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 "logging/logStream.hpp"
37 #include "memory/allocation.inline.hpp"
38 #include "memory/filemap.hpp"
39 #include "oops/oop.inline.hpp"
40 #include "os_linux.inline.hpp"
41 #include "os_share_linux.hpp"
42 #include "osContainer_linux.hpp"
43 #include "prims/jniFastGetField.hpp"
44 #include "prims/jvm_misc.hpp"
45 #include "runtime/arguments.hpp"
46 #include "runtime/atomic.hpp"
47 #include "runtime/extendedPC.hpp"
48 #include "runtime/globals.hpp"
49 #include "runtime/interfaceSupport.inline.hpp"
50 #include "runtime/init.hpp"
51 #include "runtime/java.hpp"
52 #include "runtime/javaCalls.hpp"
53 #include "runtime/mutexLocker.hpp"
54 #include "runtime/objectMonitor.hpp"
55 #include "runtime/orderAccess.hpp"
56 #include "runtime/osThread.hpp"
57 #include "runtime/perfMemory.hpp"
58 #include "runtime/sharedRuntime.hpp"
59 #include "runtime/statSampler.hpp"
60 #include "runtime/stubRoutines.hpp"
61 #include "runtime/thread.inline.hpp"
62 #include "runtime/threadCritical.hpp"
63 #include "runtime/threadSMR.hpp"
64 #include "runtime/timer.hpp"
65 #include "semaphore_posix.hpp"
66 #include "services/attachListener.hpp"
67 #include "services/memTracker.hpp"
68 #include "services/runtimeService.hpp"
69 #include "utilities/align.hpp"
70 #include "utilities/decoder.hpp"
71 #include "utilities/defaultStream.hpp"
72 #include "utilities/events.hpp"
73 #include "utilities/elfFile.hpp"
74 #include "utilities/growableArray.hpp"
75 #include "utilities/macros.hpp"
76 #include "utilities/vmError.hpp"
77
78 // put OS-includes here
79 # include <sys/types.h>
80 # include <sys/mman.h>
81 # include <sys/stat.h>
82 # include <sys/select.h>
83 # include <pthread.h>
84 # include <signal.h>
85 # include <errno.h>
86 # include <dlfcn.h>
87 # include <stdio.h>
88 # include <unistd.h>
89 # include <sys/resource.h>
90 # include <pthread.h>
91 # include <sys/stat.h>
92 # include <sys/time.h>
93 # include <sys/times.h>
94 # include <sys/utsname.h>
95 # include <sys/socket.h>
96 # include <sys/wait.h>
97 # include <pwd.h>
98 # include <poll.h>
99 # include <fcntl.h>
100 # include <string.h>
101 # include <syscall.h>
102 # include <sys/sysinfo.h>
103 # include <gnu/libc-version.h>
104 # include <sys/ipc.h>
105 # include <sys/shm.h>
106 # include <link.h>
107 # include <stdint.h>
108 # include <inttypes.h>
109 # include <sys/ioctl.h>
110
111 #ifndef _GNU_SOURCE
112 #define _GNU_SOURCE
113 #include <sched.h>
114 #undef _GNU_SOURCE
115 #else
116 #include <sched.h>
117 #endif
118
119 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
120 // getrusage() is prepared to handle the associated failure.
121 #ifndef RUSAGE_THREAD
122 #define RUSAGE_THREAD (1) /* only the calling thread */
123 #endif
124
125 #define MAX_PATH (2 * K)
126
127 #define MAX_SECS 100000000
128
129 // for timer info max values which include all bits
130 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
131
132 enum CoredumpFilterBit {
133 FILE_BACKED_PVT_BIT = 1 << 2,
134 LARGEPAGES_BIT = 1 << 6,
135 DAX_SHARED_BIT = 1 << 8
136 };
137
138 ////////////////////////////////////////////////////////////////////////////////
139 // global variables
140 julong os::Linux::_physical_memory = 0;
141
142 address os::Linux::_initial_thread_stack_bottom = NULL;
143 uintptr_t os::Linux::_initial_thread_stack_size = 0;
144
145 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
146 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
147 Mutex* os::Linux::_createThread_lock = NULL;
148 pthread_t os::Linux::_main_thread;
149 int os::Linux::_page_size = -1;
150 bool os::Linux::_supports_fast_thread_cpu_time = false;
151 uint32_t os::Linux::_os_version = 0;
152 const char * os::Linux::_glibc_version = NULL;
153 const char * os::Linux::_libpthread_version = NULL;
154
155 static jlong initial_time_count=0;
156
157 static int clock_tics_per_sec = 100;
158
159 // If the VM might have been created on the primordial thread, we need to resolve the
160 // primordial thread stack bounds and check if the current thread might be the
161 // primordial thread in places. If we know that the primordial thread is never used,
162 // such as when the VM was created by one of the standard java launchers, we can
163 // avoid this
164 static bool suppress_primordial_thread_resolution = false;
165
166 // For diagnostics to print a message once. see run_periodic_checks
167 static sigset_t check_signal_done;
168 static bool check_signals = true;
169
170 // Signal number used to suspend/resume a thread
171
172 // do not use any signal number less than SIGSEGV, see 4355769
173 static int SR_signum = SIGUSR2;
174 sigset_t SR_sigset;
175
176 // utility functions
177
178 static int SR_initialize();
179
180 julong os::available_memory() {
181 return Linux::available_memory();
182 }
183
184 julong os::Linux::available_memory() {
185 // values in struct sysinfo are "unsigned long"
186 struct sysinfo si;
187 julong avail_mem;
188
189 if (OSContainer::is_containerized()) {
190 jlong mem_limit, mem_usage;
191 if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
192 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
193 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
194 }
195 if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
196 log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
197 }
198 if (mem_limit > 0 && mem_usage > 0 ) {
199 avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
200 log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
201 return avail_mem;
202 }
203 }
204
205 sysinfo(&si);
206 avail_mem = (julong)si.freeram * si.mem_unit;
207 log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
208 return avail_mem;
209 }
210
211 julong os::physical_memory() {
212 jlong phys_mem = 0;
213 if (OSContainer::is_containerized()) {
214 jlong mem_limit;
215 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
216 log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
217 return mem_limit;
218 }
219 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
220 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
221 }
222
223 phys_mem = Linux::physical_memory();
224 log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
225 return phys_mem;
226 }
227
228 // Return true if user is running as root.
229
230 bool os::have_special_privileges() {
231 static bool init = false;
232 static bool privileges = false;
233 if (!init) {
234 privileges = (getuid() != geteuid()) || (getgid() != getegid());
235 init = true;
236 }
237 return privileges;
238 }
239
240
241 #ifndef SYS_gettid
242 // i386: 224, ia64: 1105, amd64: 186, sparc 143
243 #ifdef __ia64__
244 #define SYS_gettid 1105
245 #else
246 #ifdef __i386__
247 #define SYS_gettid 224
248 #else
249 #ifdef __amd64__
250 #define SYS_gettid 186
251 #else
252 #ifdef __sparc__
253 #define SYS_gettid 143
254 #else
255 #error define gettid for the arch
256 #endif
257 #endif
258 #endif
259 #endif
260 #endif
261
262
263 // pid_t gettid()
264 //
265 // Returns the kernel thread id of the currently running thread. Kernel
266 // thread id is used to access /proc.
267 pid_t os::Linux::gettid() {
268 int rslt = syscall(SYS_gettid);
269 assert(rslt != -1, "must be."); // old linuxthreads implementation?
270 return (pid_t)rslt;
271 }
272
273 // Most versions of linux have a bug where the number of processors are
274 // determined by looking at the /proc file system. In a chroot environment,
275 // the system call returns 1.
276 static bool unsafe_chroot_detected = false;
277 static const char *unstable_chroot_error = "/proc file system not found.\n"
278 "Java may be unstable running multithreaded in a chroot "
279 "environment on Linux when /proc filesystem is not mounted.";
280
281 void os::Linux::initialize_system_info() {
282 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
283 if (processor_count() == 1) {
284 pid_t pid = os::Linux::gettid();
285 char fname[32];
286 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
287 FILE *fp = fopen(fname, "r");
288 if (fp == NULL) {
289 unsafe_chroot_detected = true;
290 } else {
291 fclose(fp);
292 }
293 }
294 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
295 assert(processor_count() > 0, "linux error");
296 }
297
298 void os::init_system_properties_values() {
299 // The next steps are taken in the product version:
300 //
301 // Obtain the JAVA_HOME value from the location of libjvm.so.
302 // This library should be located at:
303 // <JAVA_HOME>/lib/{client|server}/libjvm.so.
304 //
305 // If "/jre/lib/" appears at the right place in the path, then we
306 // assume libjvm.so is installed in a JDK and we use this path.
307 //
308 // Otherwise exit with message: "Could not create the Java virtual machine."
309 //
310 // The following extra steps are taken in the debugging version:
311 //
312 // If "/jre/lib/" does NOT appear at the right place in the path
313 // instead of exit check for $JAVA_HOME environment variable.
314 //
315 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
316 // then we append a fake suffix "hotspot/libjvm.so" to this path so
317 // it looks like libjvm.so is installed there
318 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
319 //
320 // Otherwise exit.
321 //
322 // Important note: if the location of libjvm.so changes this
323 // code needs to be changed accordingly.
324
325 // See ld(1):
326 // The linker uses the following search paths to locate required
327 // shared libraries:
328 // 1: ...
329 // ...
330 // 7: The default directories, normally /lib and /usr/lib.
331 #ifndef OVERRIDE_LIBPATH
332 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
333 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
334 #else
335 #define DEFAULT_LIBPATH "/lib:/usr/lib"
336 #endif
337 #else
338 #define DEFAULT_LIBPATH OVERRIDE_LIBPATH
339 #endif
340
341 // Base path of extensions installed on the system.
342 #define SYS_EXT_DIR "/usr/java/packages"
343 #define EXTENSIONS_DIR "/lib/ext"
344
345 // Buffer that fits several sprintfs.
346 // Note that the space for the colon and the trailing null are provided
347 // by the nulls included by the sizeof operator.
348 const size_t bufsize =
349 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
350 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
351 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
352
353 // sysclasspath, java_home, dll_dir
354 {
355 char *pslash;
356 os::jvm_path(buf, bufsize);
357
358 // Found the full path to libjvm.so.
359 // Now cut the path to <java_home>/jre if we can.
360 pslash = strrchr(buf, '/');
361 if (pslash != NULL) {
362 *pslash = '\0'; // Get rid of /libjvm.so.
363 }
364 pslash = strrchr(buf, '/');
365 if (pslash != NULL) {
366 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
367 }
368 Arguments::set_dll_dir(buf);
369
370 if (pslash != NULL) {
371 pslash = strrchr(buf, '/');
372 if (pslash != NULL) {
373 *pslash = '\0'; // Get rid of /lib.
374 }
375 }
376 Arguments::set_java_home(buf);
377 if (!set_boot_path('/', ':')) {
378 vm_exit_during_initialization("Failed setting boot class path.", NULL);
379 }
380 }
381
382 // Where to look for native libraries.
383 //
384 // Note: Due to a legacy implementation, most of the library path
385 // is set in the launcher. This was to accomodate linking restrictions
386 // on legacy Linux implementations (which are no longer supported).
387 // Eventually, all the library path setting will be done here.
388 //
389 // However, to prevent the proliferation of improperly built native
390 // libraries, the new path component /usr/java/packages is added here.
391 // Eventually, all the library path setting will be done here.
392 {
393 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
394 // should always exist (until the legacy problem cited above is
395 // addressed).
396 const char *v = ::getenv("LD_LIBRARY_PATH");
397 const char *v_colon = ":";
398 if (v == NULL) { v = ""; v_colon = ""; }
399 // That's +1 for the colon and +1 for the trailing '\0'.
400 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
401 strlen(v) + 1 +
402 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
403 mtInternal);
404 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
405 Arguments::set_library_path(ld_library_path);
406 FREE_C_HEAP_ARRAY(char, ld_library_path);
407 }
408
409 // Extensions directories.
410 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
411 Arguments::set_ext_dirs(buf);
412
413 FREE_C_HEAP_ARRAY(char, buf);
414
415 #undef DEFAULT_LIBPATH
416 #undef SYS_EXT_DIR
417 #undef EXTENSIONS_DIR
418 }
419
420 ////////////////////////////////////////////////////////////////////////////////
421 // breakpoint support
422
423 void os::breakpoint() {
424 BREAKPOINT;
425 }
426
427 extern "C" void breakpoint() {
428 // use debugger to set breakpoint here
429 }
430
431 ////////////////////////////////////////////////////////////////////////////////
432 // signal support
433
434 debug_only(static bool signal_sets_initialized = false);
435 static sigset_t unblocked_sigs, vm_sigs;
436
437 void os::Linux::signal_sets_init() {
438 // Should also have an assertion stating we are still single-threaded.
439 assert(!signal_sets_initialized, "Already initialized");
440 // Fill in signals that are necessarily unblocked for all threads in
441 // the VM. Currently, we unblock the following signals:
442 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
443 // by -Xrs (=ReduceSignalUsage));
444 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
445 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
446 // the dispositions or masks wrt these signals.
447 // Programs embedding the VM that want to use the above signals for their
448 // own purposes must, at this time, use the "-Xrs" option to prevent
449 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
450 // (See bug 4345157, and other related bugs).
451 // In reality, though, unblocking these signals is really a nop, since
452 // these signals are not blocked by default.
453 sigemptyset(&unblocked_sigs);
454 sigaddset(&unblocked_sigs, SIGILL);
455 sigaddset(&unblocked_sigs, SIGSEGV);
456 sigaddset(&unblocked_sigs, SIGBUS);
457 sigaddset(&unblocked_sigs, SIGFPE);
458 #if defined(PPC64)
459 sigaddset(&unblocked_sigs, SIGTRAP);
460 #endif
461 sigaddset(&unblocked_sigs, SR_signum);
462
463 if (!ReduceSignalUsage) {
464 if (!os::Posix::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
465 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
466 }
467 if (!os::Posix::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
468 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
469 }
470 if (!os::Posix::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
471 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
472 }
473 }
474 // Fill in signals that are blocked by all but the VM thread.
475 sigemptyset(&vm_sigs);
476 if (!ReduceSignalUsage) {
477 sigaddset(&vm_sigs, BREAK_SIGNAL);
478 }
479 debug_only(signal_sets_initialized = true);
480
481 }
482
483 // These are signals that are unblocked while a thread is running Java.
484 // (For some reason, they get blocked by default.)
485 sigset_t* os::Linux::unblocked_signals() {
486 assert(signal_sets_initialized, "Not initialized");
487 return &unblocked_sigs;
488 }
489
490 // These are the signals that are blocked while a (non-VM) thread is
491 // running Java. Only the VM thread handles these signals.
492 sigset_t* os::Linux::vm_signals() {
493 assert(signal_sets_initialized, "Not initialized");
494 return &vm_sigs;
495 }
496
497 void os::Linux::hotspot_sigmask(Thread* thread) {
498
499 //Save caller's signal mask before setting VM signal mask
500 sigset_t caller_sigmask;
501 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
502
503 OSThread* osthread = thread->osthread();
504 osthread->set_caller_sigmask(caller_sigmask);
505
506 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
507
508 if (!ReduceSignalUsage) {
509 if (thread->is_VM_thread()) {
510 // Only the VM thread handles BREAK_SIGNAL ...
511 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
512 } else {
513 // ... all other threads block BREAK_SIGNAL
514 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
515 }
516 }
517 }
518
519 //////////////////////////////////////////////////////////////////////////////
520 // detecting pthread library
521
522 void os::Linux::libpthread_init() {
523 // Save glibc and pthread version strings.
524 #if !defined(_CS_GNU_LIBC_VERSION) || \
525 !defined(_CS_GNU_LIBPTHREAD_VERSION)
526 #error "glibc too old (< 2.3.2)"
527 #endif
528
529 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
530 assert(n > 0, "cannot retrieve glibc version");
531 char *str = (char *)malloc(n, mtInternal);
532 confstr(_CS_GNU_LIBC_VERSION, str, n);
533 os::Linux::set_glibc_version(str);
534
535 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
536 assert(n > 0, "cannot retrieve pthread version");
537 str = (char *)malloc(n, mtInternal);
538 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
539 os::Linux::set_libpthread_version(str);
540 }
541
542 /////////////////////////////////////////////////////////////////////////////
543 // thread stack expansion
544
545 // os::Linux::manually_expand_stack() takes care of expanding the thread
546 // stack. Note that this is normally not needed: pthread stacks allocate
547 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
548 // committed. Therefore it is not necessary to expand the stack manually.
549 //
550 // Manually expanding the stack was historically needed on LinuxThreads
551 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
552 // it is kept to deal with very rare corner cases:
553 //
554 // For one, user may run the VM on an own implementation of threads
555 // whose stacks are - like the old LinuxThreads - implemented using
556 // mmap(MAP_GROWSDOWN).
557 //
558 // Also, this coding may be needed if the VM is running on the primordial
559 // thread. Normally we avoid running on the primordial thread; however,
560 // user may still invoke the VM on the primordial thread.
561 //
562 // The following historical comment describes the details about running
563 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
564
565
566 // Force Linux kernel to expand current thread stack. If "bottom" is close
567 // to the stack guard, caller should block all signals.
568 //
569 // MAP_GROWSDOWN:
570 // A special mmap() flag that is used to implement thread stacks. It tells
571 // kernel that the memory region should extend downwards when needed. This
572 // allows early versions of LinuxThreads to only mmap the first few pages
573 // when creating a new thread. Linux kernel will automatically expand thread
574 // stack as needed (on page faults).
575 //
576 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
577 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
578 // region, it's hard to tell if the fault is due to a legitimate stack
579 // access or because of reading/writing non-exist memory (e.g. buffer
580 // overrun). As a rule, if the fault happens below current stack pointer,
581 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
582 // application (see Linux kernel fault.c).
583 //
584 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
585 // stack overflow detection.
586 //
587 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
588 // not use MAP_GROWSDOWN.
589 //
590 // To get around the problem and allow stack banging on Linux, we need to
591 // manually expand thread stack after receiving the SIGSEGV.
592 //
593 // There are two ways to expand thread stack to address "bottom", we used
594 // both of them in JVM before 1.5:
595 // 1. adjust stack pointer first so that it is below "bottom", and then
596 // touch "bottom"
597 // 2. mmap() the page in question
598 //
599 // Now alternate signal stack is gone, it's harder to use 2. For instance,
600 // if current sp is already near the lower end of page 101, and we need to
601 // call mmap() to map page 100, it is possible that part of the mmap() frame
602 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
603 // That will destroy the mmap() frame and cause VM to crash.
604 //
605 // The following code works by adjusting sp first, then accessing the "bottom"
606 // page to force a page fault. Linux kernel will then automatically expand the
607 // stack mapping.
608 //
609 // _expand_stack_to() assumes its frame size is less than page size, which
610 // should always be true if the function is not inlined.
611
612 static void NOINLINE _expand_stack_to(address bottom) {
613 address sp;
614 size_t size;
615 volatile char *p;
616
617 // Adjust bottom to point to the largest address within the same page, it
618 // gives us a one-page buffer if alloca() allocates slightly more memory.
619 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
620 bottom += os::Linux::page_size() - 1;
621
622 // sp might be slightly above current stack pointer; if that's the case, we
623 // will alloca() a little more space than necessary, which is OK. Don't use
624 // os::current_stack_pointer(), as its result can be slightly below current
625 // stack pointer, causing us to not alloca enough to reach "bottom".
626 sp = (address)&sp;
627
628 if (sp > bottom) {
629 size = sp - bottom;
630 p = (volatile char *)alloca(size);
631 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
632 p[0] = '\0';
633 }
634 }
635
636 void os::Linux::expand_stack_to(address bottom) {
637 _expand_stack_to(bottom);
638 }
639
640 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
641 assert(t!=NULL, "just checking");
642 assert(t->osthread()->expanding_stack(), "expand should be set");
643 assert(t->stack_base() != NULL, "stack_base was not initialized");
644
645 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
646 sigset_t mask_all, old_sigset;
647 sigfillset(&mask_all);
648 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
649 _expand_stack_to(addr);
650 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
651 return true;
652 }
653 return false;
654 }
655
656 //////////////////////////////////////////////////////////////////////////////
657 // create new thread
658
659 // Thread start routine for all newly created threads
660 static void *thread_native_entry(Thread *thread) {
661
662 thread->record_stack_base_and_size();
663
664 // Try to randomize the cache line index of hot stack frames.
665 // This helps when threads of the same stack traces evict each other's
666 // cache lines. The threads can be either from the same JVM instance, or
667 // from different JVM instances. The benefit is especially true for
668 // processors with hyperthreading technology.
669 static int counter = 0;
670 int pid = os::current_process_id();
671 alloca(((pid ^ counter++) & 7) * 128);
672
673 thread->initialize_thread_current();
674
675 OSThread* osthread = thread->osthread();
676 Monitor* sync = osthread->startThread_lock();
677
678 osthread->set_thread_id(os::current_thread_id());
679
680 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
681 os::current_thread_id(), (uintx) pthread_self());
682
683 if (UseNUMA) {
684 int lgrp_id = os::numa_get_group_id();
685 if (lgrp_id != -1) {
686 thread->set_lgrp_id(lgrp_id);
687 }
688 }
689 // initialize signal mask for this thread
690 os::Linux::hotspot_sigmask(thread);
691
692 // initialize floating point control register
693 os::Linux::init_thread_fpu_state();
694
695 // handshaking with parent thread
696 {
697 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
698
699 // notify parent thread
700 osthread->set_state(INITIALIZED);
701 sync->notify_all();
702
703 // wait until os::start_thread()
704 while (osthread->get_state() == INITIALIZED) {
705 sync->wait(Mutex::_no_safepoint_check_flag);
706 }
707 }
708
709 // call one more level start routine
710 thread->call_run();
711
712 // Note: at this point the thread object may already have deleted itself.
713 // Prevent dereferencing it from here on out.
714 thread = NULL;
715
716 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
717 os::current_thread_id(), (uintx) pthread_self());
718
719 return 0;
720 }
721
722 bool os::create_thread(Thread* thread, ThreadType thr_type,
723 size_t req_stack_size) {
724 assert(thread->osthread() == NULL, "caller responsible");
725
726 // Allocate the OSThread object
727 OSThread* osthread = new OSThread(NULL, NULL);
728 if (osthread == NULL) {
729 return false;
730 }
731
732 // set the correct thread state
733 osthread->set_thread_type(thr_type);
734
735 // Initial state is ALLOCATED but not INITIALIZED
736 osthread->set_state(ALLOCATED);
737
738 thread->set_osthread(osthread);
739
740 // init thread attributes
741 pthread_attr_t attr;
742 pthread_attr_init(&attr);
743 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
744
745 // Calculate stack size if it's not specified by caller.
746 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
747 // In the Linux NPTL pthread implementation the guard size mechanism
748 // is not implemented properly. The posix standard requires adding
749 // the size of the guard pages to the stack size, instead Linux
750 // takes the space out of 'stacksize'. Thus we adapt the requested
751 // stack_size by the size of the guard pages to mimick proper
752 // behaviour. However, be careful not to end up with a size
753 // of zero due to overflow. Don't add the guard page in that case.
