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.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::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
141 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
142 Mutex* os::Linux::_createThread_lock = NULL;
143 pthread_t os::Linux::_main_thread;
144 int os::Linux::_page_size = -1;
145 bool os::Linux::_supports_fast_thread_cpu_time = false;
146 uint32_t os::Linux::_os_version = 0;
147 const char * os::Linux::_glibc_version = NULL;
148 const char * os::Linux::_libpthread_version = NULL;
149
150 static jlong initial_time_count=0;
151
152 static int clock_tics_per_sec = 100;
153
154 // If the VM might have been created on the primordial thread, we need to resolve the
155 // primordial thread stack bounds and check if the current thread might be the
156 // primordial thread in places. If we know that the primordial thread is never used,
157 // such as when the VM was created by one of the standard java launchers, we can
158 // avoid this
159 static bool suppress_primordial_thread_resolution = false;
160
161 // For diagnostics to print a message once. see run_periodic_checks
162 static sigset_t check_signal_done;
163 static bool check_signals = true;
164
165 // Signal number used to suspend/resume a thread
166
167 // do not use any signal number less than SIGSEGV, see 4355769
168 static int SR_signum = SIGUSR2;
169 sigset_t SR_sigset;
170
171 // utility functions
172
173 static int SR_initialize();
174
175 julong os::available_memory() {
176 return Linux::available_memory();
177 }
178
179 julong os::Linux::available_memory() {
180 // values in struct sysinfo are "unsigned long"
181 struct sysinfo si;
182 julong avail_mem;
183
184 if (OSContainer::is_containerized()) {
185 jlong mem_limit, mem_usage;
186 if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
187 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
188 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
189 }
190 if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
191 log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
192 }
193 if (mem_limit > 0 && mem_usage > 0 ) {
194 avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
195 log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
196 return avail_mem;
197 }
198 }
199
200 sysinfo(&si);
201 avail_mem = (julong)si.freeram * si.mem_unit;
202 log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
203 return avail_mem;
204 }
205
206 julong os::physical_memory() {
207 jlong phys_mem = 0;
208 if (OSContainer::is_containerized()) {
209 jlong mem_limit;
210 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
211 log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
212 return mem_limit;
213 }
214 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
215 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
216 }
217
218 phys_mem = Linux::physical_memory();
219 log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
220 return phys_mem;
221 }
222
223 // Return true if user is running as root.
224
225 bool os::have_special_privileges() {
226 static bool init = false;
227 static bool privileges = false;
228 if (!init) {
229 privileges = (getuid() != geteuid()) || (getgid() != getegid());
230 init = true;
231 }
232 return privileges;
233 }
234
235
236 #ifndef SYS_gettid
237 // i386: 224, ia64: 1105, amd64: 186, sparc 143
238 #ifdef __ia64__
239 #define SYS_gettid 1105
240 #else
241 #ifdef __i386__
242 #define SYS_gettid 224
243 #else
244 #ifdef __amd64__
245 #define SYS_gettid 186
246 #else
247 #ifdef __sparc__
248 #define SYS_gettid 143
249 #else
250 #error define gettid for the arch
251 #endif
252 #endif
253 #endif
254 #endif
255 #endif
256
257
258 // pid_t gettid()
259 //
260 // Returns the kernel thread id of the currently running thread. Kernel
261 // thread id is used to access /proc.
262 pid_t os::Linux::gettid() {
263 int rslt = syscall(SYS_gettid);
264 assert(rslt != -1, "must be."); // old linuxthreads implementation?
265 return (pid_t)rslt;
266 }
267
268 // Most versions of linux have a bug where the number of processors are
269 // determined by looking at the /proc file system. In a chroot environment,
270 // the system call returns 1.
271 static bool unsafe_chroot_detected = false;
272 static const char *unstable_chroot_error = "/proc file system not found.\n"
273 "Java may be unstable running multithreaded in a chroot "
274 "environment on Linux when /proc filesystem is not mounted.";
275
276 void os::Linux::initialize_system_info() {
277 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
278 if (processor_count() == 1) {
279 pid_t pid = os::Linux::gettid();
280 char fname[32];
281 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
282 FILE *fp = fopen(fname, "r");
283 if (fp == NULL) {
284 unsafe_chroot_detected = true;
285 } else {
286 fclose(fp);
287 }
288 }
289 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
290 assert(processor_count() > 0, "linux error");
291 }
292
293 void os::init_system_properties_values() {
294 // The next steps are taken in the product version:
295 //
296 // Obtain the JAVA_HOME value from the location of libjvm.so.
297 // This library should be located at:
298 // <JAVA_HOME>/lib/{client|server}/libjvm.so.
299 //
300 // If "/jre/lib/" appears at the right place in the path, then we
301 // assume libjvm.so is installed in a JDK and we use this path.
302 //
303 // Otherwise exit with message: "Could not create the Java virtual machine."
304 //
305 // The following extra steps are taken in the debugging version:
306 //
307 // If "/jre/lib/" does NOT appear at the right place in the path
308 // instead of exit check for $JAVA_HOME environment variable.
309 //
310 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
311 // then we append a fake suffix "hotspot/libjvm.so" to this path so
312 // it looks like libjvm.so is installed there
313 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
314 //
315 // Otherwise exit.
316 //
317 // Important note: if the location of libjvm.so changes this
318 // code needs to be changed accordingly.
319
320 // See ld(1):
321 // The linker uses the following search paths to locate required
322 // shared libraries:
323 // 1: ...
324 // ...
325 // 7: The default directories, normally /lib and /usr/lib.
326 #ifndef OVERRIDE_LIBPATH
327 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
328 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
329 #else
330 #define DEFAULT_LIBPATH "/lib:/usr/lib"
331 #endif
332 #else
333 #define DEFAULT_LIBPATH OVERRIDE_LIBPATH
334 #endif
335
336 // Base path of extensions installed on the system.
337 #define SYS_EXT_DIR "/usr/java/packages"
338 #define EXTENSIONS_DIR "/lib/ext"
339
340 // Buffer that fits several sprintfs.
341 // Note that the space for the colon and the trailing null are provided
342 // by the nulls included by the sizeof operator.
343 const size_t bufsize =
344 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
345 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
346 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
347
348 // sysclasspath, java_home, dll_dir
349 {
350 char *pslash;
351 os::jvm_path(buf, bufsize);
352
353 // Found the full path to libjvm.so.
354 // Now cut the path to <java_home>/jre if we can.
355 pslash = strrchr(buf, '/');
356 if (pslash != NULL) {
357 *pslash = '\0'; // Get rid of /libjvm.so.
358 }
359 pslash = strrchr(buf, '/');
360 if (pslash != NULL) {
361 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
362 }
363 Arguments::set_dll_dir(buf);
364
365 if (pslash != NULL) {
366 pslash = strrchr(buf, '/');
367 if (pslash != NULL) {
368 *pslash = '\0'; // Get rid of /lib.
369 }
370 }
371 Arguments::set_java_home(buf);
372 if (!set_boot_path('/', ':')) {
373 vm_exit_during_initialization("Failed setting boot class path.", NULL);
374 }
375 }
376
377 // Where to look for native libraries.
378 //
379 // Note: Due to a legacy implementation, most of the library path
380 // is set in the launcher. This was to accomodate linking restrictions
381 // on legacy Linux implementations (which are no longer supported).
382 // Eventually, all the library path setting will be done here.
383 //
384 // However, to prevent the proliferation of improperly built native
385 // libraries, the new path component /usr/java/packages is added here.
386 // Eventually, all the library path setting will be done here.
387 {
388 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
389 // should always exist (until the legacy problem cited above is
390 // addressed).
391 const char *v = ::getenv("LD_LIBRARY_PATH");
392 const char *v_colon = ":";
393 if (v == NULL) { v = ""; v_colon = ""; }
394 // That's +1 for the colon and +1 for the trailing '\0'.
395 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
396 strlen(v) + 1 +
397 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
398 mtInternal);
399 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
400 Arguments::set_library_path(ld_library_path);
401 FREE_C_HEAP_ARRAY(char, ld_library_path);
402 }
403
404 // Extensions directories.
405 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
406 Arguments::set_ext_dirs(buf);
407
408 FREE_C_HEAP_ARRAY(char, buf);
409
410 #undef DEFAULT_LIBPATH
411 #undef SYS_EXT_DIR
412 #undef EXTENSIONS_DIR
413 }
414
415 ////////////////////////////////////////////////////////////////////////////////
416 // breakpoint support
417
418 void os::breakpoint() {
419 BREAKPOINT;
420 }
421
422 extern "C" void breakpoint() {
423 // use debugger to set breakpoint here
424 }
425
426 ////////////////////////////////////////////////////////////////////////////////
427 // signal support
428
429 debug_only(static bool signal_sets_initialized = false);
430 static sigset_t unblocked_sigs, vm_sigs;
431
432 void os::Linux::signal_sets_init() {
433 // Should also have an assertion stating we are still single-threaded.
434 assert(!signal_sets_initialized, "Already initialized");
435 // Fill in signals that are necessarily unblocked for all threads in
436 // the VM. Currently, we unblock the following signals:
437 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
438 // by -Xrs (=ReduceSignalUsage));
439 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
440 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
441 // the dispositions or masks wrt these signals.
442 // Programs embedding the VM that want to use the above signals for their
443 // own purposes must, at this time, use the "-Xrs" option to prevent
444 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
445 // (See bug 4345157, and other related bugs).
446 // In reality, though, unblocking these signals is really a nop, since
447 // these signals are not blocked by default.
448 sigemptyset(&unblocked_sigs);
449 sigaddset(&unblocked_sigs, SIGILL);
450 sigaddset(&unblocked_sigs, SIGSEGV);
451 sigaddset(&unblocked_sigs, SIGBUS);
452 sigaddset(&unblocked_sigs, SIGFPE);
453 #if defined(PPC64)
454 sigaddset(&unblocked_sigs, SIGTRAP);
455 #endif
456 sigaddset(&unblocked_sigs, SR_signum);
457
458 if (!ReduceSignalUsage) {
459 if (!os::Posix::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
460 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
461 }
462 if (!os::Posix::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
463 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
464 }
465 if (!os::Posix::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
466 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
467 }
468 }
469 // Fill in signals that are blocked by all but the VM thread.
470 sigemptyset(&vm_sigs);
471 if (!ReduceSignalUsage) {
472 sigaddset(&vm_sigs, BREAK_SIGNAL);
473 }
474 debug_only(signal_sets_initialized = true);
475
476 }
477
478 // These are signals that are unblocked while a thread is running Java.
479 // (For some reason, they get blocked by default.)
480 sigset_t* os::Linux::unblocked_signals() {
481 assert(signal_sets_initialized, "Not initialized");
482 return &unblocked_sigs;
483 }
484
485 // These are the signals that are blocked while a (non-VM) thread is
486 // running Java. Only the VM thread handles these signals.
487 sigset_t* os::Linux::vm_signals() {
488 assert(signal_sets_initialized, "Not initialized");
489 return &vm_sigs;
490 }
491
492 void os::Linux::hotspot_sigmask(Thread* thread) {
493
494 //Save caller's signal mask before setting VM signal mask
495 sigset_t caller_sigmask;
496 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
497
498 OSThread* osthread = thread->osthread();
499 osthread->set_caller_sigmask(caller_sigmask);
500
501 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
502
503 if (!ReduceSignalUsage) {
504 if (thread->is_VM_thread()) {
505 // Only the VM thread handles BREAK_SIGNAL ...
506 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
507 } else {
508 // ... all other threads block BREAK_SIGNAL
509 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
510 }
511 }
512 }
513
514 //////////////////////////////////////////////////////////////////////////////
515 // detecting pthread library
516
517 void os::Linux::libpthread_init() {
518 // Save glibc and pthread version strings.
519 #if !defined(_CS_GNU_LIBC_VERSION) || \
520 !defined(_CS_GNU_LIBPTHREAD_VERSION)
521 #error "glibc too old (< 2.3.2)"
522 #endif
523
524 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
525 assert(n > 0, "cannot retrieve glibc version");
526 char *str = (char *)malloc(n, mtInternal);
527 confstr(_CS_GNU_LIBC_VERSION, str, n);
528 os::Linux::set_glibc_version(str);
529
530 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
531 assert(n > 0, "cannot retrieve pthread version");
532 str = (char *)malloc(n, mtInternal);
533 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
534 os::Linux::set_libpthread_version(str);
535 }
536
537 /////////////////////////////////////////////////////////////////////////////
538 // thread stack expansion
539
540 // os::Linux::manually_expand_stack() takes care of expanding the thread
541 // stack. Note that this is normally not needed: pthread stacks allocate
542 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
543 // committed. Therefore it is not necessary to expand the stack manually.
544 //
545 // Manually expanding the stack was historically needed on LinuxThreads
546 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
547 // it is kept to deal with very rare corner cases:
548 //
549 // For one, user may run the VM on an own implementation of threads
550 // whose stacks are - like the old LinuxThreads - implemented using
551 // mmap(MAP_GROWSDOWN).
552 //
553 // Also, this coding may be needed if the VM is running on the primordial
554 // thread. Normally we avoid running on the primordial thread; however,
555 // user may still invoke the VM on the primordial thread.
556 //
557 // The following historical comment describes the details about running
558 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
559
560
561 // Force Linux kernel to expand current thread stack. If "bottom" is close
562 // to the stack guard, caller should block all signals.
563 //
564 // MAP_GROWSDOWN:
565 // A special mmap() flag that is used to implement thread stacks. It tells
566 // kernel that the memory region should extend downwards when needed. This
567 // allows early versions of LinuxThreads to only mmap the first few pages
568 // when creating a new thread. Linux kernel will automatically expand thread
569 // stack as needed (on page faults).
570 //
571 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
572 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
573 // region, it's hard to tell if the fault is due to a legitimate stack
574 // access or because of reading/writing non-exist memory (e.g. buffer
575 // overrun). As a rule, if the fault happens below current stack pointer,
576 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
577 // application (see Linux kernel fault.c).
578 //
579 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
580 // stack overflow detection.
581 //
582 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
583 // not use MAP_GROWSDOWN.
584 //
585 // To get around the problem and allow stack banging on Linux, we need to
586 // manually expand thread stack after receiving the SIGSEGV.
587 //
588 // There are two ways to expand thread stack to address "bottom", we used
589 // both of them in JVM before 1.5:
590 // 1. adjust stack pointer first so that it is below "bottom", and then
591 // touch "bottom"
592 // 2. mmap() the page in question
593 //
594 // Now alternate signal stack is gone, it's harder to use 2. For instance,
595 // if current sp is already near the lower end of page 101, and we need to
596 // call mmap() to map page 100, it is possible that part of the mmap() frame
597 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
598 // That will destroy the mmap() frame and cause VM to crash.
599 //
600 // The following code works by adjusting sp first, then accessing the "bottom"
601 // page to force a page fault. Linux kernel will then automatically expand the
602 // stack mapping.
603 //
604 // _expand_stack_to() assumes its frame size is less than page size, which
605 // should always be true if the function is not inlined.
606
607 static void NOINLINE _expand_stack_to(address bottom) {
608 address sp;
609 size_t size;
610 volatile char *p;
611
612 // Adjust bottom to point to the largest address within the same page, it
613 // gives us a one-page buffer if alloca() allocates slightly more memory.
614 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
615 bottom += os::Linux::page_size() - 1;
616
617 // sp might be slightly above current stack pointer; if that's the case, we
618 // will alloca() a little more space than necessary, which is OK. Don't use
619 // os::current_stack_pointer(), as its result can be slightly below current
620 // stack pointer, causing us to not alloca enough to reach "bottom".
621 sp = (address)&sp;
622
623 if (sp > bottom) {
624 size = sp - bottom;
625 p = (volatile char *)alloca(size);
626 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
627 p[0] = '\0';
628 }
629 }
630
631 void os::Linux::expand_stack_to(address bottom) {
632 _expand_stack_to(bottom);
633 }
634
635 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
636 assert(t!=NULL, "just checking");
637 assert(t->osthread()->expanding_stack(), "expand should be set");
638 assert(t->stack_base() != NULL, "stack_base was not initialized");
639
640 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
641 sigset_t mask_all, old_sigset;
642 sigfillset(&mask_all);
643 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
644 _expand_stack_to(addr);
645 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
646 return true;
647 }
648 return false;
649 }
650
651 //////////////////////////////////////////////////////////////////////////////
652 // create new thread
653
654 // Thread start routine for all newly created threads
655 static void *thread_native_entry(Thread *thread) {
656
657 thread->record_stack_base_and_size();
658
659 // Try to randomize the cache line index of hot stack frames.
660 // This helps when threads of the same stack traces evict each other's
661 // cache lines. The threads can be either from the same JVM instance, or
662 // from different JVM instances. The benefit is especially true for
663 // processors with hyperthreading technology.
664 static int counter = 0;
665 int pid = os::current_process_id();
666 alloca(((pid ^ counter++) & 7) * 128);
667
668 thread->initialize_thread_current();
669
670 OSThread* osthread = thread->osthread();
671 Monitor* sync = osthread->startThread_lock();
672
673 osthread->set_thread_id(os::current_thread_id());
674
675 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
676 os::current_thread_id(), (uintx) pthread_self());
677
678 if (UseNUMA) {
679 int lgrp_id = os::numa_get_group_id();
680 if (lgrp_id != -1) {
681 thread->set_lgrp_id(lgrp_id);
682 }
683 }
684 // initialize signal mask for this thread
685 os::Linux::hotspot_sigmask(thread);
686
687 // initialize floating point control register
688 os::Linux::init_thread_fpu_state();
689
690 // handshaking with parent thread
691 {
692 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
693
694 // notify parent thread
695 osthread->set_state(INITIALIZED);
696 sync->notify_all();
697
698 // wait until os::start_thread()
699 while (osthread->get_state() == INITIALIZED) {
700 sync->wait(Mutex::_no_safepoint_check_flag);
701 }
702 }
703
704 // call one more level start routine
705 thread->call_run();
706
707 // Note: at this point the thread object may already have deleted itself.
708 // Prevent dereferencing it from here on out.
709 thread = NULL;
710
711 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
712 os::current_thread_id(), (uintx) pthread_self());
713
714 return 0;
715 }
716
717 bool os::create_thread(Thread* thread, ThreadType thr_type,
718 size_t req_stack_size) {
719 assert(thread->osthread() == NULL, "caller responsible");
720
721 // Allocate the OSThread object
722 OSThread* osthread = new OSThread(NULL, NULL);
723 if (osthread == NULL) {
724 return false;
725 }
726
727 // set the correct thread state
728 osthread->set_thread_type(thr_type);
729
730 // Initial state is ALLOCATED but not INITIALIZED
731 osthread->set_state(ALLOCATED);
732
733 thread->set_osthread(osthread);
734
735 // init thread attributes
736 pthread_attr_t attr;
737 pthread_attr_init(&attr);
738 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
739
740 // Calculate stack size if it's not specified by caller.
741 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
742 // In the Linux NPTL pthread implementation the guard size mechanism
743 // is not implemented properly. The posix standard requires adding
744 // the size of the guard pages to the stack size, instead Linux
745 // takes the space out of 'stacksize'. Thus we adapt the requested
746 // stack_size by the size of the guard pages to mimick proper
747 // behaviour. However, be careful not to end up with a size
748 // of zero due to overflow. Don't add the guard page in that case.
749 size_t guard_size = os::Linux::default_guard_size(thr_type);
750 if (stack_size <= SIZE_MAX - guard_size) {
751 stack_size += guard_size;
752 }
753 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
754
755 int status = pthread_attr_setstacksize(&attr, stack_size);
756 assert_status(status == 0, status, "pthread_attr_setstacksize");
757
758 // Configure glibc guard page.
759 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
760
761 ThreadState state;
762
763 {
764 pthread_t tid;
765 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
766
767 char buf[64];
768 if (ret == 0) {
769 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
770 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
771 } else {
772 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
773 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
774 }
775
776 pthread_attr_destroy(&attr);
777
778 if (ret != 0) {
779 // Need to clean up stuff we've allocated so far
780 thread->set_osthread(NULL);
781 delete osthread;
782 return false;
783 }
784
785 // Store pthread info into the OSThread
786 osthread->set_pthread_id(tid);
787
788 // Wait until child thread is either initialized or aborted
789 {
790 Monitor* sync_with_child = osthread->startThread_lock();
791 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
792 while ((state = osthread->get_state()) == ALLOCATED) {
793 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
794 }
795 }
796 }
797
798 // Aborted due to thread limit being reached
799 if (state == ZOMBIE) {
800 thread->set_osthread(NULL);
801 delete osthread;
802 return false;
803 }
804
805 // The thread is returned suspended (in state INITIALIZED),
806 // and is started higher up in the call chain
807 assert(state == INITIALIZED, "race condition");
808 return true;
809 }
810
811 /////////////////////////////////////////////////////////////////////////////
812 // attach existing thread
813
814 // bootstrap the main thread
815 bool os::create_main_thread(JavaThread* thread) {
816 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
817 return create_attached_thread(thread);
818 }
819
820 bool os::create_attached_thread(JavaThread* thread) {
821 #ifdef ASSERT
822 thread->verify_not_published();
823 #endif
824
825 // Allocate the OSThread object
826 OSThread* osthread = new OSThread(NULL, NULL);
827
828 if (osthread == NULL) {
829 return false;
830 }
831
832 // Store pthread info into the OSThread
833 osthread->set_thread_id(os::Linux::gettid());
834 osthread->set_pthread_id(::pthread_self());
835
836 // initialize floating point control register
837 os::Linux::init_thread_fpu_state();
838
839 // Initial thread state is RUNNABLE
840 osthread->set_state(RUNNABLE);
841
842 thread->set_osthread(osthread);
843
844 if (UseNUMA) {
845 int lgrp_id = os::numa_get_group_id();
846 if (lgrp_id != -1) {
847 thread->set_lgrp_id(lgrp_id);
848 }
849 }
850
851 if (os::is_primordial_thread()) {
852 // If current thread is primordial thread, its stack is mapped on demand,
853 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
854 // the entire stack region to avoid SEGV in stack banging.
