 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2000 // file that also contains the OS version string. Some have more than one
2001 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2002 // /etc/redhat-release.), so the order is important.
2003 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2004 // their own specific XXX-release file as well as a redhat-release file.
2005 // Because of this the XXX-release file needs to be searched for before the
2006 // redhat-release file.
2007 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2008 // search for redhat-release / SuSE-release needs to be before lsb-release.
2009 // Since the lsb-release file is the new standard it needs to be searched
2010 // before the older style release files.
2011 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2012 // next to last resort. The os-release file is a new standard that contains
2013 // distribution information and the system-release file seems to be an old
2014 // standard that has been replaced by the lsb-release and os-release files.
2015 // Searching for the debian_version file is the last resort. It contains
2016 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2017 // "Debian " is printed before the contents of the debian_version file.
2018
2019 const char* distro_files[] = {
2020 "/etc/oracle-release",
2021 "/etc/mandriva-release",
2022 "/etc/mandrake-release",
2023 "/etc/sun-release",
2024 "/etc/redhat-release",
2025 "/etc/SuSE-release",
2026 "/etc/lsb-release",
2027 "/etc/turbolinux-release",
2028 "/etc/gentoo-release",
2029 "/etc/ltib-release",
2030 "/etc/angstrom-version",
2031 "/etc/system-release",
2032 "/etc/os-release",
2033 NULL };
2034
2035 void os::Linux::print_distro_info(outputStream* st) {
2036 for (int i = 0;; i++) {
2037 const char* file = distro_files[i];
2038 if (file == NULL) {
2039 break; // done
2040 }
2041 // If file prints, we found it.
2042 if (_print_ascii_file(file, st)) {
2043 return;
2044 }
2045 }
2046
2047 if (file_exists("/etc/debian_version")) {
2048 st->print("Debian ");
2049 _print_ascii_file("/etc/debian_version", st);
2050 } else {
2051 st->print("Linux");
2052 }
2053 st->cr();
2054 }
2055
2056 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2057 char buf[256];
2058 while (fgets(buf, sizeof(buf), fp)) {
2059 // Edit out extra stuff in expected format
2060 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2061 char* ptr = strstr(buf, "\""); // the name is in quotes
2062 if (ptr != NULL) {
2063 ptr++; // go beyond first quote
2064 char* nl = strchr(ptr, '\"');
2065 if (nl != NULL) *nl = '\0';
2066 strncpy(distro, ptr, length);
2067 } else {
2068 ptr = strstr(buf, "=");
2069 ptr++; // go beyond equals then
2070 char* nl = strchr(ptr, '\n');
2071 if (nl != NULL) *nl = '\0';
2072 strncpy(distro, ptr, length);
2073 }
2074 return;
2075 } else if (get_first_line) {
2076 char* nl = strchr(buf, '\n');
2077 if (nl != NULL) *nl = '\0';
2078 strncpy(distro, buf, length);
2079 return;
2080 }
2081 }
2082 // print last line and close
2083 char* nl = strchr(buf, '\n');
2084 if (nl != NULL) *nl = '\0';
2085 strncpy(distro, buf, length);
2086 }
2087
2088 static void parse_os_info(char* distro, size_t length, const char* file) {
2089 FILE* fp = fopen(file, "r");
2090 if (fp != NULL) {
2091 // if suse format, print out first line
2092 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2093 parse_os_info_helper(fp, distro, length, get_first_line);
2094 fclose(fp);
2095 }
2096 }
2097
2098 void os::get_summary_os_info(char* buf, size_t buflen) {
2099 for (int i = 0;; i++) {
2100 const char* file = distro_files[i];
2101 if (file == NULL) {
2102 break; // ran out of distro_files
2103 }
2104 if (file_exists(file)) {
2105 parse_os_info(buf, buflen, file);
2106 return;
2107 }
2108 }
2109 // special case for debian
2110 if (file_exists("/etc/debian_version")) {
2111 strncpy(buf, "Debian ", buflen);
2112 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2113 } else {
2114 strncpy(buf, "Linux", buflen);
2115 }
2116 }
2117
2118 void os::Linux::print_libversion_info(outputStream* st) {
2119 // libc, pthread
2120 st->print("libc:");
2121 st->print("%s ", os::Linux::glibc_version());
2122 st->print("%s ", os::Linux::libpthread_version());
2123 st->cr();
2124 }
2125
2126 void os::Linux::print_full_memory_info(outputStream* st) {
2127 st->print("\n/proc/meminfo:\n");
2128 _print_ascii_file("/proc/meminfo", st);
2129 st->cr();
2130 }
2131
2132 void os::Linux::print_container_info(outputStream* st) {
2133 if (OSContainer::is_containerized()) {
2134 st->print("container (cgroup) information:\n");
2135
2136 char *p = OSContainer::container_type();
2137 if (p == NULL)
2138 st->print("container_type() failed\n");
2139 else {
2140 st->print("container_type: %s\n", p);
2141 }
2142
2143 p = OSContainer::cpu_cpuset_cpus();
2144 if (p == NULL)
2145 st->print("cpu_cpuset_cpus() failed\n");
2146 else {
2147 st->print("cpu_cpuset_cpus: %s\n", p);
2148 free(p);
2149 }
2150
2151 p = OSContainer::cpu_cpuset_memory_nodes();
2152 if (p < 0)
2153 st->print("cpu_memory_nodes() failed\n");
2154 else {
2155 st->print("cpu_memory_nodes: %s\n", p);
2156 free(p);
2157 }
2158
2159 int i = OSContainer::active_processor_count();
2160 if (i < 0)
2161 st->print("active_processor_count() failed\n");
2162 else
2163 st->print("active_processor_count: %d\n", i);
2164
2165 i = OSContainer::cpu_quota();
2166 st->print("cpu_quota: %d\n", i);
2167
2168 i = OSContainer::cpu_period();
2169 st->print("cpu_period: %d\n", i);
2170
2171 i = OSContainer::cpu_shares();
2172 st->print("cpu_shares: %d\n", i);
2173
2174 jlong j = OSContainer::memory_limit_in_bytes();
2175 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2176
2177 j = OSContainer::memory_and_swap_limit_in_bytes();
2178 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2179
2180 j = OSContainer::memory_soft_limit_in_bytes();
2181 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2182
2183 j = OSContainer::OSContainer::memory_usage_in_bytes();
2184 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2185
2186 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2187 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2188 st->cr();
2189 }
2190 }
2191
2192 void os::print_memory_info(outputStream* st) {
2193
2194 st->print("Memory:");
2195 st->print(" %dk page", os::vm_page_size()>>10);
2196
2197 // values in struct sysinfo are "unsigned long"
2198 struct sysinfo si;
2199 sysinfo(&si);
2200
2201 st->print(", physical " UINT64_FORMAT "k",
2202 os::physical_memory() >> 10);
2203 st->print("(" UINT64_FORMAT "k free)",
2204 os::available_memory() >> 10);
2205 st->print(", swap " UINT64_FORMAT "k",
2206 ((jlong)si.totalswap * si.mem_unit) >> 10);
2207 st->print("(" UINT64_FORMAT "k free)",
2208 ((jlong)si.freeswap * si.mem_unit) >> 10);
2209 st->cr();
2210 }
2211
2212 // Print the first "model name" line and the first "flags" line
2213 // that we find and nothing more. We assume "model name" comes
2214 // before "flags" so if we find a second "model name", then the
2215 // "flags" field is considered missing.
2216 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2217 #if defined(IA32) || defined(AMD64)
2218 // Other platforms have less repetitive cpuinfo files
2219 FILE *fp = fopen("/proc/cpuinfo", "r");
2220 if (fp) {
2221 while (!feof(fp)) {
2222 if (fgets(buf, buflen, fp)) {
2223 // Assume model name comes before flags
2224 bool model_name_printed = false;
2225 if (strstr(buf, "model name") != NULL) {
2226 if (!model_name_printed) {
2227 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2228 st->print_raw(buf);
2229 model_name_printed = true;
2230 } else {
2231 // model name printed but not flags? Odd, just return
2232 fclose(fp);
2233 return true;
2234 }
2235 }
2236 // print the flags line too
2237 if (strstr(buf, "flags") != NULL) {
2238 st->print_raw(buf);
2239 fclose(fp);
2240 return true;
2241 }
2242 }
2243 }
2244 fclose(fp);
2245 }
2246 #endif // x86 platforms
2247 return false;
2248 }
2249
2250 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2251 // Only print the model name if the platform provides this as a summary
2252 if (!print_model_name_and_flags(st, buf, buflen)) {
2253 st->print("\n/proc/cpuinfo:\n");
2254 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2255 st->print_cr(" <Not Available>");
2256 }
2257 }
2258 }
2259
2260 #if defined(AMD64) || defined(IA32) || defined(X32)
2261 const char* search_string = "model name";
2262 #elif defined(M68K)
2263 const char* search_string = "CPU";
2264 #elif defined(PPC64)
2265 const char* search_string = "cpu";
2266 #elif defined(S390)
2267 const char* search_string = "processor";
2268 #elif defined(SPARC)
2269 const char* search_string = "cpu";
2270 #else
2271 const char* search_string = "Processor";
2272 #endif
2273
2274 // Parses the cpuinfo file for string representing the model name.
2275 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2276 FILE* fp = fopen("/proc/cpuinfo", "r");
2277 if (fp != NULL) {
2278 while (!feof(fp)) {
2279 char buf[256];
2280 if (fgets(buf, sizeof(buf), fp)) {
2281 char* start = strstr(buf, search_string);
2282 if (start != NULL) {
2283 char *ptr = start + strlen(search_string);
2284 char *end = buf + strlen(buf);
2285 while (ptr != end) {
2286 // skip whitespace and colon for the rest of the name.
2287 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2288 break;
2289 }
2290 ptr++;
2291 }
2292 if (ptr != end) {
2293 // reasonable string, get rid of newline and keep the rest
2294 char* nl = strchr(buf, '\n');
2295 if (nl != NULL) *nl = '\0';
2296 strncpy(cpuinfo, ptr, length);
2297 fclose(fp);
2298 return;
2299 }
2300 }
2301 }
2302 }
2303 fclose(fp);
2304 }
2305 // cpuinfo not found or parsing failed, just print generic string. The entire
2306 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2307 #if defined(AARCH64)
2308 strncpy(cpuinfo, "AArch64", length);
2309 #elif defined(AMD64)
2310 strncpy(cpuinfo, "x86_64", length);
2311 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2312 strncpy(cpuinfo, "ARM", length);
2313 #elif defined(IA32)
2314 strncpy(cpuinfo, "x86_32", length);
2315 #elif defined(IA64)
2316 strncpy(cpuinfo, "IA64", length);
2317 #elif defined(PPC)
2318 strncpy(cpuinfo, "PPC64", length);
2319 #elif defined(S390)
2320 strncpy(cpuinfo, "S390", length);
2321 #elif defined(SPARC)
2322 strncpy(cpuinfo, "sparcv9", length);
2323 #elif defined(ZERO_LIBARCH)
2324 strncpy(cpuinfo, ZERO_LIBARCH, length);
2325 #else
2326 strncpy(cpuinfo, "unknown", length);
2327 #endif
2328 }
2329
2330 static void print_signal_handler(outputStream* st, int sig,
2331 char* buf, size_t buflen);
2332
2333 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2334 st->print_cr("Signal Handlers:");
2335 print_signal_handler(st, SIGSEGV, buf, buflen);
2336 print_signal_handler(st, SIGBUS , buf, buflen);
2337 print_signal_handler(st, SIGFPE , buf, buflen);
2338 print_signal_handler(st, SIGPIPE, buf, buflen);
2339 print_signal_handler(st, SIGXFSZ, buf, buflen);
2340 print_signal_handler(st, SIGILL , buf, buflen);
2341 print_signal_handler(st, SR_signum, buf, buflen);
2342 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2343 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2344 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2345 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2346 #if defined(PPC64)
2347 print_signal_handler(st, SIGTRAP, buf, buflen);
2348 #endif
2349 }
2350
2351 static char saved_jvm_path[MAXPATHLEN] = {0};
2352
2353 // Find the full path to the current module, libjvm.so
2354 void os::jvm_path(char *buf, jint buflen) {
2355 // Error checking.
2356 if (buflen < MAXPATHLEN) {
2357 assert(false, "must use a large-enough buffer");
2358 buf[0] = '\0';
2359 return;
2360 }
2361 // Lazy resolve the path to current module.
2362 if (saved_jvm_path[0] != 0) {
2363 strcpy(buf, saved_jvm_path);
2364 return;
2365 }
2366
2367 char dli_fname[MAXPATHLEN];
2368 bool ret = dll_address_to_library_name(
2369 CAST_FROM_FN_PTR(address, os::jvm_path),
2370 dli_fname, sizeof(dli_fname), NULL);
2371 assert(ret, "cannot locate libjvm");
2372 char *rp = NULL;
2373 if (ret && dli_fname[0] != '\0') {
2374 rp = os::Posix::realpath(dli_fname, buf, buflen);
2375 }
2376 if (rp == NULL) {
2377 return;
2378 }
2379
2380 if (Arguments::sun_java_launcher_is_altjvm()) {
2381 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2382 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2383 // If "/jre/lib/" appears at the right place in the string, then
2384 // assume we are installed in a JDK and we're done. Otherwise, check
2385 // for a JAVA_HOME environment variable and fix up the path so it
2386 // looks like libjvm.so is installed there (append a fake suffix
2387 // hotspot/libjvm.so).
2388 const char *p = buf + strlen(buf) - 1;
2389 for (int count = 0; p > buf && count < 5; ++count) {
2390 for (--p; p > buf && *p != '/'; --p)
2391 /* empty */ ;
2392 }
2393
2394 if (strncmp(p, "/jre/lib/", 9) != 0) {
2395 // Look for JAVA_HOME in the environment.
2396 char* java_home_var = ::getenv("JAVA_HOME");
2397 if (java_home_var != NULL && java_home_var[0] != 0) {
2398 char* jrelib_p;
2399 int len;
2400
2401 // Check the current module name "libjvm.so".
2402 p = strrchr(buf, '/');
2403 if (p == NULL) {
2404 return;
2405 }
2406 assert(strstr(p, "/libjvm") == p, "invalid library name");
2407
2408 rp = os::Posix::realpath(java_home_var, buf, buflen);
2409 if (rp == NULL) {
2410 return;
2411 }
2412
2413 // determine if this is a legacy image or modules image
2414 // modules image doesn't have "jre" subdirectory
2415 len = strlen(buf);
2416 assert(len < buflen, "Ran out of buffer room");
2417 jrelib_p = buf + len;
2418 snprintf(jrelib_p, buflen-len, "/jre/lib");
2419 if (0 != access(buf, F_OK)) {
2420 snprintf(jrelib_p, buflen-len, "/lib");
2421 }
2422
2423 if (0 == access(buf, F_OK)) {
2424 // Use current module name "libjvm.so"
2425 len = strlen(buf);
2426 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2427 } else {
2428 // Go back to path of .so
2429 rp = os::Posix::realpath(dli_fname, buf, buflen);
2430 if (rp == NULL) {
2431 return;
2432 }
2433 }
2434 }
2435 }
2436 }
2437
2438 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2439 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2440 }
2441
2442 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2443 // no prefix required, not even "_"
2444 }
2445
2446 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2447 // no suffix required
2448 }
2449
2450 ////////////////////////////////////////////////////////////////////////////////
2451 // sun.misc.Signal support
2452
2453 static volatile jint sigint_count = 0;
2454
2455 static void UserHandler(int sig, void *siginfo, void *context) {
2456 // 4511530 - sem_post is serialized and handled by the manager thread. When
2457 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2458 // don't want to flood the manager thread with sem_post requests.