754 size_t guard_size = os::Linux::default_guard_size(thr_type);
755 if (stack_size <= SIZE_MAX - guard_size) {
756 stack_size += guard_size;
757 }
758 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
759
760 int status = pthread_attr_setstacksize(&attr, stack_size);
761 assert_status(status == 0, status, "pthread_attr_setstacksize");
762
763 // Configure glibc guard page.
764 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
765
766 ThreadState state;
767
768 {
769 pthread_t tid;
770 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
771
772 char buf[64];
773 if (ret == 0) {
774 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
775 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
776 } else {
777 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
778 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
779 }
780
781 pthread_attr_destroy(&attr);
782
783 if (ret != 0) {
784 // Need to clean up stuff we've allocated so far
785 thread->set_osthread(NULL);
786 delete osthread;
787 return false;
788 }
789
790 // Store pthread info into the OSThread
791 osthread->set_pthread_id(tid);
792
793 // Wait until child thread is either initialized or aborted
794 {
795 Monitor* sync_with_child = osthread->startThread_lock();
796 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
797 while ((state = osthread->get_state()) == ALLOCATED) {
798 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
799 }
800 }
801 }
802
803 // Aborted due to thread limit being reached
804 if (state == ZOMBIE) {
805 thread->set_osthread(NULL);
806 delete osthread;
807 return false;
808 }
809
810 // The thread is returned suspended (in state INITIALIZED),
811 // and is started higher up in the call chain
812 assert(state == INITIALIZED, "race condition");
813 return true;
814 }
815
816 /////////////////////////////////////////////////////////////////////////////
817 // attach existing thread
818
819 // bootstrap the main thread
820 bool os::create_main_thread(JavaThread* thread) {
821 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
822 return create_attached_thread(thread);
823 }
824
825 bool os::create_attached_thread(JavaThread* thread) {
826 #ifdef ASSERT
827 thread->verify_not_published();
828 #endif
829
830 // Allocate the OSThread object
831 OSThread* osthread = new OSThread(NULL, NULL);
832
833 if (osthread == NULL) {
834 return false;
835 }
836
837 // Store pthread info into the OSThread
838 osthread->set_thread_id(os::Linux::gettid());
839 osthread->set_pthread_id(::pthread_self());
840
841 // initialize floating point control register
842 os::Linux::init_thread_fpu_state();
843
844 // Initial thread state is RUNNABLE
845 osthread->set_state(RUNNABLE);
846
847 thread->set_osthread(osthread);
848
849 if (UseNUMA) {
850 int lgrp_id = os::numa_get_group_id();
851 if (lgrp_id != -1) {
852 thread->set_lgrp_id(lgrp_id);
853 }
854 }
855
856 if (os::is_primordial_thread()) {
857 // If current thread is primordial thread, its stack is mapped on demand,
858 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
859 // the entire stack region to avoid SEGV in stack banging.
860 // It is also useful to get around the heap-stack-gap problem on SuSE
861 // kernel (see 4821821 for details). We first expand stack to the top
862 // of yellow zone, then enable stack yellow zone (order is significant,
863 // enabling yellow zone first will crash JVM on SuSE Linux), so there
864 // is no gap between the last two virtual memory regions.
865
866 JavaThread *jt = (JavaThread *)thread;
867 address addr = jt->stack_reserved_zone_base();
868 assert(addr != NULL, "initialization problem?");
869 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
870
871 osthread->set_expanding_stack();
872 os::Linux::manually_expand_stack(jt, addr);
873 osthread->clear_expanding_stack();
874 }
875
876 // initialize signal mask for this thread
877 // and save the caller's signal mask
878 os::Linux::hotspot_sigmask(thread);
879
880 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
881 os::current_thread_id(), (uintx) pthread_self());
882
883 return true;
884 }
885
886 void os::pd_start_thread(Thread* thread) {
887 OSThread * osthread = thread->osthread();
888 assert(osthread->get_state() != INITIALIZED, "just checking");
889 Monitor* sync_with_child = osthread->startThread_lock();
890 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
891 sync_with_child->notify();
892 }
893
894 // Free Linux resources related to the OSThread
895 void os::free_thread(OSThread* osthread) {
896 assert(osthread != NULL, "osthread not set");
897
898 // We are told to free resources of the argument thread,
899 // but we can only really operate on the current thread.
900 assert(Thread::current()->osthread() == osthread,
901 "os::free_thread but not current thread");
902
903 #ifdef ASSERT
904 sigset_t current;
905 sigemptyset(¤t);
906 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
907 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
908 #endif
909
910 // Restore caller's signal mask
911 sigset_t sigmask = osthread->caller_sigmask();
912 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
913
914 delete osthread;
915 }
916
917 //////////////////////////////////////////////////////////////////////////////
918 // primordial thread
919
920 // Check if current thread is the primordial thread, similar to Solaris thr_main.
921 bool os::is_primordial_thread(void) {
922 if (suppress_primordial_thread_resolution) {
923 return false;
924 }
925 char dummy;
926 // If called before init complete, thread stack bottom will be null.
927 // Can be called if fatal error occurs before initialization.
928 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
929 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
930 os::Linux::initial_thread_stack_size() != 0,
931 "os::init did not locate primordial thread's stack region");
932 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
933 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
934 os::Linux::initial_thread_stack_size()) {
935 return true;
936 } else {
937 return false;
938 }
939 }
940
941 // Find the virtual memory area that contains addr
942 static bool find_vma(address addr, address* vma_low, address* vma_high) {
943 FILE *fp = fopen("/proc/self/maps", "r");
944 if (fp) {
945 address low, high;
946 while (!feof(fp)) {
947 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
948 if (low <= addr && addr < high) {
949 if (vma_low) *vma_low = low;
950 if (vma_high) *vma_high = high;
951 fclose(fp);
952 return true;
953 }
954 }
955 for (;;) {
956 int ch = fgetc(fp);
957 if (ch == EOF || ch == (int)'\n') break;
958 }
959 }
960 fclose(fp);
961 }
962 return false;
963 }
964
965 // Locate primordial thread stack. This special handling of primordial thread stack
966 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
967 // bogus value for the primordial process thread. While the launcher has created
968 // the VM in a new thread since JDK 6, we still have to allow for the use of the
969 // JNI invocation API from a primordial thread.
970 void os::Linux::capture_initial_stack(size_t max_size) {
971
972 // max_size is either 0 (which means accept OS default for thread stacks) or
973 // a user-specified value known to be at least the minimum needed. If we
974 // are actually on the primordial thread we can make it appear that we have a
975 // smaller max_size stack by inserting the guard pages at that location. But we
976 // cannot do anything to emulate a larger stack than what has been provided by
977 // the OS or threading library. In fact if we try to use a stack greater than
978 // what is set by rlimit then we will crash the hosting process.
979
980 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
981 // If this is "unlimited" then it will be a huge value.
982 struct rlimit rlim;
983 getrlimit(RLIMIT_STACK, &rlim);
984 size_t stack_size = rlim.rlim_cur;
985
986 // 6308388: a bug in ld.so will relocate its own .data section to the
987 // lower end of primordial stack; reduce ulimit -s value a little bit
988 // so we won't install guard page on ld.so's data section.
989 // But ensure we don't underflow the stack size - allow 1 page spare
990 if (stack_size >= (size_t)(3 * page_size())) {
991 stack_size -= 2 * page_size();
992 }
993
994 // Try to figure out where the stack base (top) is. This is harder.
995 //
996 // When an application is started, glibc saves the initial stack pointer in
997 // a global variable "__libc_stack_end", which is then used by system
998 // libraries. __libc_stack_end should be pretty close to stack top. The
999 // variable is available since the very early days. However, because it is
1000 // a private interface, it could disappear in the future.
1001 //
1002 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1003 // to __libc_stack_end, it is very close to stack top, but isn't the real
1004 // stack top. Note that /proc may not exist if VM is running as a chroot
1005 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1006 // /proc/<pid>/stat could change in the future (though unlikely).
1007 //
1008 // We try __libc_stack_end first. If that doesn't work, look for
1009 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1010 // as a hint, which should work well in most cases.
1011
1012 uintptr_t stack_start;
1013
1014 // try __libc_stack_end first
1015 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1016 if (p && *p) {
1017 stack_start = *p;
1018 } else {
1019 // see if we can get the start_stack field from /proc/self/stat
1020 FILE *fp;
1021 int pid;
1022 char state;
1023 int ppid;
1024 int pgrp;
1025 int session;
1026 int nr;
1027 int tpgrp;
1028 unsigned long flags;
1029 unsigned long minflt;
1030 unsigned long cminflt;
1031 unsigned long majflt;
1032 unsigned long cmajflt;
1033 unsigned long utime;
1034 unsigned long stime;
1035 long cutime;
1036 long cstime;
1037 long prio;
1038 long nice;
1039 long junk;
1040 long it_real;
1041 uintptr_t start;
1042 uintptr_t vsize;
1043 intptr_t rss;
1044 uintptr_t rsslim;
1045 uintptr_t scodes;
1046 uintptr_t ecode;
1047 int i;
1048
1049 // Figure what the primordial thread stack base is. Code is inspired
1050 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1051 // followed by command name surrounded by parentheses, state, etc.
1052 char stat[2048];
1053 int statlen;
1054
1055 fp = fopen("/proc/self/stat", "r");
1056 if (fp) {
1057 statlen = fread(stat, 1, 2047, fp);
1058 stat[statlen] = '\0';
1059 fclose(fp);
1060
1061 // Skip pid and the command string. Note that we could be dealing with
1062 // weird command names, e.g. user could decide to rename java launcher
1063 // to "java 1.4.2 :)", then the stat file would look like
1064 // 1234 (java 1.4.2 :)) R ... ...
1065 // We don't really need to know the command string, just find the last
1066 // occurrence of ")" and then start parsing from there. See bug 4726580.
1067 char * s = strrchr(stat, ')');
1068
1069 i = 0;
1070 if (s) {
1071 // Skip blank chars
1072 do { s++; } while (s && isspace(*s));
1073
1074 #define _UFM UINTX_FORMAT
1075 #define _DFM INTX_FORMAT
1076
1077 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1078 // 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
1079 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,
1080 &state, // 3 %c
1081 &ppid, // 4 %d
1082 &pgrp, // 5 %d
1083 &session, // 6 %d
1084 &nr, // 7 %d
1085 &tpgrp, // 8 %d
1086 &flags, // 9 %lu
1087 &minflt, // 10 %lu
1088 &cminflt, // 11 %lu
1089 &majflt, // 12 %lu
1090 &cmajflt, // 13 %lu
1091 &utime, // 14 %lu
1092 &stime, // 15 %lu
1093 &cutime, // 16 %ld
1094 &cstime, // 17 %ld
1095 &prio, // 18 %ld
1096 &nice, // 19 %ld
1097 &junk, // 20 %ld
1098 &it_real, // 21 %ld
1099 &start, // 22 UINTX_FORMAT
1100 &vsize, // 23 UINTX_FORMAT
1101 &rss, // 24 INTX_FORMAT
1102 &rsslim, // 25 UINTX_FORMAT
1103 &scodes, // 26 UINTX_FORMAT
1104 &ecode, // 27 UINTX_FORMAT
1105 &stack_start); // 28 UINTX_FORMAT
1106 }
1107
1108 #undef _UFM
1109 #undef _DFM
1110
1111 if (i != 28 - 2) {
1112 assert(false, "Bad conversion from /proc/self/stat");
1113 // product mode - assume we are the primordial thread, good luck in the
1114 // embedded case.
1115 warning("Can't detect primordial thread stack location - bad conversion");
1116 stack_start = (uintptr_t) &rlim;
1117 }
1118 } else {
1119 // For some reason we can't open /proc/self/stat (for example, running on
1120 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1121 // most cases, so don't abort:
1122 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1123 stack_start = (uintptr_t) &rlim;
1124 }
1125 }
1126
1127 // Now we have a pointer (stack_start) very close to the stack top, the
1128 // next thing to do is to figure out the exact location of stack top. We
1129 // can find out the virtual memory area that contains stack_start by
1130 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1131 // and its upper limit is the real stack top. (again, this would fail if
1132 // running inside chroot, because /proc may not exist.)
1133
1134 uintptr_t stack_top;
1135 address low, high;
1136 if (find_vma((address)stack_start, &low, &high)) {
1137 // success, "high" is the true stack top. (ignore "low", because initial
1138 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1139 stack_top = (uintptr_t)high;
1140 } else {
1141 // failed, likely because /proc/self/maps does not exist
1142 warning("Can't detect primordial thread stack location - find_vma failed");
1143 // best effort: stack_start is normally within a few pages below the real
1144 // stack top, use it as stack top, and reduce stack size so we won't put
1145 // guard page outside stack.
1146 stack_top = stack_start;
1147 stack_size -= 16 * page_size();
1148 }
1149
1150 // stack_top could be partially down the page so align it
1151 stack_top = align_up(stack_top, page_size());
1152
1153 // Allowed stack value is minimum of max_size and what we derived from rlimit
1154 if (max_size > 0) {
1155 _initial_thread_stack_size = MIN2(max_size, stack_size);
1156 } else {
1157 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1158 // clamp it at 8MB as we do on Solaris
1159 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1160 }
1161 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1162 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1163
1164 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1165
1166 if (log_is_enabled(Info, os, thread)) {
1167 // See if we seem to be on primordial process thread
1168 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1169 uintptr_t(&rlim) < stack_top;
1170
1171 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1172 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1173 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1174 stack_top, intptr_t(_initial_thread_stack_bottom));
1175 }
1176 }
1177
1178 ////////////////////////////////////////////////////////////////////////////////
1179 // time support
1180
1181 #ifndef SUPPORTS_CLOCK_MONOTONIC
1182 #error "Build platform doesn't support clock_gettime and related functionality"
1183 #endif
1184
1185 // Time since start-up in seconds to a fine granularity.
1186 // Used by VMSelfDestructTimer and the MemProfiler.
1187 double os::elapsedTime() {
1188
1189 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1190 }
1191
1192 jlong os::elapsed_counter() {
1193 return javaTimeNanos() - initial_time_count;
1194 }
1195
1196 jlong os::elapsed_frequency() {
1197 return NANOSECS_PER_SEC; // nanosecond resolution
1198 }
1199
1200 bool os::supports_vtime() { return true; }
1201 bool os::enable_vtime() { return false; }
1202 bool os::vtime_enabled() { return false; }
1203
1204 double os::elapsedVTime() {
1205 struct rusage usage;
1206 int retval = getrusage(RUSAGE_THREAD, &usage);
1207 if (retval == 0) {
1208 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);
1209 } else {
1210 // better than nothing, but not much
1211 return elapsedTime();
1212 }
1213 }
1214
1215 jlong os::javaTimeMillis() {
1216 timeval time;
1217 int status = gettimeofday(&time, NULL);
1218 assert(status != -1, "linux error");
1219 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1220 }
1221
1222 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1223 timeval time;
1224 int status = gettimeofday(&time, NULL);
1225 assert(status != -1, "linux error");
1226 seconds = jlong(time.tv_sec);
1227 nanos = jlong(time.tv_usec) * 1000;
1228 }
1229
1230 void os::Linux::fast_thread_clock_init() {
1231 if (!UseLinuxPosixThreadCPUClocks) {
1232 return;
1233 }
1234 clockid_t clockid;
1235 struct timespec tp;
1236 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1237 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1238
1239 // Switch to using fast clocks for thread cpu time if
1240 // the clock_getres() returns 0 error code.
1241 // Note, that some kernels may support the current thread
1242 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1243 // returned by the pthread_getcpuclockid().
1244 // If the fast Posix clocks are supported then the clock_getres()
1245 // must return at least tp.tv_sec == 0 which means a resolution
1246 // better than 1 sec. This is extra check for reliability.
1247
1248 if (pthread_getcpuclockid_func &&
1249 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1250 os::Posix::clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1251 _supports_fast_thread_cpu_time = true;
1252 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1253 }
1254 }
1255
1256 jlong os::javaTimeNanos() {
1257 if (os::supports_monotonic_clock()) {
1258 struct timespec tp;
1259 int status = os::Posix::clock_gettime(CLOCK_MONOTONIC, &tp);
1260 assert(status == 0, "gettime error");
1261 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1262 return result;
1263 } else {
1264 timeval time;
1265 int status = gettimeofday(&time, NULL);
1266 assert(status != -1, "linux error");
1267 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1268 return 1000 * usecs;
1269 }
1270 }
1271
1272 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1273 if (os::supports_monotonic_clock()) {
1274 info_ptr->max_value = ALL_64_BITS;
1275
1276 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1277 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1278 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1279 } else {
1280 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1281 info_ptr->max_value = ALL_64_BITS;
1282
1283 // gettimeofday is a real time clock so it skips
1284 info_ptr->may_skip_backward = true;
1285 info_ptr->may_skip_forward = true;
1286 }
1287
1288 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1289 }
1290
1291 // Return the real, user, and system times in seconds from an
1292 // arbitrary fixed point in the past.
1293 bool os::getTimesSecs(double* process_real_time,
1294 double* process_user_time,
1295 double* process_system_time) {
1296 struct tms ticks;
1297 clock_t real_ticks = times(&ticks);
1298
1299 if (real_ticks == (clock_t) (-1)) {
1300 return false;
1301 } else {
1302 double ticks_per_second = (double) clock_tics_per_sec;
1303 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1304 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1305 *process_real_time = ((double) real_ticks) / ticks_per_second;
1306
1307 return true;
1308 }
1309 }
1310
1311
1312 char * os::local_time_string(char *buf, size_t buflen) {
1313 struct tm t;
1314 time_t long_time;
1315 time(&long_time);
1316 localtime_r(&long_time, &t);
1317 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1318 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1319 t.tm_hour, t.tm_min, t.tm_sec);
1320 return buf;
1321 }
1322
1323 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1324 return localtime_r(clock, res);
1325 }
1326
1327 ////////////////////////////////////////////////////////////////////////////////
1328 // runtime exit support
1329
1330 // Note: os::shutdown() might be called very early during initialization, or
1331 // called from signal handler. Before adding something to os::shutdown(), make
1332 // sure it is async-safe and can handle partially initialized VM.
1333 void os::shutdown() {
1334
1335 // allow PerfMemory to attempt cleanup of any persistent resources
1336 perfMemory_exit();
1337
1338 // needs to remove object in file system
1339 AttachListener::abort();
1340
1341 // flush buffered output, finish log files
1342 ostream_abort();
1343
1344 // Check for abort hook
1345 abort_hook_t abort_hook = Arguments::abort_hook();
1346 if (abort_hook != NULL) {
1347 abort_hook();
1348 }
1349
1350 }
1351
1352 // Note: os::abort() might be called very early during initialization, or
1353 // called from signal handler. Before adding something to os::abort(), make
1354 // sure it is async-safe and can handle partially initialized VM.
1355 void os::abort(bool dump_core, void* siginfo, const void* context) {
1356 os::shutdown();
1357 if (dump_core) {
1358 if (UseSharedSpaces && DumpPrivateMappingsInCore) {
1359 ClassLoader::close_jrt_image();
1360 }
1361 #ifndef PRODUCT
1362 fdStream out(defaultStream::output_fd());
1363 out.print_raw("Current thread is ");
1364 char buf[16];
1365 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1366 out.print_raw_cr(buf);
1367 out.print_raw_cr("Dumping core ...");
1368 #endif
1369 ::abort(); // dump core
1370 }
1371
1372 ::exit(1);
1373 }
1374
1375 // Die immediately, no exit hook, no abort hook, no cleanup.
1376 void os::die() {
1377 ::abort();
1378 }
1379
1380 // thread_id is kernel thread id (similar to Solaris LWP id)
1381 intx os::current_thread_id() { return os::Linux::gettid(); }
1382 int os::current_process_id() {
1383 return ::getpid();
1384 }
1385
1386 // DLL functions
1387
1388 const char* os::dll_file_extension() { return ".so"; }
1389
1390 // This must be hard coded because it's the system's temporary
1391 // directory not the java application's temp directory, ala java.io.tmpdir.
1392 const char* os::get_temp_directory() { return "/tmp"; }
1393
1394 static bool file_exists(const char* filename) {
1395 struct stat statbuf;
1396 if (filename == NULL || strlen(filename) == 0) {
1397 return false;
1398 }
1399 return os::stat(filename, &statbuf) == 0;
1400 }
1401
1402 // check if addr is inside libjvm.so
1403 bool os::address_is_in_vm(address addr) {
1404 static address libjvm_base_addr;
1405 Dl_info dlinfo;
1406
1407 if (libjvm_base_addr == NULL) {
1408 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1409 libjvm_base_addr = (address)dlinfo.dli_fbase;
1410 }
1411 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1412 }
1413
1414 if (dladdr((void *)addr, &dlinfo) != 0) {
1415 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1416 }
1417
1418 return false;
1419 }
1420
1421 bool os::dll_address_to_function_name(address addr, char *buf,
1422 int buflen, int *offset,
1423 bool demangle) {
1424 // buf is not optional, but offset is optional
1425 assert(buf != NULL, "sanity check");
1426
1427 Dl_info dlinfo;
1428
1429 if (dladdr((void*)addr, &dlinfo) != 0) {
1430 // see if we have a matching symbol
1431 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1432 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1433 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1434 }
1435 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1436 return true;
1437 }
1438 // no matching symbol so try for just file info
1439 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1440 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1441 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1442 return true;
1443 }
1444 }
1445 }
1446
1447 buf[0] = '\0';
1448 if (offset != NULL) *offset = -1;
1449 return false;
1450 }
1451
1452 struct _address_to_library_name {
1453 address addr; // input : memory address
1454 size_t buflen; // size of fname
1455 char* fname; // output: library name
1456 address base; // library base addr
1457 };
1458
1459 static int address_to_library_name_callback(struct dl_phdr_info *info,
1460 size_t size, void *data) {
1461 int i;
1462 bool found = false;
1463 address libbase = NULL;
1464 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1465
1466 // iterate through all loadable segments
1467 for (i = 0; i < info->dlpi_phnum; i++) {
1468 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1469 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1470 // base address of a library is the lowest address of its loaded
1471 // segments.
1472 if (libbase == NULL || libbase > segbase) {
1473 libbase = segbase;
1474 }
1475 // see if 'addr' is within current segment
1476 if (segbase <= d->addr &&
1477 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1478 found = true;
1479 }
1480 }
1481 }
1482
1483 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1484 // so dll_address_to_library_name() can fall through to use dladdr() which
1485 // can figure out executable name from argv[0].
1486 if (found && info->dlpi_name && info->dlpi_name[0]) {
1487 d->base = libbase;
1488 if (d->fname) {
1489 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1490 }
1491 return 1;
1492 }
1493 return 0;
1494 }
1495
1496 bool os::dll_address_to_library_name(address addr, char* buf,
1497 int buflen, int* offset) {
1498 // buf is not optional, but offset is optional
1499 assert(buf != NULL, "sanity check");
1500
1501 Dl_info dlinfo;
1502 struct _address_to_library_name data;
1503
1504 // There is a bug in old glibc dladdr() implementation that it could resolve
1505 // to wrong library name if the .so file has a base address != NULL. Here
1506 // we iterate through the program headers of all loaded libraries to find
1507 // out which library 'addr' really belongs to. This workaround can be
1508 // removed once the minimum requirement for glibc is moved to 2.3.x.