855 // It is also useful to get around the heap-stack-gap problem on SuSE
856 // kernel (see 4821821 for details). We first expand stack to the top
857 // of yellow zone, then enable stack yellow zone (order is significant,
858 // enabling yellow zone first will crash JVM on SuSE Linux), so there
859 // is no gap between the last two virtual memory regions.
860
861 JavaThread *jt = (JavaThread *)thread;
862 address addr = jt->stack_reserved_zone_base();
863 assert(addr != NULL, "initialization problem?");
864 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
865
866 osthread->set_expanding_stack();
867 os::Linux::manually_expand_stack(jt, addr);
868 osthread->clear_expanding_stack();
869 }
870
871 // initialize signal mask for this thread
872 // and save the caller's signal mask
873 os::Linux::hotspot_sigmask(thread);
874
875 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
876 os::current_thread_id(), (uintx) pthread_self());
877
878 return true;
879 }
880
881 void os::pd_start_thread(Thread* thread) {
882 OSThread * osthread = thread->osthread();
883 assert(osthread->get_state() != INITIALIZED, "just checking");
884 Monitor* sync_with_child = osthread->startThread_lock();
885 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
886 sync_with_child->notify();
887 }
888
889 // Free Linux resources related to the OSThread
890 void os::free_thread(OSThread* osthread) {
891 assert(osthread != NULL, "osthread not set");
892
893 // We are told to free resources of the argument thread,
894 // but we can only really operate on the current thread.
895 assert(Thread::current()->osthread() == osthread,
896 "os::free_thread but not current thread");
897
898 #ifdef ASSERT
899 sigset_t current;
900 sigemptyset(¤t);
901 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
902 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
903 #endif
904
905 // Restore caller's signal mask
906 sigset_t sigmask = osthread->caller_sigmask();
907 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
908
909 delete osthread;
910 }
911
912 //////////////////////////////////////////////////////////////////////////////
913 // primordial thread
914
915 // Check if current thread is the primordial thread, similar to Solaris thr_main.
916 bool os::is_primordial_thread(void) {
917 if (suppress_primordial_thread_resolution) {
918 return false;
919 }
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 #ifndef SUPPORTS_CLOCK_MONOTONIC
1177 #error "Build platform doesn't support clock_gettime and related functionality"
1178 #endif
1179
1180 // Time since start-up in seconds to a fine granularity.
1181 // Used by VMSelfDestructTimer and the MemProfiler.
1182 double os::elapsedTime() {
1183
1184 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1185 }
1186
1187 jlong os::elapsed_counter() {
1188 return javaTimeNanos() - initial_time_count;
1189 }
1190
1191 jlong os::elapsed_frequency() {
1192 return NANOSECS_PER_SEC; // nanosecond resolution
1193 }
1194
1195 bool os::supports_vtime() { return true; }
1196 bool os::enable_vtime() { return false; }
1197 bool os::vtime_enabled() { return false; }
1198
1199 double os::elapsedVTime() {
1200 struct rusage usage;
1201 int retval = getrusage(RUSAGE_THREAD, &usage);
1202 if (retval == 0) {
1203 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);
1204 } else {
1205 // better than nothing, but not much
1206 return elapsedTime();
1207 }
1208 }
1209
1210 jlong os::javaTimeMillis() {
1211 timeval time;
1212 int status = gettimeofday(&time, NULL);
1213 assert(status != -1, "linux error");
1214 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1215 }
1216
1217 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1218 timeval time;
1219 int status = gettimeofday(&time, NULL);
1220 assert(status != -1, "linux error");
1221 seconds = jlong(time.tv_sec);
1222 nanos = jlong(time.tv_usec) * 1000;
1223 }
1224
1225 void os::Linux::fast_thread_clock_init() {
1226 if (!UseLinuxPosixThreadCPUClocks) {
1227 return;
1228 }
1229 clockid_t clockid;
1230 struct timespec tp;
1231 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1232 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1233
1234 // Switch to using fast clocks for thread cpu time if
1235 // the clock_getres() returns 0 error code.
1236 // Note, that some kernels may support the current thread
1237 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1238 // returned by the pthread_getcpuclockid().
1239 // If the fast Posix clocks are supported then the clock_getres()
1240 // must return at least tp.tv_sec == 0 which means a resolution
1241 // better than 1 sec. This is extra check for reliability.
1242
1243 if (pthread_getcpuclockid_func &&
1244 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1245 os::Posix::clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1246 _supports_fast_thread_cpu_time = true;
1247 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1248 }
1249 }
1250
1251 jlong os::javaTimeNanos() {
1252 if (os::supports_monotonic_clock()) {
1253 struct timespec tp;
1254 int status = os::Posix::clock_gettime(CLOCK_MONOTONIC, &tp);
1255 assert(status == 0, "gettime error");
1256 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1257 return result;
1258 } else {
1259 timeval time;
1260 int status = gettimeofday(&time, NULL);
1261 assert(status != -1, "linux error");
1262 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1263 return 1000 * usecs;
1264 }
1265 }
1266
1267 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1268 if (os::supports_monotonic_clock()) {
1269 info_ptr->max_value = ALL_64_BITS;
1270
1271 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1272 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1273 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1274 } else {
1275 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1276 info_ptr->max_value = ALL_64_BITS;
1277
1278 // gettimeofday is a real time clock so it skips
1279 info_ptr->may_skip_backward = true;
1280 info_ptr->may_skip_forward = true;
1281 }
1282
1283 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1284 }
1285
1286 // Return the real, user, and system times in seconds from an
1287 // arbitrary fixed point in the past.
1288 bool os::getTimesSecs(double* process_real_time,
1289 double* process_user_time,
1290 double* process_system_time) {
1291 struct tms ticks;
1292 clock_t real_ticks = times(&ticks);
1293
1294 if (real_ticks == (clock_t) (-1)) {
1295 return false;
1296 } else {
1297 double ticks_per_second = (double) clock_tics_per_sec;
1298 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1299 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1300 *process_real_time = ((double) real_ticks) / ticks_per_second;
1301
1302 return true;
1303 }
1304 }
1305
1306
1307 char * os::local_time_string(char *buf, size_t buflen) {
1308 struct tm t;
1309 time_t long_time;
1310 time(&long_time);
1311 localtime_r(&long_time, &t);
1312 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1313 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1314 t.tm_hour, t.tm_min, t.tm_sec);
1315 return buf;
1316 }
1317
1318 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1319 return localtime_r(clock, res);
1320 }
1321
1322 ////////////////////////////////////////////////////////////////////////////////
1323 // runtime exit support
1324
1325 // Note: os::shutdown() might be called very early during initialization, or
1326 // called from signal handler. Before adding something to os::shutdown(), make
1327 // sure it is async-safe and can handle partially initialized VM.
1328 void os::shutdown() {
1329
1330 // allow PerfMemory to attempt cleanup of any persistent resources
1331 perfMemory_exit();
1332
1333 // needs to remove object in file system
1334 AttachListener::abort();
1335
1336 // flush buffered output, finish log files
1337 ostream_abort();
1338
1339 // Check for abort hook
1340 abort_hook_t abort_hook = Arguments::abort_hook();
1341 if (abort_hook != NULL) {
1342 abort_hook();
1343 }
1344
1345 }
1346
1347 // Note: os::abort() might be called very early during initialization, or
1348 // called from signal handler. Before adding something to os::abort(), make
1349 // sure it is async-safe and can handle partially initialized VM.
1350 void os::abort(bool dump_core, void* siginfo, const void* context) {
1351 os::shutdown();
1352 if (dump_core) {
1353 #ifndef PRODUCT
1354 fdStream out(defaultStream::output_fd());
1355 out.print_raw("Current thread is ");
1356 char buf[16];
1357 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1358 out.print_raw_cr(buf);
1359 out.print_raw_cr("Dumping core ...");
1360 #endif
1361 ::abort(); // dump core
1362 }
1363
1364 ::exit(1);
1365 }
1366
1367 // Die immediately, no exit hook, no abort hook, no cleanup.
1368 void os::die() {
1369 ::abort();
1370 }
1371
1372 // thread_id is kernel thread id (similar to Solaris LWP id)
1373 intx os::current_thread_id() { return os::Linux::gettid(); }
1374 int os::current_process_id() {
1375 return ::getpid();
1376 }
1377
1378 // DLL functions
1379
1380 const char* os::dll_file_extension() { return ".so"; }
1381
1382 // This must be hard coded because it's the system's temporary
1383 // directory not the java application's temp directory, ala java.io.tmpdir.
1384 const char* os::get_temp_directory() { return "/tmp"; }
1385
1386 static bool file_exists(const char* filename) {
1387 struct stat statbuf;
1388 if (filename == NULL || strlen(filename) == 0) {
1389 return false;
1390 }
1391 return os::stat(filename, &statbuf) == 0;
1392 }
1393
1394 // check if addr is inside libjvm.so
1395 bool os::address_is_in_vm(address addr) {
1396 static address libjvm_base_addr;
1397 Dl_info dlinfo;
1398
1399 if (libjvm_base_addr == NULL) {
1400 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1401 libjvm_base_addr = (address)dlinfo.dli_fbase;
1402 }
1403 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1404 }
1405
1406 if (dladdr((void *)addr, &dlinfo) != 0) {
1407 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1408 }
1409
1410 return false;
1411 }
1412
1413 bool os::dll_address_to_function_name(address addr, char *buf,
1414 int buflen, int *offset,
1415 bool demangle) {
1416 // buf is not optional, but offset is optional
1417 assert(buf != NULL, "sanity check");
1418
1419 Dl_info dlinfo;
1420
1421 if (dladdr((void*)addr, &dlinfo) != 0) {
1422 // see if we have a matching symbol
1423 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1424 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1425 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1426 }
1427 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1428 return true;
1429 }
1430 // no matching symbol so try for just file info
1431 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1432 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1433 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1434 return true;
1435 }
1436 }
1437 }
1438
1439 buf[0] = '\0';
1440 if (offset != NULL) *offset = -1;
1441 return false;
1442 }
1443
1444 struct _address_to_library_name {
1445 address addr; // input : memory address
1446 size_t buflen; // size of fname
1447 char* fname; // output: library name
1448 address base; // library base addr
1449 };
1450
1451 static int address_to_library_name_callback(struct dl_phdr_info *info,
1452 size_t size, void *data) {
1453 int i;
1454 bool found = false;
1455 address libbase = NULL;
1456 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1457
1458 // iterate through all loadable segments
1459 for (i = 0; i < info->dlpi_phnum; i++) {
1460 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1461 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1462 // base address of a library is the lowest address of its loaded
1463 // segments.
1464 if (libbase == NULL || libbase > segbase) {
1465 libbase = segbase;
1466 }
1467 // see if 'addr' is within current segment
1468 if (segbase <= d->addr &&
1469 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1470 found = true;
1471 }
1472 }
1473 }
1474
1475 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1476 // so dll_address_to_library_name() can fall through to use dladdr() which
1477 // can figure out executable name from argv[0].
1478 if (found && info->dlpi_name && info->dlpi_name[0]) {
1479 d->base = libbase;
1480 if (d->fname) {
1481 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1482 }
1483 return 1;
1484 }
1485 return 0;
1486 }
1487
1488 bool os::dll_address_to_library_name(address addr, char* buf,
1489 int buflen, int* offset) {
1490 // buf is not optional, but offset is optional
1491 assert(buf != NULL, "sanity check");
1492
1493 Dl_info dlinfo;
1494 struct _address_to_library_name data;
1495
1496 // There is a bug in old glibc dladdr() implementation that it could resolve
1497 // to wrong library name if the .so file has a base address != NULL. Here
1498 // we iterate through the program headers of all loaded libraries to find
1499 // out which library 'addr' really belongs to. This workaround can be
1500 // removed once the minimum requirement for glibc is moved to 2.3.x.
1501 data.addr = addr;
1502 data.fname = buf;
1503 data.buflen = buflen;
1504 data.base = NULL;
1505 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1506
1507 if (rslt) {
1508 // buf already contains library name
1509 if (offset) *offset = addr - data.base;
1510 return true;
1511 }
1512 if (dladdr((void*)addr, &dlinfo) != 0) {
1513 if (dlinfo.dli_fname != NULL) {
1514 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1515 }
1516 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1517 *offset = addr - (address)dlinfo.dli_fbase;
1518 }
1519 return true;
1520 }
1521
1522 buf[0] = '\0';
1523 if (offset) *offset = -1;
1524 return false;
1525 }
1526
1527 // Loads .dll/.so and
1528 // in case of error it checks if .dll/.so was built for the
1529 // same architecture as Hotspot is running on
1530
1531
1532 // Remember the stack's state. The Linux dynamic linker will change
1533 // the stack to 'executable' at most once, so we must safepoint only once.
1534 bool os::Linux::_stack_is_executable = false;
1535
1536 // VM operation that loads a library. This is necessary if stack protection
1537 // of the Java stacks can be lost during loading the library. If we
1538 // do not stop the Java threads, they can stack overflow before the stacks
1539 // are protected again.
1540 class VM_LinuxDllLoad: public VM_Operation {
1541 private:
1542 const char *_filename;
1543 char *_ebuf;
1544 int _ebuflen;
1545 void *_lib;
1546 public:
1547 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1548 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1549 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1550 void doit() {
1551 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1552 os::Linux::_stack_is_executable = true;
1553 }
1554 void* loaded_library() { return _lib; }
1555 };
1556
1557 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1558 void * result = NULL;
1559 bool load_attempted = false;
1560
1561 // Check whether the library to load might change execution rights
1562 // of the stack. If they are changed, the protection of the stack
1563 // guard pages will be lost. We need a safepoint to fix this.
1564 //
1565 // See Linux man page execstack(8) for more info.
1566 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1567 if (!ElfFile::specifies_noexecstack(filename)) {
1568 if (!is_init_completed()) {
1569 os::Linux::_stack_is_executable = true;
1570 // This is OK - No Java threads have been created yet, and hence no
1571 // stack guard pages to fix.
1572 //
1573 // Dynamic loader will make all stacks executable after
1574 // this function returns, and will not do that again.
1575 assert(Threads::number_of_threads() == 0, "no Java threads should exist yet.");
1576 } else {
1577 warning("You have loaded library %s which might have disabled stack guard. "
1578 "The VM will try to fix the stack guard now.\n"
1579 "It's highly recommended that you fix the library with "
1580 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1581 filename);
1582
1583 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1584 JavaThread *jt = JavaThread::current();
1585 if (jt->thread_state() != _thread_in_native) {
1586 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1587 // that requires ExecStack. Cannot enter safe point. Let's give up.
1588 warning("Unable to fix stack guard. Giving up.");
1589 } else {
1590 if (!LoadExecStackDllInVMThread) {
1591 // This is for the case where the DLL has an static
1592 // constructor function that executes JNI code. We cannot
1593 // load such DLLs in the VMThread.
1594 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1595 }
1596
1597 ThreadInVMfromNative tiv(jt);
1598 debug_only(VMNativeEntryWrapper vew;)
1599
1600 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1601 VMThread::execute(&op);
1602 if (LoadExecStackDllInVMThread) {
1603 result = op.loaded_library();
1604 }
1605 load_attempted = true;
1606 }
1607 }
1608 }
1609 }
1610
1611 if (!load_attempted) {
1612 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1613 }
1614
1615 if (result != NULL) {
1616 // Successful loading
1617 return result;
1618 }
1619
1620 Elf32_Ehdr elf_head;
1621 int diag_msg_max_length=ebuflen-strlen(ebuf);
1622 char* diag_msg_buf=ebuf+strlen(ebuf);
1623
1624 if (diag_msg_max_length==0) {
1625 // No more space in ebuf for additional diagnostics message
1626 return NULL;
1627 }
1628
1629
1630 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1631
1632 if (file_descriptor < 0) {
1633 // Can't open library, report dlerror() message
1634 return NULL;
1635 }
1636
1637 bool failed_to_read_elf_head=
1638 (sizeof(elf_head)!=
1639 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1640
1641 ::close(file_descriptor);
1642 if (failed_to_read_elf_head) {
1643 // file i/o error - report dlerror() msg
1644 return NULL;
1645 }
1646
1647 typedef struct {
1648 Elf32_Half code; // Actual value as defined in elf.h
1649 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1650 unsigned char elf_class; // 32 or 64 bit
1651 unsigned char endianess; // MSB or LSB
1652 char* name; // String representation
1653 } arch_t;
1654
1655 #ifndef EM_486
1656 #define EM_486 6 /* Intel 80486 */
1657 #endif
1658 #ifndef EM_AARCH64
1659 #define EM_AARCH64 183 /* ARM AARCH64 */
1660 #endif
1661
1662 static const arch_t arch_array[]={
1663 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1664 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1665 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1666 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1667 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1668 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1669 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1670 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1671 #if defined(VM_LITTLE_ENDIAN)
1672 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1673 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1674 #else
1675 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1676 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1677 #endif
1678 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1679 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1680 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1681 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1682 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1683 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1684 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1685 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1686 };
1687
1688 #if (defined IA32)
1689 static Elf32_Half running_arch_code=EM_386;
1690 #elif (defined AMD64) || (defined X32)
1691 static Elf32_Half running_arch_code=EM_X86_64;
1692 #elif (defined IA64)
1693 static Elf32_Half running_arch_code=EM_IA_64;
1694 #elif (defined __sparc) && (defined _LP64)
1695 static Elf32_Half running_arch_code=EM_SPARCV9;
1696 #elif (defined __sparc) && (!defined _LP64)
1697 static Elf32_Half running_arch_code=EM_SPARC;
1698 #elif (defined __powerpc64__)
1699 static Elf32_Half running_arch_code=EM_PPC64;
1700 #elif (defined __powerpc__)
1701 static Elf32_Half running_arch_code=EM_PPC;
1702 #elif (defined AARCH64)
1703 static Elf32_Half running_arch_code=EM_AARCH64;
1704 #elif (defined ARM)
1705 static Elf32_Half running_arch_code=EM_ARM;
1706 #elif (defined S390)
1707 static Elf32_Half running_arch_code=EM_S390;
1708 #elif (defined ALPHA)
1709 static Elf32_Half running_arch_code=EM_ALPHA;
1710 #elif (defined MIPSEL)
1711 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1712 #elif (defined PARISC)
1713 static Elf32_Half running_arch_code=EM_PARISC;
1714 #elif (defined MIPS)
1715 static Elf32_Half running_arch_code=EM_MIPS;
1716 #elif (defined M68K)
1717 static Elf32_Half running_arch_code=EM_68K;
1718 #elif (defined SH)
1719 static Elf32_Half running_arch_code=EM_SH;
1720 #else
1721 #error Method os::dll_load requires that one of following is defined:\
1722 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1723 #endif
1724
1725 // Identify compatability class for VM's architecture and library's architecture
1726 // Obtain string descriptions for architectures
1727
1728 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1729 int running_arch_index=-1;
1730
1731 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1732 if (running_arch_code == arch_array[i].code) {
1733 running_arch_index = i;
1734 }
1735 if (lib_arch.code == arch_array[i].code) {
1736 lib_arch.compat_class = arch_array[i].compat_class;
1737 lib_arch.name = arch_array[i].name;
1738 }
1739 }
1740
1741 assert(running_arch_index != -1,
1742 "Didn't find running architecture code (running_arch_code) in arch_array");
1743 if (running_arch_index == -1) {
1744 // Even though running architecture detection failed
1745 // we may still continue with reporting dlerror() message
1746 return NULL;
1747 }
1748
1749 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1750 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1751 return NULL;
1752 }
1753
1754 #ifndef S390
1755 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1756 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1757 return NULL;
1758 }
1759 #endif // !S390
1760
1761 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1762 if (lib_arch.name!=NULL) {
1763 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1764 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1765 lib_arch.name, arch_array[running_arch_index].name);
1766 } else {
1767 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1768 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1769 lib_arch.code,
1770 arch_array[running_arch_index].name);
1771 }
1772 }
1773
1774 return NULL;
1775 }
1776
1777 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1778 int ebuflen) {
1779 void * result = ::dlopen(filename, RTLD_LAZY);
1780 if (result == NULL) {
1781 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1782 ebuf[ebuflen-1] = '\0';
1783 }
1784 return result;
1785 }
1786
1787 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1788 int ebuflen) {
1789 void * result = NULL;
1790 if (LoadExecStackDllInVMThread) {
1791 result = dlopen_helper(filename, ebuf, ebuflen);
1792 }
1793
1794 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1795 // library that requires an executable stack, or which does not have this
1796 // stack attribute set, dlopen changes the stack attribute to executable. The
1797 // read protection of the guard pages gets lost.
1798 //
1799 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1800 // may have been queued at the same time.
1801
1802 if (!_stack_is_executable) {
1803 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
1804 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1805 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1806 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1807 warning("Attempt to reguard stack yellow zone failed.");
1808 }
1809 }
1810 }
1811 }
1812
1813 return result;
1814 }
1815
1816 void* os::dll_lookup(void* handle, const char* name) {
1817 void* res = dlsym(handle, name);
1818 return res;
1819 }
1820
1821 void* os::get_default_process_handle() {
1822 return (void*)::dlopen(NULL, RTLD_LAZY);
1823 }
1824
1825 static bool _print_ascii_file(const char* filename, outputStream* st, const char* hdr = NULL) {
1826 int fd = ::open(filename, O_RDONLY);
1827 if (fd == -1) {
1828 return false;
1829 }
1830
1831 if (hdr != NULL) {
1832 st->print_cr("%s", hdr);
1833 }
1834
1835 char buf[33];
1836 int bytes;
1837 buf[32] = '\0';
1838 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1839 st->print_raw(buf, bytes);
1840 }
1841
1842 ::close(fd);
1843
1844 return true;
1845 }
1846
1847 void os::print_dll_info(outputStream *st) {
1848 st->print_cr("Dynamic libraries:");
1849
1850 char fname[32];
1851 pid_t pid = os::Linux::gettid();
1852
1853 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1854
1855 if (!_print_ascii_file(fname, st)) {
1856 st->print("Can not get library information for pid = %d\n", pid);
1857 }
1858 }
1859
1860 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1861 FILE *procmapsFile = NULL;
1862
1863 // Open the procfs maps file for the current process
1864 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1865 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1866 char line[PATH_MAX + 100];
1867
1868 // Read line by line from 'file'
1869 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1870 u8 base, top, offset, inode;
1871 char permissions[5];
1872 char device[6];
1873 char name[PATH_MAX + 1];
1874
1875 // Parse fields from line
1876 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1877 &base, &top, permissions, &offset, device, &inode, name);
1878
1879 // Filter by device id '00:00' so that we only get file system mapped files.