2459 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2460 return;
2461 }
2462
2463 // Ctrl-C is pressed during error reporting, likely because the error
2464 // handler fails to abort. Let VM die immediately.
2465 if (sig == SIGINT && VMError::is_error_reported()) {
2466 os::die();
2467 }
2468
2469 os::signal_notify(sig);
2470 }
2471
2472 void* os::user_handler() {
2473 return CAST_FROM_FN_PTR(void*, UserHandler);
2474 }
2475
2476 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2477 struct timespec ts;
2478 // Semaphore's are always associated with CLOCK_REALTIME
2479 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2480 // see unpackTime for discussion on overflow checking
2481 if (sec >= MAX_SECS) {
2482 ts.tv_sec += MAX_SECS;
2483 ts.tv_nsec = 0;
2484 } else {
2485 ts.tv_sec += sec;
2486 ts.tv_nsec += nsec;
2487 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2488 ts.tv_nsec -= NANOSECS_PER_SEC;
2489 ++ts.tv_sec; // note: this must be <= max_secs
2490 }
2491 }
2492
2493 return ts;
2494 }
2495
2496 extern "C" {
2497 typedef void (*sa_handler_t)(int);
2498 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2499 }
2500
2501 void* os::signal(int signal_number, void* handler) {
2502 struct sigaction sigAct, oldSigAct;
2503
2504 sigfillset(&(sigAct.sa_mask));
2505 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2506 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2507
2508 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2509 // -1 means registration failed
2510 return (void *)-1;
2511 }
2512
2513 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2514 }
2515
2516 void os::signal_raise(int signal_number) {
2517 ::raise(signal_number);
2518 }
2519
2520 // The following code is moved from os.cpp for making this
2521 // code platform specific, which it is by its very nature.
2522
2523 // Will be modified when max signal is changed to be dynamic
2524 int os::sigexitnum_pd() {
2525 return NSIG;
2526 }
2527
2528 // a counter for each possible signal value
2529 static volatile jint pending_signals[NSIG+1] = { 0 };
2530
2531 // Linux(POSIX) specific hand shaking semaphore.
2532 static sem_t sig_sem;
2533 static PosixSemaphore sr_semaphore;
2534
2535 void os::signal_init_pd() {
2536 // Initialize signal structures
2537 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2538
2539 // Initialize signal semaphore
2540 ::sem_init(&sig_sem, 0, 0);
2541 }
2542
2543 void os::signal_notify(int sig) {
2544 Atomic::inc(&pending_signals[sig]);
2545 ::sem_post(&sig_sem);
2546 }
2547
2548 static int check_pending_signals(bool wait) {
2549 Atomic::store(0, &sigint_count);
2550 for (;;) {
2551 for (int i = 0; i < NSIG + 1; i++) {
2552 jint n = pending_signals[i];
2553 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2554 return i;
2555 }
2556 }
2557 if (!wait) {
2558 return -1;
2559 }
2560 JavaThread *thread = JavaThread::current();
2561 ThreadBlockInVM tbivm(thread);
2562
2563 bool threadIsSuspended;
2564 do {
2565 thread->set_suspend_equivalent();
2566 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2567 ::sem_wait(&sig_sem);
2568
2569 // were we externally suspended while we were waiting?
2570 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2571 if (threadIsSuspended) {
2572 // The semaphore has been incremented, but while we were waiting
2573 // another thread suspended us. We don't want to continue running
2574 // while suspended because that would surprise the thread that
2575 // suspended us.
2576 ::sem_post(&sig_sem);
2577
2578 thread->java_suspend_self();
2579 }
2580 } while (threadIsSuspended);
2581 }
2582 }
2583
2584 int os::signal_lookup() {
2585 return check_pending_signals(false);
2586 }
2587
2588 int os::signal_wait() {
2589 return check_pending_signals(true);
2590 }
2591
2592 ////////////////////////////////////////////////////////////////////////////////
2593 // Virtual Memory
2594
2595 int os::vm_page_size() {
2596 // Seems redundant as all get out
2597 assert(os::Linux::page_size() != -1, "must call os::init");
2598 return os::Linux::page_size();
2599 }
2600
2601 // Solaris allocates memory by pages.
2602 int os::vm_allocation_granularity() {
2603 assert(os::Linux::page_size() != -1, "must call os::init");
2604 return os::Linux::page_size();
2605 }
2606
2607 // Rationale behind this function:
2608 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2609 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2610 // samples for JITted code. Here we create private executable mapping over the code cache
2611 // and then we can use standard (well, almost, as mapping can change) way to provide
2612 // info for the reporting script by storing timestamp and location of symbol
2613 void linux_wrap_code(char* base, size_t size) {
2614 static volatile jint cnt = 0;
2615
2616 if (!UseOprofile) {
2617 return;
2618 }
2619
2620 char buf[PATH_MAX+1];
2621 int num = Atomic::add(1, &cnt);
2622
2623 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2624 os::get_temp_directory(), os::current_process_id(), num);
2625 unlink(buf);
2626
2627 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2628
2629 if (fd != -1) {
2630 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2631 if (rv != (off_t)-1) {
2632 if (::write(fd, "", 1) == 1) {
2633 mmap(base, size,
2634 PROT_READ|PROT_WRITE|PROT_EXEC,
2635 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2636 }
2637 }
2638 ::close(fd);
2639 unlink(buf);
2640 }
2641 }
2642
2643 static bool recoverable_mmap_error(int err) {
2644 // See if the error is one we can let the caller handle. This
2645 // list of errno values comes from JBS-6843484. I can't find a
2646 // Linux man page that documents this specific set of errno
2647 // values so while this list currently matches Solaris, it may
2648 // change as we gain experience with this failure mode.
2649 switch (err) {
2650 case EBADF:
2651 case EINVAL:
2652 case ENOTSUP:
2653 // let the caller deal with these errors
2654 return true;
2655
2656 default:
2657 // Any remaining errors on this OS can cause our reserved mapping
2658 // to be lost. That can cause confusion where different data
2659 // structures think they have the same memory mapped. The worst
2660 // scenario is if both the VM and a library think they have the
2661 // same memory mapped.
2662 return false;
2663 }
2664 }
2665
2666 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2667 int err) {
2668 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2669 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2670 os::strerror(err), err);
2671 }
2672
2673 static void warn_fail_commit_memory(char* addr, size_t size,
2674 size_t alignment_hint, bool exec,
2675 int err) {
2676 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2677 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2678 alignment_hint, exec, os::strerror(err), err);
2679 }
2680
2681 // NOTE: Linux kernel does not really reserve the pages for us.
2682 // All it does is to check if there are enough free pages
2683 // left at the time of mmap(). This could be a potential
2684 // problem.
2685 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2686 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2687 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2688 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2689 if (res != (uintptr_t) MAP_FAILED) {
2690 if (UseNUMAInterleaving) {
2691 numa_make_global(addr, size);
2692 }
2693 return 0;
2694 }
2695
2696 int err = errno; // save errno from mmap() call above
2697
2698 if (!recoverable_mmap_error(err)) {
2699 warn_fail_commit_memory(addr, size, exec, err);
2700 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2701 }
2702
2703 return err;
2704 }
2705
2706 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2707 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2708 }
2709
2710 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2711 const char* mesg) {
2712 assert(mesg != NULL, "mesg must be specified");
2713 int err = os::Linux::commit_memory_impl(addr, size, exec);
2714 if (err != 0) {
2715 // the caller wants all commit errors to exit with the specified mesg:
2716 warn_fail_commit_memory(addr, size, exec, err);
2717 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2718 }
2719 }
2720
2721 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2722 #ifndef MAP_HUGETLB
2723 #define MAP_HUGETLB 0x40000
2724 #endif
2725
2726 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2727 #ifndef MADV_HUGEPAGE
2728 #define MADV_HUGEPAGE 14
2729 #endif
2730
2731 int os::Linux::commit_memory_impl(char* addr, size_t size,
2732 size_t alignment_hint, bool exec) {
2733 int err = os::Linux::commit_memory_impl(addr, size, exec);
2734 if (err == 0) {
2735 realign_memory(addr, size, alignment_hint);
2736 }
2737 return err;
2738 }
2739
2740 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2741 bool exec) {
2742 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2743 }
2744
2745 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2746 size_t alignment_hint, bool exec,
2747 const char* mesg) {
2748 assert(mesg != NULL, "mesg must be specified");
2749 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2750 if (err != 0) {
2751 // the caller wants all commit errors to exit with the specified mesg:
2752 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2753 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2754 }
2755 }
2756
2757 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2758 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2759 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2760 // be supported or the memory may already be backed by huge pages.
2761 ::madvise(addr, bytes, MADV_HUGEPAGE);
2762 }
2763 }
2764
2765 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2766 // This method works by doing an mmap over an existing mmaping and effectively discarding
2767 // the existing pages. However it won't work for SHM-based large pages that cannot be
2768 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2769 // small pages on top of the SHM segment. This method always works for small pages, so we
2770 // allow that in any case.
2771 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2772 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2773 }
2774 }
2775
2776 void os::numa_make_global(char *addr, size_t bytes) {
2777 Linux::numa_interleave_memory(addr, bytes);
2778 }
2779
2780 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2781 // bind policy to MPOL_PREFERRED for the current thread.
2782 #define USE_MPOL_PREFERRED 0
2783
2784 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2785 // To make NUMA and large pages more robust when both enabled, we need to ease
2786 // the requirements on where the memory should be allocated. MPOL_BIND is the
2787 // default policy and it will force memory to be allocated on the specified
2788 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2789 // the specified node, but will not force it. Using this policy will prevent
2790 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2791 // free large pages.
2792 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2793 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2794 }
2795
2796 bool os::numa_topology_changed() { return false; }
2797
2798 size_t os::numa_get_groups_num() {
2799 // Return just the number of nodes in which it's possible to allocate memory
2800 // (in numa terminology, configured nodes).
2801 return Linux::numa_num_configured_nodes();
2802 }
2803
2804 int os::numa_get_group_id() {
2805 int cpu_id = Linux::sched_getcpu();
2806 if (cpu_id != -1) {
2807 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2808 if (lgrp_id != -1) {
2809 return lgrp_id;
2810 }
2811 }
2812 return 0;
2813 }
2814
2815 int os::Linux::get_existing_num_nodes() {
2816 size_t node;
2817 size_t highest_node_number = Linux::numa_max_node();
2818 int num_nodes = 0;
2819
2820 // Get the total number of nodes in the system including nodes without memory.
2821 for (node = 0; node <= highest_node_number; node++) {
2822 if (isnode_in_existing_nodes(node)) {
2823 num_nodes++;
2824 }
2825 }
2826 return num_nodes;
2827 }
2828
2829 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2830 size_t highest_node_number = Linux::numa_max_node();
2831 size_t i = 0;
2832
2833 // Map all node ids in which is possible to allocate memory. Also nodes are
2834 // not always consecutively available, i.e. available from 0 to the highest
2835 // node number.
2836 for (size_t node = 0; node <= highest_node_number; node++) {
2837 if (Linux::isnode_in_configured_nodes(node)) {
2838 ids[i++] = node;
2839 }
2840 }
2841 return i;
2842 }
2843
2844 bool os::get_page_info(char *start, page_info* info) {
2845 return false;
2846 }
2847
2848 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2849 page_info* page_found) {
2850 return end;
2851 }
2852
2853
2854 int os::Linux::sched_getcpu_syscall(void) {
2855 unsigned int cpu = 0;
2856 int retval = -1;
2857
2858 #if defined(IA32)
2859 #ifndef SYS_getcpu
2860 #define SYS_getcpu 318
2861 #endif
2862 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2863 #elif defined(AMD64)
2864 // Unfortunately we have to bring all these macros here from vsyscall.h
2865 // to be able to compile on old linuxes.
2866 #define __NR_vgetcpu 2
2867 #define VSYSCALL_START (-10UL << 20)
2868 #define VSYSCALL_SIZE 1024
2869 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2870 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2871 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2872 retval = vgetcpu(&cpu, NULL, NULL);
2873 #endif
2874
2875 return (retval == -1) ? retval : cpu;
2876 }
2877
2878 void os::Linux::sched_getcpu_init() {
2879 // sched_getcpu() should be in libc.
2880 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2881 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2882
2883 // If it's not, try a direct syscall.
2884 if (sched_getcpu() == -1) {
2885 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2886 (void*)&sched_getcpu_syscall));
2887 }
2888 }
2889
2890 // Something to do with the numa-aware allocator needs these symbols
2891 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2892 extern "C" JNIEXPORT void numa_error(char *where) { }
2893
2894 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2895 // load symbol from base version instead.
2896 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2897 void *f = dlvsym(handle, name, "libnuma_1.1");
2898 if (f == NULL) {
2899 f = dlsym(handle, name);
2900 }
2901 return f;
2902 }
2903
2904 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2905 // Return NULL if the symbol is not defined in this particular version.
2906 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2907 return dlvsym(handle, name, "libnuma_1.2");
2908 }
2909
2910 bool os::Linux::libnuma_init() {
2911 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
2912 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2913 if (handle != NULL) {
2914 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2915 libnuma_dlsym(handle, "numa_node_to_cpus")));
2916 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2917 libnuma_dlsym(handle, "numa_max_node")));
2918 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2919 libnuma_dlsym(handle, "numa_num_configured_nodes")));
2920 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2921 libnuma_dlsym(handle, "numa_available")));
2922 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2923 libnuma_dlsym(handle, "numa_tonode_memory")));
2924 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2925 libnuma_dlsym(handle, "numa_interleave_memory")));
2926 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2927 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2928 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2929 libnuma_dlsym(handle, "numa_set_bind_policy")));
2930 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2931 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2932 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2933 libnuma_dlsym(handle, "numa_distance")));
2934
2935 if (numa_available() != -1) {
2936 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2937 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2938 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2939 // Create an index -> node mapping, since nodes are not always consecutive
2940 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2941 rebuild_nindex_to_node_map();
2942 // Create a cpu -> node mapping
2943 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2944 rebuild_cpu_to_node_map();
2945 return true;
2946 }
2947 }
2948 }
2949 return false;
2950 }
2951
2952 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2953 // Creating guard page is very expensive. Java thread has HotSpot
2954 // guard pages, only enable glibc guard page for non-Java threads.
2955 // (Remember: compiler thread is a Java thread, too!)
2956 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2957 }
2958
2959 void os::Linux::rebuild_nindex_to_node_map() {
2960 int highest_node_number = Linux::numa_max_node();
2961
2962 nindex_to_node()->clear();
2963 for (int node = 0; node <= highest_node_number; node++) {
2964 if (Linux::isnode_in_existing_nodes(node)) {
2965 nindex_to_node()->append(node);
2966 }
2967 }
2968 }
2969
2970 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2971 // The table is later used in get_node_by_cpu().
2972 void os::Linux::rebuild_cpu_to_node_map() {
2973 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2974 // in libnuma (possible values are starting from 16,
2975 // and continuing up with every other power of 2, but less
2976 // than the maximum number of CPUs supported by kernel), and
2977 // is a subject to change (in libnuma version 2 the requirements
2978 // are more reasonable) we'll just hardcode the number they use
2979 // in the library.
2980 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2981
2982 size_t cpu_num = processor_count();
2983 size_t cpu_map_size = NCPUS / BitsPerCLong;
2984 size_t cpu_map_valid_size =
2985 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2986
2987 cpu_to_node()->clear();
2988 cpu_to_node()->at_grow(cpu_num - 1);
2989
2990 size_t node_num = get_existing_num_nodes();
2991
2992 int distance = 0;
2993 int closest_distance = INT_MAX;
2994 int closest_node = 0;
2995 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2996 for (size_t i = 0; i < node_num; i++) {
2997 // Check if node is configured (not a memory-less node). If it is not, find
2998 // the closest configured node.