1509 data.addr = addr;
1510 data.fname = buf;
1511 data.buflen = buflen;
1512 data.base = NULL;
1513 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1514
1515 if (rslt) {
1516 // buf already contains library name
1517 if (offset) *offset = addr - data.base;
1518 return true;
1519 }
1520 if (dladdr((void*)addr, &dlinfo) != 0) {
1521 if (dlinfo.dli_fname != NULL) {
1522 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1523 }
1524 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1525 *offset = addr - (address)dlinfo.dli_fbase;
1526 }
1527 return true;
1528 }
1529
1530 buf[0] = '\0';
1531 if (offset) *offset = -1;
1532 return false;
1533 }
1534
1535 // Loads .dll/.so and
1536 // in case of error it checks if .dll/.so was built for the
1537 // same architecture as Hotspot is running on
1538
1539
1540 // Remember the stack's state. The Linux dynamic linker will change
1541 // the stack to 'executable' at most once, so we must safepoint only once.
1542 bool os::Linux::_stack_is_executable = false;
1543
1544 // VM operation that loads a library. This is necessary if stack protection
1545 // of the Java stacks can be lost during loading the library. If we
1546 // do not stop the Java threads, they can stack overflow before the stacks
1547 // are protected again.
1548 class VM_LinuxDllLoad: public VM_Operation {
1549 private:
1550 const char *_filename;
1551 char *_ebuf;
1552 int _ebuflen;
1553 void *_lib;
1554 public:
1555 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1556 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1557 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1558 void doit() {
1559 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1560 os::Linux::_stack_is_executable = true;
1561 }
1562 void* loaded_library() { return _lib; }
1563 };
1564
1565 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1566 void * result = NULL;
1567 bool load_attempted = false;
1568
1569 // Check whether the library to load might change execution rights
1570 // of the stack. If they are changed, the protection of the stack
1571 // guard pages will be lost. We need a safepoint to fix this.
1572 //
1573 // See Linux man page execstack(8) for more info.
1574 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1575 if (!ElfFile::specifies_noexecstack(filename)) {
1576 if (!is_init_completed()) {
1577 os::Linux::_stack_is_executable = true;
1578 // This is OK - No Java threads have been created yet, and hence no
1579 // stack guard pages to fix.
1580 //
1581 // Dynamic loader will make all stacks executable after
1582 // this function returns, and will not do that again.
1583 assert(Threads::number_of_threads() == 0, "no Java threads should exist yet.");
1584 } else {
1585 warning("You have loaded library %s which might have disabled stack guard. "
1586 "The VM will try to fix the stack guard now.\n"
1587 "It's highly recommended that you fix the library with "
1588 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1589 filename);
1590
1591 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1592 JavaThread *jt = JavaThread::current();
1593 if (jt->thread_state() != _thread_in_native) {
1594 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1595 // that requires ExecStack. Cannot enter safe point. Let's give up.
1596 warning("Unable to fix stack guard. Giving up.");
1597 } else {
1598 if (!LoadExecStackDllInVMThread) {
1599 // This is for the case where the DLL has an static
1600 // constructor function that executes JNI code. We cannot
1601 // load such DLLs in the VMThread.
1602 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1603 }
1604
1605 ThreadInVMfromNative tiv(jt);
1606 debug_only(VMNativeEntryWrapper vew;)
1607
1608 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1609 VMThread::execute(&op);
1610 if (LoadExecStackDllInVMThread) {
1611 result = op.loaded_library();
1612 }
1613 load_attempted = true;
1614 }
1615 }
1616 }
1617 }
1618
1619 if (!load_attempted) {
1620 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1621 }
1622
1623 if (result != NULL) {
1624 // Successful loading
1625 return result;
1626 }
1627
1628 Elf32_Ehdr elf_head;
1629 int diag_msg_max_length=ebuflen-strlen(ebuf);
1630 char* diag_msg_buf=ebuf+strlen(ebuf);
1631
1632 if (diag_msg_max_length==0) {
1633 // No more space in ebuf for additional diagnostics message
1634 return NULL;
1635 }
1636
1637
1638 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1639
1640 if (file_descriptor < 0) {
1641 // Can't open library, report dlerror() message
1642 return NULL;
1643 }
1644
1645 bool failed_to_read_elf_head=
1646 (sizeof(elf_head)!=
1647 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1648
1649 ::close(file_descriptor);
1650 if (failed_to_read_elf_head) {
1651 // file i/o error - report dlerror() msg
1652 return NULL;
1653 }
1654
1655 typedef struct {
1656 Elf32_Half code; // Actual value as defined in elf.h
1657 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1658 unsigned char elf_class; // 32 or 64 bit
1659 unsigned char endianess; // MSB or LSB
1660 char* name; // String representation
1661 } arch_t;
1662
1663 #ifndef EM_486
1664 #define EM_486 6 /* Intel 80486 */
1665 #endif
1666 #ifndef EM_AARCH64
1667 #define EM_AARCH64 183 /* ARM AARCH64 */
1668 #endif
1669
1670 static const arch_t arch_array[]={
1671 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1672 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1673 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1674 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1675 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1676 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1677 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1678 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1679 #if defined(VM_LITTLE_ENDIAN)
1680 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1681 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1682 #else
1683 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1684 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1685 #endif
1686 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1687 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1688 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1689 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1690 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1691 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1692 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1693 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1694 };
1695
1696 #if (defined IA32)
1697 static Elf32_Half running_arch_code=EM_386;
1698 #elif (defined AMD64) || (defined X32)
1699 static Elf32_Half running_arch_code=EM_X86_64;
1700 #elif (defined IA64)
1701 static Elf32_Half running_arch_code=EM_IA_64;
1702 #elif (defined __sparc) && (defined _LP64)
1703 static Elf32_Half running_arch_code=EM_SPARCV9;
1704 #elif (defined __sparc) && (!defined _LP64)
1705 static Elf32_Half running_arch_code=EM_SPARC;
1706 #elif (defined __powerpc64__)
1707 static Elf32_Half running_arch_code=EM_PPC64;
1708 #elif (defined __powerpc__)
1709 static Elf32_Half running_arch_code=EM_PPC;
1710 #elif (defined AARCH64)
1711 static Elf32_Half running_arch_code=EM_AARCH64;
1712 #elif (defined ARM)
1713 static Elf32_Half running_arch_code=EM_ARM;
1714 #elif (defined S390)
1715 static Elf32_Half running_arch_code=EM_S390;
1716 #elif (defined ALPHA)
1717 static Elf32_Half running_arch_code=EM_ALPHA;
1718 #elif (defined MIPSEL)
1719 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1720 #elif (defined PARISC)
1721 static Elf32_Half running_arch_code=EM_PARISC;
1722 #elif (defined MIPS)
1723 static Elf32_Half running_arch_code=EM_MIPS;
1724 #elif (defined M68K)
1725 static Elf32_Half running_arch_code=EM_68K;
1726 #elif (defined SH)
1727 static Elf32_Half running_arch_code=EM_SH;
1728 #else
1729 #error Method os::dll_load requires that one of following is defined:\
1730 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1731 #endif
1732
1733 // Identify compatability class for VM's architecture and library's architecture
1734 // Obtain string descriptions for architectures
1735
1736 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1737 int running_arch_index=-1;
1738
1739 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1740 if (running_arch_code == arch_array[i].code) {
1741 running_arch_index = i;
1742 }
1743 if (lib_arch.code == arch_array[i].code) {
1744 lib_arch.compat_class = arch_array[i].compat_class;
1745 lib_arch.name = arch_array[i].name;
1746 }
1747 }
1748
1749 assert(running_arch_index != -1,
1750 "Didn't find running architecture code (running_arch_code) in arch_array");
1751 if (running_arch_index == -1) {
1752 // Even though running architecture detection failed
1753 // we may still continue with reporting dlerror() message
1754 return NULL;
1755 }
1756
1757 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1758 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1759 return NULL;
1760 }
1761
1762 #ifndef S390
1763 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1764 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1765 return NULL;
1766 }
1767 #endif // !S390
1768
1769 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1770 if (lib_arch.name!=NULL) {
1771 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1772 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1773 lib_arch.name, arch_array[running_arch_index].name);
1774 } else {
1775 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1776 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1777 lib_arch.code,
1778 arch_array[running_arch_index].name);
1779 }
1780 }
1781
1782 return NULL;
1783 }
1784
1785 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1786 int ebuflen) {
1787 void * result = ::dlopen(filename, RTLD_LAZY);
1788 if (result == NULL) {
1789 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1790 ebuf[ebuflen-1] = '\0';
1791 }
1792 return result;
1793 }
1794
1795 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1796 int ebuflen) {
1797 void * result = NULL;
1798 if (LoadExecStackDllInVMThread) {
1799 result = dlopen_helper(filename, ebuf, ebuflen);
1800 }
1801
1802 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1803 // library that requires an executable stack, or which does not have this
1804 // stack attribute set, dlopen changes the stack attribute to executable. The
1805 // read protection of the guard pages gets lost.
1806 //
1807 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1808 // may have been queued at the same time.
1809
1810 if (!_stack_is_executable) {
1811 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
1812 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1813 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1814 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1815 warning("Attempt to reguard stack yellow zone failed.");
1816 }
1817 }
1818 }
1819 }
1820
1821 return result;
1822 }
1823
1824 void* os::dll_lookup(void* handle, const char* name) {
1825 void* res = dlsym(handle, name);
1826 return res;
1827 }
1828
1829 void* os::get_default_process_handle() {
1830 return (void*)::dlopen(NULL, RTLD_LAZY);
1831 }
1832
1833 static bool _print_ascii_file(const char* filename, outputStream* st, const char* hdr = NULL) {
1834 int fd = ::open(filename, O_RDONLY);
1835 if (fd == -1) {
1836 return false;
1837 }
1838
1839 if (hdr != NULL) {
1840 st->print_cr("%s", hdr);
1841 }
1842
1843 char buf[33];
1844 int bytes;
1845 buf[32] = '\0';
1846 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1847 st->print_raw(buf, bytes);
1848 }
1849
1850 ::close(fd);
1851
1852 return true;
1853 }
1854
1855 void os::print_dll_info(outputStream *st) {
1856 st->print_cr("Dynamic libraries:");
1857
1858 char fname[32];
1859 pid_t pid = os::Linux::gettid();
1860
1861 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1862
1863 if (!_print_ascii_file(fname, st)) {
1864 st->print("Can not get library information for pid = %d\n", pid);
1865 }
1866 }
1867
1868 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1869 FILE *procmapsFile = NULL;
1870
1871 // Open the procfs maps file for the current process
1872 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1873 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1874 char line[PATH_MAX + 100];
1875
1876 // Read line by line from 'file'
1877 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1878 u8 base, top, offset, inode;
1879 char permissions[5];
1880 char device[6];
1881 char name[PATH_MAX + 1];
1882
1883 // Parse fields from line
1884 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1885 &base, &top, permissions, &offset, device, &inode, name);
1886
1887 // Filter by device id '00:00' so that we only get file system mapped files.
1888 if (strcmp(device, "00:00") != 0) {
1889
1890 // Call callback with the fields of interest
1891 if(callback(name, (address)base, (address)top, param)) {
1892 // Oops abort, callback aborted
1893 fclose(procmapsFile);
1894 return 1;
1895 }
1896 }
1897 }
1898 fclose(procmapsFile);
1899 }
1900 return 0;
1901 }
1902
1903 void os::print_os_info_brief(outputStream* st) {
1904 os::Linux::print_distro_info(st);
1905
1906 os::Posix::print_uname_info(st);
1907
1908 os::Linux::print_libversion_info(st);
1909
1910 }
1911
1912 void os::print_os_info(outputStream* st) {
1913 st->print("OS:");
1914
1915 os::Linux::print_distro_info(st);
1916
1917 os::Posix::print_uname_info(st);
1918
1919 // Print warning if unsafe chroot environment detected
1920 if (unsafe_chroot_detected) {
1921 st->print("WARNING!! ");
1922 st->print_cr("%s", unstable_chroot_error);
1923 }
1924
1925 os::Linux::print_libversion_info(st);
1926
1927 os::Posix::print_rlimit_info(st);
1928
1929 os::Posix::print_load_average(st);
1930
1931 os::Linux::print_full_memory_info(st);
1932
1933 os::Linux::print_proc_sys_info(st);
1934
1935 os::Linux::print_ld_preload_file(st);
1936
1937 os::Linux::print_container_info(st);
1938 }
1939
1940 // Try to identify popular distros.
1941 // Most Linux distributions have a /etc/XXX-release file, which contains
1942 // the OS version string. Newer Linux distributions have a /etc/lsb-release
1943 // file that also contains the OS version string. Some have more than one
1944 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
1945 // /etc/redhat-release.), so the order is important.
1946 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
1947 // their own specific XXX-release file as well as a redhat-release file.
1948 // Because of this the XXX-release file needs to be searched for before the
1949 // redhat-release file.
1950 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
1951 // search for redhat-release / SuSE-release needs to be before lsb-release.
1952 // Since the lsb-release file is the new standard it needs to be searched
1953 // before the older style release files.
1954 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
1955 // next to last resort. The os-release file is a new standard that contains
1956 // distribution information and the system-release file seems to be an old
1957 // standard that has been replaced by the lsb-release and os-release files.
1958 // Searching for the debian_version file is the last resort. It contains
1959 // an informative string like "6.0.6" or "wheezy/sid". Because of this
1960 // "Debian " is printed before the contents of the debian_version file.
1961
1962 const char* distro_files[] = {
1963 "/etc/oracle-release",
1964 "/etc/mandriva-release",
1965 "/etc/mandrake-release",
1966 "/etc/sun-release",
1967 "/etc/redhat-release",
1968 "/etc/SuSE-release",
1969 "/etc/lsb-release",
1970 "/etc/turbolinux-release",
1971 "/etc/gentoo-release",
1972 "/etc/ltib-release",
1973 "/etc/angstrom-version",
1974 "/etc/system-release",
1975 "/etc/os-release",
1976 NULL };
1977
1978 void os::Linux::print_distro_info(outputStream* st) {
1979 for (int i = 0;; i++) {
1980 const char* file = distro_files[i];
1981 if (file == NULL) {
1982 break; // done
1983 }
1984 // If file prints, we found it.
1985 if (_print_ascii_file(file, st)) {
1986 return;
1987 }
1988 }
1989
1990 if (file_exists("/etc/debian_version")) {
1991 st->print("Debian ");
1992 _print_ascii_file("/etc/debian_version", st);
1993 } else {
1994 st->print("Linux");
1995 }
1996 st->cr();
1997 }
1998
1999 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2000 char buf[256];
2001 while (fgets(buf, sizeof(buf), fp)) {
2002 // Edit out extra stuff in expected format
2003 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2004 char* ptr = strstr(buf, "\""); // the name is in quotes
2005 if (ptr != NULL) {
2006 ptr++; // go beyond first quote
2007 char* nl = strchr(ptr, '\"');
2008 if (nl != NULL) *nl = '\0';
2009 strncpy(distro, ptr, length);
2010 } else {
2011 ptr = strstr(buf, "=");
2012 ptr++; // go beyond equals then
2013 char* nl = strchr(ptr, '\n');
2014 if (nl != NULL) *nl = '\0';
2015 strncpy(distro, ptr, length);
2016 }
2017 return;
2018 } else if (get_first_line) {
2019 char* nl = strchr(buf, '\n');
2020 if (nl != NULL) *nl = '\0';
2021 strncpy(distro, buf, length);
2022 return;
2023 }
2024 }
2025 // print last line and close
2026 char* nl = strchr(buf, '\n');
2027 if (nl != NULL) *nl = '\0';
2028 strncpy(distro, buf, length);
2029 }
2030
2031 static void parse_os_info(char* distro, size_t length, const char* file) {
2032 FILE* fp = fopen(file, "r");
2033 if (fp != NULL) {
2034 // if suse format, print out first line
2035 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2036 parse_os_info_helper(fp, distro, length, get_first_line);
2037 fclose(fp);
2038 }
2039 }
2040
2041 void os::get_summary_os_info(char* buf, size_t buflen) {
2042 for (int i = 0;; i++) {
2043 const char* file = distro_files[i];
2044 if (file == NULL) {
2045 break; // ran out of distro_files
2046 }
2047 if (file_exists(file)) {
2048 parse_os_info(buf, buflen, file);
2049 return;
2050 }
2051 }
2052 // special case for debian
2053 if (file_exists("/etc/debian_version")) {
2054 strncpy(buf, "Debian ", buflen);
2055 if (buflen > 7) {
2056 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2057 }
2058 } else {
2059 strncpy(buf, "Linux", buflen);
2060 }
2061 }
2062
2063 void os::Linux::print_libversion_info(outputStream* st) {
2064 // libc, pthread
2065 st->print("libc:");
2066 st->print("%s ", os::Linux::glibc_version());
2067 st->print("%s ", os::Linux::libpthread_version());
2068 st->cr();
2069 }
2070
2071 void os::Linux::print_proc_sys_info(outputStream* st) {
2072 st->cr();
2073 st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):");
2074 _print_ascii_file("/proc/sys/kernel/threads-max", st);
2075 st->cr();
2076 st->cr();
2077
2078 st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):");
2079 _print_ascii_file("/proc/sys/vm/max_map_count", st);
2080 st->cr();
2081 st->cr();
2082
2083 st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):");
2084 _print_ascii_file("/proc/sys/kernel/pid_max", st);
2085 st->cr();
2086 st->cr();
2087 }
2088
2089 void os::Linux::print_full_memory_info(outputStream* st) {
2090 st->print("\n/proc/meminfo:\n");
2091 _print_ascii_file("/proc/meminfo", st);
2092 st->cr();
2093 }
2094
2095 void os::Linux::print_ld_preload_file(outputStream* st) {
2096 _print_ascii_file("/etc/ld.so.preload", st, "\n/etc/ld.so.preload:");
2097 st->cr();
2098 }
2099
2100 void os::Linux::print_container_info(outputStream* st) {
2101 if (!OSContainer::is_containerized()) {
2102 return;
2103 }
2104
2105 st->print("container (cgroup) information:\n");
2106
2107 const char *p_ct = OSContainer::container_type();
2108 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed");
2109
2110 char *p = OSContainer::cpu_cpuset_cpus();
2111 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed");
2112 free(p);
2113
2114 p = OSContainer::cpu_cpuset_memory_nodes();
2115 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed");
2116 free(p);
2117
2118 int i = OSContainer::active_processor_count();
2119 if (i > 0) {
2120 st->print("active_processor_count: %d\n", i);
2121 } else {
2122 st->print("active_processor_count: failed\n");
2123 }
2124
2125 i = OSContainer::cpu_quota();
2126 st->print("cpu_quota: %d\n", i);
2127
2128 i = OSContainer::cpu_period();
2129 st->print("cpu_period: %d\n", i);
2130
2131 i = OSContainer::cpu_shares();
2132 st->print("cpu_shares: %d\n", i);
2133
2134 jlong j = OSContainer::memory_limit_in_bytes();
2135 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2136
2137 j = OSContainer::memory_and_swap_limit_in_bytes();
2138 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2139
2140 j = OSContainer::memory_soft_limit_in_bytes();
2141 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2142
2143 j = OSContainer::OSContainer::memory_usage_in_bytes();
2144 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2145
2146 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2147 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2148 st->cr();
2149 }
2150
2151 void os::print_memory_info(outputStream* st) {
2152
2153 st->print("Memory:");
2154 st->print(" %dk page", os::vm_page_size()>>10);
2155
2156 // values in struct sysinfo are "unsigned long"
2157 struct sysinfo si;
2158 sysinfo(&si);
2159
2160 st->print(", physical " UINT64_FORMAT "k",
2161 os::physical_memory() >> 10);
2162 st->print("(" UINT64_FORMAT "k free)",
2163 os::available_memory() >> 10);
2164 st->print(", swap " UINT64_FORMAT "k",
2165 ((jlong)si.totalswap * si.mem_unit) >> 10);
2166 st->print("(" UINT64_FORMAT "k free)",
2167 ((jlong)si.freeswap * si.mem_unit) >> 10);
2168 st->cr();
2169 }
2170
2171 // Print the first "model name" line and the first "flags" line
2172 // that we find and nothing more. We assume "model name" comes
2173 // before "flags" so if we find a second "model name", then the
2174 // "flags" field is considered missing.
2175 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2176 #if defined(IA32) || defined(AMD64)
2177 // Other platforms have less repetitive cpuinfo files
2178 FILE *fp = fopen("/proc/cpuinfo", "r");
2179 if (fp) {
2180 while (!feof(fp)) {
2181 if (fgets(buf, buflen, fp)) {
2182 // Assume model name comes before flags
2183 bool model_name_printed = false;
2184 if (strstr(buf, "model name") != NULL) {
2185 if (!model_name_printed) {
2186 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2187 st->print_raw(buf);
2188 model_name_printed = true;
2189 } else {
2190 // model name printed but not flags? Odd, just return
2191 fclose(fp);
2192 return true;
2193 }
2194 }
2195 // print the flags line too
2196 if (strstr(buf, "flags") != NULL) {
2197 st->print_raw(buf);
2198 fclose(fp);
2199 return true;
2200 }
2201 }
2202 }
2203 fclose(fp);
2204 }
2205 #endif // x86 platforms
2206 return false;
2207 }
2208
2209 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2210 // Only print the model name if the platform provides this as a summary
2211 if (!print_model_name_and_flags(st, buf, buflen)) {
2212 st->print("\n/proc/cpuinfo:\n");
2213 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2214 st->print_cr(" <Not Available>");
2215 }
2216 }
2217 }
2218
2219 #if defined(AMD64) || defined(IA32) || defined(X32)
2220 const char* search_string = "model name";
2221 #elif defined(M68K)
2222 const char* search_string = "CPU";
2223 #elif defined(PPC64)
2224 const char* search_string = "cpu";
2225 #elif defined(S390)
2226 const char* search_string = "processor";
2227 #elif defined(SPARC)
2228 const char* search_string = "cpu";
2229 #else
2230 const char* search_string = "Processor";
2231 #endif
2232
2233 // Parses the cpuinfo file for string representing the model name.
2234 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2235 FILE* fp = fopen("/proc/cpuinfo", "r");
2236 if (fp != NULL) {
2237 while (!feof(fp)) {
2238 char buf[256];
2239 if (fgets(buf, sizeof(buf), fp)) {
2240 char* start = strstr(buf, search_string);
2241 if (start != NULL) {
2242 char *ptr = start + strlen(search_string);
2243 char *end = buf + strlen(buf);
2244 while (ptr != end) {
2245 // skip whitespace and colon for the rest of the name.