1880 if (strcmp(device, "00:00") != 0) {
1881
1882 // Call callback with the fields of interest
1883 if(callback(name, (address)base, (address)top, param)) {
1884 // Oops abort, callback aborted
1885 fclose(procmapsFile);
1886 return 1;
1887 }
1888 }
1889 }
1890 fclose(procmapsFile);
1891 }
1892 return 0;
1893 }
1894
1895 void os::print_os_info_brief(outputStream* st) {
1896 os::Linux::print_distro_info(st);
1897
1898 os::Posix::print_uname_info(st);
1899
1900 os::Linux::print_libversion_info(st);
1901
1902 }
1903
1904 void os::print_os_info(outputStream* st) {
1905 st->print("OS:");
1906
1907 os::Linux::print_distro_info(st);
1908
1909 os::Posix::print_uname_info(st);
1910
1911 // Print warning if unsafe chroot environment detected
1912 if (unsafe_chroot_detected) {
1913 st->print("WARNING!! ");
1914 st->print_cr("%s", unstable_chroot_error);
1915 }
1916
1917 os::Linux::print_libversion_info(st);
1918
1919 os::Posix::print_rlimit_info(st);
1920
1921 os::Posix::print_load_average(st);
1922
1923 os::Linux::print_full_memory_info(st);
1924
1925 os::Linux::print_proc_sys_info(st);
1926
1927 os::Linux::print_ld_preload_file(st);
1928
1929 os::Linux::print_container_info(st);
1930 }
1931
1932 // Try to identify popular distros.
1933 // Most Linux distributions have a /etc/XXX-release file, which contains
1934 // the OS version string. Newer Linux distributions have a /etc/lsb-release
1935 // file that also contains the OS version string. Some have more than one
1936 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
1937 // /etc/redhat-release.), so the order is important.
1938 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
1939 // their own specific XXX-release file as well as a redhat-release file.
1940 // Because of this the XXX-release file needs to be searched for before the
1941 // redhat-release file.
1942 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
1943 // search for redhat-release / SuSE-release needs to be before lsb-release.
1944 // Since the lsb-release file is the new standard it needs to be searched
1945 // before the older style release files.
1946 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
1947 // next to last resort. The os-release file is a new standard that contains
1948 // distribution information and the system-release file seems to be an old
1949 // standard that has been replaced by the lsb-release and os-release files.
1950 // Searching for the debian_version file is the last resort. It contains
1951 // an informative string like "6.0.6" or "wheezy/sid". Because of this
1952 // "Debian " is printed before the contents of the debian_version file.
1953
1954 const char* distro_files[] = {
1955 "/etc/oracle-release",
1956 "/etc/mandriva-release",
1957 "/etc/mandrake-release",
1958 "/etc/sun-release",
1959 "/etc/redhat-release",
1960 "/etc/SuSE-release",
1961 "/etc/lsb-release",
1962 "/etc/turbolinux-release",
1963 "/etc/gentoo-release",
1964 "/etc/ltib-release",
1965 "/etc/angstrom-version",
1966 "/etc/system-release",
1967 "/etc/os-release",
1968 NULL };
1969
1970 void os::Linux::print_distro_info(outputStream* st) {
1971 for (int i = 0;; i++) {
1972 const char* file = distro_files[i];
1973 if (file == NULL) {
1974 break; // done
1975 }
1976 // If file prints, we found it.
1977 if (_print_ascii_file(file, st)) {
1978 return;
1979 }
1980 }
1981
1982 if (file_exists("/etc/debian_version")) {
1983 st->print("Debian ");
1984 _print_ascii_file("/etc/debian_version", st);
1985 } else {
1986 st->print("Linux");
1987 }
1988 st->cr();
1989 }
1990
1991 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
1992 char buf[256];
1993 while (fgets(buf, sizeof(buf), fp)) {
1994 // Edit out extra stuff in expected format
1995 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
1996 char* ptr = strstr(buf, "\""); // the name is in quotes
1997 if (ptr != NULL) {
1998 ptr++; // go beyond first quote
1999 char* nl = strchr(ptr, '\"');
2000 if (nl != NULL) *nl = '\0';
2001 strncpy(distro, ptr, length);
2002 } else {
2003 ptr = strstr(buf, "=");
2004 ptr++; // go beyond equals then
2005 char* nl = strchr(ptr, '\n');
2006 if (nl != NULL) *nl = '\0';
2007 strncpy(distro, ptr, length);
2008 }
2009 return;
2010 } else if (get_first_line) {
2011 char* nl = strchr(buf, '\n');
2012 if (nl != NULL) *nl = '\0';
2013 strncpy(distro, buf, length);
2014 return;
2015 }
2016 }
2017 // print last line and close
2018 char* nl = strchr(buf, '\n');
2019 if (nl != NULL) *nl = '\0';
2020 strncpy(distro, buf, length);
2021 }
2022
2023 static void parse_os_info(char* distro, size_t length, const char* file) {
2024 FILE* fp = fopen(file, "r");
2025 if (fp != NULL) {
2026 // if suse format, print out first line
2027 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2028 parse_os_info_helper(fp, distro, length, get_first_line);
2029 fclose(fp);
2030 }
2031 }
2032
2033 void os::get_summary_os_info(char* buf, size_t buflen) {
2034 for (int i = 0;; i++) {
2035 const char* file = distro_files[i];
2036 if (file == NULL) {
2037 break; // ran out of distro_files
2038 }
2039 if (file_exists(file)) {
2040 parse_os_info(buf, buflen, file);
2041 return;
2042 }
2043 }
2044 // special case for debian
2045 if (file_exists("/etc/debian_version")) {
2046 strncpy(buf, "Debian ", buflen);
2047 if (buflen > 7) {
2048 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2049 }
2050 } else {
2051 strncpy(buf, "Linux", buflen);
2052 }
2053 }
2054
2055 void os::Linux::print_libversion_info(outputStream* st) {
2056 // libc, pthread
2057 st->print("libc:");
2058 st->print("%s ", os::Linux::glibc_version());
2059 st->print("%s ", os::Linux::libpthread_version());
2060 st->cr();
2061 }
2062
2063 void os::Linux::print_proc_sys_info(outputStream* st) {
2064 st->cr();
2065 st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):");
2066 _print_ascii_file("/proc/sys/kernel/threads-max", st);
2067 st->cr();
2068 st->cr();
2069
2070 st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):");
2071 _print_ascii_file("/proc/sys/vm/max_map_count", st);
2072 st->cr();
2073 st->cr();
2074
2075 st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):");
2076 _print_ascii_file("/proc/sys/kernel/pid_max", st);
2077 st->cr();
2078 st->cr();
2079 }
2080
2081 void os::Linux::print_full_memory_info(outputStream* st) {
2082 st->print("\n/proc/meminfo:\n");
2083 _print_ascii_file("/proc/meminfo", st);
2084 st->cr();
2085 }
2086
2087 void os::Linux::print_ld_preload_file(outputStream* st) {
2088 _print_ascii_file("/etc/ld.so.preload", st, "\n/etc/ld.so.preload:");
2089 st->cr();
2090 }
2091
2092 void os::Linux::print_container_info(outputStream* st) {
2093 if (!OSContainer::is_containerized()) {
2094 return;
2095 }
2096
2097 st->print("container (cgroup) information:\n");
2098
2099 const char *p_ct = OSContainer::container_type();
2100 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed");
2101
2102 char *p = OSContainer::cpu_cpuset_cpus();
2103 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed");
2104 free(p);
2105
2106 p = OSContainer::cpu_cpuset_memory_nodes();
2107 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed");
2108 free(p);
2109
2110 int i = OSContainer::active_processor_count();
2111 if (i > 0) {
2112 st->print("active_processor_count: %d\n", i);
2113 } else {
2114 st->print("active_processor_count: failed\n");
2115 }
2116
2117 i = OSContainer::cpu_quota();
2118 st->print("cpu_quota: %d\n", i);
2119
2120 i = OSContainer::cpu_period();
2121 st->print("cpu_period: %d\n", i);
2122
2123 i = OSContainer::cpu_shares();
2124 st->print("cpu_shares: %d\n", i);
2125
2126 jlong j = OSContainer::memory_limit_in_bytes();
2127 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2128
2129 j = OSContainer::memory_and_swap_limit_in_bytes();
2130 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2131
2132 j = OSContainer::memory_soft_limit_in_bytes();
2133 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2134
2135 j = OSContainer::OSContainer::memory_usage_in_bytes();
2136 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2137
2138 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2139 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2140 st->cr();
2141 }
2142
2143 void os::print_memory_info(outputStream* st) {
2144
2145 st->print("Memory:");
2146 st->print(" %dk page", os::vm_page_size()>>10);
2147
2148 // values in struct sysinfo are "unsigned long"
2149 struct sysinfo si;
2150 sysinfo(&si);
2151
2152 st->print(", physical " UINT64_FORMAT "k",
2153 os::physical_memory() >> 10);
2154 st->print("(" UINT64_FORMAT "k free)",
2155 os::available_memory() >> 10);
2156 st->print(", swap " UINT64_FORMAT "k",
2157 ((jlong)si.totalswap * si.mem_unit) >> 10);
2158 st->print("(" UINT64_FORMAT "k free)",
2159 ((jlong)si.freeswap * si.mem_unit) >> 10);
2160 st->cr();
2161 }
2162
2163 // Print the first "model name" line and the first "flags" line
2164 // that we find and nothing more. We assume "model name" comes
2165 // before "flags" so if we find a second "model name", then the
2166 // "flags" field is considered missing.
2167 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2168 #if defined(IA32) || defined(AMD64)
2169 // Other platforms have less repetitive cpuinfo files
2170 FILE *fp = fopen("/proc/cpuinfo", "r");
2171 if (fp) {
2172 while (!feof(fp)) {
2173 if (fgets(buf, buflen, fp)) {
2174 // Assume model name comes before flags
2175 bool model_name_printed = false;
2176 if (strstr(buf, "model name") != NULL) {
2177 if (!model_name_printed) {
2178 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2179 st->print_raw(buf);
2180 model_name_printed = true;
2181 } else {
2182 // model name printed but not flags? Odd, just return
2183 fclose(fp);
2184 return true;
2185 }
2186 }
2187 // print the flags line too
2188 if (strstr(buf, "flags") != NULL) {
2189 st->print_raw(buf);
2190 fclose(fp);
2191 return true;
2192 }
2193 }
2194 }
2195 fclose(fp);
2196 }
2197 #endif // x86 platforms
2198 return false;
2199 }
2200
2201 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2202 // Only print the model name if the platform provides this as a summary
2203 if (!print_model_name_and_flags(st, buf, buflen)) {
2204 st->print("\n/proc/cpuinfo:\n");
2205 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2206 st->print_cr(" <Not Available>");
2207 }
2208 }
2209 }
2210
2211 #if defined(AMD64) || defined(IA32) || defined(X32)
2212 const char* search_string = "model name";
2213 #elif defined(M68K)
2214 const char* search_string = "CPU";
2215 #elif defined(PPC64)
2216 const char* search_string = "cpu";
2217 #elif defined(S390)
2218 const char* search_string = "processor";
2219 #elif defined(SPARC)
2220 const char* search_string = "cpu";
2221 #else
2222 const char* search_string = "Processor";
2223 #endif
2224
2225 // Parses the cpuinfo file for string representing the model name.
2226 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2227 FILE* fp = fopen("/proc/cpuinfo", "r");
2228 if (fp != NULL) {
2229 while (!feof(fp)) {
2230 char buf[256];
2231 if (fgets(buf, sizeof(buf), fp)) {
2232 char* start = strstr(buf, search_string);
2233 if (start != NULL) {
2234 char *ptr = start + strlen(search_string);
2235 char *end = buf + strlen(buf);
2236 while (ptr != end) {
2237 // skip whitespace and colon for the rest of the name.
2238 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2239 break;
2240 }
2241 ptr++;
2242 }
2243 if (ptr != end) {
2244 // reasonable string, get rid of newline and keep the rest
2245 char* nl = strchr(buf, '\n');
2246 if (nl != NULL) *nl = '\0';
2247 strncpy(cpuinfo, ptr, length);
2248 fclose(fp);
2249 return;
2250 }
2251 }
2252 }
2253 }
2254 fclose(fp);
2255 }
2256 // cpuinfo not found or parsing failed, just print generic string. The entire
2257 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2258 #if defined(AARCH64)
2259 strncpy(cpuinfo, "AArch64", length);
2260 #elif defined(AMD64)
2261 strncpy(cpuinfo, "x86_64", length);
2262 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2263 strncpy(cpuinfo, "ARM", length);
2264 #elif defined(IA32)
2265 strncpy(cpuinfo, "x86_32", length);
2266 #elif defined(IA64)
2267 strncpy(cpuinfo, "IA64", length);
2268 #elif defined(PPC)
2269 strncpy(cpuinfo, "PPC64", length);
2270 #elif defined(S390)
2271 strncpy(cpuinfo, "S390", length);
2272 #elif defined(SPARC)
2273 strncpy(cpuinfo, "sparcv9", length);
2274 #elif defined(ZERO_LIBARCH)
2275 strncpy(cpuinfo, ZERO_LIBARCH, length);
2276 #else
2277 strncpy(cpuinfo, "unknown", length);
2278 #endif
2279 }
2280
2281 static void print_signal_handler(outputStream* st, int sig,
2282 char* buf, size_t buflen);
2283
2284 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2285 st->print_cr("Signal Handlers:");
2286 print_signal_handler(st, SIGSEGV, buf, buflen);
2287 print_signal_handler(st, SIGBUS , buf, buflen);
2288 print_signal_handler(st, SIGFPE , buf, buflen);
2289 print_signal_handler(st, SIGPIPE, buf, buflen);
2290 print_signal_handler(st, SIGXFSZ, buf, buflen);
2291 print_signal_handler(st, SIGILL , buf, buflen);
2292 print_signal_handler(st, SR_signum, buf, buflen);
2293 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2294 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2295 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2296 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2297 #if defined(PPC64)
2298 print_signal_handler(st, SIGTRAP, buf, buflen);
2299 #endif
2300 }
2301
2302 static char saved_jvm_path[MAXPATHLEN] = {0};
2303
2304 // Find the full path to the current module, libjvm.so
2305 void os::jvm_path(char *buf, jint buflen) {
2306 // Error checking.
2307 if (buflen < MAXPATHLEN) {
2308 assert(false, "must use a large-enough buffer");
2309 buf[0] = '\0';
2310 return;
2311 }
2312 // Lazy resolve the path to current module.
2313 if (saved_jvm_path[0] != 0) {
2314 strcpy(buf, saved_jvm_path);
2315 return;
2316 }
2317
2318 char dli_fname[MAXPATHLEN];
2319 bool ret = dll_address_to_library_name(
2320 CAST_FROM_FN_PTR(address, os::jvm_path),
2321 dli_fname, sizeof(dli_fname), NULL);
2322 assert(ret, "cannot locate libjvm");
2323 char *rp = NULL;
2324 if (ret && dli_fname[0] != '\0') {
2325 rp = os::Posix::realpath(dli_fname, buf, buflen);
2326 }
2327 if (rp == NULL) {
2328 return;
2329 }
2330
2331 if (Arguments::sun_java_launcher_is_altjvm()) {
2332 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2333 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2334 // If "/jre/lib/" appears at the right place in the string, then
2335 // assume we are installed in a JDK and we're done. Otherwise, check
2336 // for a JAVA_HOME environment variable and fix up the path so it
2337 // looks like libjvm.so is installed there (append a fake suffix
2338 // hotspot/libjvm.so).
2339 const char *p = buf + strlen(buf) - 1;
2340 for (int count = 0; p > buf && count < 5; ++count) {
2341 for (--p; p > buf && *p != '/'; --p)
2342 /* empty */ ;
2343 }
2344
2345 if (strncmp(p, "/jre/lib/", 9) != 0) {
2346 // Look for JAVA_HOME in the environment.
2347 char* java_home_var = ::getenv("JAVA_HOME");
2348 if (java_home_var != NULL && java_home_var[0] != 0) {
2349 char* jrelib_p;
2350 int len;
2351
2352 // Check the current module name "libjvm.so".
2353 p = strrchr(buf, '/');
2354 if (p == NULL) {
2355 return;
2356 }
2357 assert(strstr(p, "/libjvm") == p, "invalid library name");
2358
2359 rp = os::Posix::realpath(java_home_var, buf, buflen);
2360 if (rp == NULL) {
2361 return;
2362 }
2363
2364 // determine if this is a legacy image or modules image
2365 // modules image doesn't have "jre" subdirectory
2366 len = strlen(buf);
2367 assert(len < buflen, "Ran out of buffer room");
2368 jrelib_p = buf + len;
2369 snprintf(jrelib_p, buflen-len, "/jre/lib");
2370 if (0 != access(buf, F_OK)) {
2371 snprintf(jrelib_p, buflen-len, "/lib");
2372 }
2373
2374 if (0 == access(buf, F_OK)) {
2375 // Use current module name "libjvm.so"
2376 len = strlen(buf);
2377 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2378 } else {
2379 // Go back to path of .so
2380 rp = os::Posix::realpath(dli_fname, buf, buflen);
2381 if (rp == NULL) {
2382 return;
2383 }
2384 }
2385 }
2386 }
2387 }
2388
2389 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2390 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2391 }
2392
2393 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2394 // no prefix required, not even "_"
2395 }
2396
2397 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2398 // no suffix required
2399 }
2400
2401 ////////////////////////////////////////////////////////////////////////////////
2402 // sun.misc.Signal support
2403
2404 static volatile jint sigint_count = 0;
2405
2406 static void UserHandler(int sig, void *siginfo, void *context) {
2407 // 4511530 - sem_post is serialized and handled by the manager thread. When
2408 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2409 // don't want to flood the manager thread with sem_post requests.
2410 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2411 return;
2412 }
2413
2414 // Ctrl-C is pressed during error reporting, likely because the error
2415 // handler fails to abort. Let VM die immediately.
2416 if (sig == SIGINT && VMError::is_error_reported()) {
2417 os::die();
2418 }
2419
2420 os::signal_notify(sig);
2421 }
2422
2423 void* os::user_handler() {
2424 return CAST_FROM_FN_PTR(void*, UserHandler);
2425 }
2426
2427 static struct timespec create_semaphore_timespec(unsigned int sec, int nsec) {
2428 struct timespec ts;
2429 // Semaphore's are always associated with CLOCK_REALTIME
2430 os::Posix::clock_gettime(CLOCK_REALTIME, &ts);
2431 // see os_posix.cpp for discussion on overflow checking
2432 if (sec >= MAX_SECS) {
2433 ts.tv_sec += MAX_SECS;
2434 ts.tv_nsec = 0;
2435 } else {
2436 ts.tv_sec += sec;
2437 ts.tv_nsec += nsec;
2438 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2439 ts.tv_nsec -= NANOSECS_PER_SEC;
2440 ++ts.tv_sec; // note: this must be <= max_secs
2441 }
2442 }
2443
2444 return ts;
2445 }
2446
2447 extern "C" {
2448 typedef void (*sa_handler_t)(int);
2449 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2450 }
2451
2452 void* os::signal(int signal_number, void* handler) {
2453 struct sigaction sigAct, oldSigAct;
2454
2455 sigfillset(&(sigAct.sa_mask));
2456 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2457 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2458
2459 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2460 // -1 means registration failed
2461 return (void *)-1;
2462 }
2463
2464 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2465 }
2466
2467 void os::signal_raise(int signal_number) {
2468 ::raise(signal_number);
2469 }
2470
2471 // The following code is moved from os.cpp for making this
2472 // code platform specific, which it is by its very nature.
2473
2474 // Will be modified when max signal is changed to be dynamic
2475 int os::sigexitnum_pd() {
2476 return NSIG;
2477 }
2478
2479 // a counter for each possible signal value
2480 static volatile jint pending_signals[NSIG+1] = { 0 };
2481
2482 // Linux(POSIX) specific hand shaking semaphore.
2483 static Semaphore* sig_sem = NULL;
2484 static PosixSemaphore sr_semaphore;
2485
2486 static void jdk_misc_signal_init() {
2487 // Initialize signal structures
2488 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2489
2490 // Initialize signal semaphore
2491 sig_sem = new Semaphore();
2492 }
2493
2494 void os::signal_notify(int sig) {
2495 if (sig_sem != NULL) {
2496 Atomic::inc(&pending_signals[sig]);
2497 sig_sem->signal();
2498 } else {
2499 // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2500 // initialization isn't called.
2501 assert(ReduceSignalUsage, "signal semaphore should be created");
2502 }
2503 }
2504
2505 static int check_pending_signals() {
2506 Atomic::store(0, &sigint_count);
2507 for (;;) {
2508 for (int i = 0; i < NSIG + 1; i++) {
2509 jint n = pending_signals[i];
2510 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2511 return i;
2512 }
2513 }
2514 JavaThread *thread = JavaThread::current();
2515 ThreadBlockInVM tbivm(thread);
2516
2517 bool threadIsSuspended;
2518 do {
2519 thread->set_suspend_equivalent();
2520 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2521 sig_sem->wait();
2522
2523 // were we externally suspended while we were waiting?