2999 if (!isnode_in_configured_nodes(nindex_to_node()->at(i))) {
3000 closest_distance = INT_MAX;
3001 // Check distance from all remaining nodes in the system. Ignore distance
3002 // from itself and from another non-configured node.
3003 for (size_t m = 0; m < node_num; m++) {
3004 if (m != i && isnode_in_configured_nodes(nindex_to_node()->at(m))) {
3005 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3006 // If a closest node is found, update. There is always at least one
3007 // configured node in the system so there is always at least one node
3008 // close.
3009 if (distance != 0 && distance < closest_distance) {
3010 closest_distance = distance;
3011 closest_node = nindex_to_node()->at(m);
3012 }
3013 }
3014 }
3015 } else {
3016 // Current node is already a configured node.
3017 closest_node = nindex_to_node()->at(i);
3018 }
3019
3020 // Get cpus from the original node and map them to the closest node. If node
3021 // is a configured node (not a memory-less node), then original node and
3022 // closest node are the same.
3023 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3024 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3025 if (cpu_map[j] != 0) {
3026 for (size_t k = 0; k < BitsPerCLong; k++) {
3027 if (cpu_map[j] & (1UL << k)) {
3028 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3029 }
3030 }
3031 }
3032 }
3033 }
3034 }
3035 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3036 }
3037
3038 int os::Linux::get_node_by_cpu(int cpu_id) {
3039 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3040 return cpu_to_node()->at(cpu_id);
3041 }
3042 return -1;
3043 }
3044
3045 GrowableArray<int>* os::Linux::_cpu_to_node;
3046 GrowableArray<int>* os::Linux::_nindex_to_node;
3047 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3048 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3049 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3050 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3051 os::Linux::numa_available_func_t os::Linux::_numa_available;
3052 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3053 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3054 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3055 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3056 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3057 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3058 unsigned long* os::Linux::_numa_all_nodes;
3059 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3060 struct bitmask* os::Linux::_numa_nodes_ptr;
3061
3062 bool os::pd_uncommit_memory(char* addr, size_t size) {
3063 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3064 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3065 return res != (uintptr_t) MAP_FAILED;
3066 }
3067
3068 // If there is no page mapped/committed, top (bottom + size) is returned
3069 static address get_stack_mapped_bottom(address bottom,
3070 size_t size,
3071 bool committed_only /* must have backing pages */) {
3072 // address used to test if the page is mapped/committed
3073 address test_addr = bottom + size;
3074 size_t page_sz = os::vm_page_size();
3075 unsigned pages = size / page_sz;
3076
3077 unsigned char vec[1];
3078 unsigned imin = 1, imax = pages + 1, imid;
3079 int mincore_return_value = 0;
3080
3081 assert(imin <= imax, "Unexpected page size");
3082
3083 while (imin < imax) {
3084 imid = (imax + imin) / 2;
3085 test_addr = bottom + (imid * page_sz);
3086
3087 // Use a trick with mincore to check whether the page is mapped or not.
3088 // mincore sets vec to 1 if page resides in memory and to 0 if page
3089 // is swapped output but if page we are asking for is unmapped
3090 // it returns -1,ENOMEM
3091 mincore_return_value = mincore(test_addr, page_sz, vec);
3092
3093 if (mincore_return_value == -1 || (committed_only && (vec[0] & 0x01) == 0)) {
3094 // Page is not mapped/committed go up
3095 // to find first mapped/committed page
3096 if ((mincore_return_value == -1 && errno != EAGAIN)
3097 || (committed_only && (vec[0] & 0x01) == 0)) {
3098 assert(mincore_return_value != -1 || errno == ENOMEM, "Unexpected mincore errno");
3099
3100 imin = imid + 1;
3101 }
3102 } else {
3103 // mapped/committed, go down
3104 imax= imid;
3105 }
3106 }
3107
3108 // Adjust stack bottom one page up if last checked page is not mapped/committed
3109 if (mincore_return_value == -1 || (committed_only && (vec[0] & 0x01) == 0)) {
3110 assert(mincore_return_value != -1 || (errno != EAGAIN && errno != ENOMEM),
3111 "Should not get to here");
3112
3113 test_addr = test_addr + page_sz;
3114 }
3115
3116 return test_addr;
3117 }
3118
3119 // Linux uses a growable mapping for the stack, and if the mapping for
3120 // the stack guard pages is not removed when we detach a thread the
3121 // stack cannot grow beyond the pages where the stack guard was
3122 // mapped. If at some point later in the process the stack expands to
3123 // that point, the Linux kernel cannot expand the stack any further
3124 // because the guard pages are in the way, and a segfault occurs.
3125 //
3126 // However, it's essential not to split the stack region by unmapping
3127 // a region (leaving a hole) that's already part of the stack mapping,
3128 // so if the stack mapping has already grown beyond the guard pages at
3129 // the time we create them, we have to truncate the stack mapping.
3130 // So, we need to know the extent of the stack mapping when
3131 // create_stack_guard_pages() is called.
3132
3133 // We only need this for stacks that are growable: at the time of
3134 // writing thread stacks don't use growable mappings (i.e. those
3135 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3136 // only applies to the main thread.
3137
3138 // If the (growable) stack mapping already extends beyond the point
3139 // where we're going to put our guard pages, truncate the mapping at
3140 // that point by munmap()ping it. This ensures that when we later
3141 // munmap() the guard pages we don't leave a hole in the stack
3142 // mapping. This only affects the main/initial thread
3143
3144 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3145 if (os::Linux::is_initial_thread()) {
3146 // As we manually grow stack up to bottom inside create_attached_thread(),
3147 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3148 // we don't need to do anything special.
3149 // Check it first, before calling heavy function.
3150 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3151 unsigned char vec[1];
3152
3153 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3154 // Fallback to slow path on all errors, including EAGAIN
3155 stack_extent = (uintptr_t) get_stack_mapped_bottom(os::Linux::initial_thread_stack_bottom(),
3156 (size_t)addr - stack_extent,
3157 false /* committed_only */);
3158 }
3159
3160 if (stack_extent < (uintptr_t)addr) {
3161 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3162 }
3163 }
3164
3165 return os::commit_memory(addr, size, !ExecMem);
3166 }
3167
3168 // If this is a growable mapping, remove the guard pages entirely by
3169 // munmap()ping them. If not, just call uncommit_memory(). This only
3170 // affects the main/initial thread, but guard against future OS changes
3171 // It's safe to always unmap guard pages for initial thread because we
3172 // always place it right after end of the mapped region
3173
3174 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3175 uintptr_t stack_extent, stack_base;
3176
3177 if (os::Linux::is_initial_thread()) {
3178 return ::munmap(addr, size) == 0;
3179 }
3180
3181 return os::uncommit_memory(addr, size);
3182 }
3183
3184 size_t os::pd_committed_stack_size(address bottom, size_t size) {
3185 address bot = get_stack_mapped_bottom(bottom, size, true /* committed_only */);
3186 return size_t(bottom + size - bot);
3187 }
3188
3189 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3190 // at 'requested_addr'. If there are existing memory mappings at the same
3191 // location, however, they will be overwritten. If 'fixed' is false,
3192 // 'requested_addr' is only treated as a hint, the return value may or
3193 // may not start from the requested address. Unlike Linux mmap(), this
3194 // function returns NULL to indicate failure.
3195 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3196 char * addr;
3197 int flags;
3198
3199 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3200 if (fixed) {
3201 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3202 flags |= MAP_FIXED;
3203 }
3204
3205 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3206 // touch an uncommitted page. Otherwise, the read/write might
3207 // succeed if we have enough swap space to back the physical page.
3208 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3209 flags, -1, 0);
3210
3211 return addr == MAP_FAILED ? NULL : addr;
3212 }
3213
3214 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3215 // (req_addr != NULL) or with a given alignment.
3216 // - bytes shall be a multiple of alignment.
3217 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3218 // - alignment sets the alignment at which memory shall be allocated.
3219 // It must be a multiple of allocation granularity.
3220 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3221 // req_addr or NULL.
3222 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3223
3224 size_t extra_size = bytes;
3225 if (req_addr == NULL && alignment > 0) {
3226 extra_size += alignment;
3227 }
3228
3229 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3230 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3231 -1, 0);
3232 if (start == MAP_FAILED) {
3233 start = NULL;
3234 } else {
3235 if (req_addr != NULL) {
3236 if (start != req_addr) {
3237 ::munmap(start, extra_size);
3238 start = NULL;
3239 }
3240 } else {
3241 char* const start_aligned = align_up(start, alignment);
3242 char* const end_aligned = start_aligned + bytes;
3243 char* const end = start + extra_size;
3244 if (start_aligned > start) {
3245 ::munmap(start, start_aligned - start);
3246 }
3247 if (end_aligned < end) {
3248 ::munmap(end_aligned, end - end_aligned);
3249 }
3250 start = start_aligned;
3251 }
3252 }
3253 return start;
3254 }
3255
3256 static int anon_munmap(char * addr, size_t size) {
3257 return ::munmap(addr, size) == 0;
3258 }
3259
3260 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3261 size_t alignment_hint) {
3262 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3263 }
3264
3265 bool os::pd_release_memory(char* addr, size_t size) {
3266 return anon_munmap(addr, size);
3267 }
3268
3269 static bool linux_mprotect(char* addr, size_t size, int prot) {
3270 // Linux wants the mprotect address argument to be page aligned.
3271 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3272
3273 // According to SUSv3, mprotect() should only be used with mappings
3274 // established by mmap(), and mmap() always maps whole pages. Unaligned
3275 // 'addr' likely indicates problem in the VM (e.g. trying to change
3276 // protection of malloc'ed or statically allocated memory). Check the
3277 // caller if you hit this assert.
3278 assert(addr == bottom, "sanity check");
3279
3280 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3281 return ::mprotect(bottom, size, prot) == 0;
3282 }
3283
3284 // Set protections specified
3285 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3286 bool is_committed) {
3287 unsigned int p = 0;
3288 switch (prot) {
3289 case MEM_PROT_NONE: p = PROT_NONE; break;
3290 case MEM_PROT_READ: p = PROT_READ; break;
3291 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3292 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3293 default:
3294 ShouldNotReachHere();
3295 }
3296 // is_committed is unused.
3297 return linux_mprotect(addr, bytes, p);
3298 }
3299
3300 bool os::guard_memory(char* addr, size_t size) {
3301 return linux_mprotect(addr, size, PROT_NONE);
3302 }
3303
3304 bool os::unguard_memory(char* addr, size_t size) {
3305 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3306 }
3307
3308 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3309 size_t page_size) {
3310 bool result = false;
3311 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3312 MAP_ANONYMOUS|MAP_PRIVATE,
3313 -1, 0);
3314 if (p != MAP_FAILED) {
3315 void *aligned_p = align_up(p, page_size);
3316
3317 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3318
3319 munmap(p, page_size * 2);
3320 }
3321
3322 if (warn && !result) {
3323 warning("TransparentHugePages is not supported by the operating system.");
3324 }
3325
3326 return result;
3327 }
3328
3329 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3330 bool result = false;
3331 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3332 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3333 -1, 0);
3334
3335 if (p != MAP_FAILED) {
3336 // We don't know if this really is a huge page or not.
3337 FILE *fp = fopen("/proc/self/maps", "r");
3338 if (fp) {
3339 while (!feof(fp)) {
3340 char chars[257];
3341 long x = 0;
3342 if (fgets(chars, sizeof(chars), fp)) {
3343 if (sscanf(chars, "%lx-%*x", &x) == 1
3344 && x == (long)p) {
3345 if (strstr (chars, "hugepage")) {
3346 result = true;
3347 break;
3348 }
3349 }
3350 }
3351 }
3352 fclose(fp);
3353 }
3354 munmap(p, page_size);
3355 }
3356
3357 if (warn && !result) {
3358 warning("HugeTLBFS is not supported by the operating system.");
3359 }
3360
3361 return result;
3362 }
3363
3364 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3365 //
3366 // From the coredump_filter documentation:
3367 //
3368 // - (bit 0) anonymous private memory
3369 // - (bit 1) anonymous shared memory
3370 // - (bit 2) file-backed private memory
3371 // - (bit 3) file-backed shared memory
3372 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3373 // effective only if the bit 2 is cleared)
3374 // - (bit 5) hugetlb private memory
3375 // - (bit 6) hugetlb shared memory
3376 //
3377 static void set_coredump_filter(void) {
3378 FILE *f;
3379 long cdm;
3380
3381 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3382 return;
3383 }
3384
3385 if (fscanf(f, "%lx", &cdm) != 1) {
3386 fclose(f);
3387 return;
3388 }
3389
3390 rewind(f);
3391
3392 if ((cdm & LARGEPAGES_BIT) == 0) {
3393 cdm |= LARGEPAGES_BIT;
3394 fprintf(f, "%#lx", cdm);
3395 }
3396
3397 fclose(f);
3398 }
3399
3400 // Large page support
3401
3402 static size_t _large_page_size = 0;
3403
3404 size_t os::Linux::find_large_page_size() {
3405 size_t large_page_size = 0;
3406
3407 // large_page_size on Linux is used to round up heap size. x86 uses either
3408 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3409 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3410 // page as large as 256M.
3411 //
3412 // Here we try to figure out page size by parsing /proc/meminfo and looking
3413 // for a line with the following format:
3414 // Hugepagesize: 2048 kB
3415 //
3416 // If we can't determine the value (e.g. /proc is not mounted, or the text
3417 // format has been changed), we'll use the largest page size supported by
3418 // the processor.
3419
3420 #ifndef ZERO
3421 large_page_size =
3422 AARCH64_ONLY(2 * M)
3423 AMD64_ONLY(2 * M)
3424 ARM32_ONLY(2 * M)
3425 IA32_ONLY(4 * M)
3426 IA64_ONLY(256 * M)
3427 PPC_ONLY(4 * M)
3428 S390_ONLY(1 * M)
3429 SPARC_ONLY(4 * M);
3430 #endif // ZERO
3431
3432 FILE *fp = fopen("/proc/meminfo", "r");
3433 if (fp) {
3434 while (!feof(fp)) {
3435 int x = 0;
3436 char buf[16];
3437 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3438 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3439 large_page_size = x * K;
3440 break;
3441 }
3442 } else {
3443 // skip to next line
3444 for (;;) {
3445 int ch = fgetc(fp);
3446 if (ch == EOF || ch == (int)'\n') break;
3447 }
3448 }
3449 }
3450 fclose(fp);
3451 }
3452
3453 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3454 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3455 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3456 proper_unit_for_byte_size(large_page_size));
3457 }
3458
3459 return large_page_size;
3460 }
3461
3462 size_t os::Linux::setup_large_page_size() {
3463 _large_page_size = Linux::find_large_page_size();
3464 const size_t default_page_size = (size_t)Linux::page_size();
3465 if (_large_page_size > default_page_size) {
3466 _page_sizes[0] = _large_page_size;
3467 _page_sizes[1] = default_page_size;
3468 _page_sizes[2] = 0;
3469 }
3470
3471 return _large_page_size;
3472 }
3473
3474 bool os::Linux::setup_large_page_type(size_t page_size) {
3475 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3476 FLAG_IS_DEFAULT(UseSHM) &&
3477 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3478
3479 // The type of large pages has not been specified by the user.