2246 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2247 break;
2248 }
2249 ptr++;
2250 }
2251 if (ptr != end) {
2252 // reasonable string, get rid of newline and keep the rest
2253 char* nl = strchr(buf, '\n');
2254 if (nl != NULL) *nl = '\0';
2255 strncpy(cpuinfo, ptr, length);
2256 fclose(fp);
2257 return;
2258 }
2259 }
2260 }
2261 }
2262 fclose(fp);
2263 }
2264 // cpuinfo not found or parsing failed, just print generic string. The entire
2265 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2266 #if defined(AARCH64)
2267 strncpy(cpuinfo, "AArch64", length);
2268 #elif defined(AMD64)
2269 strncpy(cpuinfo, "x86_64", length);
2270 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2271 strncpy(cpuinfo, "ARM", length);
2272 #elif defined(IA32)
2273 strncpy(cpuinfo, "x86_32", length);
2274 #elif defined(IA64)
2275 strncpy(cpuinfo, "IA64", length);
2276 #elif defined(PPC)
2277 strncpy(cpuinfo, "PPC64", length);
2278 #elif defined(S390)
2279 strncpy(cpuinfo, "S390", length);
2280 #elif defined(SPARC)
2281 strncpy(cpuinfo, "sparcv9", length);
2282 #elif defined(ZERO_LIBARCH)
2283 strncpy(cpuinfo, ZERO_LIBARCH, length);
2284 #else
2285 strncpy(cpuinfo, "unknown", length);
2286 #endif
2287 }
2288
2289 static void print_signal_handler(outputStream* st, int sig,
2290 char* buf, size_t buflen);
2291
2292 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2293 st->print_cr("Signal Handlers:");
2294 print_signal_handler(st, SIGSEGV, buf, buflen);
2295 print_signal_handler(st, SIGBUS , buf, buflen);
2296 print_signal_handler(st, SIGFPE , buf, buflen);
2297 print_signal_handler(st, SIGPIPE, buf, buflen);
2298 print_signal_handler(st, SIGXFSZ, buf, buflen);
2299 print_signal_handler(st, SIGILL , buf, buflen);
2300 print_signal_handler(st, SR_signum, buf, buflen);
2301 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2302 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2303 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2304 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2305 #if defined(PPC64)
2306 print_signal_handler(st, SIGTRAP, buf, buflen);
2307 #endif
2308 }
2309
2310 static char saved_jvm_path[MAXPATHLEN] = {0};
2311
2312 // Find the full path to the current module, libjvm.so
2313 void os::jvm_path(char *buf, jint buflen) {
2314 // Error checking.
2315 if (buflen < MAXPATHLEN) {
2316 assert(false, "must use a large-enough buffer");
2317 buf[0] = '\0';
2318 return;
2319 }
2320 // Lazy resolve the path to current module.
2321 if (saved_jvm_path[0] != 0) {
2322 strcpy(buf, saved_jvm_path);
2323 return;
2324 }
2325
2326 char dli_fname[MAXPATHLEN];
2327 bool ret = dll_address_to_library_name(
2328 CAST_FROM_FN_PTR(address, os::jvm_path),
2329 dli_fname, sizeof(dli_fname), NULL);
2330 assert(ret, "cannot locate libjvm");
2331 char *rp = NULL;
2332 if (ret && dli_fname[0] != '\0') {
2333 rp = os::Posix::realpath(dli_fname, buf, buflen);
2334 }
2335 if (rp == NULL) {
2336 return;
2337 }
2338
2339 if (Arguments::sun_java_launcher_is_altjvm()) {
2340 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2341 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2342 // If "/jre/lib/" appears at the right place in the string, then
2343 // assume we are installed in a JDK and we're done. Otherwise, check
2344 // for a JAVA_HOME environment variable and fix up the path so it
2345 // looks like libjvm.so is installed there (append a fake suffix
2346 // hotspot/libjvm.so).
2347 const char *p = buf + strlen(buf) - 1;
2348 for (int count = 0; p > buf && count < 5; ++count) {
2349 for (--p; p > buf && *p != '/'; --p)
2350 /* empty */ ;
2351 }
2352
2353 if (strncmp(p, "/jre/lib/", 9) != 0) {
2354 // Look for JAVA_HOME in the environment.
2355 char* java_home_var = ::getenv("JAVA_HOME");
2356 if (java_home_var != NULL && java_home_var[0] != 0) {
2357 char* jrelib_p;
2358 int len;
2359
2360 // Check the current module name "libjvm.so".
2361 p = strrchr(buf, '/');
2362 if (p == NULL) {
2363 return;
2364 }
2365 assert(strstr(p, "/libjvm") == p, "invalid library name");
2366
2367 rp = os::Posix::realpath(java_home_var, buf, buflen);
2368 if (rp == NULL) {
2369 return;
2370 }
2371
2372 // determine if this is a legacy image or modules image
2373 // modules image doesn't have "jre" subdirectory
2374 len = strlen(buf);
2375 assert(len < buflen, "Ran out of buffer room");
2376 jrelib_p = buf + len;
2377 snprintf(jrelib_p, buflen-len, "/jre/lib");
2378 if (0 != access(buf, F_OK)) {
2379 snprintf(jrelib_p, buflen-len, "/lib");
2380 }
2381
2382 if (0 == access(buf, F_OK)) {
2383 // Use current module name "libjvm.so"
2384 len = strlen(buf);
2385 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2386 } else {
2387 // Go back to path of .so
2388 rp = os::Posix::realpath(dli_fname, buf, buflen);
2389 if (rp == NULL) {
2390 return;
2391 }
2392 }
2393 }
2394 }
2395 }
2396
2397 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2398 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2399 }
2400
2401 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2402 // no prefix required, not even "_"
2403 }
2404
2405 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2406 // no suffix required
2407 }
2408
2409 ////////////////////////////////////////////////////////////////////////////////
2410 // sun.misc.Signal support
2411
2412 static volatile jint sigint_count = 0;
2413
2414 static void UserHandler(int sig, void *siginfo, void *context) {
2415 // 4511530 - sem_post is serialized and handled by the manager thread. When
2416 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2417 // don't want to flood the manager thread with sem_post requests.
2418 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2419 return;
2420 }
2421
2422 // Ctrl-C is pressed during error reporting, likely because the error
2423 // handler fails to abort. Let VM die immediately.
2424 if (sig == SIGINT && VMError::is_error_reported()) {
2425 os::die();
2426 }
2427
2428 os::signal_notify(sig);
2429 }
2430
2431 void* os::user_handler() {
2432 return CAST_FROM_FN_PTR(void*, UserHandler);
2433 }
2434
2435 static struct timespec create_semaphore_timespec(unsigned int sec, int nsec) {
2436 struct timespec ts;
2437 // Semaphore's are always associated with CLOCK_REALTIME
2438 os::Posix::clock_gettime(CLOCK_REALTIME, &ts);
2439 // see os_posix.cpp for discussion on overflow checking
2440 if (sec >= MAX_SECS) {
2441 ts.tv_sec += MAX_SECS;
2442 ts.tv_nsec = 0;
2443 } else {
2444 ts.tv_sec += sec;
2445 ts.tv_nsec += nsec;
2446 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2447 ts.tv_nsec -= NANOSECS_PER_SEC;
2448 ++ts.tv_sec; // note: this must be <= max_secs
2449 }
2450 }
2451
2452 return ts;
2453 }
2454
2455 extern "C" {
2456 typedef void (*sa_handler_t)(int);
2457 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2458 }
2459
2460 void* os::signal(int signal_number, void* handler) {
2461 struct sigaction sigAct, oldSigAct;
2462
2463 sigfillset(&(sigAct.sa_mask));
2464 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2465 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2466
2467 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2468 // -1 means registration failed
2469 return (void *)-1;
2470 }
2471
2472 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2473 }
2474
2475 void os::signal_raise(int signal_number) {
2476 ::raise(signal_number);
2477 }
2478
2479 // The following code is moved from os.cpp for making this
2480 // code platform specific, which it is by its very nature.
2481
2482 // Will be modified when max signal is changed to be dynamic
2483 int os::sigexitnum_pd() {
2484 return NSIG;
2485 }
2486
2487 // a counter for each possible signal value
2488 static volatile jint pending_signals[NSIG+1] = { 0 };
2489
2490 // Linux(POSIX) specific hand shaking semaphore.
2491 static Semaphore* sig_sem = NULL;
2492 static PosixSemaphore sr_semaphore;
2493
2494 static void jdk_misc_signal_init() {
2495 // Initialize signal structures
2496 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2497
2498 // Initialize signal semaphore
2499 sig_sem = new Semaphore();
2500 }
2501
2502 void os::signal_notify(int sig) {
2503 if (sig_sem != NULL) {
2504 Atomic::inc(&pending_signals[sig]);
2505 sig_sem->signal();
2506 } else {
2507 // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2508 // initialization isn't called.
2509 assert(ReduceSignalUsage, "signal semaphore should be created");
2510 }
2511 }
2512
2513 static int check_pending_signals() {
2514 Atomic::store(0, &sigint_count);
2515 for (;;) {
2516 for (int i = 0; i < NSIG + 1; i++) {
2517 jint n = pending_signals[i];
2518 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2519 return i;
2520 }
2521 }
2522 JavaThread *thread = JavaThread::current();
2523 ThreadBlockInVM tbivm(thread);
2524
2525 bool threadIsSuspended;
2526 do {
2527 thread->set_suspend_equivalent();
2528 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2529 sig_sem->wait();
2530
2531 // were we externally suspended while we were waiting?
2532 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2533 if (threadIsSuspended) {
2534 // The semaphore has been incremented, but while we were waiting
2535 // another thread suspended us. We don't want to continue running
2536 // while suspended because that would surprise the thread that
2537 // suspended us.
2538 sig_sem->signal();
2539
2540 thread->java_suspend_self();
2541 }
2542 } while (threadIsSuspended);
2543 }
2544 }
2545
2546 int os::signal_wait() {
2547 return check_pending_signals();
2548 }
2549
2550 ////////////////////////////////////////////////////////////////////////////////
2551 // Virtual Memory
2552
2553 int os::vm_page_size() {
2554 // Seems redundant as all get out
2555 assert(os::Linux::page_size() != -1, "must call os::init");
2556 return os::Linux::page_size();
2557 }
2558
2559 // Solaris allocates memory by pages.
2560 int os::vm_allocation_granularity() {
2561 assert(os::Linux::page_size() != -1, "must call os::init");
2562 return os::Linux::page_size();
2563 }
2564
2565 // Rationale behind this function:
2566 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2567 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2568 // samples for JITted code. Here we create private executable mapping over the code cache
2569 // and then we can use standard (well, almost, as mapping can change) way to provide
2570 // info for the reporting script by storing timestamp and location of symbol
2571 void linux_wrap_code(char* base, size_t size) {
2572 static volatile jint cnt = 0;
2573
2574 if (!UseOprofile) {
2575 return;
2576 }
2577
2578 char buf[PATH_MAX+1];
2579 int num = Atomic::add(1, &cnt);
2580
2581 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2582 os::get_temp_directory(), os::current_process_id(), num);
2583 unlink(buf);
2584
2585 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2586
2587 if (fd != -1) {
2588 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2589 if (rv != (off_t)-1) {
2590 if (::write(fd, "", 1) == 1) {
2591 mmap(base, size,
2592 PROT_READ|PROT_WRITE|PROT_EXEC,
2593 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2594 }
2595 }
2596 ::close(fd);
2597 unlink(buf);
2598 }
2599 }
2600
2601 static bool recoverable_mmap_error(int err) {
2602 // See if the error is one we can let the caller handle. This
2603 // list of errno values comes from JBS-6843484. I can't find a
2604 // Linux man page that documents this specific set of errno
2605 // values so while this list currently matches Solaris, it may
2606 // change as we gain experience with this failure mode.
2607 switch (err) {
2608 case EBADF:
2609 case EINVAL:
2610 case ENOTSUP:
2611 // let the caller deal with these errors
2612 return true;
2613
2614 default:
2615 // Any remaining errors on this OS can cause our reserved mapping
2616 // to be lost. That can cause confusion where different data
2617 // structures think they have the same memory mapped. The worst
2618 // scenario is if both the VM and a library think they have the
2619 // same memory mapped.
2620 return false;
2621 }
2622 }
2623
2624 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2625 int err) {
2626 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2627 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2628 os::strerror(err), err);
2629 }
2630
2631 static void warn_fail_commit_memory(char* addr, size_t size,
2632 size_t alignment_hint, bool exec,
2633 int err) {
2634 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2635 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2636 alignment_hint, exec, os::strerror(err), err);
2637 }
2638
2639 // NOTE: Linux kernel does not really reserve the pages for us.
2640 // All it does is to check if there are enough free pages
2641 // left at the time of mmap(). This could be a potential
2642 // problem.
2643 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2644 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2645 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2646 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2647 if (res != (uintptr_t) MAP_FAILED) {
2648 if (UseNUMAInterleaving) {
2649 numa_make_global(addr, size);
2650 }
2651 return 0;
2652 }
2653
2654 int err = errno; // save errno from mmap() call above
2655
2656 if (!recoverable_mmap_error(err)) {
2657 warn_fail_commit_memory(addr, size, exec, err);
2658 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2659 }
2660
2661 return err;
2662 }
2663
2664 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2665 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2666 }
2667
2668 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2669 const char* mesg) {
2670 assert(mesg != NULL, "mesg must be specified");
2671 int err = os::Linux::commit_memory_impl(addr, size, exec);
2672 if (err != 0) {
2673 // the caller wants all commit errors to exit with the specified mesg:
2674 warn_fail_commit_memory(addr, size, exec, err);
2675 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2676 }
2677 }
2678
2679 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2680 #ifndef MAP_HUGETLB
2681 #define MAP_HUGETLB 0x40000
2682 #endif
2683
2684 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2685 #ifndef MADV_HUGEPAGE
2686 #define MADV_HUGEPAGE 14
2687 #endif
2688
2689 int os::Linux::commit_memory_impl(char* addr, size_t size,
2690 size_t alignment_hint, bool exec) {
2691 int err = os::Linux::commit_memory_impl(addr, size, exec);
2692 if (err == 0) {
2693 realign_memory(addr, size, alignment_hint);
2694 }
2695 return err;
2696 }
2697
2698 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2699 bool exec) {
2700 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2701 }
2702
2703 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2704 size_t alignment_hint, bool exec,
2705 const char* mesg) {
2706 assert(mesg != NULL, "mesg must be specified");
2707 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2708 if (err != 0) {
2709 // the caller wants all commit errors to exit with the specified mesg:
2710 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2711 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2712 }
2713 }
2714
2715 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2716 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2717 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2718 // be supported or the memory may already be backed by huge pages.
2719 ::madvise(addr, bytes, MADV_HUGEPAGE);
2720 }
2721 }
2722
2723 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2724 // This method works by doing an mmap over an existing mmaping and effectively discarding
2725 // the existing pages. However it won't work for SHM-based large pages that cannot be
2726 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2727 // small pages on top of the SHM segment. This method always works for small pages, so we
2728 // allow that in any case.
2729 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2730 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2731 }
2732 }
2733
2734 void os::numa_make_global(char *addr, size_t bytes) {
2735 Linux::numa_interleave_memory(addr, bytes);
2736 }
2737
2738 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2739 // bind policy to MPOL_PREFERRED for the current thread.
2740 #define USE_MPOL_PREFERRED 0
2741
2742 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2743 // To make NUMA and large pages more robust when both enabled, we need to ease
2744 // the requirements on where the memory should be allocated. MPOL_BIND is the
2745 // default policy and it will force memory to be allocated on the specified
2746 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2747 // the specified node, but will not force it. Using this policy will prevent
2748 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2749 // free large pages.
2750 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2751 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2752 }
2753
2754 bool os::numa_topology_changed() { return false; }
2755
2756 size_t os::numa_get_groups_num() {
2757 // Return just the number of nodes in which it's possible to allocate memory
2758 // (in numa terminology, configured nodes).
2759 return Linux::numa_num_configured_nodes();
2760 }
2761
2762 int os::numa_get_group_id() {
2763 int cpu_id = Linux::sched_getcpu();
2764 if (cpu_id != -1) {
2765 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2766 if (lgrp_id != -1) {
2767 return lgrp_id;
2768 }
2769 }
2770 return 0;
2771 }
2772
2773 int os::Linux::get_existing_num_nodes() {
2774 int node;
2775 int highest_node_number = Linux::numa_max_node();
2776 int num_nodes = 0;
2777
2778 // Get the total number of nodes in the system including nodes without memory.
2779 for (node = 0; node <= highest_node_number; node++) {
2780 if (is_node_in_existing_nodes(node)) {
2781 num_nodes++;
2782 }
2783 }
2784 return num_nodes;
2785 }
2786
2787 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2788 int highest_node_number = Linux::numa_max_node();
2789 size_t i = 0;
2790
2791 // Map all node ids in which it is possible to allocate memory. Also nodes are
2792 // not always consecutively available, i.e. available from 0 to the highest
2793 // node number. If the nodes have been bound explicitly using numactl membind,
2794 // then allocate memory from those nodes only.
2795 for (int node = 0; node <= highest_node_number; node++) {
2796 if (Linux::is_node_in_bound_nodes((unsigned int)node)) {
2797 ids[i++] = node;
2798 }
2799 }
2800 return i;
2801 }
2802
2803 bool os::get_page_info(char *start, page_info* info) {
2804 return false;
2805 }
2806
2807 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2808 page_info* page_found) {
2809 return end;
2810 }
2811
2812
2813 int os::Linux::sched_getcpu_syscall(void) {
2814 unsigned int cpu = 0;
2815 int retval = -1;
2816
2817 #if defined(IA32)
2818 #ifndef SYS_getcpu
2819 #define SYS_getcpu 318
2820 #endif
2821 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2822 #elif defined(AMD64)
2823 // Unfortunately we have to bring all these macros here from vsyscall.h
2824 // to be able to compile on old linuxes.
2825 #define __NR_vgetcpu 2
2826 #define VSYSCALL_START (-10UL << 20)
2827 #define VSYSCALL_SIZE 1024
2828 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2829 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2830 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2831 retval = vgetcpu(&cpu, NULL, NULL);
2832 #endif
2833
2834 return (retval == -1) ? retval : cpu;
2835 }
2836
2837 void os::Linux::sched_getcpu_init() {
2838 // sched_getcpu() should be in libc.
2839 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2840 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2841
2842 // If it's not, try a direct syscall.
2843 if (sched_getcpu() == -1) {
2844 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2845 (void*)&sched_getcpu_syscall));
2846 }
2847
2848 if (sched_getcpu() == -1) {
2849 vm_exit_during_initialization("getcpu(2) system call not supported by kernel");
2850 }
2851 }
2852
2853 // Something to do with the numa-aware allocator needs these symbols
2854 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2855 extern "C" JNIEXPORT void numa_error(char *where) { }
2856
2857 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2858 // load symbol from base version instead.
2859 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2860 void *f = dlvsym(handle, name, "libnuma_1.1");
2861 if (f == NULL) {
2862 f = dlsym(handle, name);
2863 }
2864 return f;
2865 }
2866
2867 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2868 // Return NULL if the symbol is not defined in this particular version.
2869 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2870 return dlvsym(handle, name, "libnuma_1.2");
2871 }
2872
2873 bool os::Linux::libnuma_init() {
2874 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
2875 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2876 if (handle != NULL) {
2877 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2878 libnuma_dlsym(handle, "numa_node_to_cpus")));
2879 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2880 libnuma_dlsym(handle, "numa_max_node")));
2881 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2882 libnuma_dlsym(handle, "numa_num_configured_nodes")));
2883 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2884 libnuma_dlsym(handle, "numa_available")));
2885 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2886 libnuma_dlsym(handle, "numa_tonode_memory")));
2887 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2888 libnuma_dlsym(handle, "numa_interleave_memory")));
2889 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2890 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2891 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2892 libnuma_dlsym(handle, "numa_set_bind_policy")));
2893 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2894 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2895 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2896 libnuma_dlsym(handle, "numa_distance")));
2897 set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t,
2898 libnuma_v2_dlsym(handle, "numa_get_membind")));
2899 set_numa_get_interleave_mask(CAST_TO_FN_PTR(numa_get_interleave_mask_func_t,
2900 libnuma_v2_dlsym(handle, "numa_get_interleave_mask")));
2901
2902 if (numa_available() != -1) {
2903 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2904 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2905 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2906 set_numa_interleave_bitmask(_numa_get_interleave_mask());
2907 set_numa_membind_bitmask(_numa_get_membind());
2908 // Create an index -> node mapping, since nodes are not always consecutive
2909 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2910 rebuild_nindex_to_node_map();
2911 // Create a cpu -> node mapping
2912 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2913 rebuild_cpu_to_node_map();
2914 return true;
2915 }
2916 }
2917 }
2918 return false;
2919 }
2920
2921 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2922 // Creating guard page is very expensive. Java thread has HotSpot
2923 // guard pages, only enable glibc guard page for non-Java threads.
2924 // (Remember: compiler thread is a Java thread, too!)
2925 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2926 }
2927
2928 void os::Linux::rebuild_nindex_to_node_map() {
2929 int highest_node_number = Linux::numa_max_node();
2930
2931 nindex_to_node()->clear();
2932 for (int node = 0; node <= highest_node_number; node++) {
2933 if (Linux::is_node_in_existing_nodes(node)) {
2934 nindex_to_node()->append(node);
2935 }
2936 }
2937 }
2938
2939 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2940 // The table is later used in get_node_by_cpu().
2941 void os::Linux::rebuild_cpu_to_node_map() {
2942 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2943 // in libnuma (possible values are starting from 16,
2944 // and continuing up with every other power of 2, but less
2945 // than the maximum number of CPUs supported by kernel), and
2946 // is a subject to change (in libnuma version 2 the requirements
2947 // are more reasonable) we'll just hardcode the number they use
2948 // in the library.
2949 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2950
2951 size_t cpu_num = processor_count();
2952 size_t cpu_map_size = NCPUS / BitsPerCLong;
2953 size_t cpu_map_valid_size =
2954 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2955
2956 cpu_to_node()->clear();
2957 cpu_to_node()->at_grow(cpu_num - 1);
2958
2959 size_t node_num = get_existing_num_nodes();
2960
2961 int distance = 0;
2962 int closest_distance = INT_MAX;
2963 int closest_node = 0;
2964 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2965 for (size_t i = 0; i < node_num; i++) {
2966 // Check if node is configured (not a memory-less node). If it is not, find
2967 // the closest configured node. Check also if node is bound, i.e. it's allowed
2968 // to allocate memory from the node. If it's not allowed, map cpus in that node
2969 // to the closest node from which memory allocation is allowed.
2970 if (!is_node_in_configured_nodes(nindex_to_node()->at(i)) ||
2971 !is_node_in_bound_nodes(nindex_to_node()->at(i))) {
2972 closest_distance = INT_MAX;
2973 // Check distance from all remaining nodes in the system. Ignore distance
2974 // from itself, from another non-configured node, and from another non-bound
2975 // node.
2976 for (size_t m = 0; m < node_num; m++) {
2977 if (m != i &&
2978 is_node_in_configured_nodes(nindex_to_node()->at(m)) &&
2979 is_node_in_bound_nodes(nindex_to_node()->at(m))) {
2980 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
2981 // If a closest node is found, update. There is always at least one
2982 // configured and bound node in the system so there is always at least
2983 // one node close.