2524 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2525 if (threadIsSuspended) {
2526 // The semaphore has been incremented, but while we were waiting
2527 // another thread suspended us. We don't want to continue running
2528 // while suspended because that would surprise the thread that
2529 // suspended us.
2530 sig_sem->signal();
2531
2532 thread->java_suspend_self();
2533 }
2534 } while (threadIsSuspended);
2535 }
2536 }
2537
2538 int os::signal_wait() {
2539 return check_pending_signals();
2540 }
2541
2542 ////////////////////////////////////////////////////////////////////////////////
2543 // Virtual Memory
2544
2545 int os::vm_page_size() {
2546 // Seems redundant as all get out
2547 assert(os::Linux::page_size() != -1, "must call os::init");
2548 return os::Linux::page_size();
2549 }
2550
2551 // Solaris allocates memory by pages.
2552 int os::vm_allocation_granularity() {
2553 assert(os::Linux::page_size() != -1, "must call os::init");
2554 return os::Linux::page_size();
2555 }
2556
2557 // Rationale behind this function:
2558 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2559 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2560 // samples for JITted code. Here we create private executable mapping over the code cache
2561 // and then we can use standard (well, almost, as mapping can change) way to provide
2562 // info for the reporting script by storing timestamp and location of symbol
2563 void linux_wrap_code(char* base, size_t size) {
2564 static volatile jint cnt = 0;
2565
2566 if (!UseOprofile) {
2567 return;
2568 }
2569
2570 char buf[PATH_MAX+1];
2571 int num = Atomic::add(1, &cnt);
2572
2573 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2574 os::get_temp_directory(), os::current_process_id(), num);
2575 unlink(buf);
2576
2577 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2578
2579 if (fd != -1) {
2580 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2581 if (rv != (off_t)-1) {
2582 if (::write(fd, "", 1) == 1) {
2583 mmap(base, size,
2584 PROT_READ|PROT_WRITE|PROT_EXEC,
2585 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2586 }
2587 }
2588 ::close(fd);
2589 unlink(buf);
2590 }
2591 }
2592
2593 static bool recoverable_mmap_error(int err) {
2594 // See if the error is one we can let the caller handle. This
2595 // list of errno values comes from JBS-6843484. I can't find a
2596 // Linux man page that documents this specific set of errno
2597 // values so while this list currently matches Solaris, it may
2598 // change as we gain experience with this failure mode.
2599 switch (err) {
2600 case EBADF:
2601 case EINVAL:
2602 case ENOTSUP:
2603 // let the caller deal with these errors
2604 return true;
2605
2606 default:
2607 // Any remaining errors on this OS can cause our reserved mapping
2608 // to be lost. That can cause confusion where different data
2609 // structures think they have the same memory mapped. The worst
2610 // scenario is if both the VM and a library think they have the
2611 // same memory mapped.
2612 return false;
2613 }
2614 }
2615
2616 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2617 int err) {
2618 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2619 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2620 os::strerror(err), err);
2621 }
2622
2623 static void warn_fail_commit_memory(char* addr, size_t size,
2624 size_t alignment_hint, bool exec,
2625 int err) {
2626 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2627 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2628 alignment_hint, exec, os::strerror(err), err);
2629 }
2630
2631 // NOTE: Linux kernel does not really reserve the pages for us.
2632 // All it does is to check if there are enough free pages
2633 // left at the time of mmap(). This could be a potential
2634 // problem.
2635 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2636 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2637 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2638 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2639 if (res != (uintptr_t) MAP_FAILED) {
2640 if (UseNUMAInterleaving) {
2641 numa_make_global(addr, size);
2642 }
2643 return 0;
2644 }
2645
2646 int err = errno; // save errno from mmap() call above
2647
2648 if (!recoverable_mmap_error(err)) {
2649 warn_fail_commit_memory(addr, size, exec, err);
2650 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2651 }
2652
2653 return err;
2654 }
2655
2656 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2657 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2658 }
2659
2660 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2661 const char* mesg) {
2662 assert(mesg != NULL, "mesg must be specified");
2663 int err = os::Linux::commit_memory_impl(addr, size, exec);
2664 if (err != 0) {
2665 // the caller wants all commit errors to exit with the specified mesg:
2666 warn_fail_commit_memory(addr, size, exec, err);
2667 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2668 }
2669 }
2670
2671 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2672 #ifndef MAP_HUGETLB
2673 #define MAP_HUGETLB 0x40000
2674 #endif
2675
2676 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2677 #ifndef MADV_HUGEPAGE
2678 #define MADV_HUGEPAGE 14
2679 #endif
2680
2681 int os::Linux::commit_memory_impl(char* addr, size_t size,
2682 size_t alignment_hint, bool exec) {
2683 int err = os::Linux::commit_memory_impl(addr, size, exec);
2684 if (err == 0) {
2685 realign_memory(addr, size, alignment_hint);
2686 }
2687 return err;
2688 }
2689
2690 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2691 bool exec) {
2692 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2693 }
2694
2695 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2696 size_t alignment_hint, bool exec,
2697 const char* mesg) {
2698 assert(mesg != NULL, "mesg must be specified");
2699 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2700 if (err != 0) {
2701 // the caller wants all commit errors to exit with the specified mesg:
2702 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2703 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2704 }
2705 }
2706
2707 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2708 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2709 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2710 // be supported or the memory may already be backed by huge pages.
2711 ::madvise(addr, bytes, MADV_HUGEPAGE);
2712 }
2713 }
2714
2715 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2716 // This method works by doing an mmap over an existing mmaping and effectively discarding
2717 // the existing pages. However it won't work for SHM-based large pages that cannot be
2718 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2719 // small pages on top of the SHM segment. This method always works for small pages, so we
2720 // allow that in any case.
2721 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2722 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2723 }
2724 }
2725
2726 void os::numa_make_global(char *addr, size_t bytes) {
2727 if (!UseNUMAInterleaving)
2728 return ;
2729 Linux::numa_interleave_memory(addr, bytes);
2730 }
2731
2732 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2733 // bind policy to MPOL_PREFERRED for the current thread.
2734 #define USE_MPOL_PREFERRED 0
2735
2736 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2737 // To make NUMA and large pages more robust when both enabled, we need to ease
2738 // the requirements on where the memory should be allocated. MPOL_BIND is the
2739 // default policy and it will force memory to be allocated on the specified
2740 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2741 // the specified node, but will not force it. Using this policy will prevent
2742 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2743 // free large pages.
2744 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2745 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2746 }
2747
2748 bool os::numa_topology_changed() { return false; }
2749
2750 size_t os::numa_get_groups_num() {
2751 // Return just the number of nodes in which it's possible to allocate memory
2752 // (in numa terminology, configured nodes).
2753 return Linux::numa_num_configured_nodes();
2754 }
2755
2756 int os::numa_get_group_id() {
2757 int cpu_id = Linux::sched_getcpu();
2758 if (cpu_id != -1) {
2759 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2760 if (lgrp_id != -1) {
2761 return lgrp_id;
2762 }
2763 }
2764 return 0;
2765 }
2766
2767 int os::Linux::get_existing_num_nodes() {
2768 int node;
2769 int highest_node_number = Linux::numa_max_node();
2770 int num_nodes = 0;
2771
2772 // Get the total number of nodes in the system including nodes without memory.
2773 for (node = 0; node <= highest_node_number; node++) {
2774 if (isnode_in_existing_nodes(node)) {
2775 num_nodes++;
2776 }
2777 }
2778 return num_nodes;
2779 }
2780
2781 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2782 int highest_node_number = Linux::numa_max_node();
2783 size_t i = 0;
2784
2785 // Map all node ids in which it is possible to allocate memory. Also nodes are
2786 // not always consecutively available, i.e. available from 0 to the highest
2787 // node number. If the nodes have been bound explicitly using numactl membind,
2788 // then allocate memory from those nodes only.
2789 for (int node = 0; node <= highest_node_number; node++) {
2790 if (Linux::isnode_in_bound_nodes((unsigned int)node)) {
2791 ids[i++] = node;
2792 }
2793 }
2794
2795 // If externally invoked in interleave mode then get node bitmasks from interleave mode pointer.
2796 if (Linux::_numa_interleave_ptr != NULL ) {
2797 i = 0;
2798 for (int node = 0; node <= highest_node_number; node++) {
2799 if (Linux::_numa_bitmask_isbitset(Linux::_numa_interleave_ptr, node)) {
2800 ids[i++] = node;
2801 }
2802 }
2803 }
2804 return i;
2805 }
2806
2807 bool os::get_page_info(char *start, page_info* info) {
2808 return false;
2809 }
2810
2811 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2812 page_info* page_found) {
2813 return end;
2814 }
2815
2816
2817 int os::Linux::sched_getcpu_syscall(void) {
2818 unsigned int cpu = 0;
2819 int retval = -1;
2820
2821 #if defined(IA32)
2822 #ifndef SYS_getcpu
2823 #define SYS_getcpu 318
2824 #endif
2825 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2826 #elif defined(AMD64)
2827 // Unfortunately we have to bring all these macros here from vsyscall.h
2828 // to be able to compile on old linuxes.
2829 #define __NR_vgetcpu 2
2830 #define VSYSCALL_START (-10UL << 20)
2831 #define VSYSCALL_SIZE 1024
2832 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2833 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2834 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2835 retval = vgetcpu(&cpu, NULL, NULL);
2836 #endif
2837
2838 return (retval == -1) ? retval : cpu;
2839 }
2840
2841 void os::Linux::sched_getcpu_init() {
2842 // sched_getcpu() should be in libc.
2843 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2844 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2845
2846 // If it's not, try a direct syscall.
2847 if (sched_getcpu() == -1) {
2848 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2849 (void*)&sched_getcpu_syscall));
2850 }
2851
2852 if (sched_getcpu() == -1) {
2853 vm_exit_during_initialization("getcpu(2) system call not supported by kernel");
2854 }
2855 }
2856
2857 // Something to do with the numa-aware allocator needs these symbols
2858 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2859 extern "C" JNIEXPORT void numa_error(char *where) { }
2860
2861 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2862 // load symbol from base version instead.
2863 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2864 void *f = dlvsym(handle, name, "libnuma_1.1");
2865 if (f == NULL) {
2866 f = dlsym(handle, name);
2867 }
2868 return f;
2869 }
2870
2871 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2872 // Return NULL if the symbol is not defined in this particular version.
2873 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2874 return dlvsym(handle, name, "libnuma_1.2");
2875 }
2876
2877 bool os::Linux::libnuma_init() {
2878 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
2879 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2880 if (handle != NULL) {
2881 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2882 libnuma_dlsym(handle, "numa_node_to_cpus")));
2883 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2884 libnuma_dlsym(handle, "numa_max_node")));
2885 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2886 libnuma_dlsym(handle, "numa_num_configured_nodes")));
2887 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2888 libnuma_dlsym(handle, "numa_available")));
2889 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2890 libnuma_dlsym(handle, "numa_tonode_memory")));
2891 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2892 libnuma_dlsym(handle, "numa_interleave_memory")));
2893 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2894 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2895 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2896 libnuma_dlsym(handle, "numa_set_bind_policy")));
2897 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2898 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2899 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2900 libnuma_dlsym(handle, "numa_distance")));
2901 set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t,
2902 libnuma_v2_dlsym(handle, "numa_get_membind")));
2903 set_numa_get_interleave_mask(CAST_TO_FN_PTR(numa_get_interleave_mask_func_t,
2904 libnuma_v2_dlsym(handle, "numa_get_interleave_mask")));
2905
2906 if (numa_available() != -1) {
2907 struct bitmask *bmp;
2908 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2909 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2910 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2911 bmp = _numa_get_interleave_mask();
2912 set_numa_interleave_ptr(&bmp);
2913 bmp = _numa_get_membind();
2914 set_numa_membind_ptr(&bmp);
2915 // Create an index -> node mapping, since nodes are not always consecutive
2916 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2917 rebuild_nindex_to_node_map();
2918 // Create a cpu -> node mapping
2919 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2920 rebuild_cpu_to_node_map();
2921 return true;
2922 }
2923 }
2924 }
2925 return false;
2926 }
2927
2928 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2929 // Creating guard page is very expensive. Java thread has HotSpot
2930 // guard pages, only enable glibc guard page for non-Java threads.
2931 // (Remember: compiler thread is a Java thread, too!)
2932 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2933 }
2934
2935 void os::Linux::rebuild_nindex_to_node_map() {
2936 int highest_node_number = Linux::numa_max_node();
2937
2938 nindex_to_node()->clear();
2939 for (int node = 0; node <= highest_node_number; node++) {
2940 if (Linux::isnode_in_existing_nodes(node)) {
2941 nindex_to_node()->append(node);
2942 }
2943 }
2944 }
2945
2946 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2947 // The table is later used in get_node_by_cpu().
2948 void os::Linux::rebuild_cpu_to_node_map() {
2949 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2950 // in libnuma (possible values are starting from 16,
2951 // and continuing up with every other power of 2, but less
2952 // than the maximum number of CPUs supported by kernel), and
2953 // is a subject to change (in libnuma version 2 the requirements
2954 // are more reasonable) we'll just hardcode the number they use
2955 // in the library.
2956 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2957
2958 size_t cpu_num = processor_count();
2959 size_t cpu_map_size = NCPUS / BitsPerCLong;
2960 size_t cpu_map_valid_size =
2961 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2962
2963 cpu_to_node()->clear();
2964 cpu_to_node()->at_grow(cpu_num - 1);
2965
2966 size_t node_num = get_existing_num_nodes();
2967
2968 int distance = 0;
2969 int closest_distance = INT_MAX;
2970 int closest_node = 0;
2971 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2972 for (size_t i = 0; i < node_num; i++) {
2973 // Check if node is configured (not a memory-less node). If it is not, find
2974 // the closest configured node. Check also if node is bound, i.e. it's allowed
2975 // to allocate memory from the node. If it's not allowed, map cpus in that node
2976 // to the closest node from which memory allocation is allowed.
2977 if (!isnode_in_configured_nodes(nindex_to_node()->at(i)) ||
2978 !isnode_in_bound_nodes(nindex_to_node()->at(i))) {
2979 closest_distance = INT_MAX;
2980 // Check distance from all remaining nodes in the system. Ignore distance
2981 // from itself, from another non-configured node, and from another non-bound
2982 // node.
2983 for (size_t m = 0; m < node_num; m++) {
2984 if (m != i &&
2985 isnode_in_configured_nodes(nindex_to_node()->at(m)) &&
2986 isnode_in_bound_nodes(nindex_to_node()->at(m))) {
2987 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
2988 // If a closest node is found, update. There is always at least one
2989 // configured and bound node in the system so there is always at least
2990 // one node close.
2991 if (distance != 0 && distance < closest_distance) {
2992 closest_distance = distance;
2993 closest_node = nindex_to_node()->at(m);
2994 }
2995 }
2996 }
2997 } else {
2998 // Current node is already a configured node.
2999 closest_node = nindex_to_node()->at(i);
3000 }
3001
3002 // Get cpus from the original node and map them to the closest node. If node
3003 // is a configured node (not a memory-less node), then original node and
3004 // closest node are the same.
3005 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3006 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3007 if (cpu_map[j] != 0) {
3008 for (size_t k = 0; k < BitsPerCLong; k++) {
3009 if (cpu_map[j] & (1UL << k)) {
3010 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3011 }
3012 }
3013 }
3014 }
3015 }
3016 }
3017 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3018 }
3019
3020 int os::Linux::get_node_by_cpu(int cpu_id) {
3021 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3022 return cpu_to_node()->at(cpu_id);
3023 }
3024 return -1;
3025 }
3026
3027 GrowableArray<int>* os::Linux::_cpu_to_node;
3028 GrowableArray<int>* os::Linux::_nindex_to_node;
3029 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3030 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3031 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3032 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3033 os::Linux::numa_available_func_t os::Linux::_numa_available;
3034 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3035 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3036 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3037 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3038 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3039 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3040 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind;
3041 os::Linux::numa_get_interleave_mask_func_t os::Linux::_numa_get_interleave_mask;
3042 unsigned long* os::Linux::_numa_all_nodes;
3043 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3044 struct bitmask* os::Linux::_numa_nodes_ptr;
3045 struct bitmask* os::Linux::_numa_interleave_ptr;
3046 struct bitmask* os::Linux::_numa_membind_ptr;
3047
3048 bool os::pd_uncommit_memory(char* addr, size_t size) {
3049 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3050 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3051 return res != (uintptr_t) MAP_FAILED;
3052 }
3053
3054 static address get_stack_commited_bottom(address bottom, size_t size) {
3055 address nbot = bottom;
3056 address ntop = bottom + size;
3057
3058 size_t page_sz = os::vm_page_size();
3059 unsigned pages = size / page_sz;
3060
3061 unsigned char vec[1];
3062 unsigned imin = 1, imax = pages + 1, imid;
3063 int mincore_return_value = 0;
3064
3065 assert(imin <= imax, "Unexpected page size");
3066
3067 while (imin < imax) {
3068 imid = (imax + imin) / 2;
3069 nbot = ntop - (imid * page_sz);
3070
3071 // Use a trick with mincore to check whether the page is mapped or not.
3072 // mincore sets vec to 1 if page resides in memory and to 0 if page
3073 // is swapped output but if page we are asking for is unmapped
3074 // it returns -1,ENOMEM
3075 mincore_return_value = mincore(nbot, page_sz, vec);
3076
3077 if (mincore_return_value == -1) {
3078 // Page is not mapped go up
3079 // to find first mapped page
3080 if (errno != EAGAIN) {
3081 assert(errno == ENOMEM, "Unexpected mincore errno");
3082 imax = imid;
3083 }
3084 } else {
3085 // Page is mapped go down
3086 // to find first not mapped page
3087 imin = imid + 1;
3088 }
3089 }
3090
3091 nbot = nbot + page_sz;
3092
3093 // Adjust stack bottom one page up if last checked page is not mapped
3094 if (mincore_return_value == -1) {
3095 nbot = nbot + page_sz;
3096 }
3097
3098 return nbot;
3099 }
3100
3101 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3102 int mincore_return_value;
3103 const size_t stripe = 1024; // query this many pages each time
3104 unsigned char vec[stripe + 1];
3105 // set a guard
3106 vec[stripe] = 'X';
3107
3108 const size_t page_sz = os::vm_page_size();
3109 size_t pages = size / page_sz;
3110
3111 assert(is_aligned(start, page_sz), "Start address must be page aligned");
3112 assert(is_aligned(size, page_sz), "Size must be page aligned");
3113
3114 committed_start = NULL;
3115
3116 int loops = (pages + stripe - 1) / stripe;
3117 int committed_pages = 0;
3118 address loop_base = start;
3119 bool found_range = false;
3120
3121 for (int index = 0; index < loops && !found_range; index ++) {
3122 assert(pages > 0, "Nothing to do");
3123 int pages_to_query = (pages >= stripe) ? stripe : pages;
3124 pages -= pages_to_query;
3125
3126 // Get stable read
3127 while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3128
3129 // During shutdown, some memory goes away without properly notifying NMT,
3130 // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3131 // Bailout and return as not committed for now.
3132 if (mincore_return_value == -1 && errno == ENOMEM) {
3133 return false;
3134 }
3135
3136 assert(vec[stripe] == 'X', "overflow guard");
3137 assert(mincore_return_value == 0, "Range must be valid");
3138 // Process this stripe
3139 for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3140 if ((vec[vecIdx] & 0x01) == 0) { // not committed
3141 // End of current contiguous region
3142 if (committed_start != NULL) {
3143 found_range = true;
3144 break;
3145 }
3146 } else { // committed
3147 // Start of region
3148 if (committed_start == NULL) {
3149 committed_start = loop_base + page_sz * vecIdx;
3150 }
3151 committed_pages ++;
3152 }
3153 }
3154
3155 loop_base += pages_to_query * page_sz;
3156 }
3157
3158 if (committed_start != NULL) {
3159 assert(committed_pages > 0, "Must have committed region");
3160 assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3161 assert(committed_start >= start && committed_start < start + size, "Out of range");
3162 committed_size = page_sz * committed_pages;
3163 return true;
3164 } else {
3165 assert(committed_pages == 0, "Should not have committed region");
3166 return false;
3167 }
3168 }
3169
3170
3171 // Linux uses a growable mapping for the stack, and if the mapping for
3172 // the stack guard pages is not removed when we detach a thread the
3173 // stack cannot grow beyond the pages where the stack guard was
3174 // mapped. If at some point later in the process the stack expands to
3175 // that point, the Linux kernel cannot expand the stack any further
3176 // because the guard pages are in the way, and a segfault occurs.
3177 //
3178 // However, it's essential not to split the stack region by unmapping
3179 // a region (leaving a hole) that's already part of the stack mapping,
3180 // so if the stack mapping has already grown beyond the guard pages at
3181 // the time we create them, we have to truncate the stack mapping.
3182 // So, we need to know the extent of the stack mapping when
3183 // create_stack_guard_pages() is called.
3184
3185 // We only need this for stacks that are growable: at the time of
3186 // writing thread stacks don't use growable mappings (i.e. those
3187 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3188 // only applies to the main thread.
3189
3190 // If the (growable) stack mapping already extends beyond the point
3191 // where we're going to put our guard pages, truncate the mapping at
3192 // that point by munmap()ping it. This ensures that when we later
3193 // munmap() the guard pages we don't leave a hole in the stack
3194 // mapping. This only affects the main/primordial thread
3195
3196 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3197 if (os::is_primordial_thread()) {
3198 // As we manually grow stack up to bottom inside create_attached_thread(),
3199 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3200 // we don't need to do anything special.
3201 // Check it first, before calling heavy function.