3480
3481 // Try UseHugeTLBFS and then UseSHM.
3482 UseHugeTLBFS = UseSHM = true;
3483
3484 // Don't try UseTransparentHugePages since there are known
3485 // performance issues with it turned on. This might change in the future.
3486 UseTransparentHugePages = false;
3487 }
3488
3489 if (UseTransparentHugePages) {
3490 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3491 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3492 UseHugeTLBFS = false;
3493 UseSHM = false;
3494 return true;
3495 }
3496 UseTransparentHugePages = false;
3497 }
3498
3499 if (UseHugeTLBFS) {
3500 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3501 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3502 UseSHM = false;
3503 return true;
3504 }
3505 UseHugeTLBFS = false;
3506 }
3507
3508 return UseSHM;
3509 }
3510
3511 void os::large_page_init() {
3512 if (!UseLargePages &&
3513 !UseTransparentHugePages &&
3514 !UseHugeTLBFS &&
3515 !UseSHM) {
3516 // Not using large pages.
3517 return;
3518 }
3519
3520 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3521 // The user explicitly turned off large pages.
3522 // Ignore the rest of the large pages flags.
3523 UseTransparentHugePages = false;
3524 UseHugeTLBFS = false;
3525 UseSHM = false;
3526 return;
3527 }
3528
3529 size_t large_page_size = Linux::setup_large_page_size();
3530 UseLargePages = Linux::setup_large_page_type(large_page_size);
3531
3532 set_coredump_filter();
3533 }
3534
3535 #ifndef SHM_HUGETLB
3536 #define SHM_HUGETLB 04000
3537 #endif
3538
3539 #define shm_warning_format(format, ...) \
3540 do { \
3541 if (UseLargePages && \
3542 (!FLAG_IS_DEFAULT(UseLargePages) || \
3543 !FLAG_IS_DEFAULT(UseSHM) || \
3544 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3545 warning(format, __VA_ARGS__); \
3546 } \
3547 } while (0)
3548
3549 #define shm_warning(str) shm_warning_format("%s", str)
3550
3551 #define shm_warning_with_errno(str) \
3552 do { \
3553 int err = errno; \
3554 shm_warning_format(str " (error = %d)", err); \
3555 } while (0)
3556
3557 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3558 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3559
3560 if (!is_aligned(alignment, SHMLBA)) {
3561 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3562 return NULL;
3563 }
3564
3565 // To ensure that we get 'alignment' aligned memory from shmat,
3566 // we pre-reserve aligned virtual memory and then attach to that.
3567
3568 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3569 if (pre_reserved_addr == NULL) {
3570 // Couldn't pre-reserve aligned memory.
3571 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3572 return NULL;
3573 }
3574
3575 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3576 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3577
3578 if ((intptr_t)addr == -1) {
3579 int err = errno;
3580 shm_warning_with_errno("Failed to attach shared memory.");
3581
3582 assert(err != EACCES, "Unexpected error");
3583 assert(err != EIDRM, "Unexpected error");
3584 assert(err != EINVAL, "Unexpected error");
3585
3586 // Since we don't know if the kernel unmapped the pre-reserved memory area
3587 // we can't unmap it, since that would potentially unmap memory that was
3588 // mapped from other threads.
3589 return NULL;
3590 }
3591
3592 return addr;
3593 }
3594
3595 static char* shmat_at_address(int shmid, char* req_addr) {
3596 if (!is_aligned(req_addr, SHMLBA)) {
3597 assert(false, "Requested address needs to be SHMLBA aligned");
3598 return NULL;
3599 }
3600
3601 char* addr = (char*)shmat(shmid, req_addr, 0);
3602
3603 if ((intptr_t)addr == -1) {
3604 shm_warning_with_errno("Failed to attach shared memory.");
3605 return NULL;
3606 }
3607
3608 return addr;
3609 }
3610
3611 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3612 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3613 if (req_addr != NULL) {
3614 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3615 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3616 return shmat_at_address(shmid, req_addr);
3617 }
3618
3619 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3620 // return large page size aligned memory addresses when req_addr == NULL.
3621 // However, if the alignment is larger than the large page size, we have
3622 // to manually ensure that the memory returned is 'alignment' aligned.
3623 if (alignment > os::large_page_size()) {
3624 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3625 return shmat_with_alignment(shmid, bytes, alignment);
3626 } else {
3627 return shmat_at_address(shmid, NULL);
3628 }
3629 }
3630
3631 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3632 char* req_addr, bool exec) {
3633 // "exec" is passed in but not used. Creating the shared image for
3634 // the code cache doesn't have an SHM_X executable permission to check.
3635 assert(UseLargePages && UseSHM, "only for SHM large pages");
3636 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3637 assert(is_aligned(req_addr, alignment), "Unaligned address");
3638
3639 if (!is_aligned(bytes, os::large_page_size())) {
3640 return NULL; // Fallback to small pages.
3641 }
3642
3643 // Create a large shared memory region to attach to based on size.
3644 // Currently, size is the total size of the heap.
3645 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3646 if (shmid == -1) {
3647 // Possible reasons for shmget failure:
3648 // 1. shmmax is too small for Java heap.
3649 // > check shmmax value: cat /proc/sys/kernel/shmmax
3650 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3651 // 2. not enough large page memory.
3652 // > check available large pages: cat /proc/meminfo
3653 // > increase amount of large pages:
3654 // echo new_value > /proc/sys/vm/nr_hugepages
3655 // Note 1: different Linux may use different name for this property,
3656 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3657 // Note 2: it's possible there's enough physical memory available but
3658 // they are so fragmented after a long run that they can't
3659 // coalesce into large pages. Try to reserve large pages when
3660 // the system is still "fresh".
3661 shm_warning_with_errno("Failed to reserve shared memory.");
3662 return NULL;
3663 }
3664
3665 // Attach to the region.
3666 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3667
3668 // Remove shmid. If shmat() is successful, the actual shared memory segment
3669 // will be deleted when it's detached by shmdt() or when the process
3670 // terminates. If shmat() is not successful this will remove the shared
3671 // segment immediately.
3672 shmctl(shmid, IPC_RMID, NULL);
3673
3674 return addr;
3675 }
3676
3677 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3678 int error) {
3679 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3680
3681 bool warn_on_failure = UseLargePages &&
3682 (!FLAG_IS_DEFAULT(UseLargePages) ||
3683 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3684 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3685
3686 if (warn_on_failure) {
3687 char msg[128];
3688 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3689 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3690 warning("%s", msg);
3691 }
3692 }
3693
3694 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3695 char* req_addr,
3696 bool exec) {
3697 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3698 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3699 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3700
3701 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3702 char* addr = (char*)::mmap(req_addr, bytes, prot,
3703 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3704 -1, 0);
3705
3706 if (addr == MAP_FAILED) {
3707 warn_on_large_pages_failure(req_addr, bytes, errno);
3708 return NULL;
3709 }
3710
3711 assert(is_aligned(addr, os::large_page_size()), "Must be");
3712
3713 return addr;
3714 }
3715
3716 // Reserve memory using mmap(MAP_HUGETLB).
3717 // - bytes shall be a multiple of alignment.
3718 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3719 // - alignment sets the alignment at which memory shall be allocated.
3720 // It must be a multiple of allocation granularity.
3721 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3722 // req_addr or NULL.
3723 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3724 size_t alignment,
3725 char* req_addr,
3726 bool exec) {
3727 size_t large_page_size = os::large_page_size();
3728 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3729
3730 assert(is_aligned(req_addr, alignment), "Must be");
3731 assert(is_aligned(bytes, alignment), "Must be");
3732
3733 // First reserve - but not commit - the address range in small pages.
3734 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3735
3736 if (start == NULL) {
3737 return NULL;
3738 }
3739
3740 assert(is_aligned(start, alignment), "Must be");
3741
3742 char* end = start + bytes;
3743
3744 // Find the regions of the allocated chunk that can be promoted to large pages.
3745 char* lp_start = align_up(start, large_page_size);
3746 char* lp_end = align_down(end, large_page_size);
3747
3748 size_t lp_bytes = lp_end - lp_start;
3749
3750 assert(is_aligned(lp_bytes, large_page_size), "Must be");
3751
3752 if (lp_bytes == 0) {
3753 // The mapped region doesn't even span the start and the end of a large page.
3754 // Fall back to allocate a non-special area.
3755 ::munmap(start, end - start);
3756 return NULL;
3757 }
3758
3759 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3760
3761 void* result;
3762
3763 // Commit small-paged leading area.
3764 if (start != lp_start) {
3765 result = ::mmap(start, lp_start - start, prot,
3766 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3767 -1, 0);
3768 if (result == MAP_FAILED) {
3769 ::munmap(lp_start, end - lp_start);
3770 return NULL;
3771 }
3772 }
3773
3774 // Commit large-paged area.
3775 result = ::mmap(lp_start, lp_bytes, prot,
3776 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3777 -1, 0);
3778 if (result == MAP_FAILED) {
3779 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3780 // If the mmap above fails, the large pages region will be unmapped and we
3781 // have regions before and after with small pages. Release these regions.
3782 //
3783 // | mapped | unmapped | mapped |
3784 // ^ ^ ^ ^
3785 // start lp_start lp_end end
3786 //
3787 ::munmap(start, lp_start - start);
3788 ::munmap(lp_end, end - lp_end);
3789 return NULL;
3790 }
3791
3792 // Commit small-paged trailing area.
3793 if (lp_end != end) {
3794 result = ::mmap(lp_end, end - lp_end, prot,
3795 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3796 -1, 0);
3797 if (result == MAP_FAILED) {
3798 ::munmap(start, lp_end - start);
3799 return NULL;
3800 }
3801 }
3802
3803 return start;
3804 }
3805
3806 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3807 size_t alignment,
3808 char* req_addr,
3809 bool exec) {
3810 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3811 assert(is_aligned(req_addr, alignment), "Must be");
3812 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3813 assert(is_power_of_2(os::large_page_size()), "Must be");
3814 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3815
3816 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3817 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3818 } else {
3819 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3820 }
3821 }
3822
3823 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3824 char* req_addr, bool exec) {
3825 assert(UseLargePages, "only for large pages");
3826
3827 char* addr;
3828 if (UseSHM) {
3829 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3830 } else {
3831 assert(UseHugeTLBFS, "must be");
3832 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3833 }
3834
3835 if (addr != NULL) {
3836 if (UseNUMAInterleaving) {
3837 numa_make_global(addr, bytes);
3838 }
3839
3840 // The memory is committed
3841 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3842 }
3843
3844 return addr;
3845 }
3846
3847 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3848 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3849 return shmdt(base) == 0;
3850 }
3851
3852 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3853 return pd_release_memory(base, bytes);
3854 }
3855
3856 bool os::release_memory_special(char* base, size_t bytes) {
3857 bool res;
3858 if (MemTracker::tracking_level() > NMT_minimal) {
3859 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3860 res = os::Linux::release_memory_special_impl(base, bytes);
3861 if (res) {
3862 tkr.record((address)base, bytes);
3863 }
3864
3865 } else {
3866 res = os::Linux::release_memory_special_impl(base, bytes);
3867 }
3868 return res;
3869 }
3870
3871 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3872 assert(UseLargePages, "only for large pages");
3873 bool res;
3874
3875 if (UseSHM) {
3876 res = os::Linux::release_memory_special_shm(base, bytes);
3877 } else {
3878 assert(UseHugeTLBFS, "must be");
3879 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3880 }
3881 return res;
3882 }
3883
3884 size_t os::large_page_size() {
3885 return _large_page_size;
3886 }
3887
3888 // With SysV SHM the entire memory region must be allocated as shared
3889 // memory.
3890 // HugeTLBFS allows application to commit large page memory on demand.
3891 // However, when committing memory with HugeTLBFS fails, the region
3892 // that was supposed to be committed will lose the old reservation
3893 // and allow other threads to steal that memory region. Because of this
3894 // behavior we can't commit HugeTLBFS memory.
3895 bool os::can_commit_large_page_memory() {
3896 return UseTransparentHugePages;
3897 }
3898
3899 bool os::can_execute_large_page_memory() {
3900 return UseTransparentHugePages || UseHugeTLBFS;
3901 }
3902
3903 // Reserve memory at an arbitrary address, only if that area is
3904 // available (and not reserved for something else).
3905
3906 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3907 const int max_tries = 10;
3908 char* base[max_tries];
3909 size_t size[max_tries];
3910 const size_t gap = 0x000000;
3911
3912 // Assert only that the size is a multiple of the page size, since
3913 // that's all that mmap requires, and since that's all we really know
3914 // about at this low abstraction level. If we need higher alignment,
3915 // we can either pass an alignment to this method or verify alignment
3916 // in one of the methods further up the call chain. See bug 5044738.
3917 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3918
3919 // Repeatedly allocate blocks until the block is allocated at the
3920 // right spot.
3921
3922 // Linux mmap allows caller to pass an address as hint; give it a try first,
3923 // if kernel honors the hint then we can return immediately.
3924 char * addr = anon_mmap(requested_addr, bytes, false);
3925 if (addr == requested_addr) {
3926 return requested_addr;
3927 }
3928
3929 if (addr != NULL) {
3930 // mmap() is successful but it fails to reserve at the requested address
3931 anon_munmap(addr, bytes);
3932 }
3933
3934 int i;
3935 for (i = 0; i < max_tries; ++i) {
3936 base[i] = reserve_memory(bytes);
3937
3938 if (base[i] != NULL) {
3939 // Is this the block we wanted?
3940 if (base[i] == requested_addr) {
3941 size[i] = bytes;
3942 break;
3943 }
3944
3945 // Does this overlap the block we wanted? Give back the overlapped
3946 // parts and try again.
3947
3948 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3949 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3950 unmap_memory(base[i], top_overlap);
3951 base[i] += top_overlap;
3952 size[i] = bytes - top_overlap;
3953 } else {
3954 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3955 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3956 unmap_memory(requested_addr, bottom_overlap);
3957 size[i] = bytes - bottom_overlap;
3958 } else {
3959 size[i] = bytes;
3960 }
3961 }
3962 }
3963 }
3964
3965 // Give back the unused reserved pieces.
3966
3967 for (int j = 0; j < i; ++j) {
3968 if (base[j] != NULL) {
3969 unmap_memory(base[j], size[j]);
3970 }
3971 }
3972
3973 if (i < max_tries) {
3974 return requested_addr;
3975 } else {
3976 return NULL;
3977 }
3978 }
3979
3980 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3981 return ::read(fd, buf, nBytes);
3982 }
3983
3984 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3985 return ::pread(fd, buf, nBytes, offset);
3986 }
3987
3988 // Short sleep, direct OS call.
3989 //
3990 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3991 // sched_yield(2) will actually give up the CPU:
3992 //
3993 // * Alone on this pariticular CPU, keeps running.
3994 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3995 // (pre 2.6.39).
3996 //
3997 // So calling this with 0 is an alternative.
3998 //
3999 void os::naked_short_sleep(jlong ms) {
4000 struct timespec req;
4001
4002 assert(ms < 1000, "Un-interruptable sleep, short time use only");
4003 req.tv_sec = 0;
4004 if (ms > 0) {
4005 req.tv_nsec = (ms % 1000) * 1000000;
4006 } else {
4007 req.tv_nsec = 1;
4008 }
4009
4010 nanosleep(&req, NULL);
4011
4012 return;
4013 }
4014
4015 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4016 void os::infinite_sleep() {
4017 while (true) { // sleep forever ...