2984 if (distance != 0 && distance < closest_distance) {
2985 closest_distance = distance;
2986 closest_node = nindex_to_node()->at(m);
2987 }
2988 }
2989 }
2990 } else {
2991 // Current node is already a configured node.
2992 closest_node = nindex_to_node()->at(i);
2993 }
2994
2995 // Get cpus from the original node and map them to the closest node. If node
2996 // is a configured node (not a memory-less node), then original node and
2997 // closest node are the same.
2998 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2999 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3000 if (cpu_map[j] != 0) {
3001 for (size_t k = 0; k < BitsPerCLong; k++) {
3002 if (cpu_map[j] & (1UL << k)) {
3003 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3004 }
3005 }
3006 }
3007 }
3008 }
3009 }
3010 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3011 }
3012
3013 int os::Linux::get_node_by_cpu(int cpu_id) {
3014 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3015 return cpu_to_node()->at(cpu_id);
3016 }
3017 return -1;
3018 }
3019
3020 GrowableArray<int>* os::Linux::_cpu_to_node;
3021 GrowableArray<int>* os::Linux::_nindex_to_node;
3022 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3023 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3024 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3025 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3026 os::Linux::numa_available_func_t os::Linux::_numa_available;
3027 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3028 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3029 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3030 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3031 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3032 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3033 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind;
3034 os::Linux::numa_get_interleave_mask_func_t os::Linux::_numa_get_interleave_mask;
3035 os::Linux::Numa_allocation_policy os::Linux::_current_numa_policy;
3036 unsigned long* os::Linux::_numa_all_nodes;
3037 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3038 struct bitmask* os::Linux::_numa_nodes_ptr;
3039 struct bitmask* os::Linux::_numa_interleave_bitmask;
3040 struct bitmask* os::Linux::_numa_membind_bitmask;
3041
3042 bool os::pd_uncommit_memory(char* addr, size_t size) {
3043 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3044 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3045 return res != (uintptr_t) MAP_FAILED;
3046 }
3047
3048 static address get_stack_commited_bottom(address bottom, size_t size) {
3049 address nbot = bottom;
3050 address ntop = bottom + size;
3051
3052 size_t page_sz = os::vm_page_size();
3053 unsigned pages = size / page_sz;
3054
3055 unsigned char vec[1];
3056 unsigned imin = 1, imax = pages + 1, imid;
3057 int mincore_return_value = 0;
3058
3059 assert(imin <= imax, "Unexpected page size");
3060
3061 while (imin < imax) {
3062 imid = (imax + imin) / 2;
3063 nbot = ntop - (imid * page_sz);
3064
3065 // Use a trick with mincore to check whether the page is mapped or not.
3066 // mincore sets vec to 1 if page resides in memory and to 0 if page
3067 // is swapped output but if page we are asking for is unmapped
3068 // it returns -1,ENOMEM
3069 mincore_return_value = mincore(nbot, page_sz, vec);
3070
3071 if (mincore_return_value == -1) {
3072 // Page is not mapped go up
3073 // to find first mapped page
3074 if (errno != EAGAIN) {
3075 assert(errno == ENOMEM, "Unexpected mincore errno");
3076 imax = imid;
3077 }
3078 } else {
3079 // Page is mapped go down
3080 // to find first not mapped page
3081 imin = imid + 1;
3082 }
3083 }
3084
3085 nbot = nbot + page_sz;
3086
3087 // Adjust stack bottom one page up if last checked page is not mapped
3088 if (mincore_return_value == -1) {
3089 nbot = nbot + page_sz;
3090 }
3091
3092 return nbot;
3093 }
3094
3095 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3096 int mincore_return_value;
3097 const size_t stripe = 1024; // query this many pages each time
3098 unsigned char vec[stripe + 1];
3099 // set a guard
3100 vec[stripe] = 'X';
3101
3102 const size_t page_sz = os::vm_page_size();
3103 size_t pages = size / page_sz;
3104
3105 assert(is_aligned(start, page_sz), "Start address must be page aligned");
3106 assert(is_aligned(size, page_sz), "Size must be page aligned");
3107
3108 committed_start = NULL;
3109
3110 int loops = (pages + stripe - 1) / stripe;
3111 int committed_pages = 0;
3112 address loop_base = start;
3113 bool found_range = false;
3114
3115 for (int index = 0; index < loops && !found_range; index ++) {
3116 assert(pages > 0, "Nothing to do");
3117 int pages_to_query = (pages >= stripe) ? stripe : pages;
3118 pages -= pages_to_query;
3119
3120 // Get stable read
3121 while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3122
3123 // During shutdown, some memory goes away without properly notifying NMT,
3124 // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3125 // Bailout and return as not committed for now.
3126 if (mincore_return_value == -1 && errno == ENOMEM) {
3127 return false;
3128 }
3129
3130 assert(vec[stripe] == 'X', "overflow guard");
3131 assert(mincore_return_value == 0, "Range must be valid");
3132 // Process this stripe
3133 for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3134 if ((vec[vecIdx] & 0x01) == 0) { // not committed
3135 // End of current contiguous region
3136 if (committed_start != NULL) {
3137 found_range = true;
3138 break;
3139 }
3140 } else { // committed
3141 // Start of region
3142 if (committed_start == NULL) {
3143 committed_start = loop_base + page_sz * vecIdx;
3144 }
3145 committed_pages ++;
3146 }
3147 }
3148
3149 loop_base += pages_to_query * page_sz;
3150 }
3151
3152 if (committed_start != NULL) {
3153 assert(committed_pages > 0, "Must have committed region");
3154 assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3155 assert(committed_start >= start && committed_start < start + size, "Out of range");
3156 committed_size = page_sz * committed_pages;
3157 return true;
3158 } else {
3159 assert(committed_pages == 0, "Should not have committed region");
3160 return false;
3161 }
3162 }
3163
3164
3165 // Linux uses a growable mapping for the stack, and if the mapping for
3166 // the stack guard pages is not removed when we detach a thread the
3167 // stack cannot grow beyond the pages where the stack guard was
3168 // mapped. If at some point later in the process the stack expands to
3169 // that point, the Linux kernel cannot expand the stack any further
3170 // because the guard pages are in the way, and a segfault occurs.
3171 //
3172 // However, it's essential not to split the stack region by unmapping
3173 // a region (leaving a hole) that's already part of the stack mapping,
3174 // so if the stack mapping has already grown beyond the guard pages at
3175 // the time we create them, we have to truncate the stack mapping.
3176 // So, we need to know the extent of the stack mapping when
3177 // create_stack_guard_pages() is called.
3178
3179 // We only need this for stacks that are growable: at the time of
3180 // writing thread stacks don't use growable mappings (i.e. those
3181 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3182 // only applies to the main thread.
3183
3184 // If the (growable) stack mapping already extends beyond the point
3185 // where we're going to put our guard pages, truncate the mapping at
3186 // that point by munmap()ping it. This ensures that when we later
3187 // munmap() the guard pages we don't leave a hole in the stack
3188 // mapping. This only affects the main/primordial thread
3189
3190 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3191 if (os::is_primordial_thread()) {
3192 // As we manually grow stack up to bottom inside create_attached_thread(),
3193 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3194 // we don't need to do anything special.
3195 // Check it first, before calling heavy function.
3196 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3197 unsigned char vec[1];
3198
3199 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3200 // Fallback to slow path on all errors, including EAGAIN
3201 stack_extent = (uintptr_t) get_stack_commited_bottom(
3202 os::Linux::initial_thread_stack_bottom(),
3203 (size_t)addr - stack_extent);
3204 }
3205
3206 if (stack_extent < (uintptr_t)addr) {
3207 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3208 }
3209 }
3210
3211 return os::commit_memory(addr, size, !ExecMem);
3212 }
3213
3214 // If this is a growable mapping, remove the guard pages entirely by
3215 // munmap()ping them. If not, just call uncommit_memory(). This only
3216 // affects the main/primordial thread, but guard against future OS changes.
3217 // It's safe to always unmap guard pages for primordial thread because we
3218 // always place it right after end of the mapped region.
3219
3220 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3221 uintptr_t stack_extent, stack_base;
3222
3223 if (os::is_primordial_thread()) {
3224 return ::munmap(addr, size) == 0;
3225 }
3226
3227 return os::uncommit_memory(addr, size);
3228 }
3229
3230 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3231 // at 'requested_addr'. If there are existing memory mappings at the same
3232 // location, however, they will be overwritten. If 'fixed' is false,
3233 // 'requested_addr' is only treated as a hint, the return value may or
3234 // may not start from the requested address. Unlike Linux mmap(), this
3235 // function returns NULL to indicate failure.
3236 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3237 char * addr;
3238 int flags;
3239
3240 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3241 if (fixed) {
3242 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3243 flags |= MAP_FIXED;
3244 }
3245
3246 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3247 // touch an uncommitted page. Otherwise, the read/write might
3248 // succeed if we have enough swap space to back the physical page.
3249 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3250 flags, -1, 0);
3251
3252 return addr == MAP_FAILED ? NULL : addr;
3253 }
3254
3255 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3256 // (req_addr != NULL) or with a given alignment.
3257 // - bytes shall be a multiple of alignment.
3258 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3259 // - alignment sets the alignment at which memory shall be allocated.
3260 // It must be a multiple of allocation granularity.
3261 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3262 // req_addr or NULL.
3263 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3264
3265 size_t extra_size = bytes;
3266 if (req_addr == NULL && alignment > 0) {
3267 extra_size += alignment;
3268 }
3269
3270 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3271 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3272 -1, 0);
3273 if (start == MAP_FAILED) {
3274 start = NULL;
3275 } else {
3276 if (req_addr != NULL) {
3277 if (start != req_addr) {
3278 ::munmap(start, extra_size);
3279 start = NULL;
3280 }
3281 } else {
3282 char* const start_aligned = align_up(start, alignment);
3283 char* const end_aligned = start_aligned + bytes;
3284 char* const end = start + extra_size;
3285 if (start_aligned > start) {
3286 ::munmap(start, start_aligned - start);
3287 }
3288 if (end_aligned < end) {
3289 ::munmap(end_aligned, end - end_aligned);
3290 }
3291 start = start_aligned;
3292 }
3293 }
3294 return start;
3295 }
3296
3297 static int anon_munmap(char * addr, size_t size) {
3298 return ::munmap(addr, size) == 0;
3299 }
3300
3301 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3302 size_t alignment_hint) {
3303 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3304 }
3305
3306 bool os::pd_release_memory(char* addr, size_t size) {
3307 return anon_munmap(addr, size);
3308 }
3309
3310 static bool linux_mprotect(char* addr, size_t size, int prot) {
3311 // Linux wants the mprotect address argument to be page aligned.
3312 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3313
3314 // According to SUSv3, mprotect() should only be used with mappings
3315 // established by mmap(), and mmap() always maps whole pages. Unaligned
3316 // 'addr' likely indicates problem in the VM (e.g. trying to change
3317 // protection of malloc'ed or statically allocated memory). Check the
3318 // caller if you hit this assert.
3319 assert(addr == bottom, "sanity check");
3320
3321 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3322 return ::mprotect(bottom, size, prot) == 0;
3323 }
3324
3325 // Set protections specified
3326 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3327 bool is_committed) {
3328 unsigned int p = 0;
3329 switch (prot) {
3330 case MEM_PROT_NONE: p = PROT_NONE; break;
3331 case MEM_PROT_READ: p = PROT_READ; break;
3332 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3333 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3334 default:
3335 ShouldNotReachHere();
3336 }
3337 // is_committed is unused.
3338 return linux_mprotect(addr, bytes, p);
3339 }
3340
3341 bool os::guard_memory(char* addr, size_t size) {
3342 return linux_mprotect(addr, size, PROT_NONE);
3343 }
3344
3345 bool os::unguard_memory(char* addr, size_t size) {
3346 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3347 }
3348
3349 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3350 size_t page_size) {
3351 bool result = false;
3352 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3353 MAP_ANONYMOUS|MAP_PRIVATE,
3354 -1, 0);
3355 if (p != MAP_FAILED) {
3356 void *aligned_p = align_up(p, page_size);
3357
3358 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3359
3360 munmap(p, page_size * 2);
3361 }
3362
3363 if (warn && !result) {
3364 warning("TransparentHugePages is not supported by the operating system.");
3365 }
3366
3367 return result;
3368 }
3369
3370 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3371 bool result = false;
3372 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3373 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3374 -1, 0);
3375
3376 if (p != MAP_FAILED) {
3377 // We don't know if this really is a huge page or not.
3378 FILE *fp = fopen("/proc/self/maps", "r");
3379 if (fp) {
3380 while (!feof(fp)) {
3381 char chars[257];
3382 long x = 0;
3383 if (fgets(chars, sizeof(chars), fp)) {
3384 if (sscanf(chars, "%lx-%*x", &x) == 1
3385 && x == (long)p) {
3386 if (strstr (chars, "hugepage")) {
3387 result = true;
3388 break;
3389 }
3390 }
3391 }
3392 }
3393 fclose(fp);
3394 }
3395 munmap(p, page_size);
3396 }
3397
3398 if (warn && !result) {
3399 warning("HugeTLBFS is not supported by the operating system.");
3400 }
3401
3402 return result;
3403 }
3404
3405 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3406 //
3407 // From the coredump_filter documentation:
3408 //
3409 // - (bit 0) anonymous private memory
3410 // - (bit 1) anonymous shared memory
3411 // - (bit 2) file-backed private memory
3412 // - (bit 3) file-backed shared memory
3413 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3414 // effective only if the bit 2 is cleared)
3415 // - (bit 5) hugetlb private memory
3416 // - (bit 6) hugetlb shared memory
3417 // - (bit 7) dax private memory
3418 // - (bit 8) dax shared memory
3419 //
3420 static void set_coredump_filter(CoredumpFilterBit bit) {
3421 FILE *f;
3422 long cdm;
3423
3424 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3425 return;
3426 }
3427
3428 if (fscanf(f, "%lx", &cdm) != 1) {
3429 fclose(f);
3430 return;
3431 }
3432
3433 long saved_cdm = cdm;
3434 rewind(f);
3435 cdm |= bit;
3436
3437 if (cdm != saved_cdm) {
3438 fprintf(f, "%#lx", cdm);
3439 }
3440
3441 fclose(f);
3442 }
3443
3444 // Large page support
3445
3446 static size_t _large_page_size = 0;
3447
3448 size_t os::Linux::find_large_page_size() {
3449 size_t large_page_size = 0;
3450
3451 // large_page_size on Linux is used to round up heap size. x86 uses either
3452 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3453 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3454 // page as large as 256M.
3455 //
3456 // Here we try to figure out page size by parsing /proc/meminfo and looking
3457 // for a line with the following format:
3458 // Hugepagesize: 2048 kB
3459 //
3460 // If we can't determine the value (e.g. /proc is not mounted, or the text
3461 // format has been changed), we'll use the largest page size supported by
3462 // the processor.
3463
3464 #ifndef ZERO
3465 large_page_size =
3466 AARCH64_ONLY(2 * M)
3467 AMD64_ONLY(2 * M)
3468 ARM32_ONLY(2 * M)
3469 IA32_ONLY(4 * M)
3470 IA64_ONLY(256 * M)
3471 PPC_ONLY(4 * M)
3472 S390_ONLY(1 * M)
3473 SPARC_ONLY(4 * M);
3474 #endif // ZERO
3475
3476 FILE *fp = fopen("/proc/meminfo", "r");
3477 if (fp) {
3478 while (!feof(fp)) {
3479 int x = 0;
3480 char buf[16];
3481 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3482 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3483 large_page_size = x * K;
3484 break;
3485 }
3486 } else {
3487 // skip to next line
3488 for (;;) {
3489 int ch = fgetc(fp);
3490 if (ch == EOF || ch == (int)'\n') break;
3491 }
3492 }
3493 }
3494 fclose(fp);
3495 }
3496
3497 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3498 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3499 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3500 proper_unit_for_byte_size(large_page_size));
3501 }
3502
3503 return large_page_size;
3504 }
3505
3506 size_t os::Linux::setup_large_page_size() {
3507 _large_page_size = Linux::find_large_page_size();
3508 const size_t default_page_size = (size_t)Linux::page_size();
3509 if (_large_page_size > default_page_size) {
3510 _page_sizes[0] = _large_page_size;
3511 _page_sizes[1] = default_page_size;
3512 _page_sizes[2] = 0;
3513 }
3514
3515 return _large_page_size;
3516 }
3517
3518 bool os::Linux::setup_large_page_type(size_t page_size) {
3519 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3520 FLAG_IS_DEFAULT(UseSHM) &&
3521 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3522
3523 // The type of large pages has not been specified by the user.
3524
3525 // Try UseHugeTLBFS and then UseSHM.
3526 UseHugeTLBFS = UseSHM = true;
3527
3528 // Don't try UseTransparentHugePages since there are known
3529 // performance issues with it turned on. This might change in the future.
3530 UseTransparentHugePages = false;
3531 }
3532
3533 if (UseTransparentHugePages) {
3534 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3535 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3536 UseHugeTLBFS = false;
3537 UseSHM = false;
3538 return true;
3539 }
3540 UseTransparentHugePages = false;
3541 }
3542
3543 if (UseHugeTLBFS) {
3544 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3545 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3546 UseSHM = false;
3547 return true;
3548 }
3549 UseHugeTLBFS = false;
3550 }
3551
3552 return UseSHM;
3553 }
3554
3555 void os::large_page_init() {
3556 if (!UseLargePages &&
3557 !UseTransparentHugePages &&
3558 !UseHugeTLBFS &&
3559 !UseSHM) {
3560 // Not using large pages.
3561 return;
3562 }
3563
3564 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3565 // The user explicitly turned off large pages.
3566 // Ignore the rest of the large pages flags.
3567 UseTransparentHugePages = false;
3568 UseHugeTLBFS = false;
3569 UseSHM = false;
3570 return;
3571 }
3572
3573 size_t large_page_size = Linux::setup_large_page_size();
3574 UseLargePages = Linux::setup_large_page_type(large_page_size);
3575
3576 set_coredump_filter(LARGEPAGES_BIT);
3577 }
3578
3579 #ifndef SHM_HUGETLB
3580 #define SHM_HUGETLB 04000
3581 #endif
3582
3583 #define shm_warning_format(format, ...) \
3584 do { \
3585 if (UseLargePages && \
3586 (!FLAG_IS_DEFAULT(UseLargePages) || \
3587 !FLAG_IS_DEFAULT(UseSHM) || \
3588 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3589 warning(format, __VA_ARGS__); \
3590 } \
3591 } while (0)
3592
3593 #define shm_warning(str) shm_warning_format("%s", str)
3594
3595 #define shm_warning_with_errno(str) \
3596 do { \
3597 int err = errno; \
3598 shm_warning_format(str " (error = %d)", err); \
3599 } while (0)
3600
3601 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3602 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3603
3604 if (!is_aligned(alignment, SHMLBA)) {
3605 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3606 return NULL;
3607 }
3608
3609 // To ensure that we get 'alignment' aligned memory from shmat,
3610 // we pre-reserve aligned virtual memory and then attach to that.
3611
3612 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3613 if (pre_reserved_addr == NULL) {
3614 // Couldn't pre-reserve aligned memory.
3615 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3616 return NULL;
3617 }
3618
3619 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3620 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3621
3622 if ((intptr_t)addr == -1) {
3623 int err = errno;
3624 shm_warning_with_errno("Failed to attach shared memory.");
3625
3626 assert(err != EACCES, "Unexpected error");
3627 assert(err != EIDRM, "Unexpected error");
3628 assert(err != EINVAL, "Unexpected error");
3629
3630 // Since we don't know if the kernel unmapped the pre-reserved memory area
3631 // we can't unmap it, since that would potentially unmap memory that was
3632 // mapped from other threads.
3633 return NULL;
3634 }
3635
3636 return addr;
3637 }
3638
3639 static char* shmat_at_address(int shmid, char* req_addr) {
3640 if (!is_aligned(req_addr, SHMLBA)) {
3641 assert(false, "Requested address needs to be SHMLBA aligned");
3642 return NULL;
3643 }
3644
3645 char* addr = (char*)shmat(shmid, req_addr, 0);
3646
3647 if ((intptr_t)addr == -1) {
3648 shm_warning_with_errno("Failed to attach shared memory.");
3649 return NULL;
3650 }
3651
3652 return addr;
3653 }
3654
3655 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3656 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3657 if (req_addr != NULL) {
3658 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3659 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3660 return shmat_at_address(shmid, req_addr);
3661 }
3662
3663 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3664 // return large page size aligned memory addresses when req_addr == NULL.
3665 // However, if the alignment is larger than the large page size, we have
3666 // to manually ensure that the memory returned is 'alignment' aligned.
3667 if (alignment > os::large_page_size()) {
3668 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3669 return shmat_with_alignment(shmid, bytes, alignment);
3670 } else {
3671 return shmat_at_address(shmid, NULL);
3672 }
3673 }
3674
3675 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3676 char* req_addr, bool exec) {
3677 // "exec" is passed in but not used. Creating the shared image for
3678 // the code cache doesn't have an SHM_X executable permission to check.
3679 assert(UseLargePages && UseSHM, "only for SHM large pages");
3680 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3681 assert(is_aligned(req_addr, alignment), "Unaligned address");
3682
3683 if (!is_aligned(bytes, os::large_page_size())) {
3684 return NULL; // Fallback to small pages.
3685 }
3686
3687 // Create a large shared memory region to attach to based on size.
3688 // Currently, size is the total size of the heap.
3689 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3690 if (shmid == -1) {
3691 // Possible reasons for shmget failure:
3692 // 1. shmmax is too small for Java heap.
3693 // > check shmmax value: cat /proc/sys/kernel/shmmax
3694 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3695 // 2. not enough large page memory.
3696 // > check available large pages: cat /proc/meminfo
3697 // > increase amount of large pages:
3698 // echo new_value > /proc/sys/vm/nr_hugepages
3699 // Note 1: different Linux may use different name for this property,
3700 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3701 // Note 2: it's possible there's enough physical memory available but
3702 // they are so fragmented after a long run that they can't
3703 // coalesce into large pages. Try to reserve large pages when
3704 // the system is still "fresh".
3705 shm_warning_with_errno("Failed to reserve shared memory.");
3706 return NULL;
3707 }
3708
3709 // Attach to the region.
3710 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3711
3712 // Remove shmid. If shmat() is successful, the actual shared memory segment
3713 // will be deleted when it's detached by shmdt() or when the process
3714 // terminates. If shmat() is not successful this will remove the shared
3715 // segment immediately.
3716 shmctl(shmid, IPC_RMID, NULL);
3717
3718 return addr;
3719 }
3720
3721 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3722 int error) {
3723 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3724
3725 bool warn_on_failure = UseLargePages &&
3726 (!FLAG_IS_DEFAULT(UseLargePages) ||
3727 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3728 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3729
3730 if (warn_on_failure) {
3731 char msg[128];
3732 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3733 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3734 warning("%s", msg);
3735 }
3736 }
3737
3738 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3739 char* req_addr,
3740 bool exec) {
3741 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3742 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3743 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3744
3745 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3746 char* addr = (char*)::mmap(req_addr, bytes, prot,
3747 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3748 -1, 0);
3749
3750 if (addr == MAP_FAILED) {
3751 warn_on_large_pages_failure(req_addr, bytes, errno);
3752 return NULL;
3753 }
3754
3755 assert(is_aligned(addr, os::large_page_size()), "Must be");
3756
3757 return addr;
3758 }
3759
3760 // Reserve memory using mmap(MAP_HUGETLB).