3202 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3203 unsigned char vec[1];
3204
3205 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3206 // Fallback to slow path on all errors, including EAGAIN
3207 stack_extent = (uintptr_t) get_stack_commited_bottom(
3208 os::Linux::initial_thread_stack_bottom(),
3209 (size_t)addr - stack_extent);
3210 }
3211
3212 if (stack_extent < (uintptr_t)addr) {
3213 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3214 }
3215 }
3216
3217 return os::commit_memory(addr, size, !ExecMem);
3218 }
3219
3220 // If this is a growable mapping, remove the guard pages entirely by
3221 // munmap()ping them. If not, just call uncommit_memory(). This only
3222 // affects the main/primordial thread, but guard against future OS changes.
3223 // It's safe to always unmap guard pages for primordial thread because we
3224 // always place it right after end of the mapped region.
3225
3226 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3227 uintptr_t stack_extent, stack_base;
3228
3229 if (os::is_primordial_thread()) {
3230 return ::munmap(addr, size) == 0;
3231 }
3232
3233 return os::uncommit_memory(addr, size);
3234 }
3235
3236 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3237 // at 'requested_addr'. If there are existing memory mappings at the same
3238 // location, however, they will be overwritten. If 'fixed' is false,
3239 // 'requested_addr' is only treated as a hint, the return value may or
3240 // may not start from the requested address. Unlike Linux mmap(), this
3241 // function returns NULL to indicate failure.
3242 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3243 char * addr;
3244 int flags;
3245
3246 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3247 if (fixed) {
3248 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3249 flags |= MAP_FIXED;
3250 }
3251
3252 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3253 // touch an uncommitted page. Otherwise, the read/write might
3254 // succeed if we have enough swap space to back the physical page.
3255 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3256 flags, -1, 0);
3257
3258 return addr == MAP_FAILED ? NULL : addr;
3259 }
3260
3261 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3262 // (req_addr != NULL) or with a given alignment.
3263 // - bytes shall be a multiple of alignment.
3264 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3265 // - alignment sets the alignment at which memory shall be allocated.
3266 // It must be a multiple of allocation granularity.
3267 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3268 // req_addr or NULL.
3269 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3270
3271 size_t extra_size = bytes;
3272 if (req_addr == NULL && alignment > 0) {
3273 extra_size += alignment;
3274 }
3275
3276 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3277 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3278 -1, 0);
3279 if (start == MAP_FAILED) {
3280 start = NULL;
3281 } else {
3282 if (req_addr != NULL) {
3283 if (start != req_addr) {
3284 ::munmap(start, extra_size);
3285 start = NULL;
3286 }
3287 } else {
3288 char* const start_aligned = align_up(start, alignment);
3289 char* const end_aligned = start_aligned + bytes;
3290 char* const end = start + extra_size;
3291 if (start_aligned > start) {
3292 ::munmap(start, start_aligned - start);
3293 }
3294 if (end_aligned < end) {
3295 ::munmap(end_aligned, end - end_aligned);
3296 }
3297 start = start_aligned;
3298 }
3299 }
3300 return start;
3301 }
3302
3303 static int anon_munmap(char * addr, size_t size) {
3304 return ::munmap(addr, size) == 0;
3305 }
3306
3307 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3308 size_t alignment_hint) {
3309 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3310 }
3311
3312 bool os::pd_release_memory(char* addr, size_t size) {
3313 return anon_munmap(addr, size);
3314 }
3315
3316 static bool linux_mprotect(char* addr, size_t size, int prot) {
3317 // Linux wants the mprotect address argument to be page aligned.
3318 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3319
3320 // According to SUSv3, mprotect() should only be used with mappings
3321 // established by mmap(), and mmap() always maps whole pages. Unaligned
3322 // 'addr' likely indicates problem in the VM (e.g. trying to change
3323 // protection of malloc'ed or statically allocated memory). Check the
3324 // caller if you hit this assert.
3325 assert(addr == bottom, "sanity check");
3326
3327 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3328 return ::mprotect(bottom, size, prot) == 0;
3329 }
3330
3331 // Set protections specified
3332 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3333 bool is_committed) {
3334 unsigned int p = 0;
3335 switch (prot) {
3336 case MEM_PROT_NONE: p = PROT_NONE; break;
3337 case MEM_PROT_READ: p = PROT_READ; break;
3338 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3339 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3340 default:
3341 ShouldNotReachHere();
3342 }
3343 // is_committed is unused.
3344 return linux_mprotect(addr, bytes, p);
3345 }
3346
3347 bool os::guard_memory(char* addr, size_t size) {
3348 return linux_mprotect(addr, size, PROT_NONE);
3349 }
3350
3351 bool os::unguard_memory(char* addr, size_t size) {
3352 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3353 }
3354
3355 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3356 size_t page_size) {
3357 bool result = false;
3358 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3359 MAP_ANONYMOUS|MAP_PRIVATE,
3360 -1, 0);
3361 if (p != MAP_FAILED) {
3362 void *aligned_p = align_up(p, page_size);
3363
3364 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3365
3366 munmap(p, page_size * 2);
3367 }
3368
3369 if (warn && !result) {
3370 warning("TransparentHugePages is not supported by the operating system.");
3371 }
3372
3373 return result;
3374 }
3375
3376 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3377 bool result = false;
3378 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3379 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3380 -1, 0);
3381
3382 if (p != MAP_FAILED) {
3383 // We don't know if this really is a huge page or not.
3384 FILE *fp = fopen("/proc/self/maps", "r");
3385 if (fp) {
3386 while (!feof(fp)) {
3387 char chars[257];
3388 long x = 0;
3389 if (fgets(chars, sizeof(chars), fp)) {
3390 if (sscanf(chars, "%lx-%*x", &x) == 1
3391 && x == (long)p) {
3392 if (strstr (chars, "hugepage")) {
3393 result = true;
3394 break;
3395 }
3396 }
3397 }
3398 }
3399 fclose(fp);
3400 }
3401 munmap(p, page_size);
3402 }
3403
3404 if (warn && !result) {
3405 warning("HugeTLBFS is not supported by the operating system.");
3406 }
3407
3408 return result;
3409 }
3410
3411 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3412 //
3413 // From the coredump_filter documentation:
3414 //
3415 // - (bit 0) anonymous private memory
3416 // - (bit 1) anonymous shared memory
3417 // - (bit 2) file-backed private memory
3418 // - (bit 3) file-backed shared memory
3419 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3420 // effective only if the bit 2 is cleared)
3421 // - (bit 5) hugetlb private memory
3422 // - (bit 6) hugetlb shared memory
3423 // - (bit 7) dax private memory
3424 // - (bit 8) dax shared memory
3425 //
3426 static void set_coredump_filter(bool largepages, bool dax_shared) {
3427 FILE *f;
3428 long cdm;
3429 bool filter_changed = false;
3430
3431 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3432 return;
3433 }
3434
3435 if (fscanf(f, "%lx", &cdm) != 1) {
3436 fclose(f);
3437 return;
3438 }
3439
3440 rewind(f);
3441
3442 if (largepages && (cdm & LARGEPAGES_BIT) == 0) {
3443 cdm |= LARGEPAGES_BIT;
3444 filter_changed = true;
3445 }
3446 if (dax_shared && (cdm & DAX_SHARED_BIT) == 0) {
3447 cdm |= DAX_SHARED_BIT;
3448 filter_changed = true;
3449 }
3450 if (filter_changed) {
3451 fprintf(f, "%#lx", cdm);
3452 }
3453
3454 fclose(f);
3455 }
3456
3457 // Large page support
3458
3459 static size_t _large_page_size = 0;
3460
3461 size_t os::Linux::find_large_page_size() {
3462 size_t large_page_size = 0;
3463
3464 // large_page_size on Linux is used to round up heap size. x86 uses either
3465 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3466 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3467 // page as large as 256M.
3468 //
3469 // Here we try to figure out page size by parsing /proc/meminfo and looking
3470 // for a line with the following format:
3471 // Hugepagesize: 2048 kB
3472 //
3473 // If we can't determine the value (e.g. /proc is not mounted, or the text
3474 // format has been changed), we'll use the largest page size supported by
3475 // the processor.
3476
3477 #ifndef ZERO
3478 large_page_size =
3479 AARCH64_ONLY(2 * M)
3480 AMD64_ONLY(2 * M)
3481 ARM32_ONLY(2 * M)
3482 IA32_ONLY(4 * M)
3483 IA64_ONLY(256 * M)
3484 PPC_ONLY(4 * M)
3485 S390_ONLY(1 * M)
3486 SPARC_ONLY(4 * M);
3487 #endif // ZERO
3488
3489 FILE *fp = fopen("/proc/meminfo", "r");
3490 if (fp) {
3491 while (!feof(fp)) {
3492 int x = 0;
3493 char buf[16];
3494 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3495 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3496 large_page_size = x * K;
3497 break;
3498 }
3499 } else {
3500 // skip to next line
3501 for (;;) {
3502 int ch = fgetc(fp);
3503 if (ch == EOF || ch == (int)'\n') break;
3504 }
3505 }
3506 }
3507 fclose(fp);
3508 }
3509
3510 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3511 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3512 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3513 proper_unit_for_byte_size(large_page_size));
3514 }
3515
3516 return large_page_size;
3517 }
3518
3519 size_t os::Linux::setup_large_page_size() {
3520 _large_page_size = Linux::find_large_page_size();
3521 const size_t default_page_size = (size_t)Linux::page_size();
3522 if (_large_page_size > default_page_size) {
3523 _page_sizes[0] = _large_page_size;
3524 _page_sizes[1] = default_page_size;
3525 _page_sizes[2] = 0;
3526 }
3527
3528 return _large_page_size;
3529 }
3530
3531 bool os::Linux::setup_large_page_type(size_t page_size) {
3532 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3533 FLAG_IS_DEFAULT(UseSHM) &&
3534 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3535
3536 // The type of large pages has not been specified by the user.
3537
3538 // Try UseHugeTLBFS and then UseSHM.
3539 UseHugeTLBFS = UseSHM = true;
3540
3541 // Don't try UseTransparentHugePages since there are known
3542 // performance issues with it turned on. This might change in the future.
3543 UseTransparentHugePages = false;
3544 }
3545
3546 if (UseTransparentHugePages) {
3547 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3548 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3549 UseHugeTLBFS = false;
3550 UseSHM = false;
3551 return true;
3552 }
3553 UseTransparentHugePages = false;
3554 }
3555
3556 if (UseHugeTLBFS) {
3557 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3558 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3559 UseSHM = false;
3560 return true;
3561 }
3562 UseHugeTLBFS = false;
3563 }
3564
3565 return UseSHM;
3566 }
3567
3568 void os::large_page_init() {
3569 if (!UseLargePages &&
3570 !UseTransparentHugePages &&
3571 !UseHugeTLBFS &&
3572 !UseSHM) {
3573 // Not using large pages.
3574 return;
3575 }
3576
3577 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3578 // The user explicitly turned off large pages.
3579 // Ignore the rest of the large pages flags.
3580 UseTransparentHugePages = false;
3581 UseHugeTLBFS = false;
3582 UseSHM = false;
3583 return;
3584 }
3585
3586 size_t large_page_size = Linux::setup_large_page_size();
3587 UseLargePages = Linux::setup_large_page_type(large_page_size);
3588
3589 set_coredump_filter(true /*largepages*/, false /*dax_shared*/);
3590 }
3591
3592 #ifndef SHM_HUGETLB
3593 #define SHM_HUGETLB 04000
3594 #endif
3595
3596 #define shm_warning_format(format, ...) \
3597 do { \
3598 if (UseLargePages && \
3599 (!FLAG_IS_DEFAULT(UseLargePages) || \
3600 !FLAG_IS_DEFAULT(UseSHM) || \
3601 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3602 warning(format, __VA_ARGS__); \
3603 } \
3604 } while (0)
3605
3606 #define shm_warning(str) shm_warning_format("%s", str)
3607
3608 #define shm_warning_with_errno(str) \
3609 do { \
3610 int err = errno; \
3611 shm_warning_format(str " (error = %d)", err); \
3612 } while (0)
3613
3614 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3615 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3616
3617 if (!is_aligned(alignment, SHMLBA)) {
3618 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3619 return NULL;
3620 }
3621
3622 // To ensure that we get 'alignment' aligned memory from shmat,
3623 // we pre-reserve aligned virtual memory and then attach to that.
3624
3625 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3626 if (pre_reserved_addr == NULL) {
3627 // Couldn't pre-reserve aligned memory.
3628 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3629 return NULL;
3630 }
3631
3632 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3633 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3634
3635 if ((intptr_t)addr == -1) {
3636 int err = errno;
3637 shm_warning_with_errno("Failed to attach shared memory.");
3638
3639 assert(err != EACCES, "Unexpected error");
3640 assert(err != EIDRM, "Unexpected error");
3641 assert(err != EINVAL, "Unexpected error");
3642
3643 // Since we don't know if the kernel unmapped the pre-reserved memory area
3644 // we can't unmap it, since that would potentially unmap memory that was
3645 // mapped from other threads.
3646 return NULL;
3647 }
3648
3649 return addr;
3650 }
3651
3652 static char* shmat_at_address(int shmid, char* req_addr) {
3653 if (!is_aligned(req_addr, SHMLBA)) {
3654 assert(false, "Requested address needs to be SHMLBA aligned");
3655 return NULL;
3656 }
3657
3658 char* addr = (char*)shmat(shmid, req_addr, 0);
3659
3660 if ((intptr_t)addr == -1) {
3661 shm_warning_with_errno("Failed to attach shared memory.");
3662 return NULL;
3663 }
3664
3665 return addr;
3666 }
3667
3668 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3669 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3670 if (req_addr != NULL) {
3671 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3672 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3673 return shmat_at_address(shmid, req_addr);
3674 }
3675
3676 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3677 // return large page size aligned memory addresses when req_addr == NULL.
3678 // However, if the alignment is larger than the large page size, we have
3679 // to manually ensure that the memory returned is 'alignment' aligned.
3680 if (alignment > os::large_page_size()) {
3681 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3682 return shmat_with_alignment(shmid, bytes, alignment);
3683 } else {
3684 return shmat_at_address(shmid, NULL);
3685 }
3686 }
3687
3688 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3689 char* req_addr, bool exec) {
3690 // "exec" is passed in but not used. Creating the shared image for
3691 // the code cache doesn't have an SHM_X executable permission to check.
3692 assert(UseLargePages && UseSHM, "only for SHM large pages");
3693 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3694 assert(is_aligned(req_addr, alignment), "Unaligned address");
3695
3696 if (!is_aligned(bytes, os::large_page_size())) {
3697 return NULL; // Fallback to small pages.
3698 }
3699
3700 // Create a large shared memory region to attach to based on size.
3701 // Currently, size is the total size of the heap.
3702 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3703 if (shmid == -1) {
3704 // Possible reasons for shmget failure:
3705 // 1. shmmax is too small for Java heap.
3706 // > check shmmax value: cat /proc/sys/kernel/shmmax
3707 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3708 // 2. not enough large page memory.
3709 // > check available large pages: cat /proc/meminfo
3710 // > increase amount of large pages:
3711 // echo new_value > /proc/sys/vm/nr_hugepages
3712 // Note 1: different Linux may use different name for this property,
3713 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3714 // Note 2: it's possible there's enough physical memory available but
3715 // they are so fragmented after a long run that they can't
3716 // coalesce into large pages. Try to reserve large pages when
3717 // the system is still "fresh".
3718 shm_warning_with_errno("Failed to reserve shared memory.");
3719 return NULL;
3720 }
3721
3722 // Attach to the region.
3723 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3724
3725 // Remove shmid. If shmat() is successful, the actual shared memory segment
3726 // will be deleted when it's detached by shmdt() or when the process
3727 // terminates. If shmat() is not successful this will remove the shared
3728 // segment immediately.
3729 shmctl(shmid, IPC_RMID, NULL);
3730
3731 return addr;
3732 }
3733
3734 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3735 int error) {
3736 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3737
3738 bool warn_on_failure = UseLargePages &&
3739 (!FLAG_IS_DEFAULT(UseLargePages) ||
3740 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3741 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3742
3743 if (warn_on_failure) {
3744 char msg[128];
3745 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3746 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3747 warning("%s", msg);
3748 }
3749 }
3750
3751 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3752 char* req_addr,
3753 bool exec) {
3754 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3755 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3756 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3757
3758 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3759 char* addr = (char*)::mmap(req_addr, bytes, prot,
3760 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3761 -1, 0);
3762
3763 if (addr == MAP_FAILED) {
3764 warn_on_large_pages_failure(req_addr, bytes, errno);
3765 return NULL;
3766 }
3767
3768 assert(is_aligned(addr, os::large_page_size()), "Must be");
3769
3770 return addr;
3771 }
3772
3773 // Reserve memory using mmap(MAP_HUGETLB).
3774 // - bytes shall be a multiple of alignment.
3775 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3776 // - alignment sets the alignment at which memory shall be allocated.
3777 // It must be a multiple of allocation granularity.
3778 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3779 // req_addr or NULL.
3780 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3781 size_t alignment,
3782 char* req_addr,
3783 bool exec) {
3784 size_t large_page_size = os::large_page_size();
3785 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3786
3787 assert(is_aligned(req_addr, alignment), "Must be");
3788 assert(is_aligned(bytes, alignment), "Must be");
3789
3790 // First reserve - but not commit - the address range in small pages.
3791 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3792
3793 if (start == NULL) {
3794 return NULL;
3795 }
3796
3797 assert(is_aligned(start, alignment), "Must be");
3798
3799 char* end = start + bytes;
3800
3801 // Find the regions of the allocated chunk that can be promoted to large pages.
3802 char* lp_start = align_up(start, large_page_size);
3803 char* lp_end = align_down(end, large_page_size);
3804
3805 size_t lp_bytes = lp_end - lp_start;
3806
3807 assert(is_aligned(lp_bytes, large_page_size), "Must be");
3808
3809 if (lp_bytes == 0) {
3810 // The mapped region doesn't even span the start and the end of a large page.
3811 // Fall back to allocate a non-special area.
3812 ::munmap(start, end - start);
3813 return NULL;
3814 }
3815
3816 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3817
3818 void* result;
3819
3820 // Commit small-paged leading area.
3821 if (start != lp_start) {
3822 result = ::mmap(start, lp_start - start, prot,
3823 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3824 -1, 0);
3825 if (result == MAP_FAILED) {
3826 ::munmap(lp_start, end - lp_start);
3827 return NULL;
3828 }
3829 }
3830
3831 // Commit large-paged area.
3832 result = ::mmap(lp_start, lp_bytes, prot,
3833 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3834 -1, 0);
3835 if (result == MAP_FAILED) {
3836 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3837 // If the mmap above fails, the large pages region will be unmapped and we
3838 // have regions before and after with small pages. Release these regions.
3839 //
3840 // | mapped | unmapped | mapped |
3841 // ^ ^ ^ ^
3842 // start lp_start lp_end end
3843 //
3844 ::munmap(start, lp_start - start);
3845 ::munmap(lp_end, end - lp_end);
3846 return NULL;
3847 }
3848
3849 // Commit small-paged trailing area.
3850 if (lp_end != end) {
3851 result = ::mmap(lp_end, end - lp_end, prot,
3852 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3853 -1, 0);
3854 if (result == MAP_FAILED) {
3855 ::munmap(start, lp_end - start);
3856 return NULL;
3857 }
3858 }
3859
3860 return start;
3861 }
3862
3863 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3864 size_t alignment,
3865 char* req_addr,
3866 bool exec) {
3867 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3868 assert(is_aligned(req_addr, alignment), "Must be");
3869 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3870 assert(is_power_of_2(os::large_page_size()), "Must be");
3871 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3872
3873 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3874 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3875 } else {
3876 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3877 }
3878 }
3879
3880 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3881 char* req_addr, bool exec) {
3882 assert(UseLargePages, "only for large pages");
3883
3884 char* addr;
3885 if (UseSHM) {
3886 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3887 } else {
3888 assert(UseHugeTLBFS, "must be");
3889 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3890 }
3891
3892 if (addr != NULL) {
3893 if (UseNUMAInterleaving) {
3894 numa_make_global(addr, bytes);
3895 }
3896
3897 // The memory is committed
3898 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3899 }
3900
3901 return addr;
3902 }
3903
3904 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3905 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3906 return shmdt(base) == 0;
3907 }
3908
3909 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3910 return pd_release_memory(base, bytes);
3911 }
3912
3913 bool os::release_memory_special(char* base, size_t bytes) {
3914 bool res;
3915 if (MemTracker::tracking_level() > NMT_minimal) {
3916 Tracker tkr(Tracker::release);
3917 res = os::Linux::release_memory_special_impl(base, bytes);
3918 if (res) {
3919 tkr.record((address)base, bytes);
3920 }
3921
3922 } else {
3923 res = os::Linux::release_memory_special_impl(base, bytes);
3924 }
3925 return res;
3926 }
3927
3928 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3929 assert(UseLargePages, "only for large pages");
3930 bool res;
3931
3932 if (UseSHM) {
3933 res = os::Linux::release_memory_special_shm(base, bytes);
3934 } else {
3935 assert(UseHugeTLBFS, "must be");
3936 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3937 }
3938 return res;
3939 }
3940
3941 size_t os::large_page_size() {
3942 return _large_page_size;
3943 }
3944
3945 // With SysV SHM the entire memory region must be allocated as shared
3946 // memory.
3947 // HugeTLBFS allows application to commit large page memory on demand.
3948 // However, when committing memory with HugeTLBFS fails, the region
3949 // that was supposed to be committed will lose the old reservation
3950 // and allow other threads to steal that memory region. Because of this
3951 // behavior we can't commit HugeTLBFS memory.
3952 bool os::can_commit_large_page_memory() {
3953 return UseTransparentHugePages;
3954 }
3955
3956 bool os::can_execute_large_page_memory() {
3957 return UseTransparentHugePages || UseHugeTLBFS;
3958 }
3959
3960 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
3961 assert(file_desc >= 0, "file_desc is not valid");
3962 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
3963 if (result != NULL) {
3964 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
3965 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
3966 }
3967 }
3968 return result;
3969 }
3970
3971 // Reserve memory at an arbitrary address, only if that area is
3972 // available (and not reserved for something else).