4018 ::sleep(100); // ... 100 seconds at a time
4019 }
4020 }
4021
4022 // Used to convert frequent JVM_Yield() to nops
4023 bool os::dont_yield() {
4024 return DontYieldALot;
4025 }
4026
4027 void os::naked_yield() {
4028 sched_yield();
4029 }
4030
4031 ////////////////////////////////////////////////////////////////////////////////
4032 // thread priority support
4033
4034 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4035 // only supports dynamic priority, static priority must be zero. For real-time
4036 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4037 // However, for large multi-threaded applications, SCHED_RR is not only slower
4038 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4039 // of 5 runs - Sep 2005).
4040 //
4041 // The following code actually changes the niceness of kernel-thread/LWP. It
4042 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4043 // not the entire user process, and user level threads are 1:1 mapped to kernel
4044 // threads. It has always been the case, but could change in the future. For
4045 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4046 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4047
4048 int os::java_to_os_priority[CriticalPriority + 1] = {
4049 19, // 0 Entry should never be used
4050
4051 4, // 1 MinPriority
4052 3, // 2
4053 2, // 3
4054
4055 1, // 4
4056 0, // 5 NormPriority
4057 -1, // 6
4058
4059 -2, // 7
4060 -3, // 8
4061 -4, // 9 NearMaxPriority
4062
4063 -5, // 10 MaxPriority
4064
4065 -5 // 11 CriticalPriority
4066 };
4067
4068 static int prio_init() {
4069 if (ThreadPriorityPolicy == 1) {
4070 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4071 // if effective uid is not root. Perhaps, a more elegant way of doing
4072 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4073 if (geteuid() != 0) {
4074 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4075 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4076 }
4077 ThreadPriorityPolicy = 0;
4078 }
4079 }
4080 if (UseCriticalJavaThreadPriority) {
4081 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4082 }
4083 return 0;
4084 }
4085
4086 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4087 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4088
4089 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4090 return (ret == 0) ? OS_OK : OS_ERR;
4091 }
4092
4093 OSReturn os::get_native_priority(const Thread* const thread,
4094 int *priority_ptr) {
4095 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4096 *priority_ptr = java_to_os_priority[NormPriority];
4097 return OS_OK;
4098 }
4099
4100 errno = 0;
4101 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4102 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4103 }
4104
4105 // Hint to the underlying OS that a task switch would not be good.
4106 // Void return because it's a hint and can fail.
4107 void os::hint_no_preempt() {}
4108
4109 ////////////////////////////////////////////////////////////////////////////////
4110 // suspend/resume support
4111
4112 // The low-level signal-based suspend/resume support is a remnant from the
4113 // old VM-suspension that used to be for java-suspension, safepoints etc,
4114 // within hotspot. Currently used by JFR's OSThreadSampler
4115 //
4116 // The remaining code is greatly simplified from the more general suspension
4117 // code that used to be used.
4118 //
4119 // The protocol is quite simple:
4120 // - suspend:
4121 // - sends a signal to the target thread
4122 // - polls the suspend state of the osthread using a yield loop
4123 // - target thread signal handler (SR_handler) sets suspend state
4124 // and blocks in sigsuspend until continued
4125 // - resume:
4126 // - sets target osthread state to continue
4127 // - sends signal to end the sigsuspend loop in the SR_handler
4128 //
4129 // Note that the SR_lock plays no role in this suspend/resume protocol,
4130 // but is checked for NULL in SR_handler as a thread termination indicator.
4131 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4132 //
4133 // Note that resume_clear_context() and suspend_save_context() are needed
4134 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4135 // which in part is used by:
4136 // - Forte Analyzer: AsyncGetCallTrace()
4137 // - StackBanging: get_frame_at_stack_banging_point()
4138
4139 static void resume_clear_context(OSThread *osthread) {
4140 osthread->set_ucontext(NULL);
4141 osthread->set_siginfo(NULL);
4142 }
4143
4144 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4145 ucontext_t* context) {
4146 osthread->set_ucontext(context);
4147 osthread->set_siginfo(siginfo);
4148 }
4149
4150 // Handler function invoked when a thread's execution is suspended or
4151 // resumed. We have to be careful that only async-safe functions are
4152 // called here (Note: most pthread functions are not async safe and
4153 // should be avoided.)
4154 //
4155 // Note: sigwait() is a more natural fit than sigsuspend() from an
4156 // interface point of view, but sigwait() prevents the signal hander
4157 // from being run. libpthread would get very confused by not having
4158 // its signal handlers run and prevents sigwait()'s use with the
4159 // mutex granting granting signal.
4160 //
4161 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4162 //
4163 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4164 // Save and restore errno to avoid confusing native code with EINTR
4165 // after sigsuspend.
4166 int old_errno = errno;
4167
4168 Thread* thread = Thread::current_or_null_safe();
4169 assert(thread != NULL, "Missing current thread in SR_handler");
4170
4171 // On some systems we have seen signal delivery get "stuck" until the signal
4172 // mask is changed as part of thread termination. Check that the current thread
4173 // has not already terminated (via SR_lock()) - else the following assertion
4174 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4175 // destructor has completed.
4176
4177 if (thread->SR_lock() == NULL) {
4178 return;
4179 }
4180
4181 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4182
4183 OSThread* osthread = thread->osthread();
4184
4185 os::SuspendResume::State current = osthread->sr.state();
4186 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4187 suspend_save_context(osthread, siginfo, context);
4188
4189 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4190 os::SuspendResume::State state = osthread->sr.suspended();
4191 if (state == os::SuspendResume::SR_SUSPENDED) {
4192 sigset_t suspend_set; // signals for sigsuspend()
4193 sigemptyset(&suspend_set);
4194 // get current set of blocked signals and unblock resume signal
4195 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4196 sigdelset(&suspend_set, SR_signum);
4197
4198 sr_semaphore.signal();
4199 // wait here until we are resumed
4200 while (1) {
4201 sigsuspend(&suspend_set);
4202
4203 os::SuspendResume::State result = osthread->sr.running();
4204 if (result == os::SuspendResume::SR_RUNNING) {
4205 sr_semaphore.signal();
4206 break;
4207 }
4208 }
4209
4210 } else if (state == os::SuspendResume::SR_RUNNING) {
4211 // request was cancelled, continue
4212 } else {
4213 ShouldNotReachHere();
4214 }
4215
4216 resume_clear_context(osthread);
4217 } else if (current == os::SuspendResume::SR_RUNNING) {
4218 // request was cancelled, continue
4219 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4220 // ignore
4221 } else {
4222 // ignore
4223 }
4224
4225 errno = old_errno;
4226 }
4227
4228 static int SR_initialize() {
4229 struct sigaction act;
4230 char *s;
4231
4232 // Get signal number to use for suspend/resume
4233 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4234 int sig = ::strtol(s, 0, 10);
4235 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4236 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4237 SR_signum = sig;
4238 } else {
4239 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4240 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4241 }
4242 }
4243
4244 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4245 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4246
4247 sigemptyset(&SR_sigset);
4248 sigaddset(&SR_sigset, SR_signum);
4249
4250 // Set up signal handler for suspend/resume
4251 act.sa_flags = SA_RESTART|SA_SIGINFO;
4252 act.sa_handler = (void (*)(int)) SR_handler;
4253
4254 // SR_signum is blocked by default.
4255 // 4528190 - We also need to block pthread restart signal (32 on all
4256 // supported Linux platforms). Note that LinuxThreads need to block
4257 // this signal for all threads to work properly. So we don't have
4258 // to use hard-coded signal number when setting up the mask.
4259 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4260
4261 if (sigaction(SR_signum, &act, 0) == -1) {
4262 return -1;
4263 }
4264
4265 // Save signal flag
4266 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4267 return 0;
4268 }
4269
4270 static int sr_notify(OSThread* osthread) {
4271 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4272 assert_status(status == 0, status, "pthread_kill");
4273 return status;
4274 }
4275
4276 // "Randomly" selected value for how long we want to spin
4277 // before bailing out on suspending a thread, also how often
4278 // we send a signal to a thread we want to resume
4279 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4280 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4281
4282 // returns true on success and false on error - really an error is fatal
4283 // but this seems the normal response to library errors
4284 static bool do_suspend(OSThread* osthread) {
4285 assert(osthread->sr.is_running(), "thread should be running");
4286 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4287
4288 // mark as suspended and send signal
4289 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4290 // failed to switch, state wasn't running?
4291 ShouldNotReachHere();
4292 return false;
4293 }
4294
4295 if (sr_notify(osthread) != 0) {
4296 ShouldNotReachHere();
4297 }
4298
4299 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4300 while (true) {
4301 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4302 break;
4303 } else {
4304 // timeout
4305 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4306 if (cancelled == os::SuspendResume::SR_RUNNING) {
4307 return false;
4308 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4309 // make sure that we consume the signal on the semaphore as well
4310 sr_semaphore.wait();
4311 break;
4312 } else {
4313 ShouldNotReachHere();
4314 return false;
4315 }
4316 }
4317 }
4318
4319 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4320 return true;
4321 }
4322
4323 static void do_resume(OSThread* osthread) {
4324 assert(osthread->sr.is_suspended(), "thread should be suspended");
4325 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4326
4327 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4328 // failed to switch to WAKEUP_REQUEST
4329 ShouldNotReachHere();
4330 return;
4331 }
4332
4333 while (true) {
4334 if (sr_notify(osthread) == 0) {
4335 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4336 if (osthread->sr.is_running()) {
4337 return;
4338 }
4339 }
4340 } else {
4341 ShouldNotReachHere();
4342 }
4343 }
4344
4345 guarantee(osthread->sr.is_running(), "Must be running!");
4346 }
4347
4348 ///////////////////////////////////////////////////////////////////////////////////
4349 // signal handling (except suspend/resume)
4350
4351 // This routine may be used by user applications as a "hook" to catch signals.
4352 // The user-defined signal handler must pass unrecognized signals to this
4353 // routine, and if it returns true (non-zero), then the signal handler must
4354 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4355 // routine will never retun false (zero), but instead will execute a VM panic
4356 // routine kill the process.
4357 //
4358 // If this routine returns false, it is OK to call it again. This allows
4359 // the user-defined signal handler to perform checks either before or after
4360 // the VM performs its own checks. Naturally, the user code would be making
4361 // a serious error if it tried to handle an exception (such as a null check
4362 // or breakpoint) that the VM was generating for its own correct operation.
4363 //
4364 // This routine may recognize any of the following kinds of signals:
4365 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4366 // It should be consulted by handlers for any of those signals.
4367 //
4368 // The caller of this routine must pass in the three arguments supplied
4369 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4370 // field of the structure passed to sigaction(). This routine assumes that
4371 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4372 //
4373 // Note that the VM will print warnings if it detects conflicting signal
4374 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4375 //
4376 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4377 siginfo_t* siginfo,
4378 void* ucontext,
4379 int abort_if_unrecognized);
4380
4381 void signalHandler(int sig, siginfo_t* info, void* uc) {
4382 assert(info != NULL && uc != NULL, "it must be old kernel");
4383 int orig_errno = errno; // Preserve errno value over signal handler.
4384 JVM_handle_linux_signal(sig, info, uc, true);
4385 errno = orig_errno;
4386 }
4387
4388
4389 // This boolean allows users to forward their own non-matching signals
4390 // to JVM_handle_linux_signal, harmlessly.
4391 bool os::Linux::signal_handlers_are_installed = false;
4392
4393 // For signal-chaining
4394 struct sigaction sigact[NSIG];
4395 uint64_t sigs = 0;
4396 #if (64 < NSIG-1)
4397 #error "Not all signals can be encoded in sigs. Adapt its type!"
4398 #endif
4399 bool os::Linux::libjsig_is_loaded = false;
4400 typedef struct sigaction *(*get_signal_t)(int);
4401 get_signal_t os::Linux::get_signal_action = NULL;
4402
4403 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4404 struct sigaction *actp = NULL;
4405
4406 if (libjsig_is_loaded) {
4407 // Retrieve the old signal handler from libjsig
4408 actp = (*get_signal_action)(sig);
4409 }
4410 if (actp == NULL) {
4411 // Retrieve the preinstalled signal handler from jvm
4412 actp = get_preinstalled_handler(sig);
4413 }
4414
4415 return actp;
4416 }
4417
4418 static bool call_chained_handler(struct sigaction *actp, int sig,
4419 siginfo_t *siginfo, void *context) {
4420 // Call the old signal handler
4421 if (actp->sa_handler == SIG_DFL) {
4422 // It's more reasonable to let jvm treat it as an unexpected exception
4423 // instead of taking the default action.
4424 return false;
4425 } else if (actp->sa_handler != SIG_IGN) {
4426 if ((actp->sa_flags & SA_NODEFER) == 0) {
4427 // automaticlly block the signal
4428 sigaddset(&(actp->sa_mask), sig);
4429 }
4430
4431 sa_handler_t hand = NULL;
4432 sa_sigaction_t sa = NULL;
4433 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4434 // retrieve the chained handler
4435 if (siginfo_flag_set) {
4436 sa = actp->sa_sigaction;
4437 } else {
4438 hand = actp->sa_handler;
4439 }
4440
4441 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4442 actp->sa_handler = SIG_DFL;
4443 }
4444
4445 // try to honor the signal mask
4446 sigset_t oset;
4447 sigemptyset(&oset);
4448 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4449
4450 // call into the chained handler
4451 if (siginfo_flag_set) {
4452 (*sa)(sig, siginfo, context);
4453 } else {
4454 (*hand)(sig);
4455 }
4456
4457 // restore the signal mask
4458 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4459 }
4460 // Tell jvm's signal handler the signal is taken care of.
4461 return true;
4462 }
4463
4464 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4465 bool chained = false;
4466 // signal-chaining
4467 if (UseSignalChaining) {
4468 struct sigaction *actp = get_chained_signal_action(sig);
4469 if (actp != NULL) {
4470 chained = call_chained_handler(actp, sig, siginfo, context);
4471 }
4472 }
4473 return chained;
4474 }
4475
4476 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4477 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4478 return &sigact[sig];
4479 }
4480 return NULL;
4481 }
4482
4483 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4484 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4485 sigact[sig] = oldAct;
4486 sigs |= (uint64_t)1 << (sig-1);
4487 }
4488
4489 // for diagnostic
4490 int sigflags[NSIG];
4491
4492 int os::Linux::get_our_sigflags(int sig) {
4493 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4494 return sigflags[sig];
4495 }
4496
4497 void os::Linux::set_our_sigflags(int sig, int flags) {
4498 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4499 if (sig > 0 && sig < NSIG) {
4500 sigflags[sig] = flags;
4501 }
4502 }
4503
4504 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4505 // Check for overwrite.
4506 struct sigaction oldAct;
4507 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4508
4509 void* oldhand = oldAct.sa_sigaction
4510 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4511 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4512 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4513 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4514 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4515 if (AllowUserSignalHandlers || !set_installed) {
4516 // Do not overwrite; user takes responsibility to forward to us.
4517 return;
4518 } else if (UseSignalChaining) {
4519 // save the old handler in jvm
4520 save_preinstalled_handler(sig, oldAct);
4521 // libjsig also interposes the sigaction() call below and saves the
4522 // old sigaction on it own.
4523 } else {
4524 fatal("Encountered unexpected pre-existing sigaction handler "
4525 "%#lx for signal %d.", (long)oldhand, sig);
4526 }
4527 }
4528
4529 struct sigaction sigAct;
4530 sigfillset(&(sigAct.sa_mask));
4531 sigAct.sa_handler = SIG_DFL;
4532 if (!set_installed) {
4533 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4534 } else {
4535 sigAct.sa_sigaction = signalHandler;
4536 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4537 }
4538 // Save flags, which are set by ours
4539 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4540 sigflags[sig] = sigAct.sa_flags;
4541
4542 int ret = sigaction(sig, &sigAct, &oldAct);
4543 assert(ret == 0, "check");
4544
4545 void* oldhand2 = oldAct.sa_sigaction
4546 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4547 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4548 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4549 }
4550
4551 // install signal handlers for signals that HotSpot needs to
4552 // handle in order to support Java-level exception handling.