3761 // - bytes shall be a multiple of alignment.
3762 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3763 // - alignment sets the alignment at which memory shall be allocated.
3764 // It must be a multiple of allocation granularity.
3765 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3766 // req_addr or NULL.
3767 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3768 size_t alignment,
3769 char* req_addr,
3770 bool exec) {
3771 size_t large_page_size = os::large_page_size();
3772 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3773
3774 assert(is_aligned(req_addr, alignment), "Must be");
3775 assert(is_aligned(bytes, alignment), "Must be");
3776
3777 // First reserve - but not commit - the address range in small pages.
3778 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3779
3780 if (start == NULL) {
3781 return NULL;
3782 }
3783
3784 assert(is_aligned(start, alignment), "Must be");
3785
3786 char* end = start + bytes;
3787
3788 // Find the regions of the allocated chunk that can be promoted to large pages.
3789 char* lp_start = align_up(start, large_page_size);
3790 char* lp_end = align_down(end, large_page_size);
3791
3792 size_t lp_bytes = lp_end - lp_start;
3793
3794 assert(is_aligned(lp_bytes, large_page_size), "Must be");
3795
3796 if (lp_bytes == 0) {
3797 // The mapped region doesn't even span the start and the end of a large page.
3798 // Fall back to allocate a non-special area.
3799 ::munmap(start, end - start);
3800 return NULL;
3801 }
3802
3803 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3804
3805 void* result;
3806
3807 // Commit small-paged leading area.
3808 if (start != lp_start) {
3809 result = ::mmap(start, lp_start - start, prot,
3810 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3811 -1, 0);
3812 if (result == MAP_FAILED) {
3813 ::munmap(lp_start, end - lp_start);
3814 return NULL;
3815 }
3816 }
3817
3818 // Commit large-paged area.
3819 result = ::mmap(lp_start, lp_bytes, prot,
3820 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3821 -1, 0);
3822 if (result == MAP_FAILED) {
3823 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3824 // If the mmap above fails, the large pages region will be unmapped and we
3825 // have regions before and after with small pages. Release these regions.
3826 //
3827 // | mapped | unmapped | mapped |
3828 // ^ ^ ^ ^
3829 // start lp_start lp_end end
3830 //
3831 ::munmap(start, lp_start - start);
3832 ::munmap(lp_end, end - lp_end);
3833 return NULL;
3834 }
3835
3836 // Commit small-paged trailing area.
3837 if (lp_end != end) {
3838 result = ::mmap(lp_end, end - lp_end, prot,
3839 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3840 -1, 0);
3841 if (result == MAP_FAILED) {
3842 ::munmap(start, lp_end - start);
3843 return NULL;
3844 }
3845 }
3846
3847 return start;
3848 }
3849
3850 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3851 size_t alignment,
3852 char* req_addr,
3853 bool exec) {
3854 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3855 assert(is_aligned(req_addr, alignment), "Must be");
3856 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3857 assert(is_power_of_2(os::large_page_size()), "Must be");
3858 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3859
3860 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3861 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3862 } else {
3863 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3864 }
3865 }
3866
3867 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3868 char* req_addr, bool exec) {
3869 assert(UseLargePages, "only for large pages");
3870
3871 char* addr;
3872 if (UseSHM) {
3873 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3874 } else {
3875 assert(UseHugeTLBFS, "must be");
3876 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3877 }
3878
3879 if (addr != NULL) {
3880 if (UseNUMAInterleaving) {
3881 numa_make_global(addr, bytes);
3882 }
3883
3884 // The memory is committed
3885 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3886 }
3887
3888 return addr;
3889 }
3890
3891 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3892 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3893 return shmdt(base) == 0;
3894 }
3895
3896 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3897 return pd_release_memory(base, bytes);
3898 }
3899
3900 bool os::release_memory_special(char* base, size_t bytes) {
3901 bool res;
3902 if (MemTracker::tracking_level() > NMT_minimal) {
3903 Tracker tkr(Tracker::release);
3904 res = os::Linux::release_memory_special_impl(base, bytes);
3905 if (res) {
3906 tkr.record((address)base, bytes);
3907 }
3908
3909 } else {
3910 res = os::Linux::release_memory_special_impl(base, bytes);
3911 }
3912 return res;
3913 }
3914
3915 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3916 assert(UseLargePages, "only for large pages");
3917 bool res;
3918
3919 if (UseSHM) {
3920 res = os::Linux::release_memory_special_shm(base, bytes);
3921 } else {
3922 assert(UseHugeTLBFS, "must be");
3923 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3924 }
3925 return res;
3926 }
3927
3928 size_t os::large_page_size() {
3929 return _large_page_size;
3930 }
3931
3932 // With SysV SHM the entire memory region must be allocated as shared
3933 // memory.
3934 // HugeTLBFS allows application to commit large page memory on demand.
3935 // However, when committing memory with HugeTLBFS fails, the region
3936 // that was supposed to be committed will lose the old reservation
3937 // and allow other threads to steal that memory region. Because of this
3938 // behavior we can't commit HugeTLBFS memory.
3939 bool os::can_commit_large_page_memory() {
3940 return UseTransparentHugePages;
3941 }
3942
3943 bool os::can_execute_large_page_memory() {
3944 return UseTransparentHugePages || UseHugeTLBFS;
3945 }
3946
3947 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
3948 assert(file_desc >= 0, "file_desc is not valid");
3949 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
3950 if (result != NULL) {
3951 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
3952 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
3953 }
3954 }
3955 return result;
3956 }
3957
3958 // Reserve memory at an arbitrary address, only if that area is
3959 // available (and not reserved for something else).
3960
3961 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3962 const int max_tries = 10;
3963 char* base[max_tries];
3964 size_t size[max_tries];
3965 const size_t gap = 0x000000;
3966
3967 // Assert only that the size is a multiple of the page size, since
3968 // that's all that mmap requires, and since that's all we really know
3969 // about at this low abstraction level. If we need higher alignment,
3970 // we can either pass an alignment to this method or verify alignment
3971 // in one of the methods further up the call chain. See bug 5044738.
3972 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3973
3974 // Repeatedly allocate blocks until the block is allocated at the
3975 // right spot.
3976
3977 // Linux mmap allows caller to pass an address as hint; give it a try first,
3978 // if kernel honors the hint then we can return immediately.
3979 char * addr = anon_mmap(requested_addr, bytes, false);
3980 if (addr == requested_addr) {
3981 return requested_addr;
3982 }
3983
3984 if (addr != NULL) {
3985 // mmap() is successful but it fails to reserve at the requested address
3986 anon_munmap(addr, bytes);
3987 }
3988
3989 int i;
3990 for (i = 0; i < max_tries; ++i) {
3991 base[i] = reserve_memory(bytes);
3992
3993 if (base[i] != NULL) {
3994 // Is this the block we wanted?
3995 if (base[i] == requested_addr) {
3996 size[i] = bytes;
3997 break;
3998 }
3999
4000 // Does this overlap the block we wanted? Give back the overlapped
4001 // parts and try again.
4002
4003 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
4004 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
4005 unmap_memory(base[i], top_overlap);
4006 base[i] += top_overlap;
4007 size[i] = bytes - top_overlap;
4008 } else {
4009 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
4010 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
4011 unmap_memory(requested_addr, bottom_overlap);
4012 size[i] = bytes - bottom_overlap;
4013 } else {
4014 size[i] = bytes;
4015 }
4016 }
4017 }
4018 }
4019
4020 // Give back the unused reserved pieces.
4021
4022 for (int j = 0; j < i; ++j) {
4023 if (base[j] != NULL) {
4024 unmap_memory(base[j], size[j]);
4025 }
4026 }
4027
4028 if (i < max_tries) {
4029 return requested_addr;
4030 } else {
4031 return NULL;
4032 }
4033 }
4034
4035 size_t os::read(int fd, void *buf, unsigned int nBytes) {
4036 return ::read(fd, buf, nBytes);
4037 }
4038
4039 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
4040 return ::pread(fd, buf, nBytes, offset);
4041 }
4042
4043 // Short sleep, direct OS call.
4044 //
4045 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
4046 // sched_yield(2) will actually give up the CPU:
4047 //
4048 // * Alone on this pariticular CPU, keeps running.
4049 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
4050 // (pre 2.6.39).
4051 //
4052 // So calling this with 0 is an alternative.
4053 //
4054 void os::naked_short_sleep(jlong ms) {
4055 struct timespec req;
4056
4057 assert(ms < 1000, "Un-interruptable sleep, short time use only");
4058 req.tv_sec = 0;
4059 if (ms > 0) {
4060 req.tv_nsec = (ms % 1000) * 1000000;
4061 } else {
4062 req.tv_nsec = 1;
4063 }
4064
4065 nanosleep(&req, NULL);
4066
4067 return;
4068 }
4069
4070 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4071 void os::infinite_sleep() {
4072 while (true) { // sleep forever ...
4073 ::sleep(100); // ... 100 seconds at a time
4074 }
4075 }
4076
4077 // Used to convert frequent JVM_Yield() to nops
4078 bool os::dont_yield() {
4079 return DontYieldALot;
4080 }
4081
4082 void os::naked_yield() {
4083 sched_yield();
4084 }
4085
4086 ////////////////////////////////////////////////////////////////////////////////
4087 // thread priority support
4088
4089 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4090 // only supports dynamic priority, static priority must be zero. For real-time
4091 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4092 // However, for large multi-threaded applications, SCHED_RR is not only slower
4093 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4094 // of 5 runs - Sep 2005).
4095 //
4096 // The following code actually changes the niceness of kernel-thread/LWP. It
4097 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4098 // not the entire user process, and user level threads are 1:1 mapped to kernel
4099 // threads. It has always been the case, but could change in the future. For
4100 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4101 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4102
4103 int os::java_to_os_priority[CriticalPriority + 1] = {
4104 19, // 0 Entry should never be used
4105
4106 4, // 1 MinPriority
4107 3, // 2
4108 2, // 3
4109
4110 1, // 4
4111 0, // 5 NormPriority
4112 -1, // 6
4113
4114 -2, // 7
4115 -3, // 8
4116 -4, // 9 NearMaxPriority
4117
4118 -5, // 10 MaxPriority
4119
4120 -5 // 11 CriticalPriority
4121 };
4122
4123 static int prio_init() {
4124 if (ThreadPriorityPolicy == 1) {
4125 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4126 // if effective uid is not root. Perhaps, a more elegant way of doing
4127 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4128 if (geteuid() != 0) {
4129 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4130 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4131 }
4132 ThreadPriorityPolicy = 0;
4133 }
4134 }
4135 if (UseCriticalJavaThreadPriority) {
4136 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4137 }
4138 return 0;
4139 }
4140
4141 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4142 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4143
4144 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4145 return (ret == 0) ? OS_OK : OS_ERR;
4146 }
4147
4148 OSReturn os::get_native_priority(const Thread* const thread,
4149 int *priority_ptr) {
4150 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4151 *priority_ptr = java_to_os_priority[NormPriority];
4152 return OS_OK;
4153 }
4154
4155 errno = 0;
4156 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4157 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4158 }
4159
4160 ////////////////////////////////////////////////////////////////////////////////
4161 // suspend/resume support
4162
4163 // The low-level signal-based suspend/resume support is a remnant from the
4164 // old VM-suspension that used to be for java-suspension, safepoints etc,
4165 // within hotspot. Currently used by JFR's OSThreadSampler
4166 //
4167 // The remaining code is greatly simplified from the more general suspension
4168 // code that used to be used.
4169 //
4170 // The protocol is quite simple:
4171 // - suspend:
4172 // - sends a signal to the target thread
4173 // - polls the suspend state of the osthread using a yield loop
4174 // - target thread signal handler (SR_handler) sets suspend state
4175 // and blocks in sigsuspend until continued
4176 // - resume:
4177 // - sets target osthread state to continue
4178 // - sends signal to end the sigsuspend loop in the SR_handler
4179 //
4180 // Note that the SR_lock plays no role in this suspend/resume protocol,
4181 // but is checked for NULL in SR_handler as a thread termination indicator.
4182 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4183 //
4184 // Note that resume_clear_context() and suspend_save_context() are needed
4185 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4186 // which in part is used by:
4187 // - Forte Analyzer: AsyncGetCallTrace()
4188 // - StackBanging: get_frame_at_stack_banging_point()
4189
4190 static void resume_clear_context(OSThread *osthread) {
4191 osthread->set_ucontext(NULL);
4192 osthread->set_siginfo(NULL);
4193 }
4194
4195 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4196 ucontext_t* context) {
4197 osthread->set_ucontext(context);
4198 osthread->set_siginfo(siginfo);
4199 }
4200
4201 // Handler function invoked when a thread's execution is suspended or
4202 // resumed. We have to be careful that only async-safe functions are
4203 // called here (Note: most pthread functions are not async safe and
4204 // should be avoided.)
4205 //
4206 // Note: sigwait() is a more natural fit than sigsuspend() from an
4207 // interface point of view, but sigwait() prevents the signal hander
4208 // from being run. libpthread would get very confused by not having
4209 // its signal handlers run and prevents sigwait()'s use with the
4210 // mutex granting granting signal.
4211 //
4212 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4213 //
4214 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4215 // Save and restore errno to avoid confusing native code with EINTR
4216 // after sigsuspend.
4217 int old_errno = errno;
4218
4219 Thread* thread = Thread::current_or_null_safe();
4220 assert(thread != NULL, "Missing current thread in SR_handler");
4221
4222 // On some systems we have seen signal delivery get "stuck" until the signal
4223 // mask is changed as part of thread termination. Check that the current thread
4224 // has not already terminated (via SR_lock()) - else the following assertion
4225 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4226 // destructor has completed.
4227
4228 if (thread->SR_lock() == NULL) {
4229 return;
4230 }
4231
4232 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4233
4234 OSThread* osthread = thread->osthread();
4235
4236 os::SuspendResume::State current = osthread->sr.state();
4237 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4238 suspend_save_context(osthread, siginfo, context);
4239
4240 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4241 os::SuspendResume::State state = osthread->sr.suspended();
4242 if (state == os::SuspendResume::SR_SUSPENDED) {
4243 sigset_t suspend_set; // signals for sigsuspend()
4244 sigemptyset(&suspend_set);
4245 // get current set of blocked signals and unblock resume signal
4246 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4247 sigdelset(&suspend_set, SR_signum);
4248
4249 sr_semaphore.signal();
4250 // wait here until we are resumed
4251 while (1) {
4252 sigsuspend(&suspend_set);
4253
4254 os::SuspendResume::State result = osthread->sr.running();
4255 if (result == os::SuspendResume::SR_RUNNING) {
4256 sr_semaphore.signal();
4257 break;
4258 }
4259 }
4260
4261 } else if (state == os::SuspendResume::SR_RUNNING) {
4262 // request was cancelled, continue
4263 } else {
4264 ShouldNotReachHere();
4265 }
4266
4267 resume_clear_context(osthread);
4268 } else if (current == os::SuspendResume::SR_RUNNING) {
4269 // request was cancelled, continue
4270 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4271 // ignore
4272 } else {
4273 // ignore
4274 }
4275
4276 errno = old_errno;
4277 }
4278
4279 static int SR_initialize() {
4280 struct sigaction act;
4281 char *s;
4282
4283 // Get signal number to use for suspend/resume
4284 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4285 int sig = ::strtol(s, 0, 10);
4286 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4287 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4288 SR_signum = sig;
4289 } else {
4290 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4291 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4292 }
4293 }
4294
4295 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4296 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4297
4298 sigemptyset(&SR_sigset);
4299 sigaddset(&SR_sigset, SR_signum);
4300
4301 // Set up signal handler for suspend/resume
4302 act.sa_flags = SA_RESTART|SA_SIGINFO;
4303 act.sa_handler = (void (*)(int)) SR_handler;
4304
4305 // SR_signum is blocked by default.
4306 // 4528190 - We also need to block pthread restart signal (32 on all
4307 // supported Linux platforms). Note that LinuxThreads need to block
4308 // this signal for all threads to work properly. So we don't have
4309 // to use hard-coded signal number when setting up the mask.
4310 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4311
4312 if (sigaction(SR_signum, &act, 0) == -1) {
4313 return -1;
4314 }
4315
4316 // Save signal flag
4317 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4318 return 0;
4319 }
4320
4321 static int sr_notify(OSThread* osthread) {
4322 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4323 assert_status(status == 0, status, "pthread_kill");
4324 return status;
4325 }
4326
4327 // "Randomly" selected value for how long we want to spin
4328 // before bailing out on suspending a thread, also how often
4329 // we send a signal to a thread we want to resume
4330 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4331 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4332
4333 // returns true on success and false on error - really an error is fatal
4334 // but this seems the normal response to library errors
4335 static bool do_suspend(OSThread* osthread) {
4336 assert(osthread->sr.is_running(), "thread should be running");
4337 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4338
4339 // mark as suspended and send signal
4340 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4341 // failed to switch, state wasn't running?
4342 ShouldNotReachHere();
4343 return false;
4344 }
4345
4346 if (sr_notify(osthread) != 0) {
4347 ShouldNotReachHere();
4348 }
4349
4350 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4351 while (true) {
4352 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4353 break;
4354 } else {
4355 // timeout
4356 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4357 if (cancelled == os::SuspendResume::SR_RUNNING) {
4358 return false;
4359 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4360 // make sure that we consume the signal on the semaphore as well
4361 sr_semaphore.wait();
4362 break;
4363 } else {
4364 ShouldNotReachHere();
4365 return false;
4366 }
4367 }
4368 }
4369
4370 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4371 return true;
4372 }
4373
4374 static void do_resume(OSThread* osthread) {
4375 assert(osthread->sr.is_suspended(), "thread should be suspended");
4376 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4377
4378 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4379 // failed to switch to WAKEUP_REQUEST
4380 ShouldNotReachHere();
4381 return;
4382 }
4383
4384 while (true) {
4385 if (sr_notify(osthread) == 0) {
4386 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4387 if (osthread->sr.is_running()) {
4388 return;
4389 }
4390 }
4391 } else {
4392 ShouldNotReachHere();
4393 }
4394 }
4395
4396 guarantee(osthread->sr.is_running(), "Must be running!");
4397 }
4398
4399 ///////////////////////////////////////////////////////////////////////////////////
4400 // signal handling (except suspend/resume)
4401
4402 // This routine may be used by user applications as a "hook" to catch signals.
4403 // The user-defined signal handler must pass unrecognized signals to this
4404 // routine, and if it returns true (non-zero), then the signal handler must
4405 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4406 // routine will never retun false (zero), but instead will execute a VM panic
4407 // routine kill the process.
4408 //
4409 // If this routine returns false, it is OK to call it again. This allows
4410 // the user-defined signal handler to perform checks either before or after
4411 // the VM performs its own checks. Naturally, the user code would be making
4412 // a serious error if it tried to handle an exception (such as a null check
4413 // or breakpoint) that the VM was generating for its own correct operation.
4414 //
4415 // This routine may recognize any of the following kinds of signals:
4416 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4417 // It should be consulted by handlers for any of those signals.
4418 //
4419 // The caller of this routine must pass in the three arguments supplied
4420 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4421 // field of the structure passed to sigaction(). This routine assumes that
4422 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4423 //
4424 // Note that the VM will print warnings if it detects conflicting signal
4425 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4426 //
4427 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4428 siginfo_t* siginfo,
4429 void* ucontext,
4430 int abort_if_unrecognized);
4431
4432 static void signalHandler(int sig, siginfo_t* info, void* uc) {
4433 assert(info != NULL && uc != NULL, "it must be old kernel");
4434 int orig_errno = errno; // Preserve errno value over signal handler.
4435 JVM_handle_linux_signal(sig, info, uc, true);
4436 errno = orig_errno;
4437 }
4438
4439
4440 // This boolean allows users to forward their own non-matching signals
4441 // to JVM_handle_linux_signal, harmlessly.
4442 bool os::Linux::signal_handlers_are_installed = false;
4443
4444 // For signal-chaining
4445 struct sigaction sigact[NSIG];
4446 uint64_t sigs = 0;
4447 #if (64 < NSIG-1)
4448 #error "Not all signals can be encoded in sigs. Adapt its type!"
4449 #endif
4450 bool os::Linux::libjsig_is_loaded = false;
4451 typedef struct sigaction *(*get_signal_t)(int);
4452 get_signal_t os::Linux::get_signal_action = NULL;
4453
4454 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4455 struct sigaction *actp = NULL;
4456
4457 if (libjsig_is_loaded) {
4458 // Retrieve the old signal handler from libjsig
4459 actp = (*get_signal_action)(sig);
4460 }
4461 if (actp == NULL) {
4462 // Retrieve the preinstalled signal handler from jvm
4463 actp = get_preinstalled_handler(sig);
4464 }
4465
4466 return actp;
4467 }
4468
4469 static bool call_chained_handler(struct sigaction *actp, int sig,
4470 siginfo_t *siginfo, void *context) {
4471 // Call the old signal handler
4472 if (actp->sa_handler == SIG_DFL) {
4473 // It's more reasonable to let jvm treat it as an unexpected exception
4474 // instead of taking the default action.
4475 return false;
4476 } else if (actp->sa_handler != SIG_IGN) {
4477 if ((actp->sa_flags & SA_NODEFER) == 0) {
4478 // automaticlly block the signal
4479 sigaddset(&(actp->sa_mask), sig);
4480 }
4481
4482 sa_handler_t hand = NULL;
4483 sa_sigaction_t sa = NULL;
4484 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4485 // retrieve the chained handler
4486 if (siginfo_flag_set) {
4487 sa = actp->sa_sigaction;
4488 } else {
4489 hand = actp->sa_handler;
4490 }
4491
4492 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4493 actp->sa_handler = SIG_DFL;
4494 }
4495
4496 // try to honor the signal mask
4497 sigset_t oset;
4498 sigemptyset(&oset);
4499 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4500
4501 // call into the chained handler
4502 if (siginfo_flag_set) {
4503 (*sa)(sig, siginfo, context);
4504 } else {
4505 (*hand)(sig);
4506 }
4507
4508 // restore the signal mask
4509 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4510 }
4511 // Tell jvm's signal handler the signal is taken care of.
4512 return true;
4513 }
4514
4515 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4516 bool chained = false;
4517 // signal-chaining
4518 if (UseSignalChaining) {
4519 struct sigaction *actp = get_chained_signal_action(sig);
4520 if (actp != NULL) {
4521 chained = call_chained_handler(actp, sig, siginfo, context);
4522 }
4523 }
4524 return chained;
4525 }
4526
4527 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4528 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4529 return &sigact[sig];
4530 }
4531 return NULL;
4532 }
4533
4534 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4535 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4536 sigact[sig] = oldAct;
4537 sigs |= (uint64_t)1 << (sig-1);
4538 }
4539
4540 // for diagnostic
4541 int sigflags[NSIG];
4542
4543 int os::Linux::get_our_sigflags(int sig) {
4544 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4545 return sigflags[sig];
4546 }
4547
4548 void os::Linux::set_our_sigflags(int sig, int flags) {
4549 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4550 if (sig > 0 && sig < NSIG) {
4551 sigflags[sig] = flags;
4552 }
4553 }
4554
4555 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4556 // Check for overwrite.