3973
3974 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3975 const int max_tries = 10;
3976 char* base[max_tries];
3977 size_t size[max_tries];
3978 const size_t gap = 0x000000;
3979
3980 // Assert only that the size is a multiple of the page size, since
3981 // that's all that mmap requires, and since that's all we really know
3982 // about at this low abstraction level. If we need higher alignment,
3983 // we can either pass an alignment to this method or verify alignment
3984 // in one of the methods further up the call chain. See bug 5044738.
3985 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3986
3987 // Repeatedly allocate blocks until the block is allocated at the
3988 // right spot.
3989
3990 // Linux mmap allows caller to pass an address as hint; give it a try first,
3991 // if kernel honors the hint then we can return immediately.
3992 char * addr = anon_mmap(requested_addr, bytes, false);
3993 if (addr == requested_addr) {
3994 return requested_addr;
3995 }
3996
3997 if (addr != NULL) {
3998 // mmap() is successful but it fails to reserve at the requested address
3999 anon_munmap(addr, bytes);
4000 }
4001
4002 int i;
4003 for (i = 0; i < max_tries; ++i) {
4004 base[i] = reserve_memory(bytes);
4005
4006 if (base[i] != NULL) {
4007 // Is this the block we wanted?
4008 if (base[i] == requested_addr) {
4009 size[i] = bytes;
4010 break;
4011 }
4012
4013 // Does this overlap the block we wanted? Give back the overlapped
4014 // parts and try again.
4015
4016 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
4017 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
4018 unmap_memory(base[i], top_overlap);
4019 base[i] += top_overlap;
4020 size[i] = bytes - top_overlap;
4021 } else {
4022 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
4023 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
4024 unmap_memory(requested_addr, bottom_overlap);
4025 size[i] = bytes - bottom_overlap;
4026 } else {
4027 size[i] = bytes;
4028 }
4029 }
4030 }
4031 }
4032
4033 // Give back the unused reserved pieces.
4034
4035 for (int j = 0; j < i; ++j) {
4036 if (base[j] != NULL) {
4037 unmap_memory(base[j], size[j]);
4038 }
4039 }
4040
4041 if (i < max_tries) {
4042 return requested_addr;
4043 } else {
4044 return NULL;
4045 }
4046 }
4047
4048 size_t os::read(int fd, void *buf, unsigned int nBytes) {
4049 return ::read(fd, buf, nBytes);
4050 }
4051
4052 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
4053 return ::pread(fd, buf, nBytes, offset);
4054 }
4055
4056 // Short sleep, direct OS call.
4057 //
4058 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
4059 // sched_yield(2) will actually give up the CPU:
4060 //
4061 // * Alone on this pariticular CPU, keeps running.
4062 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
4063 // (pre 2.6.39).
4064 //
4065 // So calling this with 0 is an alternative.
4066 //
4067 void os::naked_short_sleep(jlong ms) {
4068 struct timespec req;
4069
4070 assert(ms < 1000, "Un-interruptable sleep, short time use only");
4071 req.tv_sec = 0;
4072 if (ms > 0) {
4073 req.tv_nsec = (ms % 1000) * 1000000;
4074 } else {
4075 req.tv_nsec = 1;
4076 }
4077
4078 nanosleep(&req, NULL);
4079
4080 return;
4081 }
4082
4083 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4084 void os::infinite_sleep() {
4085 while (true) { // sleep forever ...
4086 ::sleep(100); // ... 100 seconds at a time
4087 }
4088 }
4089
4090 // Used to convert frequent JVM_Yield() to nops
4091 bool os::dont_yield() {
4092 return DontYieldALot;
4093 }
4094
4095 void os::naked_yield() {
4096 sched_yield();
4097 }
4098
4099 ////////////////////////////////////////////////////////////////////////////////
4100 // thread priority support
4101
4102 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4103 // only supports dynamic priority, static priority must be zero. For real-time
4104 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4105 // However, for large multi-threaded applications, SCHED_RR is not only slower
4106 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4107 // of 5 runs - Sep 2005).
4108 //
4109 // The following code actually changes the niceness of kernel-thread/LWP. It
4110 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4111 // not the entire user process, and user level threads are 1:1 mapped to kernel
4112 // threads. It has always been the case, but could change in the future. For
4113 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4114 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4115
4116 int os::java_to_os_priority[CriticalPriority + 1] = {
4117 19, // 0 Entry should never be used
4118
4119 4, // 1 MinPriority
4120 3, // 2
4121 2, // 3
4122
4123 1, // 4
4124 0, // 5 NormPriority
4125 -1, // 6
4126
4127 -2, // 7
4128 -3, // 8
4129 -4, // 9 NearMaxPriority
4130
4131 -5, // 10 MaxPriority
4132
4133 -5 // 11 CriticalPriority
4134 };
4135
4136 static int prio_init() {
4137 if (ThreadPriorityPolicy == 1) {
4138 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4139 // if effective uid is not root. Perhaps, a more elegant way of doing
4140 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4141 if (geteuid() != 0) {
4142 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4143 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4144 }
4145 ThreadPriorityPolicy = 0;
4146 }
4147 }
4148 if (UseCriticalJavaThreadPriority) {
4149 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4150 }
4151 return 0;
4152 }
4153
4154 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4155 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4156
4157 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4158 return (ret == 0) ? OS_OK : OS_ERR;
4159 }
4160
4161 OSReturn os::get_native_priority(const Thread* const thread,
4162 int *priority_ptr) {
4163 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4164 *priority_ptr = java_to_os_priority[NormPriority];
4165 return OS_OK;
4166 }
4167
4168 errno = 0;
4169 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4170 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4171 }
4172
4173 ////////////////////////////////////////////////////////////////////////////////
4174 // suspend/resume support
4175
4176 // The low-level signal-based suspend/resume support is a remnant from the
4177 // old VM-suspension that used to be for java-suspension, safepoints etc,
4178 // within hotspot. Currently used by JFR's OSThreadSampler
4179 //
4180 // The remaining code is greatly simplified from the more general suspension
4181 // code that used to be used.
4182 //
4183 // The protocol is quite simple:
4184 // - suspend:
4185 // - sends a signal to the target thread
4186 // - polls the suspend state of the osthread using a yield loop
4187 // - target thread signal handler (SR_handler) sets suspend state
4188 // and blocks in sigsuspend until continued
4189 // - resume:
4190 // - sets target osthread state to continue
4191 // - sends signal to end the sigsuspend loop in the SR_handler
4192 //
4193 // Note that the SR_lock plays no role in this suspend/resume protocol,
4194 // but is checked for NULL in SR_handler as a thread termination indicator.
4195 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4196 //
4197 // Note that resume_clear_context() and suspend_save_context() are needed
4198 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4199 // which in part is used by:
4200 // - Forte Analyzer: AsyncGetCallTrace()
4201 // - StackBanging: get_frame_at_stack_banging_point()
4202
4203 static void resume_clear_context(OSThread *osthread) {
4204 osthread->set_ucontext(NULL);
4205 osthread->set_siginfo(NULL);
4206 }
4207
4208 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4209 ucontext_t* context) {
4210 osthread->set_ucontext(context);
4211 osthread->set_siginfo(siginfo);
4212 }
4213
4214 // Handler function invoked when a thread's execution is suspended or
4215 // resumed. We have to be careful that only async-safe functions are
4216 // called here (Note: most pthread functions are not async safe and
4217 // should be avoided.)
4218 //
4219 // Note: sigwait() is a more natural fit than sigsuspend() from an
4220 // interface point of view, but sigwait() prevents the signal hander
4221 // from being run. libpthread would get very confused by not having
4222 // its signal handlers run and prevents sigwait()'s use with the
4223 // mutex granting granting signal.
4224 //
4225 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4226 //
4227 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4228 // Save and restore errno to avoid confusing native code with EINTR
4229 // after sigsuspend.
4230 int old_errno = errno;
4231
4232 Thread* thread = Thread::current_or_null_safe();
4233 assert(thread != NULL, "Missing current thread in SR_handler");
4234
4235 // On some systems we have seen signal delivery get "stuck" until the signal
4236 // mask is changed as part of thread termination. Check that the current thread
4237 // has not already terminated (via SR_lock()) - else the following assertion
4238 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4239 // destructor has completed.
4240
4241 if (thread->SR_lock() == NULL) {
4242 return;
4243 }
4244
4245 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4246
4247 OSThread* osthread = thread->osthread();
4248
4249 os::SuspendResume::State current = osthread->sr.state();
4250 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4251 suspend_save_context(osthread, siginfo, context);
4252
4253 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4254 os::SuspendResume::State state = osthread->sr.suspended();
4255 if (state == os::SuspendResume::SR_SUSPENDED) {
4256 sigset_t suspend_set; // signals for sigsuspend()
4257 sigemptyset(&suspend_set);
4258 // get current set of blocked signals and unblock resume signal
4259 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4260 sigdelset(&suspend_set, SR_signum);
4261
4262 sr_semaphore.signal();
4263 // wait here until we are resumed
4264 while (1) {
4265 sigsuspend(&suspend_set);
4266
4267 os::SuspendResume::State result = osthread->sr.running();
4268 if (result == os::SuspendResume::SR_RUNNING) {
4269 sr_semaphore.signal();
4270 break;
4271 }
4272 }
4273
4274 } else if (state == os::SuspendResume::SR_RUNNING) {
4275 // request was cancelled, continue
4276 } else {
4277 ShouldNotReachHere();
4278 }
4279
4280 resume_clear_context(osthread);
4281 } else if (current == os::SuspendResume::SR_RUNNING) {
4282 // request was cancelled, continue
4283 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4284 // ignore
4285 } else {
4286 // ignore
4287 }
4288
4289 errno = old_errno;
4290 }
4291
4292 static int SR_initialize() {
4293 struct sigaction act;
4294 char *s;
4295
4296 // Get signal number to use for suspend/resume
4297 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4298 int sig = ::strtol(s, 0, 10);
4299 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4300 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4301 SR_signum = sig;
4302 } else {
4303 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4304 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4305 }
4306 }
4307
4308 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4309 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4310
4311 sigemptyset(&SR_sigset);
4312 sigaddset(&SR_sigset, SR_signum);
4313
4314 // Set up signal handler for suspend/resume
4315 act.sa_flags = SA_RESTART|SA_SIGINFO;
4316 act.sa_handler = (void (*)(int)) SR_handler;
4317
4318 // SR_signum is blocked by default.
4319 // 4528190 - We also need to block pthread restart signal (32 on all
4320 // supported Linux platforms). Note that LinuxThreads need to block
4321 // this signal for all threads to work properly. So we don't have
4322 // to use hard-coded signal number when setting up the mask.
4323 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4324
4325 if (sigaction(SR_signum, &act, 0) == -1) {
4326 return -1;
4327 }
4328
4329 // Save signal flag
4330 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4331 return 0;
4332 }
4333
4334 static int sr_notify(OSThread* osthread) {
4335 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4336 assert_status(status == 0, status, "pthread_kill");
4337 return status;
4338 }
4339
4340 // "Randomly" selected value for how long we want to spin
4341 // before bailing out on suspending a thread, also how often
4342 // we send a signal to a thread we want to resume
4343 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4344 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4345
4346 // returns true on success and false on error - really an error is fatal
4347 // but this seems the normal response to library errors
4348 static bool do_suspend(OSThread* osthread) {
4349 assert(osthread->sr.is_running(), "thread should be running");
4350 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4351
4352 // mark as suspended and send signal
4353 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4354 // failed to switch, state wasn't running?
4355 ShouldNotReachHere();
4356 return false;
4357 }
4358
4359 if (sr_notify(osthread) != 0) {
4360 ShouldNotReachHere();
4361 }
4362
4363 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4364 while (true) {
4365 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4366 break;
4367 } else {
4368 // timeout
4369 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4370 if (cancelled == os::SuspendResume::SR_RUNNING) {
4371 return false;
4372 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4373 // make sure that we consume the signal on the semaphore as well
4374 sr_semaphore.wait();
4375 break;
4376 } else {
4377 ShouldNotReachHere();
4378 return false;
4379 }
4380 }
4381 }
4382
4383 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4384 return true;
4385 }
4386
4387 static void do_resume(OSThread* osthread) {
4388 assert(osthread->sr.is_suspended(), "thread should be suspended");
4389 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4390
4391 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4392 // failed to switch to WAKEUP_REQUEST
4393 ShouldNotReachHere();
4394 return;
4395 }
4396
4397 while (true) {
4398 if (sr_notify(osthread) == 0) {
4399 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4400 if (osthread->sr.is_running()) {
4401 return;
4402 }
4403 }
4404 } else {
4405 ShouldNotReachHere();
4406 }
4407 }
4408
4409 guarantee(osthread->sr.is_running(), "Must be running!");
4410 }
4411
4412 ///////////////////////////////////////////////////////////////////////////////////
4413 // signal handling (except suspend/resume)
4414
4415 // This routine may be used by user applications as a "hook" to catch signals.
4416 // The user-defined signal handler must pass unrecognized signals to this
4417 // routine, and if it returns true (non-zero), then the signal handler must
4418 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4419 // routine will never retun false (zero), but instead will execute a VM panic
4420 // routine kill the process.
4421 //
4422 // If this routine returns false, it is OK to call it again. This allows
4423 // the user-defined signal handler to perform checks either before or after
4424 // the VM performs its own checks. Naturally, the user code would be making
4425 // a serious error if it tried to handle an exception (such as a null check
4426 // or breakpoint) that the VM was generating for its own correct operation.
4427 //
4428 // This routine may recognize any of the following kinds of signals:
4429 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4430 // It should be consulted by handlers for any of those signals.
4431 //
4432 // The caller of this routine must pass in the three arguments supplied
4433 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4434 // field of the structure passed to sigaction(). This routine assumes that
4435 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4436 //
4437 // Note that the VM will print warnings if it detects conflicting signal
4438 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4439 //
4440 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4441 siginfo_t* siginfo,
4442 void* ucontext,
4443 int abort_if_unrecognized);
4444
4445 static void signalHandler(int sig, siginfo_t* info, void* uc) {
4446 assert(info != NULL && uc != NULL, "it must be old kernel");
4447 int orig_errno = errno; // Preserve errno value over signal handler.
4448 JVM_handle_linux_signal(sig, info, uc, true);
4449 errno = orig_errno;
4450 }
4451
4452
4453 // This boolean allows users to forward their own non-matching signals
4454 // to JVM_handle_linux_signal, harmlessly.
4455 bool os::Linux::signal_handlers_are_installed = false;
4456
4457 // For signal-chaining
4458 struct sigaction sigact[NSIG];
4459 uint64_t sigs = 0;
4460 #if (64 < NSIG-1)
4461 #error "Not all signals can be encoded in sigs. Adapt its type!"
4462 #endif
4463 bool os::Linux::libjsig_is_loaded = false;
4464 typedef struct sigaction *(*get_signal_t)(int);
4465 get_signal_t os::Linux::get_signal_action = NULL;
4466
4467 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4468 struct sigaction *actp = NULL;
4469
4470 if (libjsig_is_loaded) {
4471 // Retrieve the old signal handler from libjsig
4472 actp = (*get_signal_action)(sig);
4473 }
4474 if (actp == NULL) {
4475 // Retrieve the preinstalled signal handler from jvm
4476 actp = get_preinstalled_handler(sig);
4477 }
4478
4479 return actp;
4480 }
4481
4482 static bool call_chained_handler(struct sigaction *actp, int sig,
4483 siginfo_t *siginfo, void *context) {
4484 // Call the old signal handler
4485 if (actp->sa_handler == SIG_DFL) {
4486 // It's more reasonable to let jvm treat it as an unexpected exception
4487 // instead of taking the default action.
4488 return false;
4489 } else if (actp->sa_handler != SIG_IGN) {
4490 if ((actp->sa_flags & SA_NODEFER) == 0) {
4491 // automaticlly block the signal
4492 sigaddset(&(actp->sa_mask), sig);
4493 }
4494
4495 sa_handler_t hand = NULL;
4496 sa_sigaction_t sa = NULL;
4497 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4498 // retrieve the chained handler
4499 if (siginfo_flag_set) {
4500 sa = actp->sa_sigaction;
4501 } else {
4502 hand = actp->sa_handler;
4503 }
4504
4505 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4506 actp->sa_handler = SIG_DFL;
4507 }
4508
4509 // try to honor the signal mask
4510 sigset_t oset;
4511 sigemptyset(&oset);
4512 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4513
4514 // call into the chained handler
4515 if (siginfo_flag_set) {
4516 (*sa)(sig, siginfo, context);
4517 } else {
4518 (*hand)(sig);
4519 }
4520
4521 // restore the signal mask
4522 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4523 }
4524 // Tell jvm's signal handler the signal is taken care of.
4525 return true;
4526 }
4527
4528 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4529 bool chained = false;
4530 // signal-chaining
4531 if (UseSignalChaining) {
4532 struct sigaction *actp = get_chained_signal_action(sig);
4533 if (actp != NULL) {
4534 chained = call_chained_handler(actp, sig, siginfo, context);
4535 }
4536 }
4537 return chained;
4538 }
4539
4540 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4541 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4542 return &sigact[sig];
4543 }
4544 return NULL;
4545 }
4546
4547 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4548 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4549 sigact[sig] = oldAct;
4550 sigs |= (uint64_t)1 << (sig-1);
4551 }
4552
4553 // for diagnostic
4554 int sigflags[NSIG];
4555
4556 int os::Linux::get_our_sigflags(int sig) {
4557 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4558 return sigflags[sig];
4559 }
4560
4561 void os::Linux::set_our_sigflags(int sig, int flags) {
4562 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4563 if (sig > 0 && sig < NSIG) {
4564 sigflags[sig] = flags;
4565 }
4566 }
4567
4568 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4569 // Check for overwrite.
4570 struct sigaction oldAct;
4571 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4572
4573 void* oldhand = oldAct.sa_sigaction
4574 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4575 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4576 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4577 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4578 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4579 if (AllowUserSignalHandlers || !set_installed) {
4580 // Do not overwrite; user takes responsibility to forward to us.
4581 return;
4582 } else if (UseSignalChaining) {
4583 // save the old handler in jvm
4584 save_preinstalled_handler(sig, oldAct);
4585 // libjsig also interposes the sigaction() call below and saves the
4586 // old sigaction on it own.
4587 } else {
4588 fatal("Encountered unexpected pre-existing sigaction handler "
4589 "%#lx for signal %d.", (long)oldhand, sig);
4590 }
4591 }
4592
4593 struct sigaction sigAct;
4594 sigfillset(&(sigAct.sa_mask));
4595 sigAct.sa_handler = SIG_DFL;
4596 if (!set_installed) {
4597 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4598 } else {
4599 sigAct.sa_sigaction = signalHandler;
4600 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4601 }
4602 // Save flags, which are set by ours
4603 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4604 sigflags[sig] = sigAct.sa_flags;
4605
4606 int ret = sigaction(sig, &sigAct, &oldAct);
4607 assert(ret == 0, "check");
4608
4609 void* oldhand2 = oldAct.sa_sigaction
4610 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4611 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4612 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4613 }
4614
4615 // install signal handlers for signals that HotSpot needs to
4616 // handle in order to support Java-level exception handling.
4617
4618 void os::Linux::install_signal_handlers() {
4619 if (!signal_handlers_are_installed) {
4620 signal_handlers_are_installed = true;
4621
4622 // signal-chaining
4623 typedef void (*signal_setting_t)();
4624 signal_setting_t begin_signal_setting = NULL;
4625 signal_setting_t end_signal_setting = NULL;
4626 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4627 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4628 if (begin_signal_setting != NULL) {
4629 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4630 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4631 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4632 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4633 libjsig_is_loaded = true;
4634 assert(UseSignalChaining, "should enable signal-chaining");
4635 }
4636 if (libjsig_is_loaded) {
4637 // Tell libjsig jvm is setting signal handlers
4638 (*begin_signal_setting)();
4639 }
4640
4641 set_signal_handler(SIGSEGV, true);
4642 set_signal_handler(SIGPIPE, true);
4643 set_signal_handler(SIGBUS, true);
4644 set_signal_handler(SIGILL, true);
4645 set_signal_handler(SIGFPE, true);
4646 #if defined(PPC64)
4647 set_signal_handler(SIGTRAP, true);
4648 #endif
4649 set_signal_handler(SIGXFSZ, true);
4650
4651 if (libjsig_is_loaded) {
4652 // Tell libjsig jvm finishes setting signal handlers
4653 (*end_signal_setting)();
4654 }
4655
4656 // We don't activate signal checker if libjsig is in place, we trust ourselves
4657 // and if UserSignalHandler is installed all bets are off.
4658 // Log that signal checking is off only if -verbose:jni is specified.
4659 if (CheckJNICalls) {
4660 if (libjsig_is_loaded) {
4661 if (PrintJNIResolving) {
4662 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4663 }
4664 check_signals = false;
4665 }
4666 if (AllowUserSignalHandlers) {
4667 if (PrintJNIResolving) {
4668 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4669 }
4670 check_signals = false;
4671 }
4672 }
4673 }
4674 }
4675
4676 // This is the fastest way to get thread cpu time on Linux.
4677 // Returns cpu time (user+sys) for any thread, not only for current.
4678 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4679 // It might work on 2.6.10+ with a special kernel/glibc patch.
4680 // For reference, please, see IEEE Std 1003.1-2004:
4681 // http://www.unix.org/single_unix_specification
4682
4683 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4684 struct timespec tp;
4685 int rc = os::Posix::clock_gettime(clockid, &tp);
4686 assert(rc == 0, "clock_gettime is expected to return 0 code");
4687
4688 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4689 }
4690
4691 void os::Linux::initialize_os_info() {
4692 assert(_os_version == 0, "OS info already initialized");
4693
4694 struct utsname _uname;
4695
4696 uint32_t major;
4697 uint32_t minor;
4698 uint32_t fix;
4699
4700 int rc;
4701
4702 // Kernel version is unknown if
4703 // verification below fails.