4553
4554 void os::Linux::install_signal_handlers() {
4555 if (!signal_handlers_are_installed) {
4556 signal_handlers_are_installed = true;
4557
4558 // signal-chaining
4559 typedef void (*signal_setting_t)();
4560 signal_setting_t begin_signal_setting = NULL;
4561 signal_setting_t end_signal_setting = NULL;
4562 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4563 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4564 if (begin_signal_setting != NULL) {
4565 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4566 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4567 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4568 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4569 libjsig_is_loaded = true;
4570 assert(UseSignalChaining, "should enable signal-chaining");
4571 }
4572 if (libjsig_is_loaded) {
4573 // Tell libjsig jvm is setting signal handlers
4574 (*begin_signal_setting)();
4575 }
4576
4577 set_signal_handler(SIGSEGV, true);
4578 set_signal_handler(SIGPIPE, true);
4579 set_signal_handler(SIGBUS, true);
4580 set_signal_handler(SIGILL, true);
4581 set_signal_handler(SIGFPE, true);
4582 #if defined(PPC64)
4583 set_signal_handler(SIGTRAP, true);
4584 #endif
4585 set_signal_handler(SIGXFSZ, true);
4586
4587 if (libjsig_is_loaded) {
4588 // Tell libjsig jvm finishes setting signal handlers
4589 (*end_signal_setting)();
4590 }
4591
4592 // We don't activate signal checker if libjsig is in place, we trust ourselves
4593 // and if UserSignalHandler is installed all bets are off.
4594 // Log that signal checking is off only if -verbose:jni is specified.
4595 if (CheckJNICalls) {
4596 if (libjsig_is_loaded) {
4597 if (PrintJNIResolving) {
4598 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4599 }
4600 check_signals = false;
4601 }
4602 if (AllowUserSignalHandlers) {
4603 if (PrintJNIResolving) {
4604 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4605 }
4606 check_signals = false;
4607 }
4608 }
4609 }
4610 }
4611
4612 // This is the fastest way to get thread cpu time on Linux.
4613 // Returns cpu time (user+sys) for any thread, not only for current.
4614 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4615 // It might work on 2.6.10+ with a special kernel/glibc patch.
4616 // For reference, please, see IEEE Std 1003.1-2004:
4617 // http://www.unix.org/single_unix_specification
4618
4619 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4620 struct timespec tp;
4621 int rc = os::Linux::clock_gettime(clockid, &tp);
4622 assert(rc == 0, "clock_gettime is expected to return 0 code");
4623
4624 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4625 }
4626
4627 void os::Linux::initialize_os_info() {
4628 assert(_os_version == 0, "OS info already initialized");
4629
4630 struct utsname _uname;
4631
4632 uint32_t major;
4633 uint32_t minor;
4634 uint32_t fix;
4635
4636 int rc;
4637
4638 // Kernel version is unknown if
4639 // verification below fails.
4640 _os_version = 0x01000000;
4641
4642 rc = uname(&_uname);
4643 if (rc != -1) {
4644
4645 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4646 if (rc == 3) {
4647
4648 if (major < 256 && minor < 256 && fix < 256) {
4649 // Kernel version format is as expected,
4650 // set it overriding unknown state.
4651 _os_version = (major << 16) |
4652 (minor << 8 ) |
4653 (fix << 0 ) ;
4654 }
4655 }
4656 }
4657 }
4658
4659 uint32_t os::Linux::os_version() {
4660 assert(_os_version != 0, "not initialized");
4661 return _os_version & 0x00FFFFFF;
4662 }
4663
4664 bool os::Linux::os_version_is_known() {
4665 assert(_os_version != 0, "not initialized");
4666 return _os_version & 0x01000000 ? false : true;
4667 }
4668
4669 /////
4670 // glibc on Linux platform uses non-documented flag
4671 // to indicate, that some special sort of signal
4672 // trampoline is used.
4673 // We will never set this flag, and we should
4674 // ignore this flag in our diagnostic
4675 #ifdef SIGNIFICANT_SIGNAL_MASK
4676 #undef SIGNIFICANT_SIGNAL_MASK
4677 #endif
4678 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4679
4680 static const char* get_signal_handler_name(address handler,
4681 char* buf, int buflen) {
4682 int offset = 0;
4683 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4684 if (found) {
4685 // skip directory names
4686 const char *p1, *p2;
4687 p1 = buf;
4688 size_t len = strlen(os::file_separator());
4689 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4690 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4691 } else {
4692 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4693 }
4694 return buf;
4695 }
4696
4697 static void print_signal_handler(outputStream* st, int sig,
4698 char* buf, size_t buflen) {
4699 struct sigaction sa;
4700
4701 sigaction(sig, NULL, &sa);
4702
4703 // See comment for SIGNIFICANT_SIGNAL_MASK define
4704 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4705
4706 st->print("%s: ", os::exception_name(sig, buf, buflen));
4707
4708 address handler = (sa.sa_flags & SA_SIGINFO)
4709 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4710 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4711
4712 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4713 st->print("SIG_DFL");
4714 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4715 st->print("SIG_IGN");
4716 } else {
4717 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4718 }
4719
4720 st->print(", sa_mask[0]=");
4721 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4722
4723 address rh = VMError::get_resetted_sighandler(sig);
4724 // May be, handler was resetted by VMError?
4725 if (rh != NULL) {
4726 handler = rh;
4727 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4728 }
4729
4730 st->print(", sa_flags=");
4731 os::Posix::print_sa_flags(st, sa.sa_flags);
4732
4733 // Check: is it our handler?
4734 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4735 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4736 // It is our signal handler
4737 // check for flags, reset system-used one!
4738 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4739 st->print(
4740 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4741 os::Linux::get_our_sigflags(sig));
4742 }
4743 }
4744 st->cr();
4745 }
4746
4747
4748 #define DO_SIGNAL_CHECK(sig) \
4749 do { \
4750 if (!sigismember(&check_signal_done, sig)) { \
4751 os::Linux::check_signal_handler(sig); \
4752 } \
4753 } while (0)
4754
4755 // This method is a periodic task to check for misbehaving JNI applications
4756 // under CheckJNI, we can add any periodic checks here
4757
4758 void os::run_periodic_checks() {
4759 if (check_signals == false) return;
4760
4761 // SEGV and BUS if overridden could potentially prevent
4762 // generation of hs*.log in the event of a crash, debugging
4763 // such a case can be very challenging, so we absolutely
4764 // check the following for a good measure:
4765 DO_SIGNAL_CHECK(SIGSEGV);
4766 DO_SIGNAL_CHECK(SIGILL);
4767 DO_SIGNAL_CHECK(SIGFPE);
4768 DO_SIGNAL_CHECK(SIGBUS);
4769 DO_SIGNAL_CHECK(SIGPIPE);
4770 DO_SIGNAL_CHECK(SIGXFSZ);
4771 #if defined(PPC64)
4772 DO_SIGNAL_CHECK(SIGTRAP);
4773 #endif
4774
4775 // ReduceSignalUsage allows the user to override these handlers
4776 // see comments at the very top and jvm_md.h
4777 if (!ReduceSignalUsage) {
4778 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4779 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4780 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4781 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4782 }
4783
4784 DO_SIGNAL_CHECK(SR_signum);
4785 }
4786
4787 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4788
4789 static os_sigaction_t os_sigaction = NULL;
4790
4791 void os::Linux::check_signal_handler(int sig) {
4792 char buf[O_BUFLEN];
4793 address jvmHandler = NULL;
4794
4795
4796 struct sigaction act;
4797 if (os_sigaction == NULL) {
4798 // only trust the default sigaction, in case it has been interposed
4799 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4800 if (os_sigaction == NULL) return;
4801 }
4802
4803 os_sigaction(sig, (struct sigaction*)NULL, &act);
4804
4805
4806 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4807
4808 address thisHandler = (act.sa_flags & SA_SIGINFO)
4809 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4810 : CAST_FROM_FN_PTR(address, act.sa_handler);
4811
4812
4813 switch (sig) {
4814 case SIGSEGV:
4815 case SIGBUS:
4816 case SIGFPE:
4817 case SIGPIPE:
4818 case SIGILL:
4819 case SIGXFSZ:
4820 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4821 break;
4822
4823 case SHUTDOWN1_SIGNAL:
4824 case SHUTDOWN2_SIGNAL:
4825 case SHUTDOWN3_SIGNAL:
4826 case BREAK_SIGNAL:
4827 jvmHandler = (address)user_handler();
4828 break;
4829
4830 default:
4831 if (sig == SR_signum) {
4832 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4833 } else {
4834 return;
4835 }
4836 break;
4837 }
4838
4839 if (thisHandler != jvmHandler) {
4840 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4841 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4842 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4843 // No need to check this sig any longer
4844 sigaddset(&check_signal_done, sig);
4845 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4846 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4847 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4848 exception_name(sig, buf, O_BUFLEN));
4849 }
4850 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4851 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4852 tty->print("expected:");
4853 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4854 tty->cr();
4855 tty->print(" found:");
4856 os::Posix::print_sa_flags(tty, act.sa_flags);
4857 tty->cr();
4858 // No need to check this sig any longer
4859 sigaddset(&check_signal_done, sig);
4860 }
4861
4862 // Dump all the signal
4863 if (sigismember(&check_signal_done, sig)) {
4864 print_signal_handlers(tty, buf, O_BUFLEN);
4865 }
4866 }
4867
4868 extern void report_error(char* file_name, int line_no, char* title,
4869 char* format, ...);
4870
4871 // this is called _before_ the most of global arguments have been parsed
4872 void os::init(void) {
4873 char dummy; // used to get a guess on initial stack address
4874 // first_hrtime = gethrtime();
4875
4876 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4877
4878 init_random(1234567);
4879
4880 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4881 if (Linux::page_size() == -1) {
4882 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4883 os::strerror(errno));
4884 }
4885 init_page_sizes((size_t) Linux::page_size());
4886
4887 Linux::initialize_system_info();
4888
4889 Linux::initialize_os_info();
4890
4891 // main_thread points to the aboriginal thread
4892 Linux::_main_thread = pthread_self();
4893
4894 Linux::clock_init();
4895 initial_time_count = javaTimeNanos();
4896
4897 // retrieve entry point for pthread_setname_np
4898 Linux::_pthread_setname_np =
4899 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4900
4901 os::Posix::init();
4902 }
4903
4904 // To install functions for atexit system call
4905 extern "C" {
4906 static void perfMemory_exit_helper() {
4907 perfMemory_exit();
4908 }
4909 }
4910
4911 void os::pd_init_container_support() {
4912 OSContainer::init();
4913 }
4914
4915 // this is called _after_ the global arguments have been parsed
4916 jint os::init_2(void) {
4917
4918 os::Posix::init_2();
4919
4920 Linux::fast_thread_clock_init();
4921
4922 // initialize suspend/resume support - must do this before signal_sets_init()
4923 if (SR_initialize() != 0) {
4924 perror("SR_initialize failed");
4925 return JNI_ERR;
4926 }
4927
4928 Linux::signal_sets_init();
4929 Linux::install_signal_handlers();
4930
4931 // Check and sets minimum stack sizes against command line options
4932 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4933 return JNI_ERR;
4934 }
4935 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4936
4937 #if defined(IA32)
4938 workaround_expand_exec_shield_cs_limit();
4939 #endif
4940
4941 Linux::libpthread_init();
4942 Linux::sched_getcpu_init();
4943 log_info(os)("HotSpot is running with %s, %s",
4944 Linux::glibc_version(), Linux::libpthread_version());
4945
4946 if (UseNUMA) {
4947 if (!Linux::libnuma_init()) {
4948 UseNUMA = false;
4949 } else {
4950 if ((Linux::numa_max_node() < 1)) {
4951 // There's only one node(they start from 0), disable NUMA.
4952 UseNUMA = false;
4953 }
4954 }
4955 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4956 // we can make the adaptive lgrp chunk resizing work. If the user specified
4957 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4958 // disable adaptive resizing.
4959 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4960 if (FLAG_IS_DEFAULT(UseNUMA)) {
4961 UseNUMA = false;
4962 } else {
4963 if (FLAG_IS_DEFAULT(UseLargePages) &&
4964 FLAG_IS_DEFAULT(UseSHM) &&
4965 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4966 UseLargePages = false;
4967 } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4968 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4969 UseAdaptiveSizePolicy = false;
4970 UseAdaptiveNUMAChunkSizing = false;
4971 }
4972 }
4973 }
4974 if (!UseNUMA && ForceNUMA) {
4975 UseNUMA = true;
4976 }
4977 }
4978
4979 if (MaxFDLimit) {
4980 // set the number of file descriptors to max. print out error
4981 // if getrlimit/setrlimit fails but continue regardless.
4982 struct rlimit nbr_files;
4983 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4984 if (status != 0) {
4985 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4986 } else {
4987 nbr_files.rlim_cur = nbr_files.rlim_max;
4988 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4989 if (status != 0) {
4990 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4991 }
4992 }
4993 }
4994
4995 // Initialize lock used to serialize thread creation (see os::create_thread)
4996 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4997
4998 // at-exit methods are called in the reverse order of their registration.
4999 // atexit functions are called on return from main or as a result of a
5000 // call to exit(3C). There can be only 32 of these functions registered
5001 // and atexit() does not set errno.
5002
5003 if (PerfAllowAtExitRegistration) {
5004 // only register atexit functions if PerfAllowAtExitRegistration is set.
5005 // atexit functions can be delayed until process exit time, which
5006 // can be problematic for embedded VM situations. Embedded VMs should
5007 // call DestroyJavaVM() to assure that VM resources are released.
5008
5009 // note: perfMemory_exit_helper atexit function may be removed in
5010 // the future if the appropriate cleanup code can be added to the
5011 // VM_Exit VMOperation's doit method.
5012 if (atexit(perfMemory_exit_helper) != 0) {
5013 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5014 }
5015 }
5016
5017 // initialize thread priority policy
5018 prio_init();
5019
5020 return JNI_OK;
5021 }
5022
5023 // Mark the polling page as unreadable
5024 void os::make_polling_page_unreadable(void) {
5025 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5026 fatal("Could not disable polling page");
5027 }
5028 }
5029
5030 // Mark the polling page as readable
5031 void os::make_polling_page_readable(void) {
5032 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5033 fatal("Could not enable polling page");
5034 }
5035 }
5036
5037 // older glibc versions don't have this macro (which expands to
5038 // an optimized bit-counting function) so we have to roll our own
5039 #ifndef CPU_COUNT
5040
5041 static int _cpu_count(const cpu_set_t* cpus) {
5042 int count = 0;
5043 // only look up to the number of configured processors
5044 for (int i = 0; i < os::processor_count(); i++) {
5045 if (CPU_ISSET(i, cpus)) {
5046 count++;
5047 }
5048 }
5049 return count;
5050 }
5051
5052 #define CPU_COUNT(cpus) _cpu_count(cpus)
5053
5054 #endif // CPU_COUNT
5055
5056 // Get the current number of available processors for this process.
5057 // This value can change at any time during a process's lifetime.
5058 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5059 // If it appears there may be more than 1024 processors then we do a
5060 // dynamic check - see 6515172 for details.
5061 // If anything goes wrong we fallback to returning the number of online
5062 // processors - which can be greater than the number available to the process.