4557 struct sigaction oldAct;
4558 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4559
4560 void* oldhand = oldAct.sa_sigaction
4561 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4562 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4563 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4564 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4565 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4566 if (AllowUserSignalHandlers || !set_installed) {
4567 // Do not overwrite; user takes responsibility to forward to us.
4568 return;
4569 } else if (UseSignalChaining) {
4570 // save the old handler in jvm
4571 save_preinstalled_handler(sig, oldAct);
4572 // libjsig also interposes the sigaction() call below and saves the
4573 // old sigaction on it own.
4574 } else {
4575 fatal("Encountered unexpected pre-existing sigaction handler "
4576 "%#lx for signal %d.", (long)oldhand, sig);
4577 }
4578 }
4579
4580 struct sigaction sigAct;
4581 sigfillset(&(sigAct.sa_mask));
4582 sigAct.sa_handler = SIG_DFL;
4583 if (!set_installed) {
4584 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4585 } else {
4586 sigAct.sa_sigaction = signalHandler;
4587 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4588 }
4589 // Save flags, which are set by ours
4590 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4591 sigflags[sig] = sigAct.sa_flags;
4592
4593 int ret = sigaction(sig, &sigAct, &oldAct);
4594 assert(ret == 0, "check");
4595
4596 void* oldhand2 = oldAct.sa_sigaction
4597 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4598 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4599 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4600 }
4601
4602 // install signal handlers for signals that HotSpot needs to
4603 // handle in order to support Java-level exception handling.
4604
4605 void os::Linux::install_signal_handlers() {
4606 if (!signal_handlers_are_installed) {
4607 signal_handlers_are_installed = true;
4608
4609 // signal-chaining
4610 typedef void (*signal_setting_t)();
4611 signal_setting_t begin_signal_setting = NULL;
4612 signal_setting_t end_signal_setting = NULL;
4613 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4614 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4615 if (begin_signal_setting != NULL) {
4616 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4617 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4618 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4619 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4620 libjsig_is_loaded = true;
4621 assert(UseSignalChaining, "should enable signal-chaining");
4622 }
4623 if (libjsig_is_loaded) {
4624 // Tell libjsig jvm is setting signal handlers
4625 (*begin_signal_setting)();
4626 }
4627
4628 set_signal_handler(SIGSEGV, true);
4629 set_signal_handler(SIGPIPE, true);
4630 set_signal_handler(SIGBUS, true);
4631 set_signal_handler(SIGILL, true);
4632 set_signal_handler(SIGFPE, true);
4633 #if defined(PPC64)
4634 set_signal_handler(SIGTRAP, true);
4635 #endif
4636 set_signal_handler(SIGXFSZ, true);
4637
4638 if (libjsig_is_loaded) {
4639 // Tell libjsig jvm finishes setting signal handlers
4640 (*end_signal_setting)();
4641 }
4642
4643 // We don't activate signal checker if libjsig is in place, we trust ourselves
4644 // and if UserSignalHandler is installed all bets are off.
4645 // Log that signal checking is off only if -verbose:jni is specified.
4646 if (CheckJNICalls) {
4647 if (libjsig_is_loaded) {
4648 if (PrintJNIResolving) {
4649 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4650 }
4651 check_signals = false;
4652 }
4653 if (AllowUserSignalHandlers) {
4654 if (PrintJNIResolving) {
4655 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4656 }
4657 check_signals = false;
4658 }
4659 }
4660 }
4661 }
4662
4663 // This is the fastest way to get thread cpu time on Linux.
4664 // Returns cpu time (user+sys) for any thread, not only for current.
4665 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4666 // It might work on 2.6.10+ with a special kernel/glibc patch.
4667 // For reference, please, see IEEE Std 1003.1-2004:
4668 // http://www.unix.org/single_unix_specification
4669
4670 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4671 struct timespec tp;
4672 int rc = os::Posix::clock_gettime(clockid, &tp);
4673 assert(rc == 0, "clock_gettime is expected to return 0 code");
4674
4675 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4676 }
4677
4678 void os::Linux::initialize_os_info() {
4679 assert(_os_version == 0, "OS info already initialized");
4680
4681 struct utsname _uname;
4682
4683 uint32_t major;
4684 uint32_t minor;
4685 uint32_t fix;
4686
4687 int rc;
4688
4689 // Kernel version is unknown if
4690 // verification below fails.
4691 _os_version = 0x01000000;
4692
4693 rc = uname(&_uname);
4694 if (rc != -1) {
4695
4696 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4697 if (rc == 3) {
4698
4699 if (major < 256 && minor < 256 && fix < 256) {
4700 // Kernel version format is as expected,
4701 // set it overriding unknown state.
4702 _os_version = (major << 16) |
4703 (minor << 8 ) |
4704 (fix << 0 ) ;
4705 }
4706 }
4707 }
4708 }
4709
4710 uint32_t os::Linux::os_version() {
4711 assert(_os_version != 0, "not initialized");
4712 return _os_version & 0x00FFFFFF;
4713 }
4714
4715 bool os::Linux::os_version_is_known() {
4716 assert(_os_version != 0, "not initialized");
4717 return _os_version & 0x01000000 ? false : true;
4718 }
4719
4720 /////
4721 // glibc on Linux platform uses non-documented flag
4722 // to indicate, that some special sort of signal
4723 // trampoline is used.
4724 // We will never set this flag, and we should
4725 // ignore this flag in our diagnostic
4726 #ifdef SIGNIFICANT_SIGNAL_MASK
4727 #undef SIGNIFICANT_SIGNAL_MASK
4728 #endif
4729 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4730
4731 static const char* get_signal_handler_name(address handler,
4732 char* buf, int buflen) {
4733 int offset = 0;
4734 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4735 if (found) {
4736 // skip directory names
4737 const char *p1, *p2;
4738 p1 = buf;
4739 size_t len = strlen(os::file_separator());
4740 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4741 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4742 } else {
4743 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4744 }
4745 return buf;
4746 }
4747
4748 static void print_signal_handler(outputStream* st, int sig,
4749 char* buf, size_t buflen) {
4750 struct sigaction sa;
4751
4752 sigaction(sig, NULL, &sa);
4753
4754 // See comment for SIGNIFICANT_SIGNAL_MASK define
4755 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4756
4757 st->print("%s: ", os::exception_name(sig, buf, buflen));
4758
4759 address handler = (sa.sa_flags & SA_SIGINFO)
4760 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4761 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4762
4763 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4764 st->print("SIG_DFL");
4765 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4766 st->print("SIG_IGN");
4767 } else {
4768 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4769 }
4770
4771 st->print(", sa_mask[0]=");
4772 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4773
4774 address rh = VMError::get_resetted_sighandler(sig);
4775 // May be, handler was resetted by VMError?
4776 if (rh != NULL) {
4777 handler = rh;
4778 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4779 }
4780
4781 st->print(", sa_flags=");
4782 os::Posix::print_sa_flags(st, sa.sa_flags);
4783
4784 // Check: is it our handler?
4785 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4786 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4787 // It is our signal handler
4788 // check for flags, reset system-used one!
4789 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4790 st->print(
4791 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4792 os::Linux::get_our_sigflags(sig));
4793 }
4794 }
4795 st->cr();
4796 }
4797
4798
4799 #define DO_SIGNAL_CHECK(sig) \
4800 do { \
4801 if (!sigismember(&check_signal_done, sig)) { \
4802 os::Linux::check_signal_handler(sig); \
4803 } \
4804 } while (0)
4805
4806 // This method is a periodic task to check for misbehaving JNI applications
4807 // under CheckJNI, we can add any periodic checks here
4808
4809 void os::run_periodic_checks() {
4810 if (check_signals == false) return;
4811
4812 // SEGV and BUS if overridden could potentially prevent
4813 // generation of hs*.log in the event of a crash, debugging
4814 // such a case can be very challenging, so we absolutely
4815 // check the following for a good measure:
4816 DO_SIGNAL_CHECK(SIGSEGV);
4817 DO_SIGNAL_CHECK(SIGILL);
4818 DO_SIGNAL_CHECK(SIGFPE);
4819 DO_SIGNAL_CHECK(SIGBUS);
4820 DO_SIGNAL_CHECK(SIGPIPE);
4821 DO_SIGNAL_CHECK(SIGXFSZ);
4822 #if defined(PPC64)
4823 DO_SIGNAL_CHECK(SIGTRAP);
4824 #endif
4825
4826 // ReduceSignalUsage allows the user to override these handlers
4827 // see comments at the very top and jvm_md.h
4828 if (!ReduceSignalUsage) {
4829 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4830 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4831 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4832 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4833 }
4834
4835 DO_SIGNAL_CHECK(SR_signum);
4836 }
4837
4838 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4839
4840 static os_sigaction_t os_sigaction = NULL;
4841
4842 void os::Linux::check_signal_handler(int sig) {
4843 char buf[O_BUFLEN];
4844 address jvmHandler = NULL;
4845
4846
4847 struct sigaction act;
4848 if (os_sigaction == NULL) {
4849 // only trust the default sigaction, in case it has been interposed
4850 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4851 if (os_sigaction == NULL) return;
4852 }
4853
4854 os_sigaction(sig, (struct sigaction*)NULL, &act);
4855
4856
4857 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4858
4859 address thisHandler = (act.sa_flags & SA_SIGINFO)
4860 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4861 : CAST_FROM_FN_PTR(address, act.sa_handler);
4862
4863
4864 switch (sig) {
4865 case SIGSEGV:
4866 case SIGBUS:
4867 case SIGFPE:
4868 case SIGPIPE:
4869 case SIGILL:
4870 case SIGXFSZ:
4871 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4872 break;
4873
4874 case SHUTDOWN1_SIGNAL:
4875 case SHUTDOWN2_SIGNAL:
4876 case SHUTDOWN3_SIGNAL:
4877 case BREAK_SIGNAL:
4878 jvmHandler = (address)user_handler();
4879 break;
4880
4881 default:
4882 if (sig == SR_signum) {
4883 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4884 } else {
4885 return;
4886 }
4887 break;
4888 }
4889
4890 if (thisHandler != jvmHandler) {
4891 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4892 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4893 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4894 // No need to check this sig any longer
4895 sigaddset(&check_signal_done, sig);
4896 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4897 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4898 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4899 exception_name(sig, buf, O_BUFLEN));
4900 }
4901 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4902 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4903 tty->print("expected:");
4904 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4905 tty->cr();
4906 tty->print(" found:");
4907 os::Posix::print_sa_flags(tty, act.sa_flags);
4908 tty->cr();
4909 // No need to check this sig any longer
4910 sigaddset(&check_signal_done, sig);
4911 }
4912
4913 // Dump all the signal
4914 if (sigismember(&check_signal_done, sig)) {
4915 print_signal_handlers(tty, buf, O_BUFLEN);
4916 }
4917 }
4918
4919 extern void report_error(char* file_name, int line_no, char* title,
4920 char* format, ...);
4921
4922 // this is called _before_ most of the global arguments have been parsed
4923 void os::init(void) {
4924 char dummy; // used to get a guess on initial stack address
4925
4926 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4927
4928 init_random(1234567);
4929
4930 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4931 if (Linux::page_size() == -1) {
4932 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4933 os::strerror(errno));
4934 }
4935 init_page_sizes((size_t) Linux::page_size());
4936
4937 Linux::initialize_system_info();
4938
4939 Linux::initialize_os_info();
4940
4941 // _main_thread points to the thread that created/loaded the JVM.
4942 Linux::_main_thread = pthread_self();
4943
4944 // retrieve entry point for pthread_setname_np
4945 Linux::_pthread_setname_np =
4946 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4947
4948 os::Posix::init();
4949
4950 initial_time_count = javaTimeNanos();
4951
4952 // Always warn if no monotonic clock available
4953 if (!os::Posix::supports_monotonic_clock()) {
4954 warning("No monotonic clock was available - timed services may " \
4955 "be adversely affected if the time-of-day clock changes");
4956 }
4957 }
4958
4959 // To install functions for atexit system call
4960 extern "C" {
4961 static void perfMemory_exit_helper() {
4962 perfMemory_exit();
4963 }
4964 }
4965
4966 void os::pd_init_container_support() {
4967 OSContainer::init();
4968 }
4969
4970 void os::Linux::numa_init() {
4971
4972 // Java can be invoked as
4973 // 1. Without numactl and heap will be allocated/configured on all nodes as
4974 // per the system policy.
4975 // 2. With numactl --interleave:
4976 // Use numa_get_interleave_mask(v2) API to get nodes bitmask. The same
4977 // API for membind case bitmask is reset.
4978 // Interleave is only hint and Kernel can fallback to other nodes if
4979 // no memory is available on the target nodes.
4980 // 3. With numactl --membind:
4981 // Use numa_get_membind(v2) API to get nodes bitmask. The same API for
4982 // interleave case returns bitmask of all nodes.
4983 // numa_all_nodes_ptr holds bitmask of all nodes.
4984 // numa_get_interleave_mask(v2) and numa_get_membind(v2) APIs returns correct
4985 // bitmask when externally configured to run on all or fewer nodes.
4986
4987 if (!Linux::libnuma_init()) {
4988 UseNUMA = false;
4989 } else {
4990 if ((Linux::numa_max_node() < 1) || Linux::is_bound_to_single_node()) {
4991 // If there's only one node (they start from 0) or if the process
4992 // is bound explicitly to a single node using membind, disable NUMA.
4993 UseNUMA = false;
4994 } else {
4995
4996 LogTarget(Info,os) log;
4997 LogStream ls(log);
4998
4999 Linux::set_configured_numa_policy (Linux::identify_numa_policy());
5000
5001 struct bitmask* bmp = Linux::_numa_membind_bitmask;
5002 const char* numa_mode = "membind";
5003
5004 if (Linux::is_running_in_interleave_mode()) {
5005 bmp = Linux::_numa_interleave_bitmask;
5006 numa_mode = "interleave";
5007 }
5008
5009 ls.print("UseNUMA is enabled and invoked in '%s' mode."
5010 " Heap will be configured using NUMA memory nodes:", numa_mode);
5011
5012 for (int node = 0; node < Linux::numa_max_node(); node++) {
5013 if (Linux::_numa_bitmask_isbitset(bmp, node)) {
5014 ls.print(" %d", node);
5015 }
5016 }
5017 }
5018 }
5019
5020 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5021 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5022 // we can make the adaptive lgrp chunk resizing work. If the user specified both
5023 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5024 // and disable adaptive resizing.
5025 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5026 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5027 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5028 UseAdaptiveSizePolicy = false;
5029 UseAdaptiveNUMAChunkSizing = false;
5030 }
5031 }
5032
5033 if (!UseNUMA && ForceNUMA) {
5034 UseNUMA = true;
5035 }
5036 }
5037
5038 // this is called _after_ the global arguments have been parsed
5039 jint os::init_2(void) {
5040
5041 // This could be set after os::Posix::init() but all platforms
5042 // have to set it the same so we have to mirror Solaris.
5043 DEBUG_ONLY(os::set_mutex_init_done();)
5044
5045 os::Posix::init_2();
5046
5047 Linux::fast_thread_clock_init();
5048
5049 // initialize suspend/resume support - must do this before signal_sets_init()
5050 if (SR_initialize() != 0) {
5051 perror("SR_initialize failed");
5052 return JNI_ERR;
5053 }
5054
5055 Linux::signal_sets_init();
5056 Linux::install_signal_handlers();
5057 // Initialize data for jdk.internal.misc.Signal
5058 if (!ReduceSignalUsage) {
5059 jdk_misc_signal_init();
5060 }
5061
5062 // Check and sets minimum stack sizes against command line options
5063 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
5064 return JNI_ERR;
5065 }
5066
5067 suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
5068 if (!suppress_primordial_thread_resolution) {
5069 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5070 }
5071
5072 #if defined(IA32)
5073 workaround_expand_exec_shield_cs_limit();
5074 #endif
5075
5076 Linux::libpthread_init();
5077 Linux::sched_getcpu_init();
5078 log_info(os)("HotSpot is running with %s, %s",
5079 Linux::glibc_version(), Linux::libpthread_version());
5080
5081 if (UseNUMA) {
5082 Linux::numa_init();
5083 }
5084
5085 if (MaxFDLimit) {
5086 // set the number of file descriptors to max. print out error
5087 // if getrlimit/setrlimit fails but continue regardless.
5088 struct rlimit nbr_files;
5089 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5090 if (status != 0) {
5091 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5092 } else {
5093 nbr_files.rlim_cur = nbr_files.rlim_max;
5094 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5095 if (status != 0) {
5096 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5097 }
5098 }
5099 }
5100
5101 // Initialize lock used to serialize thread creation (see os::create_thread)
5102 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5103
5104 // at-exit methods are called in the reverse order of their registration.
5105 // atexit functions are called on return from main or as a result of a
5106 // call to exit(3C). There can be only 32 of these functions registered
5107 // and atexit() does not set errno.
5108
5109 if (PerfAllowAtExitRegistration) {
5110 // only register atexit functions if PerfAllowAtExitRegistration is set.
5111 // atexit functions can be delayed until process exit time, which
5112 // can be problematic for embedded VM situations. Embedded VMs should
5113 // call DestroyJavaVM() to assure that VM resources are released.
5114
5115 // note: perfMemory_exit_helper atexit function may be removed in
5116 // the future if the appropriate cleanup code can be added to the
5117 // VM_Exit VMOperation's doit method.
5118 if (atexit(perfMemory_exit_helper) != 0) {
5119 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5120 }
5121 }
5122
5123 // initialize thread priority policy
5124 prio_init();
5125
5126 if (!FLAG_IS_DEFAULT(AllocateHeapAt)) {
5127 set_coredump_filter(DAX_SHARED_BIT);
5128 }
5129
5130 if (UseSharedSpaces && DumpPrivateMappingsInCore) {
5131 set_coredump_filter(FILE_BACKED_PVT_BIT);
5132 }
5133
5134 return JNI_OK;
5135 }
5136
5137 // Mark the polling page as unreadable
5138 void os::make_polling_page_unreadable(void) {
5139 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5140 fatal("Could not disable polling page");
5141 }
5142 }
5143
5144 // Mark the polling page as readable
5145 void os::make_polling_page_readable(void) {
5146 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5147 fatal("Could not enable polling page");
5148 }
5149 }
5150
5151 // older glibc versions don't have this macro (which expands to
5152 // an optimized bit-counting function) so we have to roll our own
5153 #ifndef CPU_COUNT
5154
5155 static int _cpu_count(const cpu_set_t* cpus) {
5156 int count = 0;
5157 // only look up to the number of configured processors
5158 for (int i = 0; i < os::processor_count(); i++) {
5159 if (CPU_ISSET(i, cpus)) {
5160 count++;
5161 }
5162 }
5163 return count;
5164 }
5165
5166 #define CPU_COUNT(cpus) _cpu_count(cpus)
5167
5168 #endif // CPU_COUNT
5169
5170 // Get the current number of available processors for this process.
5171 // This value can change at any time during a process's lifetime.
5172 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5173 // If it appears there may be more than 1024 processors then we do a
5174 // dynamic check - see 6515172 for details.
5175 // If anything goes wrong we fallback to returning the number of online
5176 // processors - which can be greater than the number available to the process.
5177 int os::Linux::active_processor_count() {
5178 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5179 cpu_set_t* cpus_p = &cpus;
5180 int cpus_size = sizeof(cpu_set_t);
5181
5182 int configured_cpus = os::processor_count(); // upper bound on available cpus
5183 int cpu_count = 0;
5184
5185 // old build platforms may not support dynamic cpu sets
5186 #ifdef CPU_ALLOC
5187
5188 // To enable easy testing of the dynamic path on different platforms we
5189 // introduce a diagnostic flag: UseCpuAllocPath
5190 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5191 // kernel may use a mask bigger than cpu_set_t
5192 log_trace(os)("active_processor_count: using dynamic path %s"
5193 "- configured processors: %d",
5194 UseCpuAllocPath ? "(forced) " : "",
5195 configured_cpus);
5196 cpus_p = CPU_ALLOC(configured_cpus);
5197 if (cpus_p != NULL) {
5198 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5199 // zero it just to be safe
5200 CPU_ZERO_S(cpus_size, cpus_p);
5201 }
5202 else {
5203 // failed to allocate so fallback to online cpus
5204 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5205 log_trace(os)("active_processor_count: "
5206 "CPU_ALLOC failed (%s) - using "
5207 "online processor count: %d",
5208 os::strerror(errno), online_cpus);
5209 return online_cpus;
5210 }
5211 }
5212 else {
5213 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5214 configured_cpus);
5215 }
5216 #else // CPU_ALLOC
5217 // these stubs won't be executed
5218 #define CPU_COUNT_S(size, cpus) -1
5219 #define CPU_FREE(cpus)
5220
5221 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5222 configured_cpus);
5223 #endif // CPU_ALLOC
5224
5225 // pid 0 means the current thread - which we have to assume represents the process
5226 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5227 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5228 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5229 }
5230 else {
5231 cpu_count = CPU_COUNT(cpus_p);
5232 }
5233 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5234 }
5235 else {
5236 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5237 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5238 "which may exceed available processors", os::strerror(errno), cpu_count);
5239 }
5240
5241 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5242 CPU_FREE(cpus_p);
5243 }
5244
5245 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5246 return cpu_count;
5247 }
5248
5249 // Determine the active processor count from one of
5250 // three different sources:
5251 //
5252 // 1. User option -XX:ActiveProcessorCount
5253 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5254 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5255 //
5256 // Option 1, if specified, will always override.
5257 // If the cgroup subsystem is active and configured, we
5258 // will return the min of the cgroup and option 2 results.
5259 // This is required since tools, such as numactl, that
5260 // alter cpu affinity do not update cgroup subsystem
5261 // cpuset configuration files.
5262 int os::active_processor_count() {
5263 // User has overridden the number of active processors
5264 if (ActiveProcessorCount > 0) {
5265 log_trace(os)("active_processor_count: "
5266 "active processor count set by user : %d",
5267 ActiveProcessorCount);
5268 return ActiveProcessorCount;
5269 }
5270
5271 int active_cpus;
5272 if (OSContainer::is_containerized()) {
5273 active_cpus = OSContainer::active_processor_count();
5274 log_trace(os)("active_processor_count: determined by OSContainer: %d",
5275 active_cpus);
5276 } else {
5277 active_cpus = os::Linux::active_processor_count();
5278 }
5279
5280 return active_cpus;
5281 }
5282
5283 uint os::processor_id() {
5284 const int id = Linux::sched_getcpu();
5285 assert(id >= 0 && id < _processor_count, "Invalid processor id");
5286 return (uint)id;
5287 }
5288
5289 void os::set_native_thread_name(const char *name) {
5290 if (Linux::_pthread_setname_np) {
5291 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5292 snprintf(buf, sizeof(buf), "%s", name);
5293 buf[sizeof(buf) - 1] = '\0';
5294 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5295 // ERANGE should not happen; all other errors should just be ignored.