4704 _os_version = 0x01000000;
4705
4706 rc = uname(&_uname);
4707 if (rc != -1) {
4708
4709 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4710 if (rc == 3) {
4711
4712 if (major < 256 && minor < 256 && fix < 256) {
4713 // Kernel version format is as expected,
4714 // set it overriding unknown state.
4715 _os_version = (major << 16) |
4716 (minor << 8 ) |
4717 (fix << 0 ) ;
4718 }
4719 }
4720 }
4721 }
4722
4723 uint32_t os::Linux::os_version() {
4724 assert(_os_version != 0, "not initialized");
4725 return _os_version & 0x00FFFFFF;
4726 }
4727
4728 bool os::Linux::os_version_is_known() {
4729 assert(_os_version != 0, "not initialized");
4730 return _os_version & 0x01000000 ? false : true;
4731 }
4732
4733 /////
4734 // glibc on Linux platform uses non-documented flag
4735 // to indicate, that some special sort of signal
4736 // trampoline is used.
4737 // We will never set this flag, and we should
4738 // ignore this flag in our diagnostic
4739 #ifdef SIGNIFICANT_SIGNAL_MASK
4740 #undef SIGNIFICANT_SIGNAL_MASK
4741 #endif
4742 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4743
4744 static const char* get_signal_handler_name(address handler,
4745 char* buf, int buflen) {
4746 int offset = 0;
4747 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4748 if (found) {
4749 // skip directory names
4750 const char *p1, *p2;
4751 p1 = buf;
4752 size_t len = strlen(os::file_separator());
4753 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4754 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4755 } else {
4756 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4757 }
4758 return buf;
4759 }
4760
4761 static void print_signal_handler(outputStream* st, int sig,
4762 char* buf, size_t buflen) {
4763 struct sigaction sa;
4764
4765 sigaction(sig, NULL, &sa);
4766
4767 // See comment for SIGNIFICANT_SIGNAL_MASK define
4768 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4769
4770 st->print("%s: ", os::exception_name(sig, buf, buflen));
4771
4772 address handler = (sa.sa_flags & SA_SIGINFO)
4773 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4774 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4775
4776 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4777 st->print("SIG_DFL");
4778 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4779 st->print("SIG_IGN");
4780 } else {
4781 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4782 }
4783
4784 st->print(", sa_mask[0]=");
4785 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4786
4787 address rh = VMError::get_resetted_sighandler(sig);
4788 // May be, handler was resetted by VMError?
4789 if (rh != NULL) {
4790 handler = rh;
4791 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4792 }
4793
4794 st->print(", sa_flags=");
4795 os::Posix::print_sa_flags(st, sa.sa_flags);
4796
4797 // Check: is it our handler?
4798 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4799 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4800 // It is our signal handler
4801 // check for flags, reset system-used one!
4802 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4803 st->print(
4804 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4805 os::Linux::get_our_sigflags(sig));
4806 }
4807 }
4808 st->cr();
4809 }
4810
4811
4812 #define DO_SIGNAL_CHECK(sig) \
4813 do { \
4814 if (!sigismember(&check_signal_done, sig)) { \
4815 os::Linux::check_signal_handler(sig); \
4816 } \
4817 } while (0)
4818
4819 // This method is a periodic task to check for misbehaving JNI applications
4820 // under CheckJNI, we can add any periodic checks here
4821
4822 void os::run_periodic_checks() {
4823 if (check_signals == false) return;
4824
4825 // SEGV and BUS if overridden could potentially prevent
4826 // generation of hs*.log in the event of a crash, debugging
4827 // such a case can be very challenging, so we absolutely
4828 // check the following for a good measure:
4829 DO_SIGNAL_CHECK(SIGSEGV);
4830 DO_SIGNAL_CHECK(SIGILL);
4831 DO_SIGNAL_CHECK(SIGFPE);
4832 DO_SIGNAL_CHECK(SIGBUS);
4833 DO_SIGNAL_CHECK(SIGPIPE);
4834 DO_SIGNAL_CHECK(SIGXFSZ);
4835 #if defined(PPC64)
4836 DO_SIGNAL_CHECK(SIGTRAP);
4837 #endif
4838
4839 // ReduceSignalUsage allows the user to override these handlers
4840 // see comments at the very top and jvm_md.h
4841 if (!ReduceSignalUsage) {
4842 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4843 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4844 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4845 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4846 }
4847
4848 DO_SIGNAL_CHECK(SR_signum);
4849 }
4850
4851 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4852
4853 static os_sigaction_t os_sigaction = NULL;
4854
4855 void os::Linux::check_signal_handler(int sig) {
4856 char buf[O_BUFLEN];
4857 address jvmHandler = NULL;
4858
4859
4860 struct sigaction act;
4861 if (os_sigaction == NULL) {
4862 // only trust the default sigaction, in case it has been interposed
4863 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4864 if (os_sigaction == NULL) return;
4865 }
4866
4867 os_sigaction(sig, (struct sigaction*)NULL, &act);
4868
4869
4870 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4871
4872 address thisHandler = (act.sa_flags & SA_SIGINFO)
4873 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4874 : CAST_FROM_FN_PTR(address, act.sa_handler);
4875
4876
4877 switch (sig) {
4878 case SIGSEGV:
4879 case SIGBUS:
4880 case SIGFPE:
4881 case SIGPIPE:
4882 case SIGILL:
4883 case SIGXFSZ:
4884 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4885 break;
4886
4887 case SHUTDOWN1_SIGNAL:
4888 case SHUTDOWN2_SIGNAL:
4889 case SHUTDOWN3_SIGNAL:
4890 case BREAK_SIGNAL:
4891 jvmHandler = (address)user_handler();
4892 break;
4893
4894 default:
4895 if (sig == SR_signum) {
4896 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4897 } else {
4898 return;
4899 }
4900 break;
4901 }
4902
4903 if (thisHandler != jvmHandler) {
4904 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4905 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4906 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4907 // No need to check this sig any longer
4908 sigaddset(&check_signal_done, sig);
4909 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4910 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4911 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4912 exception_name(sig, buf, O_BUFLEN));
4913 }
4914 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4915 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4916 tty->print("expected:");
4917 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4918 tty->cr();
4919 tty->print(" found:");
4920 os::Posix::print_sa_flags(tty, act.sa_flags);
4921 tty->cr();
4922 // No need to check this sig any longer
4923 sigaddset(&check_signal_done, sig);
4924 }
4925
4926 // Dump all the signal
4927 if (sigismember(&check_signal_done, sig)) {
4928 print_signal_handlers(tty, buf, O_BUFLEN);
4929 }
4930 }
4931
4932 extern void report_error(char* file_name, int line_no, char* title,
4933 char* format, ...);
4934
4935 // this is called _before_ most of the global arguments have been parsed
4936 void os::init(void) {
4937 char dummy; // used to get a guess on initial stack address
4938
4939 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4940
4941 init_random(1234567);
4942
4943 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4944 if (Linux::page_size() == -1) {
4945 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4946 os::strerror(errno));
4947 }
4948 init_page_sizes((size_t) Linux::page_size());
4949
4950 Linux::initialize_system_info();
4951
4952 Linux::initialize_os_info();
4953
4954 // _main_thread points to the thread that created/loaded the JVM.
4955 Linux::_main_thread = pthread_self();
4956
4957 // retrieve entry point for pthread_setname_np
4958 Linux::_pthread_setname_np =
4959 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4960
4961 os::Posix::init();
4962
4963 initial_time_count = javaTimeNanos();
4964
4965 // Always warn if no monotonic clock available
4966 if (!os::Posix::supports_monotonic_clock()) {
4967 warning("No monotonic clock was available - timed services may " \
4968 "be adversely affected if the time-of-day clock changes");
4969 }
4970 }
4971
4972 // To install functions for atexit system call
4973 extern "C" {
4974 static void perfMemory_exit_helper() {
4975 perfMemory_exit();
4976 }
4977 }
4978
4979 void os::pd_init_container_support() {
4980 OSContainer::init();
4981 }
4982
4983 // this is called _after_ the global arguments have been parsed
4984 jint os::init_2(void) {
4985
4986 // This could be set after os::Posix::init() but all platforms
4987 // have to set it the same so we have to mirror Solaris.
4988 DEBUG_ONLY(os::set_mutex_init_done();)
4989
4990 os::Posix::init_2();
4991
4992 Linux::fast_thread_clock_init();
4993
4994 // initialize suspend/resume support - must do this before signal_sets_init()
4995 if (SR_initialize() != 0) {
4996 perror("SR_initialize failed");
4997 return JNI_ERR;
4998 }
4999
5000 Linux::signal_sets_init();
5001 Linux::install_signal_handlers();
5002 // Initialize data for jdk.internal.misc.Signal
5003 if (!ReduceSignalUsage) {
5004 jdk_misc_signal_init();
5005 }
5006
5007 // Check and sets minimum stack sizes against command line options
5008 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
5009 return JNI_ERR;
5010 }
5011
5012 suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
5013 if (!suppress_primordial_thread_resolution) {
5014 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5015 }
5016
5017 #if defined(IA32)
5018 workaround_expand_exec_shield_cs_limit();
5019 #endif
5020
5021 Linux::libpthread_init();
5022 Linux::sched_getcpu_init();
5023 log_info(os)("HotSpot is running with %s, %s",
5024 Linux::glibc_version(), Linux::libpthread_version());
5025
5026 if (UseNUMA) {
5027 if (!Linux::libnuma_init()) {
5028 UseNUMA = false;
5029 } else {
5030
5031 // Identify whether running in membind or interleave mode.
5032 struct bitmask *bmp;
5033 bool _is_membind = false;
5034 bool _is_interleaved = false;
5035 char _buf[BUFSIZ] = {'\0'};
5036 char *_bufptr = _buf;
5037
5038 log_info(os)("UseNUMA is enabled");
5039 // Check for membind mode.
5040 bmp = Linux::_numa_membind_ptr;
5041 for (int node = 0; node <= Linux::numa_max_node() ; node++) {
5042 if (Linux::_numa_bitmask_isbitset(bmp, node)) {
5043 _is_membind = true;
5044 }
5045 }
5046
5047 // Check for interleave mode.
5048 bmp = Linux::_numa_interleave_ptr;
5049 for (int node = 0; node <= Linux::numa_max_node() ; node++) {
5050 if (Linux::_numa_bitmask_isbitset(bmp, node)) {
5051 _is_interleaved = true;
5052 // Set membind to false as interleave mode allows all nodes to be used.
5053 _is_membind = false;
5054 }
5055 }
5056
5057 if (_is_membind) {
5058 bmp = Linux::_numa_membind_ptr;
5059 Linux::set_numa_interleave_ptr (NULL);
5060 log_info(os) (" Java is configured to run in membind mode");
5061 }
5062
5063 if (_is_interleaved) {
5064 bmp = Linux::_numa_interleave_ptr;
5065 Linux::set_numa_membind_ptr (NULL);
5066 log_info(os) (" Java is configured to run in interleave mode");
5067 }
5068
5069 for (int node = 0; node <= Linux::numa_max_node() ; node++) {
5070 if (Linux::_numa_bitmask_isbitset(bmp, node)) {
5071 _bufptr += sprintf (_bufptr, "%d, ", node);
5072 }
5073 }
5074 _bufptr[-2 ] = '\0';
5075 log_info(os) (" Heap will be configured using NUMA memory nodes: %s", _buf);
5076 }
5077
5078
5079 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5080 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5081 // we can make the adaptive lgrp chunk resizing work. If the user specified both
5082 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5083 // and disable adaptive resizing.
5084 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5085 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5086 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5087 UseAdaptiveSizePolicy = false;
5088 UseAdaptiveNUMAChunkSizing = false;
5089 }
5090 }
5091
5092 if (!UseNUMA && ForceNUMA) {
5093 UseNUMA = true;
5094 }
5095 }
5096
5097 if (MaxFDLimit) {
5098 // set the number of file descriptors to max. print out error
5099 // if getrlimit/setrlimit fails but continue regardless.
5100 struct rlimit nbr_files;
5101 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5102 if (status != 0) {
5103 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5104 } else {
5105 nbr_files.rlim_cur = nbr_files.rlim_max;
5106 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5107 if (status != 0) {
5108 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5109 }
5110 }
5111 }
5112
5113 // Initialize lock used to serialize thread creation (see os::create_thread)
5114 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5115
5116 // at-exit methods are called in the reverse order of their registration.
5117 // atexit functions are called on return from main or as a result of a
5118 // call to exit(3C). There can be only 32 of these functions registered
5119 // and atexit() does not set errno.
5120
5121 if (PerfAllowAtExitRegistration) {
5122 // only register atexit functions if PerfAllowAtExitRegistration is set.
5123 // atexit functions can be delayed until process exit time, which
5124 // can be problematic for embedded VM situations. Embedded VMs should
5125 // call DestroyJavaVM() to assure that VM resources are released.
5126
5127 // note: perfMemory_exit_helper atexit function may be removed in
5128 // the future if the appropriate cleanup code can be added to the
5129 // VM_Exit VMOperation's doit method.
5130 if (atexit(perfMemory_exit_helper) != 0) {
5131 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5132 }
5133 }
5134
5135 // initialize thread priority policy
5136 prio_init();
5137
5138 if (!FLAG_IS_DEFAULT(AllocateHeapAt)) {
5139 set_coredump_filter(false /*largepages*/, true /*dax_shared*/);
5140 }
5141 return JNI_OK;
5142 }
5143
5144 // Mark the polling page as unreadable
5145 void os::make_polling_page_unreadable(void) {
5146 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5147 fatal("Could not disable polling page");
5148 }
5149 }
5150
5151 // Mark the polling page as readable
5152 void os::make_polling_page_readable(void) {
5153 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5154 fatal("Could not enable polling page");
5155 }
5156 }
5157
5158 // older glibc versions don't have this macro (which expands to
5159 // an optimized bit-counting function) so we have to roll our own
5160 #ifndef CPU_COUNT
5161
5162 static int _cpu_count(const cpu_set_t* cpus) {
5163 int count = 0;
5164 // only look up to the number of configured processors
5165 for (int i = 0; i < os::processor_count(); i++) {
5166 if (CPU_ISSET(i, cpus)) {
5167 count++;
5168 }
5169 }
5170 return count;
5171 }
5172
5173 #define CPU_COUNT(cpus) _cpu_count(cpus)
5174
5175 #endif // CPU_COUNT
5176
5177 // Get the current number of available processors for this process.
5178 // This value can change at any time during a process's lifetime.
5179 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5180 // If it appears there may be more than 1024 processors then we do a
5181 // dynamic check - see 6515172 for details.
5182 // If anything goes wrong we fallback to returning the number of online
5183 // processors - which can be greater than the number available to the process.
5184 int os::Linux::active_processor_count() {
5185 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5186 cpu_set_t* cpus_p = &cpus;
5187 int cpus_size = sizeof(cpu_set_t);
5188
5189 int configured_cpus = os::processor_count(); // upper bound on available cpus
5190 int cpu_count = 0;
5191
5192 // old build platforms may not support dynamic cpu sets
5193 #ifdef CPU_ALLOC
5194
5195 // To enable easy testing of the dynamic path on different platforms we
5196 // introduce a diagnostic flag: UseCpuAllocPath
5197 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5198 // kernel may use a mask bigger than cpu_set_t
5199 log_trace(os)("active_processor_count: using dynamic path %s"
5200 "- configured processors: %d",
5201 UseCpuAllocPath ? "(forced) " : "",
5202 configured_cpus);
5203 cpus_p = CPU_ALLOC(configured_cpus);
5204 if (cpus_p != NULL) {
5205 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5206 // zero it just to be safe
5207 CPU_ZERO_S(cpus_size, cpus_p);
5208 }
5209 else {
5210 // failed to allocate so fallback to online cpus
5211 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5212 log_trace(os)("active_processor_count: "
5213 "CPU_ALLOC failed (%s) - using "
5214 "online processor count: %d",
5215 os::strerror(errno), online_cpus);
5216 return online_cpus;
5217 }
5218 }
5219 else {
5220 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5221 configured_cpus);
5222 }
5223 #else // CPU_ALLOC
5224 // these stubs won't be executed
5225 #define CPU_COUNT_S(size, cpus) -1
5226 #define CPU_FREE(cpus)
5227
5228 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5229 configured_cpus);
5230 #endif // CPU_ALLOC
5231
5232 // pid 0 means the current thread - which we have to assume represents the process
5233 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5234 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5235 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5236 }
5237 else {
5238 cpu_count = CPU_COUNT(cpus_p);
5239 }
5240 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5241 }
5242 else {
5243 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5244 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5245 "which may exceed available processors", os::strerror(errno), cpu_count);
5246 }
5247
5248 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5249 CPU_FREE(cpus_p);
5250 }
5251
5252 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5253 return cpu_count;
5254 }
5255
5256 // Determine the active processor count from one of
5257 // three different sources:
5258 //
5259 // 1. User option -XX:ActiveProcessorCount
5260 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5261 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5262 //
5263 // Option 1, if specified, will always override.
5264 // If the cgroup subsystem is active and configured, we
5265 // will return the min of the cgroup and option 2 results.
5266 // This is required since tools, such as numactl, that
5267 // alter cpu affinity do not update cgroup subsystem
5268 // cpuset configuration files.
5269 int os::active_processor_count() {
5270 // User has overridden the number of active processors
5271 if (ActiveProcessorCount > 0) {
5272 log_trace(os)("active_processor_count: "
5273 "active processor count set by user : %d",
5274 ActiveProcessorCount);
5275 return ActiveProcessorCount;
5276 }
5277
5278 int active_cpus;
5279 if (OSContainer::is_containerized()) {
5280 active_cpus = OSContainer::active_processor_count();
5281 log_trace(os)("active_processor_count: determined by OSContainer: %d",
5282 active_cpus);
5283 } else {
5284 active_cpus = os::Linux::active_processor_count();
5285 }
5286
5287 return active_cpus;
5288 }
5289
5290 uint os::processor_id() {
5291 const int id = Linux::sched_getcpu();
5292 assert(id >= 0 && id < _processor_count, "Invalid processor id");
5293 return (uint)id;
5294 }
5295
5296 void os::set_native_thread_name(const char *name) {
5297 if (Linux::_pthread_setname_np) {
5298 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5299 snprintf(buf, sizeof(buf), "%s", name);
5300 buf[sizeof(buf) - 1] = '\0';
5301 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5302 // ERANGE should not happen; all other errors should just be ignored.
5303 assert(rc != ERANGE, "pthread_setname_np failed");
5304 }
5305 }
5306
5307 bool os::distribute_processes(uint length, uint* distribution) {
5308 // Not yet implemented.
5309 return false;
5310 }
5311
5312 bool os::bind_to_processor(uint processor_id) {
5313 // Not yet implemented.
5314 return false;
5315 }
5316
5317 ///
5318
5319 void os::SuspendedThreadTask::internal_do_task() {
5320 if (do_suspend(_thread->osthread())) {
5321 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5322 do_task(context);
5323 do_resume(_thread->osthread());
5324 }
5325 }
5326
5327 ////////////////////////////////////////////////////////////////////////////////
5328 // debug support
5329
5330 bool os::find(address addr, outputStream* st) {
5331 Dl_info dlinfo;
5332 memset(&dlinfo, 0, sizeof(dlinfo));
5333 if (dladdr(addr, &dlinfo) != 0) {
5334 st->print(PTR_FORMAT ": ", p2i(addr));
5335 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5336 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5337 p2i(addr) - p2i(dlinfo.dli_saddr));
5338 } else if (dlinfo.dli_fbase != NULL) {
5339 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5340 } else {
5341 st->print("<absolute address>");
5342 }
5343 if (dlinfo.dli_fname != NULL) {
5344 st->print(" in %s", dlinfo.dli_fname);
5345 }
5346 if (dlinfo.dli_fbase != NULL) {
5347 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5348 }
5349 st->cr();
5350
5351 if (Verbose) {
5352 // decode some bytes around the PC
5353 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5354 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5355 address lowest = (address) dlinfo.dli_sname;
5356 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5357 if (begin < lowest) begin = lowest;
5358 Dl_info dlinfo2;
5359 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5360 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5361 end = (address) dlinfo2.dli_saddr;
5362 }
5363 Disassembler::decode(begin, end, st);
5364 }
5365 return true;
5366 }
5367 return false;
5368 }
5369
5370 ////////////////////////////////////////////////////////////////////////////////
5371 // misc
5372
5373 // This does not do anything on Linux. This is basically a hook for being
5374 // able to use structured exception handling (thread-local exception filters)
5375 // on, e.g., Win32.
5376 void
5377 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5378 JavaCallArguments* args, Thread* thread) {
5379 f(value, method, args, thread);
5380 }
5381
5382 void os::print_statistics() {
5383 }
5384
5385 bool os::message_box(const char* title, const char* message) {
5386 int i;
5387 fdStream err(defaultStream::error_fd());
5388 for (i = 0; i < 78; i++) err.print_raw("=");
5389 err.cr();
5390 err.print_raw_cr(title);
5391 for (i = 0; i < 78; i++) err.print_raw("-");
5392 err.cr();
5393 err.print_raw_cr(message);
5394 for (i = 0; i < 78; i++) err.print_raw("=");
5395 err.cr();
5396
5397 char buf[16];
5398 // Prevent process from exiting upon "read error" without consuming all CPU
5399 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5400
5401 return buf[0] == 'y' || buf[0] == 'Y';
5402 }
5403
5404 // Is a (classpath) directory empty?