5063 int os::Linux::active_processor_count() {
5064 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5065 cpu_set_t* cpus_p = &cpus;
5066 int cpus_size = sizeof(cpu_set_t);
5067
5068 int configured_cpus = os::processor_count(); // upper bound on available cpus
5069 int cpu_count = 0;
5070
5071 // old build platforms may not support dynamic cpu sets
5072 #ifdef CPU_ALLOC
5073
5074 // To enable easy testing of the dynamic path on different platforms we
5075 // introduce a diagnostic flag: UseCpuAllocPath
5076 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5077 // kernel may use a mask bigger than cpu_set_t
5078 log_trace(os)("active_processor_count: using dynamic path %s"
5079 "- configured processors: %d",
5080 UseCpuAllocPath ? "(forced) " : "",
5081 configured_cpus);
5082 cpus_p = CPU_ALLOC(configured_cpus);
5083 if (cpus_p != NULL) {
5084 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5085 // zero it just to be safe
5086 CPU_ZERO_S(cpus_size, cpus_p);
5087 }
5088 else {
5089 // failed to allocate so fallback to online cpus
5090 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5091 log_trace(os)("active_processor_count: "
5092 "CPU_ALLOC failed (%s) - using "
5093 "online processor count: %d",
5094 os::strerror(errno), online_cpus);
5095 return online_cpus;
5096 }
5097 }
5098 else {
5099 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5100 configured_cpus);
5101 }
5102 #else // CPU_ALLOC
5103 // these stubs won't be executed
5104 #define CPU_COUNT_S(size, cpus) -1
5105 #define CPU_FREE(cpus)
5106
5107 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5108 configured_cpus);
5109 #endif // CPU_ALLOC
5110
5111 // pid 0 means the current thread - which we have to assume represents the process
5112 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5113 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5114 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5115 }
5116 else {
5117 cpu_count = CPU_COUNT(cpus_p);
5118 }
5119 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5120 }
5121 else {
5122 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5123 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5124 "which may exceed available processors", os::strerror(errno), cpu_count);
5125 }
5126
5127 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5128 CPU_FREE(cpus_p);
5129 }
5130
5131 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5132 return cpu_count;
5133 }
5134
5135 // Determine the active processor count from one of
5136 // three different sources:
5137 //
5138 // 1. User option -XX:ActiveProcessorCount
5139 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5140 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5141 //
5142 // Option 1, if specified, will always override.
5143 // If the cgroup subsystem is active and configured, we
5144 // will return the min of the cgroup and option 2 results.
5145 // This is required since tools, such as numactl, that
5146 // alter cpu affinity do not update cgroup subsystem
5147 // cpuset configuration files.
5148 int os::active_processor_count() {
5149 // User has overridden the number of active processors
5150 if (ActiveProcessorCount > 0) {
5151 log_trace(os)("active_processor_count: "
5152 "active processor count set by user : %d",
5153 ActiveProcessorCount);
5154 return ActiveProcessorCount;
5155 }
5156
5157 int active_cpus;
5158 if (OSContainer::is_containerized()) {
5159 active_cpus = OSContainer::active_processor_count();
5160 log_trace(os)("active_processor_count: determined by OSContainer: %d",
5161 active_cpus);
5162 } else {
5163 active_cpus = os::Linux::active_processor_count();
5164 }
5165
5166 return active_cpus;
5167 }
5168
5169 void os::set_native_thread_name(const char *name) {
5170 if (Linux::_pthread_setname_np) {
5171 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5172 snprintf(buf, sizeof(buf), "%s", name);
5173 buf[sizeof(buf) - 1] = '\0';
5174 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5175 // ERANGE should not happen; all other errors should just be ignored.
5176 assert(rc != ERANGE, "pthread_setname_np failed");
5177 }
5178 }
5179
5180 bool os::distribute_processes(uint length, uint* distribution) {
5181 // Not yet implemented.
5182 return false;
5183 }
5184
5185 bool os::bind_to_processor(uint processor_id) {
5186 // Not yet implemented.
5187 return false;
5188 }
5189
5190 ///
5191
5192 void os::SuspendedThreadTask::internal_do_task() {
5193 if (do_suspend(_thread->osthread())) {
5194 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5195 do_task(context);
5196 do_resume(_thread->osthread());
5197 }
5198 }
5199
5200 ////////////////////////////////////////////////////////////////////////////////
5201 // debug support
5202
5203 bool os::find(address addr, outputStream* st) {
5204 Dl_info dlinfo;
5205 memset(&dlinfo, 0, sizeof(dlinfo));
5206 if (dladdr(addr, &dlinfo) != 0) {
5207 st->print(PTR_FORMAT ": ", p2i(addr));
5208 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5209 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5210 p2i(addr) - p2i(dlinfo.dli_saddr));
5211 } else if (dlinfo.dli_fbase != NULL) {
5212 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5213 } else {
5214 st->print("<absolute address>");
5215 }
5216 if (dlinfo.dli_fname != NULL) {
5217 st->print(" in %s", dlinfo.dli_fname);
5218 }
5219 if (dlinfo.dli_fbase != NULL) {
5220 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5221 }
5222 st->cr();
5223
5224 if (Verbose) {
5225 // decode some bytes around the PC
5226 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5227 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5228 address lowest = (address) dlinfo.dli_sname;
5229 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5230 if (begin < lowest) begin = lowest;
5231 Dl_info dlinfo2;
5232 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5233 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5234 end = (address) dlinfo2.dli_saddr;
5235 }
5236 Disassembler::decode(begin, end, st);
5237 }
5238 return true;
5239 }
5240 return false;
5241 }
5242
5243 ////////////////////////////////////////////////////////////////////////////////
5244 // misc
5245
5246 // This does not do anything on Linux. This is basically a hook for being
5247 // able to use structured exception handling (thread-local exception filters)
5248 // on, e.g., Win32.
5249 void
5250 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5251 JavaCallArguments* args, Thread* thread) {
5252 f(value, method, args, thread);
5253 }
5254
5255 void os::print_statistics() {
5256 }
5257
5258 bool os::message_box(const char* title, const char* message) {
5259 int i;
5260 fdStream err(defaultStream::error_fd());
5261 for (i = 0; i < 78; i++) err.print_raw("=");
5262 err.cr();
5263 err.print_raw_cr(title);
5264 for (i = 0; i < 78; i++) err.print_raw("-");
5265 err.cr();
5266 err.print_raw_cr(message);
5267 for (i = 0; i < 78; i++) err.print_raw("=");
5268 err.cr();
5269
5270 char buf[16];
5271 // Prevent process from exiting upon "read error" without consuming all CPU
5272 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5273
5274 return buf[0] == 'y' || buf[0] == 'Y';
5275 }
5276
5277 int os::stat(const char *path, struct stat *sbuf) {
5278 char pathbuf[MAX_PATH];
5279 if (strlen(path) > MAX_PATH - 1) {
5280 errno = ENAMETOOLONG;
5281 return -1;
5282 }
5283 os::native_path(strcpy(pathbuf, path));
5284 return ::stat(pathbuf, sbuf);
5285 }
5286
5287 // Is a (classpath) directory empty?
5288 bool os::dir_is_empty(const char* path) {
5289 DIR *dir = NULL;
5290 struct dirent *ptr;
5291
5292 dir = opendir(path);
5293 if (dir == NULL) return true;
5294
5295 // Scan the directory
5296 bool result = true;
5297 char buf[sizeof(struct dirent) + MAX_PATH];
5298 while (result && (ptr = ::readdir(dir)) != NULL) {
5299 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5300 result = false;
5301 }
5302 }
5303 closedir(dir);
5304 return result;
5305 }
5306
5307 // This code originates from JDK's sysOpen and open64_w
5308 // from src/solaris/hpi/src/system_md.c
5309
5310 int os::open(const char *path, int oflag, int mode) {
5311 if (strlen(path) > MAX_PATH - 1) {
5312 errno = ENAMETOOLONG;
5313 return -1;
5314 }
5315
5316 // All file descriptors that are opened in the Java process and not
5317 // specifically destined for a subprocess should have the close-on-exec
5318 // flag set. If we don't set it, then careless 3rd party native code
5319 // might fork and exec without closing all appropriate file descriptors
5320 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5321 // turn might:
5322 //
5323 // - cause end-of-file to fail to be detected on some file
5324 // descriptors, resulting in mysterious hangs, or
5325 //
5326 // - might cause an fopen in the subprocess to fail on a system
5327 // suffering from bug 1085341.
5328 //
5329 // (Yes, the default setting of the close-on-exec flag is a Unix
5330 // design flaw)
5331 //
5332 // See:
5333 // 1085341: 32-bit stdio routines should support file descriptors >255
5334 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5335 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5336 //
5337 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5338 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5339 // because it saves a system call and removes a small window where the flag
5340 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5341 // and we fall back to using FD_CLOEXEC (see below).
5342 #ifdef O_CLOEXEC
5343 oflag |= O_CLOEXEC;
5344 #endif
5345
5346 int fd = ::open64(path, oflag, mode);
5347 if (fd == -1) return -1;
5348
5349 //If the open succeeded, the file might still be a directory
5350 {
5351 struct stat64 buf64;
5352 int ret = ::fstat64(fd, &buf64);
5353 int st_mode = buf64.st_mode;
5354
5355 if (ret != -1) {
5356 if ((st_mode & S_IFMT) == S_IFDIR) {
5357 errno = EISDIR;
5358 ::close(fd);
5359 return -1;
5360 }
5361 } else {
5362 ::close(fd);
5363 return -1;
5364 }
5365 }
5366
5367 #ifdef FD_CLOEXEC
5368 // Validate that the use of the O_CLOEXEC flag on open above worked.
5369 // With recent kernels, we will perform this check exactly once.
5370 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5371 if (!O_CLOEXEC_is_known_to_work) {
5372 int flags = ::fcntl(fd, F_GETFD);
5373 if (flags != -1) {
5374 if ((flags & FD_CLOEXEC) != 0)
5375 O_CLOEXEC_is_known_to_work = 1;
5376 else
5377 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5378 }
5379 }
5380 #endif
5381
5382 return fd;
5383 }
5384
5385
5386 // create binary file, rewriting existing file if required
5387 int os::create_binary_file(const char* path, bool rewrite_existing) {
5388 int oflags = O_WRONLY | O_CREAT;
5389 if (!rewrite_existing) {
5390 oflags |= O_EXCL;
5391 }
5392 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5393 }
5394
5395 // return current position of file pointer
5396 jlong os::current_file_offset(int fd) {
5397 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5398 }
5399
5400 // move file pointer to the specified offset
5401 jlong os::seek_to_file_offset(int fd, jlong offset) {
5402 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5403 }
5404
5405 // This code originates from JDK's sysAvailable
5406 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5407
5408 int os::available(int fd, jlong *bytes) {
5409 jlong cur, end;
5410 int mode;
5411 struct stat64 buf64;
5412
5413 if (::fstat64(fd, &buf64) >= 0) {
5414 mode = buf64.st_mode;
5415 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5416 int n;
5417 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5418 *bytes = n;
5419 return 1;
5420 }
5421 }
5422 }
5423 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5424 return 0;
5425 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5426 return 0;
5427 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5428 return 0;
5429 }
5430 *bytes = end - cur;
5431 return 1;
5432 }
5433
5434 // Map a block of memory.
5435 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5436 char *addr, size_t bytes, bool read_only,
5437 bool allow_exec) {
5438 int prot;
5439 int flags = MAP_PRIVATE;
5440
5441 if (read_only) {
5442 prot = PROT_READ;
5443 } else {
5444 prot = PROT_READ | PROT_WRITE;
5445 }
5446
5447 if (allow_exec) {
5448 prot |= PROT_EXEC;
5449 }
5450
5451 if (addr != NULL) {
5452 flags |= MAP_FIXED;
5453 }
5454
5455 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5456 fd, file_offset);
5457 if (mapped_address == MAP_FAILED) {
5458 return NULL;
5459 }
5460 return mapped_address;
5461 }
5462
5463
5464 // Remap a block of memory.
5465 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5466 char *addr, size_t bytes, bool read_only,
5467 bool allow_exec) {
5468 // same as map_memory() on this OS
5469 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5470 allow_exec);
5471 }
5472
5473
5474 // Unmap a block of memory.
5475 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5476 return munmap(addr, bytes) == 0;
5477 }
5478
5479 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5480
5481 static clockid_t thread_cpu_clockid(Thread* thread) {
5482 pthread_t tid = thread->osthread()->pthread_id();
5483 clockid_t clockid;
5484
5485 // Get thread clockid
5486 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5487 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5488 return clockid;
5489 }
5490
5491 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5492 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5493 // of a thread.
5494 //
5495 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5496 // the fast estimate available on the platform.
5497
5498 jlong os::current_thread_cpu_time() {
5499 if (os::Linux::supports_fast_thread_cpu_time()) {
5500 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5501 } else {
5502 // return user + sys since the cost is the same
5503 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5504 }
5505 }
5506
5507 jlong os::thread_cpu_time(Thread* thread) {
5508 // consistent with what current_thread_cpu_time() returns
5509 if (os::Linux::supports_fast_thread_cpu_time()) {
5510 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5511 } else {
5512 return slow_thread_cpu_time(thread, true /* user + sys */);
5513 }
5514 }
5515
5516 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5517 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5518 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5519 } else {
5520 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5521 }
5522 }
5523
5524 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5525 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5526 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5527 } else {
5528 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5529 }
5530 }
5531
5532 // -1 on error.
5533 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5534 pid_t tid = thread->osthread()->thread_id();
5535 char *s;
5536 char stat[2048];
5537 int statlen;
5538 char proc_name[64];
5539 int count;
5540 long sys_time, user_time;
5541 char cdummy;
5542 int idummy;
5543 long ldummy;
5544 FILE *fp;
5545
5546 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5547 fp = fopen(proc_name, "r");
5548 if (fp == NULL) return -1;
5549 statlen = fread(stat, 1, 2047, fp);
5550 stat[statlen] = '\0';
5551 fclose(fp);
5552
5553 // Skip pid and the command string. Note that we could be dealing with
5554 // weird command names, e.g. user could decide to rename java launcher
5555 // to "java 1.4.2 :)", then the stat file would look like
5556 // 1234 (java 1.4.2 :)) R ... ...
5557 // We don't really need to know the command string, just find the last
5558 // occurrence of ")" and then start parsing from there. See bug 4726580.
5559 s = strrchr(stat, ')');
5560 if (s == NULL) return -1;
5561
5562 // Skip blank chars
5563 do { s++; } while (s && isspace(*s));
5564
5565 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5566 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5567 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5568 &user_time, &sys_time);
5569 if (count != 13) return -1;
5570 if (user_sys_cpu_time) {
5571 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5572 } else {
5573 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5574 }
5575 }
5576
5577 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5578 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5579 info_ptr->may_skip_backward = false; // elapsed time not wall time
5580 info_ptr->may_skip_forward = false; // elapsed time not wall time
5581 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5582 }
5583
5584 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5585 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5586 info_ptr->may_skip_backward = false; // elapsed time not wall time
5587 info_ptr->may_skip_forward = false; // elapsed time not wall time
5588 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5589 }
5590
5591 bool os::is_thread_cpu_time_supported() {
5592 return true;
5593 }
5594
5595 // System loadavg support. Returns -1 if load average cannot be obtained.
5596 // Linux doesn't yet have a (official) notion of processor sets,
5597 // so just return the system wide load average.
