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