5405 bool os::dir_is_empty(const char* path) {
5406 DIR *dir = NULL;
5407 struct dirent *ptr;
5408
5409 dir = opendir(path);
5410 if (dir == NULL) return true;
5411
5412 // Scan the directory
5413 bool result = true;
5414 while (result && (ptr = readdir(dir)) != NULL) {
5415 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5416 result = false;
5417 }
5418 }
5419 closedir(dir);
5420 return result;
5421 }
5422
5423 // This code originates from JDK's sysOpen and open64_w
5424 // from src/solaris/hpi/src/system_md.c
5425
5426 int os::open(const char *path, int oflag, int mode) {
5427 if (strlen(path) > MAX_PATH - 1) {
5428 errno = ENAMETOOLONG;
5429 return -1;
5430 }
5431
5432 // All file descriptors that are opened in the Java process and not
5433 // specifically destined for a subprocess should have the close-on-exec
5434 // flag set. If we don't set it, then careless 3rd party native code
5435 // might fork and exec without closing all appropriate file descriptors
5436 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5437 // turn might:
5438 //
5439 // - cause end-of-file to fail to be detected on some file
5440 // descriptors, resulting in mysterious hangs, or
5441 //
5442 // - might cause an fopen in the subprocess to fail on a system
5443 // suffering from bug 1085341.
5444 //
5445 // (Yes, the default setting of the close-on-exec flag is a Unix
5446 // design flaw)
5447 //
5448 // See:
5449 // 1085341: 32-bit stdio routines should support file descriptors >255
5450 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5451 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5452 //
5453 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5454 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5455 // because it saves a system call and removes a small window where the flag
5456 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5457 // and we fall back to using FD_CLOEXEC (see below).
5458 #ifdef O_CLOEXEC
5459 oflag |= O_CLOEXEC;
5460 #endif
5461
5462 int fd = ::open64(path, oflag, mode);
5463 if (fd == -1) return -1;
5464
5465 //If the open succeeded, the file might still be a directory
5466 {
5467 struct stat64 buf64;
5468 int ret = ::fstat64(fd, &buf64);
5469 int st_mode = buf64.st_mode;
5470
5471 if (ret != -1) {
5472 if ((st_mode & S_IFMT) == S_IFDIR) {
5473 errno = EISDIR;
5474 ::close(fd);
5475 return -1;
5476 }
5477 } else {
5478 ::close(fd);
5479 return -1;
5480 }
5481 }
5482
5483 #ifdef FD_CLOEXEC
5484 // Validate that the use of the O_CLOEXEC flag on open above worked.
5485 // With recent kernels, we will perform this check exactly once.
5486 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5487 if (!O_CLOEXEC_is_known_to_work) {
5488 int flags = ::fcntl(fd, F_GETFD);
5489 if (flags != -1) {
5490 if ((flags & FD_CLOEXEC) != 0)
5491 O_CLOEXEC_is_known_to_work = 1;
5492 else
5493 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5494 }
5495 }
5496 #endif
5497
5498 return fd;
5499 }
5500
5501
5502 // create binary file, rewriting existing file if required
5503 int os::create_binary_file(const char* path, bool rewrite_existing) {
5504 int oflags = O_WRONLY | O_CREAT;
5505 if (!rewrite_existing) {
5506 oflags |= O_EXCL;
5507 }
5508 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5509 }
5510
5511 // return current position of file pointer
5512 jlong os::current_file_offset(int fd) {
5513 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5514 }
5515
5516 // move file pointer to the specified offset
5517 jlong os::seek_to_file_offset(int fd, jlong offset) {
5518 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5519 }
5520
5521 // This code originates from JDK's sysAvailable
5522 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5523
5524 int os::available(int fd, jlong *bytes) {
5525 jlong cur, end;
5526 int mode;
5527 struct stat64 buf64;
5528
5529 if (::fstat64(fd, &buf64) >= 0) {
5530 mode = buf64.st_mode;
5531 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5532 int n;
5533 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5534 *bytes = n;
5535 return 1;
5536 }
5537 }
5538 }
5539 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5540 return 0;
5541 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5542 return 0;
5543 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5544 return 0;
5545 }
5546 *bytes = end - cur;
5547 return 1;
5548 }
5549
5550 // Map a block of memory.
5551 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5552 char *addr, size_t bytes, bool read_only,
5553 bool allow_exec) {
5554 int prot;
5555 int flags = MAP_PRIVATE;
5556
5557 if (read_only) {
5558 prot = PROT_READ;
5559 } else {
5560 prot = PROT_READ | PROT_WRITE;
5561 }
5562
5563 if (allow_exec) {
5564 prot |= PROT_EXEC;
5565 }
5566
5567 if (addr != NULL) {
5568 flags |= MAP_FIXED;
5569 }
5570
5571 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5572 fd, file_offset);
5573 if (mapped_address == MAP_FAILED) {
5574 return NULL;
5575 }
5576 return mapped_address;
5577 }
5578
5579
5580 // Remap a block of memory.
5581 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5582 char *addr, size_t bytes, bool read_only,
5583 bool allow_exec) {
5584 // same as map_memory() on this OS
5585 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5586 allow_exec);
5587 }
5588
5589
5590 // Unmap a block of memory.
5591 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5592 return munmap(addr, bytes) == 0;
5593 }
5594
5595 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5596
5597 static jlong fast_cpu_time(Thread *thread) {
5598 clockid_t clockid;
5599 int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(),
5600 &clockid);
5601 if (rc == 0) {
5602 return os::Linux::fast_thread_cpu_time(clockid);
5603 } else {
5604 // It's possible to encounter a terminated native thread that failed
5605 // to detach itself from the VM - which should result in ESRCH.
5606 assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed");
5607 return -1;
5608 }
5609 }
5610
5611 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5612 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5613 // of a thread.
5614 //
5615 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5616 // the fast estimate available on the platform.
5617
5618 jlong os::current_thread_cpu_time() {
5619 if (os::Linux::supports_fast_thread_cpu_time()) {
5620 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5621 } else {
5622 // return user + sys since the cost is the same
5623 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5624 }
5625 }
5626
5627 jlong os::thread_cpu_time(Thread* thread) {
5628 // consistent with what current_thread_cpu_time() returns
5629 if (os::Linux::supports_fast_thread_cpu_time()) {
5630 return fast_cpu_time(thread);
5631 } else {
5632 return slow_thread_cpu_time(thread, true /* user + sys */);
5633 }
5634 }
5635
5636 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5637 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5638 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5639 } else {
5640 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5641 }
5642 }
5643
5644 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5645 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5646 return fast_cpu_time(thread);
5647 } else {
5648 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5649 }
5650 }
5651
5652 // -1 on error.
5653 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5654 pid_t tid = thread->osthread()->thread_id();
5655 char *s;
5656 char stat[2048];
5657 int statlen;
5658 char proc_name[64];
5659 int count;
5660 long sys_time, user_time;
5661 char cdummy;
5662 int idummy;
5663 long ldummy;
5664 FILE *fp;
5665
5666 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5667 fp = fopen(proc_name, "r");
5668 if (fp == NULL) return -1;
5669 statlen = fread(stat, 1, 2047, fp);
5670 stat[statlen] = '\0';
5671 fclose(fp);
5672
5673 // Skip pid and the command string. Note that we could be dealing with
5674 // weird command names, e.g. user could decide to rename java launcher
5675 // to "java 1.4.2 :)", then the stat file would look like
5676 // 1234 (java 1.4.2 :)) R ... ...
5677 // We don't really need to know the command string, just find the last
5678 // occurrence of ")" and then start parsing from there. See bug 4726580.
5679 s = strrchr(stat, ')');
5680 if (s == NULL) return -1;
5681
5682 // Skip blank chars
5683 do { s++; } while (s && isspace(*s));
5684
5685 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5686 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5687 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5688 &user_time, &sys_time);
5689 if (count != 13) return -1;
5690 if (user_sys_cpu_time) {
5691 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5692 } else {
5693 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5694 }
5695 }
5696
5697 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5698 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5699 info_ptr->may_skip_backward = false; // elapsed time not wall time
5700 info_ptr->may_skip_forward = false; // elapsed time not wall time
5701 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5702 }
5703
5704 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5705 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5706 info_ptr->may_skip_backward = false; // elapsed time not wall time
5707 info_ptr->may_skip_forward = false; // elapsed time not wall time
5708 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5709 }
5710
5711 bool os::is_thread_cpu_time_supported() {
5712 return true;
5713 }
5714
5715 // System loadavg support. Returns -1 if load average cannot be obtained.
5716 // Linux doesn't yet have a (official) notion of processor sets,
5717 // so just return the system wide load average.
5718 int os::loadavg(double loadavg[], int nelem) {
5719 return ::getloadavg(loadavg, nelem);
5720 }
5721
5722 void os::pause() {
5723 char filename[MAX_PATH];
5724 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5725 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5726 } else {
5727 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5728 }
5729
5730 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5731 if (fd != -1) {
5732 struct stat buf;
5733 ::close(fd);
5734 while (::stat(filename, &buf) == 0) {
5735 (void)::poll(NULL, 0, 100);
5736 }
5737 } else {
5738 jio_fprintf(stderr,
5739 "Could not open pause file '%s', continuing immediately.\n", filename);
5740 }
5741 }
5742
5743 extern char** environ;
5744
5745 // Run the specified command in a separate process. Return its exit value,
5746 // or -1 on failure (e.g. can't fork a new process).
5747 // Unlike system(), this function can be called from signal handler. It
5748 // doesn't block SIGINT et al.
5749 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) {
5750 const char * argv[4] = {"sh", "-c", cmd, NULL};
5751
5752 pid_t pid ;
5753
5754 if (use_vfork_if_available) {
5755 pid = vfork();
5756 } else {
5757 pid = fork();
5758 }
5759
5760 if (pid < 0) {
5761 // fork failed
5762 return -1;
5763
5764 } else if (pid == 0) {
5765 // child process
5766
5767 execve("/bin/sh", (char* const*)argv, environ);
5768
5769 // execve failed
5770 _exit(-1);
5771
5772 } else {
5773 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5774 // care about the actual exit code, for now.
5775
5776 int status;
5777
5778 // Wait for the child process to exit. This returns immediately if
5779 // the child has already exited. */
5780 while (waitpid(pid, &status, 0) < 0) {
5781 switch (errno) {
5782 case ECHILD: return 0;
5783 case EINTR: break;
5784 default: return -1;
5785 }
5786 }
5787
5788 if (WIFEXITED(status)) {
5789 // The child exited normally; get its exit code.
5790 return WEXITSTATUS(status);
5791 } else if (WIFSIGNALED(status)) {
5792 // The child exited because of a signal
5793 // The best value to return is 0x80 + signal number,
5794 // because that is what all Unix shells do, and because
5795 // it allows callers to distinguish between process exit and
5796 // process death by signal.
5797 return 0x80 + WTERMSIG(status);
5798 } else {
5799 // Unknown exit code; pass it through
5800 return status;
5801 }
5802 }
5803 }
5804
5805 // Get the default path to the core file
5806 // Returns the length of the string
5807 int os::get_core_path(char* buffer, size_t bufferSize) {
5808 /*
5809 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5810 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5811 */
5812 const int core_pattern_len = 129;
5813 char core_pattern[core_pattern_len] = {0};
5814
5815 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5816 if (core_pattern_file == -1) {
5817 return -1;
5818 }
5819
5820 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5821 ::close(core_pattern_file);
5822 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5823 return -1;
5824 }
5825 if (core_pattern[ret-1] == '\n') {
5826 core_pattern[ret-1] = '\0';
5827 } else {
5828 core_pattern[ret] = '\0';
5829 }
5830
5831 // Replace the %p in the core pattern with the process id. NOTE: we do this
5832 // only if the pattern doesn't start with "|", and we support only one %p in
5833 // the pattern.
5834 char *pid_pos = strstr(core_pattern, "%p");
5835 const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : ""; // skip over the "%p"
5836 int written;
5837
5838 if (core_pattern[0] == '/') {
5839 if (pid_pos != NULL) {
5840 *pid_pos = '\0';
5841 written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern,
5842 current_process_id(), tail);
5843 } else {
5844 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5845 }
5846 } else {
5847 char cwd[PATH_MAX];
5848
5849 const char* p = get_current_directory(cwd, PATH_MAX);
5850 if (p == NULL) {
5851 return -1;
5852 }
5853
5854 if (core_pattern[0] == '|') {
5855 written = jio_snprintf(buffer, bufferSize,
5856 "\"%s\" (or dumping to %s/core.%d)",
5857 &core_pattern[1], p, current_process_id());
5858 } else if (pid_pos != NULL) {
5859 *pid_pos = '\0';
5860 written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern,
5861 current_process_id(), tail);
5862 } else {
5863 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5864 }
5865 }
5866
5867 if (written < 0) {
5868 return -1;
5869 }
5870
5871 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5872 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5873
5874 if (core_uses_pid_file != -1) {
5875 char core_uses_pid = 0;
5876 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5877 ::close(core_uses_pid_file);
5878
5879 if (core_uses_pid == '1') {
5880 jio_snprintf(buffer + written, bufferSize - written,
5881 ".%d", current_process_id());
5882 }
5883 }
5884 }
5885
5886 return strlen(buffer);
5887 }
5888
5889 bool os::start_debugging(char *buf, int buflen) {
5890 int len = (int)strlen(buf);
5891 char *p = &buf[len];
5892
5893 jio_snprintf(p, buflen-len,
5894 "\n\n"
5895 "Do you want to debug the problem?\n\n"
5896 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5897 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5898 "Otherwise, press RETURN to abort...",
5899 os::current_process_id(), os::current_process_id(),
5900 os::current_thread_id(), os::current_thread_id());
5901
5902 bool yes = os::message_box("Unexpected Error", buf);
5903
5904 if (yes) {
5905 // yes, user asked VM to launch debugger
5906 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5907 os::current_process_id(), os::current_process_id());
5908
5909 os::fork_and_exec(buf);
5910 yes = false;
5911 }
5912 return yes;
5913 }
5914
5915
5916 // Java/Compiler thread:
5917 //
5918 // Low memory addresses
5919 // P0 +------------------------+
5920 // | |\ Java thread created by VM does not have glibc
5921 // | glibc guard page | - guard page, attached Java thread usually has
5922 // | |/ 1 glibc guard page.
5923 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5924 // | |\
5925 // | HotSpot Guard Pages | - red, yellow and reserved pages
5926 // | |/
5927 // +------------------------+ JavaThread::stack_reserved_zone_base()
5928 // | |\
5929 // | Normal Stack | -
5930 // | |/
5931 // P2 +------------------------+ Thread::stack_base()
5932 //
5933 // Non-Java thread:
5934 //
5935 // Low memory addresses
5936 // P0 +------------------------+
5937 // | |\
5938 // | glibc guard page | - usually 1 page
5939 // | |/
5940 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5941 // | |\
5942 // | Normal Stack | -
5943 // | |/
5944 // P2 +------------------------+ Thread::stack_base()
5945 //
5946 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
5947 // returned from pthread_attr_getstack().
5948 // ** Due to NPTL implementation error, linux takes the glibc guard page out
5949 // of the stack size given in pthread_attr. We work around this for
5950 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
5951 //
5952 #ifndef ZERO
5953 static void current_stack_region(address * bottom, size_t * size) {
5954 if (os::is_primordial_thread()) {
5955 // primordial thread needs special handling because pthread_getattr_np()
5956 // may return bogus value.
5957 *bottom = os::Linux::initial_thread_stack_bottom();
5958 *size = os::Linux::initial_thread_stack_size();
5959 } else {
5960 pthread_attr_t attr;
5961
5962 int rslt = pthread_getattr_np(pthread_self(), &attr);
5963
5964 // JVM needs to know exact stack location, abort if it fails
5965 if (rslt != 0) {
5966 if (rslt == ENOMEM) {
5967 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
5968 } else {
5969 fatal("pthread_getattr_np failed with error = %d", rslt);
5970 }
5971 }
5972
5973 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
5974 fatal("Cannot locate current stack attributes!");
5975 }
5976
5977 // Work around NPTL stack guard error.
5978 size_t guard_size = 0;
5979 rslt = pthread_attr_getguardsize(&attr, &guard_size);
5980 if (rslt != 0) {
5981 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
5982 }
5983 *bottom += guard_size;
5984 *size -= guard_size;
5985
5986 pthread_attr_destroy(&attr);
5987
5988 }
5989 assert(os::current_stack_pointer() >= *bottom &&
5990 os::current_stack_pointer() < *bottom + *size, "just checking");
5991 }
5992
5993 address os::current_stack_base() {
5994 address bottom;
5995 size_t size;
5996 current_stack_region(&bottom, &size);
5997 return (bottom + size);
5998 }
5999
6000 size_t os::current_stack_size() {
6001 // This stack size includes the usable stack and HotSpot guard pages
6002 // (for the threads that have Hotspot guard pages).
6003 address bottom;
6004 size_t size;
6005 current_stack_region(&bottom, &size);
6006 return size;
6007 }
6008 #endif
6009
6010 static inline struct timespec get_mtime(const char* filename) {
6011 struct stat st;
6012 int ret = os::stat(filename, &st);
6013 assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
6014 return st.st_mtim;
6015 }
6016
6017 int os::compare_file_modified_times(const char* file1, const char* file2) {
6018 struct timespec filetime1 = get_mtime(file1);
6019 struct timespec filetime2 = get_mtime(file2);
6020 int diff = filetime1.tv_sec - filetime2.tv_sec;
6021 if (diff == 0) {
6022 return filetime1.tv_nsec - filetime2.tv_nsec;
6023 }
6024 return diff;
6025 }
6026
6027 /////////////// Unit tests ///////////////
6028
6029 #ifndef PRODUCT
6030
6031 class TestReserveMemorySpecial : AllStatic {
6032 public:
6033 static void small_page_write(void* addr, size_t size) {
6034 size_t page_size = os::vm_page_size();
6035
6036 char* end = (char*)addr + size;
6037 for (char* p = (char*)addr; p < end; p += page_size) {
6038 *p = 1;
6039 }
6040 }
6041
6042 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6043 if (!UseHugeTLBFS) {
6044 return;
6045 }
6046
6047 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6048
6049 if (addr != NULL) {
6050 small_page_write(addr, size);
6051
6052 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6053 }
6054 }
6055
6056 static void test_reserve_memory_special_huge_tlbfs_only() {
6057 if (!UseHugeTLBFS) {
6058 return;
6059 }
6060
6061 size_t lp = os::large_page_size();
6062
6063 for (size_t size = lp; size <= lp * 10; size += lp) {
6064 test_reserve_memory_special_huge_tlbfs_only(size);
6065 }
6066 }
6067
6068 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6069 size_t lp = os::large_page_size();
6070 size_t ag = os::vm_allocation_granularity();
6071
6072 // sizes to test
6073 const size_t sizes[] = {
6074 lp, lp + ag, lp + lp / 2, lp * 2,
6075 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6076 lp * 10, lp * 10 + lp / 2
6077 };
6078 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6079
6080 // For each size/alignment combination, we test three scenarios:
6081 // 1) with req_addr == NULL
6082 // 2) with a non-null req_addr at which we expect to successfully allocate
6083 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6084 // expect the allocation to either fail or to ignore req_addr
6085
6086 // Pre-allocate two areas; they shall be as large as the largest allocation
6087 // and aligned to the largest alignment we will be testing.
6088 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6089 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6090 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6091 -1, 0);
6092 assert(mapping1 != MAP_FAILED, "should work");
6093
6094 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6095 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6096 -1, 0);
6097 assert(mapping2 != MAP_FAILED, "should work");
6098
6099 // Unmap the first mapping, but leave the second mapping intact: the first
6100 // mapping will serve as a value for a "good" req_addr (case 2). The second
6101 // mapping, still intact, as "bad" req_addr (case 3).
6102 ::munmap(mapping1, mapping_size);
6103
6104 // Case 1
6105 for (int i = 0; i < num_sizes; i++) {
6106 const size_t size = sizes[i];
6107 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6108 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6109 if (p != NULL) {
6110 assert(is_aligned(p, alignment), "must be");
6111 small_page_write(p, size);
6112 os::Linux::release_memory_special_huge_tlbfs(p, size);
6113 }
6114 }
6115 }
6116
6117 // Case 2
6118 for (int i = 0; i < num_sizes; i++) {
6119 const size_t size = sizes[i];
6120 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6121 char* const req_addr = align_up(mapping1, alignment);
6122 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6123 if (p != NULL) {
6124 assert(p == req_addr, "must be");
6125 small_page_write(p, size);
6126 os::Linux::release_memory_special_huge_tlbfs(p, size);
6127 }
6128 }
6129 }
6130
6131 // Case 3
6132 for (int i = 0; i < num_sizes; i++) {
6133 const size_t size = sizes[i];
6134 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6135 char* const req_addr = align_up(mapping2, alignment);
6136 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6137 // as the area around req_addr contains already existing mappings, the API should always
6138 // return NULL (as per contract, it cannot return another address)
6139 assert(p == NULL, "must be");
6140 }
6141 }
6142
6143 ::munmap(mapping2, mapping_size);
6144
6145 }
6146
6147 static void test_reserve_memory_special_huge_tlbfs() {
6148 if (!UseHugeTLBFS) {
6149 return;
6150 }
6151
6152 test_reserve_memory_special_huge_tlbfs_only();
6153 test_reserve_memory_special_huge_tlbfs_mixed();
6154 }
6155
6156 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6157 if (!UseSHM) {
6158 return;
6159 }
6160
6161 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6162
6163 if (addr != NULL) {
6164 assert(is_aligned(addr, alignment), "Check");
6165 assert(is_aligned(addr, os::large_page_size()), "Check");
6166
6167 small_page_write(addr, size);
6168
6169 os::Linux::release_memory_special_shm(addr, size);
6170 }
6171 }
6172
6173 static void test_reserve_memory_special_shm() {
6174 size_t lp = os::large_page_size();
6175 size_t ag = os::vm_allocation_granularity();
6176
6177 for (size_t size = ag; size < lp * 3; size += ag) {
6178 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6179 test_reserve_memory_special_shm(size, alignment);
6180 }
6181 }
6182 }
6183
6184 static void test() {
6185 test_reserve_memory_special_huge_tlbfs();
6186 test_reserve_memory_special_shm();
6187 }
6188 };
6189
6190 void TestReserveMemorySpecial_test() {
6191 TestReserveMemorySpecial::test();
6192 }
6193
6194 #endif