5598 int os::loadavg(double loadavg[], int nelem) {
5599 return ::getloadavg(loadavg, nelem);
5600 }
5601
5602 void os::pause() {
5603 char filename[MAX_PATH];
5604 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5605 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5606 } else {
5607 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5608 }
5609
5610 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5611 if (fd != -1) {
5612 struct stat buf;
5613 ::close(fd);
5614 while (::stat(filename, &buf) == 0) {
5615 (void)::poll(NULL, 0, 100);
5616 }
5617 } else {
5618 jio_fprintf(stderr,
5619 "Could not open pause file '%s', continuing immediately.\n", filename);
5620 }
5621 }
5622
5623 extern char** environ;
5624
5625 // Run the specified command in a separate process. Return its exit value,
5626 // or -1 on failure (e.g. can't fork a new process).
5627 // Unlike system(), this function can be called from signal handler. It
5628 // doesn't block SIGINT et al.
5629 int os::fork_and_exec(char* cmd) {
5630 const char * argv[4] = {"sh", "-c", cmd, NULL};
5631
5632 pid_t pid = fork();
5633
5634 if (pid < 0) {
5635 // fork failed
5636 return -1;
5637
5638 } else if (pid == 0) {
5639 // child process
5640
5641 execve("/bin/sh", (char* const*)argv, environ);
5642
5643 // execve failed
5644 _exit(-1);
5645
5646 } else {
5647 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5648 // care about the actual exit code, for now.
5649
5650 int status;
5651
5652 // Wait for the child process to exit. This returns immediately if
5653 // the child has already exited. */
5654 while (waitpid(pid, &status, 0) < 0) {
5655 switch (errno) {
5656 case ECHILD: return 0;
5657 case EINTR: break;
5658 default: return -1;
5659 }
5660 }
5661
5662 if (WIFEXITED(status)) {
5663 // The child exited normally; get its exit code.
5664 return WEXITSTATUS(status);
5665 } else if (WIFSIGNALED(status)) {
5666 // The child exited because of a signal
5667 // The best value to return is 0x80 + signal number,
5668 // because that is what all Unix shells do, and because
5669 // it allows callers to distinguish between process exit and
5670 // process death by signal.
5671 return 0x80 + WTERMSIG(status);
5672 } else {
5673 // Unknown exit code; pass it through
5674 return status;
5675 }
5676 }
5677 }
5678
5679 // is_headless_jre()
5680 //
5681 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5682 // in order to report if we are running in a headless jre
5683 //
5684 // Since JDK8 xawt/libmawt.so was moved into the same directory
5685 // as libawt.so, and renamed libawt_xawt.so
5686 //
5687 bool os::is_headless_jre() {
5688 struct stat statbuf;
5689 char buf[MAXPATHLEN];
5690 char libmawtpath[MAXPATHLEN];
5691 const char *xawtstr = "/xawt/libmawt.so";
5692 const char *new_xawtstr = "/libawt_xawt.so";
5693 char *p;
5694
5695 // Get path to libjvm.so
5696 os::jvm_path(buf, sizeof(buf));
5697
5698 // Get rid of libjvm.so
5699 p = strrchr(buf, '/');
5700 if (p == NULL) {
5701 return false;
5702 } else {
5703 *p = '\0';
5704 }
5705
5706 // Get rid of client or server
5707 p = strrchr(buf, '/');
5708 if (p == NULL) {
5709 return false;
5710 } else {
5711 *p = '\0';
5712 }
5713
5714 // check xawt/libmawt.so
5715 strcpy(libmawtpath, buf);
5716 strcat(libmawtpath, xawtstr);
5717 if (::stat(libmawtpath, &statbuf) == 0) return false;
5718
5719 // check libawt_xawt.so
5720 strcpy(libmawtpath, buf);
5721 strcat(libmawtpath, new_xawtstr);
5722 if (::stat(libmawtpath, &statbuf) == 0) return false;
5723
5724 return true;
5725 }
5726
5727 // Get the default path to the core file
5728 // Returns the length of the string
5729 int os::get_core_path(char* buffer, size_t bufferSize) {
5730 /*
5731 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5732 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5733 */
5734 const int core_pattern_len = 129;
5735 char core_pattern[core_pattern_len] = {0};
5736
5737 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5738 if (core_pattern_file == -1) {
5739 return -1;
5740 }
5741
5742 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5743 ::close(core_pattern_file);
5744 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5745 return -1;
5746 }
5747 if (core_pattern[ret-1] == '\n') {
5748 core_pattern[ret-1] = '\0';
5749 } else {
5750 core_pattern[ret] = '\0';
5751 }
5752
5753 char *pid_pos = strstr(core_pattern, "%p");
5754 int written;
5755
5756 if (core_pattern[0] == '/') {
5757 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5758 } else {
5759 char cwd[PATH_MAX];
5760
5761 const char* p = get_current_directory(cwd, PATH_MAX);
5762 if (p == NULL) {
5763 return -1;
5764 }
5765
5766 if (core_pattern[0] == '|') {
5767 written = jio_snprintf(buffer, bufferSize,
5768 "\"%s\" (or dumping to %s/core.%d)",
5769 &core_pattern[1], p, current_process_id());
5770 } else {
5771 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5772 }
5773 }
5774
5775 if (written < 0) {
5776 return -1;
5777 }
5778
5779 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5780 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5781
5782 if (core_uses_pid_file != -1) {
5783 char core_uses_pid = 0;
5784 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5785 ::close(core_uses_pid_file);
5786
5787 if (core_uses_pid == '1') {
5788 jio_snprintf(buffer + written, bufferSize - written,
5789 ".%d", current_process_id());
5790 }
5791 }
5792 }
5793
5794 return strlen(buffer);
5795 }
5796
5797 bool os::start_debugging(char *buf, int buflen) {
5798 int len = (int)strlen(buf);
5799 char *p = &buf[len];
5800
5801 jio_snprintf(p, buflen-len,
5802 "\n\n"
5803 "Do you want to debug the problem?\n\n"
5804 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5805 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5806 "Otherwise, press RETURN to abort...",
5807 os::current_process_id(), os::current_process_id(),
5808 os::current_thread_id(), os::current_thread_id());
5809
5810 bool yes = os::message_box("Unexpected Error", buf);
5811
5812 if (yes) {
5813 // yes, user asked VM to launch debugger
5814 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5815 os::current_process_id(), os::current_process_id());
5816
5817 os::fork_and_exec(buf);
5818 yes = false;
5819 }
5820 return yes;
5821 }
5822
5823
5824 // Java/Compiler thread:
5825 //
5826 // Low memory addresses
5827 // P0 +------------------------+
5828 // | |\ Java thread created by VM does not have glibc
5829 // | glibc guard page | - guard page, attached Java thread usually has
5830 // | |/ 1 glibc guard page.
5831 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5832 // | |\
5833 // | HotSpot Guard Pages | - red, yellow and reserved pages
5834 // | |/
5835 // +------------------------+ JavaThread::stack_reserved_zone_base()
5836 // | |\
5837 // | Normal Stack | -
5838 // | |/
5839 // P2 +------------------------+ Thread::stack_base()
5840 //
5841 // Non-Java thread:
5842 //
5843 // Low memory addresses
5844 // P0 +------------------------+
5845 // | |\
5846 // | glibc guard page | - usually 1 page
5847 // | |/
5848 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5849 // | |\
5850 // | Normal Stack | -
5851 // | |/
5852 // P2 +------------------------+ Thread::stack_base()
5853 //
5854 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
5855 // returned from pthread_attr_getstack().
5856 // ** Due to NPTL implementation error, linux takes the glibc guard page out
5857 // of the stack size given in pthread_attr. We work around this for
5858 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
5859 //
5860 #ifndef ZERO
5861 static void current_stack_region(address * bottom, size_t * size) {
5862 if (os::Linux::is_initial_thread()) {
5863 // initial thread needs special handling because pthread_getattr_np()
5864 // may return bogus value.
5865 *bottom = os::Linux::initial_thread_stack_bottom();
5866 *size = os::Linux::initial_thread_stack_size();
5867 } else {
5868 pthread_attr_t attr;
5869
5870 int rslt = pthread_getattr_np(pthread_self(), &attr);
5871
5872 // JVM needs to know exact stack location, abort if it fails
5873 if (rslt != 0) {
5874 if (rslt == ENOMEM) {
5875 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
5876 } else {
5877 fatal("pthread_getattr_np failed with error = %d", rslt);
5878 }
5879 }
5880
5881 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
5882 fatal("Cannot locate current stack attributes!");
5883 }
5884
5885 // Work around NPTL stack guard error.
5886 size_t guard_size = 0;
5887 rslt = pthread_attr_getguardsize(&attr, &guard_size);
5888 if (rslt != 0) {
5889 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
5890 }
5891 *bottom += guard_size;
5892 *size -= guard_size;
5893
5894 pthread_attr_destroy(&attr);
5895
5896 }
5897 assert(os::current_stack_pointer() >= *bottom &&
5898 os::current_stack_pointer() < *bottom + *size, "just checking");
5899 }
5900
5901 address os::current_stack_base() {
5902 address bottom;
5903 size_t size;
5904 current_stack_region(&bottom, &size);
5905 return (bottom + size);
5906 }
5907
5908 size_t os::current_stack_size() {
5909 // This stack size includes the usable stack and HotSpot guard pages
5910 // (for the threads that have Hotspot guard pages).
5911 address bottom;
5912 size_t size;
5913 current_stack_region(&bottom, &size);
5914 return size;
5915 }
5916 #endif
5917
5918 static inline struct timespec get_mtime(const char* filename) {
5919 struct stat st;
5920 int ret = os::stat(filename, &st);
5921 assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
5922 return st.st_mtim;
5923 }
5924
5925 int os::compare_file_modified_times(const char* file1, const char* file2) {
5926 struct timespec filetime1 = get_mtime(file1);
5927 struct timespec filetime2 = get_mtime(file2);
5928 int diff = filetime1.tv_sec - filetime2.tv_sec;
5929 if (diff == 0) {
5930 return filetime1.tv_nsec - filetime2.tv_nsec;
5931 }
5932 return diff;
5933 }
5934
5935 /////////////// Unit tests ///////////////
5936
5937 #ifndef PRODUCT
5938
5939 #define test_log(...) \
5940 do { \
5941 if (VerboseInternalVMTests) { \
5942 tty->print_cr(__VA_ARGS__); \
5943 tty->flush(); \
5944 } \
5945 } while (false)
5946
5947 class TestReserveMemorySpecial : AllStatic {
5948 public:
5949 static void small_page_write(void* addr, size_t size) {
5950 size_t page_size = os::vm_page_size();
5951
5952 char* end = (char*)addr + size;
5953 for (char* p = (char*)addr; p < end; p += page_size) {
5954 *p = 1;
5955 }
5956 }
5957
5958 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
5959 if (!UseHugeTLBFS) {
5960 return;
5961 }
5962
5963 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
5964
5965 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
5966
5967 if (addr != NULL) {
5968 small_page_write(addr, size);
5969
5970 os::Linux::release_memory_special_huge_tlbfs(addr, size);
5971 }
5972 }
5973
5974 static void test_reserve_memory_special_huge_tlbfs_only() {
5975 if (!UseHugeTLBFS) {
5976 return;
5977 }
5978
5979 size_t lp = os::large_page_size();
5980
5981 for (size_t size = lp; size <= lp * 10; size += lp) {
5982 test_reserve_memory_special_huge_tlbfs_only(size);
5983 }
5984 }
5985
5986 static void test_reserve_memory_special_huge_tlbfs_mixed() {
5987 size_t lp = os::large_page_size();
5988 size_t ag = os::vm_allocation_granularity();
5989
5990 // sizes to test
5991 const size_t sizes[] = {
5992 lp, lp + ag, lp + lp / 2, lp * 2,
5993 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
5994 lp * 10, lp * 10 + lp / 2
5995 };
5996 const int num_sizes = sizeof(sizes) / sizeof(size_t);
5997
5998 // For each size/alignment combination, we test three scenarios:
5999 // 1) with req_addr == NULL
6000 // 2) with a non-null req_addr at which we expect to successfully allocate
6001 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6002 // expect the allocation to either fail or to ignore req_addr
6003
6004 // Pre-allocate two areas; they shall be as large as the largest allocation
6005 // and aligned to the largest alignment we will be testing.
6006 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6007 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6008 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6009 -1, 0);
6010 assert(mapping1 != MAP_FAILED, "should work");
6011
6012 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6013 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6014 -1, 0);
6015 assert(mapping2 != MAP_FAILED, "should work");
6016
6017 // Unmap the first mapping, but leave the second mapping intact: the first
6018 // mapping will serve as a value for a "good" req_addr (case 2). The second
6019 // mapping, still intact, as "bad" req_addr (case 3).
6020 ::munmap(mapping1, mapping_size);
6021
6022 // Case 1
6023 test_log("%s, req_addr NULL:", __FUNCTION__);
6024 test_log("size align result");
6025
6026 for (int i = 0; i < num_sizes; i++) {
6027 const size_t size = sizes[i];
6028 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6029 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6030 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6031 size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6032 if (p != NULL) {
6033 assert(is_aligned(p, alignment), "must be");
6034 small_page_write(p, size);
6035 os::Linux::release_memory_special_huge_tlbfs(p, size);
6036 }
6037 }
6038 }
6039
6040 // Case 2
6041 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6042 test_log("size align req_addr result");
6043
6044 for (int i = 0; i < num_sizes; i++) {
6045 const size_t size = sizes[i];
6046 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6047 char* const req_addr = align_up(mapping1, alignment);
6048 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6049 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6050 size, alignment, p2i(req_addr), p2i(p),
6051 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6052 if (p != NULL) {
6053 assert(p == req_addr, "must be");
6054 small_page_write(p, size);
6055 os::Linux::release_memory_special_huge_tlbfs(p, size);
6056 }
6057 }
6058 }
6059
6060 // Case 3
6061 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6062 test_log("size align req_addr result");
6063
6064 for (int i = 0; i < num_sizes; i++) {
6065 const size_t size = sizes[i];
6066 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6067 char* const req_addr = align_up(mapping2, alignment);
6068 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6069 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6070 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6071 // as the area around req_addr contains already existing mappings, the API should always
6072 // return NULL (as per contract, it cannot return another address)
6073 assert(p == NULL, "must be");
6074 }
6075 }
6076
6077 ::munmap(mapping2, mapping_size);
6078
6079 }
6080
6081 static void test_reserve_memory_special_huge_tlbfs() {
6082 if (!UseHugeTLBFS) {
6083 return;
6084 }
6085
6086 test_reserve_memory_special_huge_tlbfs_only();
6087 test_reserve_memory_special_huge_tlbfs_mixed();
6088 }
6089
6090 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6091 if (!UseSHM) {
6092 return;
6093 }
6094
6095 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6096
6097 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6098
6099 if (addr != NULL) {
6100 assert(is_aligned(addr, alignment), "Check");
6101 assert(is_aligned(addr, os::large_page_size()), "Check");
6102
6103 small_page_write(addr, size);
6104
6105 os::Linux::release_memory_special_shm(addr, size);
6106 }
6107 }
6108
6109 static void test_reserve_memory_special_shm() {
6110 size_t lp = os::large_page_size();
6111 size_t ag = os::vm_allocation_granularity();
6112
6113 for (size_t size = ag; size < lp * 3; size += ag) {
6114 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6115 test_reserve_memory_special_shm(size, alignment);
6116 }
6117 }
6118 }
6119
6120 static void test() {
6121 test_reserve_memory_special_huge_tlbfs();
6122 test_reserve_memory_special_shm();
6123 }
6124 };
6125
6126 void TestReserveMemorySpecial_test() {
6127 TestReserveMemorySpecial::test();
6128 }
6129
6130 #endif