5296 assert(rc != ERANGE, "pthread_setname_np failed");
5297 }
5298 }
5299
5300 bool os::distribute_processes(uint length, uint* distribution) {
5301 // Not yet implemented.
5302 return false;
5303 }
5304
5305 bool os::bind_to_processor(uint processor_id) {
5306 // Not yet implemented.
5307 return false;
5308 }
5309
5310 ///
5311
5312 void os::SuspendedThreadTask::internal_do_task() {
5313 if (do_suspend(_thread->osthread())) {
5314 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5315 do_task(context);
5316 do_resume(_thread->osthread());
5317 }
5318 }
5319
5320 ////////////////////////////////////////////////////////////////////////////////
5321 // debug support
5322
5323 bool os::find(address addr, outputStream* st) {
5324 Dl_info dlinfo;
5325 memset(&dlinfo, 0, sizeof(dlinfo));
5326 if (dladdr(addr, &dlinfo) != 0) {
5327 st->print(PTR_FORMAT ": ", p2i(addr));
5328 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5329 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5330 p2i(addr) - p2i(dlinfo.dli_saddr));
5331 } else if (dlinfo.dli_fbase != NULL) {
5332 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5333 } else {
5334 st->print("<absolute address>");
5335 }
5336 if (dlinfo.dli_fname != NULL) {
5337 st->print(" in %s", dlinfo.dli_fname);
5338 }
5339 if (dlinfo.dli_fbase != NULL) {
5340 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5341 }
5342 st->cr();
5343
5344 if (Verbose) {
5345 // decode some bytes around the PC
5346 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5347 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5348 address lowest = (address) dlinfo.dli_sname;
5349 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5350 if (begin < lowest) begin = lowest;
5351 Dl_info dlinfo2;
5352 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5353 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5354 end = (address) dlinfo2.dli_saddr;
5355 }
5356 Disassembler::decode(begin, end, st);
5357 }
5358 return true;
5359 }
5360 return false;
5361 }
5362
5363 ////////////////////////////////////////////////////////////////////////////////
5364 // misc
5365
5366 // This does not do anything on Linux. This is basically a hook for being
5367 // able to use structured exception handling (thread-local exception filters)
5368 // on, e.g., Win32.
5369 void
5370 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5371 JavaCallArguments* args, Thread* thread) {
5372 f(value, method, args, thread);
5373 }
5374
5375 void os::print_statistics() {
5376 }
5377
5378 bool os::message_box(const char* title, const char* message) {
5379 int i;
5380 fdStream err(defaultStream::error_fd());
5381 for (i = 0; i < 78; i++) err.print_raw("=");
5382 err.cr();
5383 err.print_raw_cr(title);
5384 for (i = 0; i < 78; i++) err.print_raw("-");
5385 err.cr();
5386 err.print_raw_cr(message);
5387 for (i = 0; i < 78; i++) err.print_raw("=");
5388 err.cr();
5389
5390 char buf[16];
5391 // Prevent process from exiting upon "read error" without consuming all CPU
5392 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5393
5394 return buf[0] == 'y' || buf[0] == 'Y';
5395 }
5396
5397 // Is a (classpath) directory empty?
5398 bool os::dir_is_empty(const char* path) {
5399 DIR *dir = NULL;
5400 struct dirent *ptr;
5401
5402 dir = opendir(path);
5403 if (dir == NULL) return true;
5404
5405 // Scan the directory
5406 bool result = true;
5407 while (result && (ptr = readdir(dir)) != NULL) {
5408 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5409 result = false;
5410 }
5411 }
5412 closedir(dir);
5413 return result;
5414 }
5415
5416 // This code originates from JDK's sysOpen and open64_w
5417 // from src/solaris/hpi/src/system_md.c
5418
5419 int os::open(const char *path, int oflag, int mode) {
5420 if (strlen(path) > MAX_PATH - 1) {
5421 errno = ENAMETOOLONG;
5422 return -1;
5423 }
5424
5425 // All file descriptors that are opened in the Java process and not
5426 // specifically destined for a subprocess should have the close-on-exec
5427 // flag set. If we don't set it, then careless 3rd party native code
5428 // might fork and exec without closing all appropriate file descriptors
5429 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5430 // turn might:
5431 //
5432 // - cause end-of-file to fail to be detected on some file
5433 // descriptors, resulting in mysterious hangs, or
5434 //
5435 // - might cause an fopen in the subprocess to fail on a system
5436 // suffering from bug 1085341.
5437 //
5438 // (Yes, the default setting of the close-on-exec flag is a Unix
5439 // design flaw)
5440 //
5441 // See:
5442 // 1085341: 32-bit stdio routines should support file descriptors >255
5443 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5444 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5445 //
5446 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5447 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5448 // because it saves a system call and removes a small window where the flag
5449 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5450 // and we fall back to using FD_CLOEXEC (see below).
5451 #ifdef O_CLOEXEC
5452 oflag |= O_CLOEXEC;
5453 #endif
5454
5455 int fd = ::open64(path, oflag, mode);
5456 if (fd == -1) return -1;
5457
5458 //If the open succeeded, the file might still be a directory
5459 {
5460 struct stat64 buf64;
5461 int ret = ::fstat64(fd, &buf64);
5462 int st_mode = buf64.st_mode;
5463
5464 if (ret != -1) {
5465 if ((st_mode & S_IFMT) == S_IFDIR) {
5466 errno = EISDIR;
5467 ::close(fd);
5468 return -1;
5469 }
5470 } else {
5471 ::close(fd);
5472 return -1;
5473 }
5474 }
5475
5476 #ifdef FD_CLOEXEC
5477 // Validate that the use of the O_CLOEXEC flag on open above worked.
5478 // With recent kernels, we will perform this check exactly once.
5479 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5480 if (!O_CLOEXEC_is_known_to_work) {
5481 int flags = ::fcntl(fd, F_GETFD);
5482 if (flags != -1) {
5483 if ((flags & FD_CLOEXEC) != 0)
5484 O_CLOEXEC_is_known_to_work = 1;
5485 else
5486 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5487 }
5488 }
5489 #endif
5490
5491 return fd;
5492 }
5493
5494
5495 // create binary file, rewriting existing file if required
5496 int os::create_binary_file(const char* path, bool rewrite_existing) {
5497 int oflags = O_WRONLY | O_CREAT;
5498 if (!rewrite_existing) {
5499 oflags |= O_EXCL;
5500 }
5501 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5502 }
5503
5504 // return current position of file pointer
5505 jlong os::current_file_offset(int fd) {
5506 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5507 }
5508
5509 // move file pointer to the specified offset
5510 jlong os::seek_to_file_offset(int fd, jlong offset) {
5511 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5512 }
5513
5514 // This code originates from JDK's sysAvailable
5515 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5516
5517 int os::available(int fd, jlong *bytes) {
5518 jlong cur, end;
5519 int mode;
5520 struct stat64 buf64;
5521
5522 if (::fstat64(fd, &buf64) >= 0) {
5523 mode = buf64.st_mode;
5524 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5525 int n;
5526 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5527 *bytes = n;
5528 return 1;
5529 }
5530 }
5531 }
5532 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5533 return 0;
5534 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5535 return 0;
5536 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5537 return 0;
5538 }
5539 *bytes = end - cur;
5540 return 1;
5541 }
5542
5543 // Map a block of memory.
5544 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5545 char *addr, size_t bytes, bool read_only,
5546 bool allow_exec) {
5547 int prot;
5548 int flags = MAP_PRIVATE;
5549
5550 if (read_only) {
5551 prot = PROT_READ;
5552 } else {
5553 prot = PROT_READ | PROT_WRITE;
5554 }
5555
5556 if (allow_exec) {
5557 prot |= PROT_EXEC;
5558 }
5559
5560 if (addr != NULL) {
5561 flags |= MAP_FIXED;
5562 }
5563
5564 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5565 fd, file_offset);
5566 if (mapped_address == MAP_FAILED) {
5567 return NULL;
5568 }
5569 return mapped_address;
5570 }
5571
5572
5573 // Remap a block of memory.
5574 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5575 char *addr, size_t bytes, bool read_only,
5576 bool allow_exec) {
5577 // same as map_memory() on this OS
5578 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5579 allow_exec);
5580 }
5581
5582
5583 // Unmap a block of memory.
5584 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5585 return munmap(addr, bytes) == 0;
5586 }
5587
5588 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5589
5590 static jlong fast_cpu_time(Thread *thread) {
5591 clockid_t clockid;
5592 int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(),
5593 &clockid);
5594 if (rc == 0) {
5595 return os::Linux::fast_thread_cpu_time(clockid);
5596 } else {
5597 // It's possible to encounter a terminated native thread that failed
5598 // to detach itself from the VM - which should result in ESRCH.
5599 assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed");
5600 return -1;
5601 }
5602 }
5603
5604 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5605 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5606 // of a thread.
5607 //
5608 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5609 // the fast estimate available on the platform.
5610
5611 jlong os::current_thread_cpu_time() {
5612 if (os::Linux::supports_fast_thread_cpu_time()) {
5613 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5614 } else {
5615 // return user + sys since the cost is the same
5616 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5617 }
5618 }
5619
5620 jlong os::thread_cpu_time(Thread* thread) {
5621 // consistent with what current_thread_cpu_time() returns
5622 if (os::Linux::supports_fast_thread_cpu_time()) {
5623 return fast_cpu_time(thread);
5624 } else {
5625 return slow_thread_cpu_time(thread, true /* user + sys */);
5626 }
5627 }
5628
5629 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5630 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5631 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5632 } else {
5633 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5634 }
5635 }
5636
5637 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5638 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5639 return fast_cpu_time(thread);
5640 } else {
5641 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5642 }
5643 }
5644
5645 // -1 on error.
5646 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5647 pid_t tid = thread->osthread()->thread_id();
5648 char *s;
5649 char stat[2048];
5650 int statlen;
5651 char proc_name[64];
5652 int count;
5653 long sys_time, user_time;
5654 char cdummy;
5655 int idummy;
5656 long ldummy;
5657 FILE *fp;
5658
5659 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5660 fp = fopen(proc_name, "r");
5661 if (fp == NULL) return -1;
5662 statlen = fread(stat, 1, 2047, fp);
5663 stat[statlen] = '\0';
5664 fclose(fp);
5665
5666 // Skip pid and the command string. Note that we could be dealing with
5667 // weird command names, e.g. user could decide to rename java launcher
5668 // to "java 1.4.2 :)", then the stat file would look like
5669 // 1234 (java 1.4.2 :)) R ... ...
5670 // We don't really need to know the command string, just find the last
5671 // occurrence of ")" and then start parsing from there. See bug 4726580.
5672 s = strrchr(stat, ')');
5673 if (s == NULL) return -1;
5674
5675 // Skip blank chars
5676 do { s++; } while (s && isspace(*s));
5677
5678 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5679 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5680 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5681 &user_time, &sys_time);
5682 if (count != 13) return -1;
5683 if (user_sys_cpu_time) {
5684 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5685 } else {
5686 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5687 }
5688 }
5689
5690 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5691 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5692 info_ptr->may_skip_backward = false; // elapsed time not wall time
5693 info_ptr->may_skip_forward = false; // elapsed time not wall time
5694 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5695 }
5696
5697 void os::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 bool os::is_thread_cpu_time_supported() {
5705 return true;
5706 }
5707
5708 // System loadavg support. Returns -1 if load average cannot be obtained.
5709 // Linux doesn't yet have a (official) notion of processor sets,
5710 // so just return the system wide load average.
5711 int os::loadavg(double loadavg[], int nelem) {
5712 return ::getloadavg(loadavg, nelem);
5713 }
5714
5715 void os::pause() {
5716 char filename[MAX_PATH];
5717 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5718 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5719 } else {
5720 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5721 }
5722
5723 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5724 if (fd != -1) {
5725 struct stat buf;
5726 ::close(fd);
5727 while (::stat(filename, &buf) == 0) {
5728 (void)::poll(NULL, 0, 100);
5729 }
5730 } else {
5731 jio_fprintf(stderr,
5732 "Could not open pause file '%s', continuing immediately.\n", filename);
5733 }
5734 }
5735
5736 extern char** environ;
5737
5738 // Run the specified command in a separate process. Return its exit value,
5739 // or -1 on failure (e.g. can't fork a new process).
5740 // Unlike system(), this function can be called from signal handler. It
5741 // doesn't block SIGINT et al.
5742 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) {
5743 const char * argv[4] = {"sh", "-c", cmd, NULL};
5744
5745 pid_t pid ;
5746
5747 if (use_vfork_if_available) {
5748 pid = vfork();
5749 } else {
5750 pid = fork();
5751 }
5752
5753 if (pid < 0) {
5754 // fork failed
5755 return -1;
5756
5757 } else if (pid == 0) {
5758 // child process
5759
5760 execve("/bin/sh", (char* const*)argv, environ);
5761
5762 // execve failed
5763 _exit(-1);
5764
5765 } else {
5766 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5767 // care about the actual exit code, for now.
5768
5769 int status;
5770
5771 // Wait for the child process to exit. This returns immediately if
5772 // the child has already exited. */
5773 while (waitpid(pid, &status, 0) < 0) {
5774 switch (errno) {
5775 case ECHILD: return 0;
5776 case EINTR: break;
5777 default: return -1;
5778 }
5779 }
5780
5781 if (WIFEXITED(status)) {
5782 // The child exited normally; get its exit code.
5783 return WEXITSTATUS(status);
5784 } else if (WIFSIGNALED(status)) {
5785 // The child exited because of a signal
5786 // The best value to return is 0x80 + signal number,
5787 // because that is what all Unix shells do, and because
5788 // it allows callers to distinguish between process exit and
5789 // process death by signal.
5790 return 0x80 + WTERMSIG(status);
5791 } else {
5792 // Unknown exit code; pass it through
5793 return status;
5794 }
5795 }
5796 }
5797
5798 // Get the default path to the core file
5799 // Returns the length of the string
5800 int os::get_core_path(char* buffer, size_t bufferSize) {
5801 /*
5802 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5803 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5804 */
5805 const int core_pattern_len = 129;
5806 char core_pattern[core_pattern_len] = {0};
5807
5808 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5809 if (core_pattern_file == -1) {
5810 return -1;
5811 }
5812
5813 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5814 ::close(core_pattern_file);
5815 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5816 return -1;
5817 }
5818 if (core_pattern[ret-1] == '\n') {
5819 core_pattern[ret-1] = '\0';
5820 } else {
5821 core_pattern[ret] = '\0';
5822 }
5823
5824 // Replace the %p in the core pattern with the process id. NOTE: we do this
5825 // only if the pattern doesn't start with "|", and we support only one %p in
5826 // the pattern.
5827 char *pid_pos = strstr(core_pattern, "%p");
5828 const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : ""; // skip over the "%p"
5829 int written;
5830
5831 if (core_pattern[0] == '/') {
5832 if (pid_pos != NULL) {
5833 *pid_pos = '\0';
5834 written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern,
5835 current_process_id(), tail);
5836 } else {
5837 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5838 }
5839 } else {
5840 char cwd[PATH_MAX];
5841
5842 const char* p = get_current_directory(cwd, PATH_MAX);
5843 if (p == NULL) {
5844 return -1;
5845 }
5846
5847 if (core_pattern[0] == '|') {
5848 written = jio_snprintf(buffer, bufferSize,
5849 "\"%s\" (or dumping to %s/core.%d)",
5850 &core_pattern[1], p, current_process_id());
5851 } else if (pid_pos != NULL) {
5852 *pid_pos = '\0';
5853 written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern,
5854 current_process_id(), tail);
5855 } else {
5856 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5857 }
5858 }
5859
5860 if (written < 0) {
5861 return -1;
5862 }
5863
5864 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5865 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5866
5867 if (core_uses_pid_file != -1) {
5868 char core_uses_pid = 0;
5869 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5870 ::close(core_uses_pid_file);
5871
5872 if (core_uses_pid == '1') {
5873 jio_snprintf(buffer + written, bufferSize - written,
5874 ".%d", current_process_id());
5875 }
5876 }
5877 }
5878
5879 return strlen(buffer);
5880 }
5881
5882 bool os::start_debugging(char *buf, int buflen) {
5883 int len = (int)strlen(buf);
5884 char *p = &buf[len];
5885
5886 jio_snprintf(p, buflen-len,
5887 "\n\n"
5888 "Do you want to debug the problem?\n\n"
5889 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5890 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5891 "Otherwise, press RETURN to abort...",
5892 os::current_process_id(), os::current_process_id(),
5893 os::current_thread_id(), os::current_thread_id());
5894
5895 bool yes = os::message_box("Unexpected Error", buf);
5896
5897 if (yes) {
5898 // yes, user asked VM to launch debugger
5899 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5900 os::current_process_id(), os::current_process_id());
5901
5902 os::fork_and_exec(buf);
5903 yes = false;
5904 }
5905 return yes;
5906 }
5907
5908
5909 // Java/Compiler thread:
5910 //
5911 // Low memory addresses
5912 // P0 +------------------------+
5913 // | |\ Java thread created by VM does not have glibc
5914 // | glibc guard page | - guard page, attached Java thread usually has
5915 // | |/ 1 glibc guard page.
5916 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5917 // | |\
5918 // | HotSpot Guard Pages | - red, yellow and reserved pages
5919 // | |/
5920 // +------------------------+ JavaThread::stack_reserved_zone_base()
5921 // | |\
5922 // | Normal Stack | -
5923 // | |/
5924 // P2 +------------------------+ Thread::stack_base()
5925 //
5926 // Non-Java thread:
5927 //
5928 // Low memory addresses
5929 // P0 +------------------------+
5930 // | |\
5931 // | glibc guard page | - usually 1 page
5932 // | |/
5933 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5934 // | |\
5935 // | Normal Stack | -
5936 // | |/
5937 // P2 +------------------------+ Thread::stack_base()
5938 //
5939 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
5940 // returned from pthread_attr_getstack().
5941 // ** Due to NPTL implementation error, linux takes the glibc guard page out
5942 // of the stack size given in pthread_attr. We work around this for
5943 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
5944 //
5945 #ifndef ZERO
5946 static void current_stack_region(address * bottom, size_t * size) {
5947 if (os::is_primordial_thread()) {
5948 // primordial thread needs special handling because pthread_getattr_np()
5949 // may return bogus value.
5950 *bottom = os::Linux::initial_thread_stack_bottom();
5951 *size = os::Linux::initial_thread_stack_size();
5952 } else {
5953 pthread_attr_t attr;
5954
5955 int rslt = pthread_getattr_np(pthread_self(), &attr);
5956
5957 // JVM needs to know exact stack location, abort if it fails
5958 if (rslt != 0) {
5959 if (rslt == ENOMEM) {
5960 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
5961 } else {
5962 fatal("pthread_getattr_np failed with error = %d", rslt);
5963 }
5964 }
5965
5966 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
5967 fatal("Cannot locate current stack attributes!");
5968 }
5969
5970 // Work around NPTL stack guard error.
5971 size_t guard_size = 0;
5972 rslt = pthread_attr_getguardsize(&attr, &guard_size);
5973 if (rslt != 0) {
5974 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
5975 }
5976 *bottom += guard_size;
5977 *size -= guard_size;
5978
5979 pthread_attr_destroy(&attr);
5980
5981 }
5982 assert(os::current_stack_pointer() >= *bottom &&
5983 os::current_stack_pointer() < *bottom + *size, "just checking");
5984 }
5985
5986 address os::current_stack_base() {
5987 address bottom;
5988 size_t size;
5989 current_stack_region(&bottom, &size);
5990 return (bottom + size);
5991 }
5992
5993 size_t os::current_stack_size() {
5994 // This stack size includes the usable stack and HotSpot guard pages
5995 // (for the threads that have Hotspot guard pages).
5996 address bottom;
5997 size_t size;
5998 current_stack_region(&bottom, &size);
5999 return size;
6000 }
6001 #endif
6002
6003 static inline struct timespec get_mtime(const char* filename) {
6004 struct stat st;
6005 int ret = os::stat(filename, &st);
6006 assert(ret == 0, "failed to stat() file '%s': %s", filename, os::strerror(errno));
6007 return st.st_mtim;
6008 }
6009
6010 int os::compare_file_modified_times(const char* file1, const char* file2) {
6011 struct timespec filetime1 = get_mtime(file1);
6012 struct timespec filetime2 = get_mtime(file2);
6013 int diff = filetime1.tv_sec - filetime2.tv_sec;
6014 if (diff == 0) {
6015 return filetime1.tv_nsec - filetime2.tv_nsec;
6016 }
6017 return diff;
6018 }
6019
6020 /////////////// Unit tests ///////////////
6021
6022 #ifndef PRODUCT
6023
6024 class TestReserveMemorySpecial : AllStatic {
6025 public:
6026 static void small_page_write(void* addr, size_t size) {
6027 size_t page_size = os::vm_page_size();
6028
6029 char* end = (char*)addr + size;
6030 for (char* p = (char*)addr; p < end; p += page_size) {
6031 *p = 1;
6032 }
6033 }
6034
6035 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6036 if (!UseHugeTLBFS) {
6037 return;
6038 }
6039
6040 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6041
6042 if (addr != NULL) {
6043 small_page_write(addr, size);
6044
6045 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6046 }
6047 }
6048
6049 static void test_reserve_memory_special_huge_tlbfs_only() {
6050 if (!UseHugeTLBFS) {
6051 return;
6052 }
6053
6054 size_t lp = os::large_page_size();
6055
6056 for (size_t size = lp; size <= lp * 10; size += lp) {
6057 test_reserve_memory_special_huge_tlbfs_only(size);
6058 }
6059 }
6060
6061 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6062 size_t lp = os::large_page_size();
6063 size_t ag = os::vm_allocation_granularity();
6064
6065 // sizes to test
6066 const size_t sizes[] = {
6067 lp, lp + ag, lp + lp / 2, lp * 2,
6068 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6069 lp * 10, lp * 10 + lp / 2
6070 };
6071 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6072
6073 // For each size/alignment combination, we test three scenarios:
6074 // 1) with req_addr == NULL
6075 // 2) with a non-null req_addr at which we expect to successfully allocate
6076 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6077 // expect the allocation to either fail or to ignore req_addr
6078
6079 // Pre-allocate two areas; they shall be as large as the largest allocation
6080 // and aligned to the largest alignment we will be testing.
6081 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6082 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6083 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6084 -1, 0);
6085 assert(mapping1 != MAP_FAILED, "should work");
6086
6087 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6088 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6089 -1, 0);
6090 assert(mapping2 != MAP_FAILED, "should work");
6091
6092 // Unmap the first mapping, but leave the second mapping intact: the first
6093 // mapping will serve as a value for a "good" req_addr (case 2). The second
6094 // mapping, still intact, as "bad" req_addr (case 3).
6095 ::munmap(mapping1, mapping_size);
6096
6097 // Case 1
6098 for (int i = 0; i < num_sizes; i++) {
6099 const size_t size = sizes[i];
6100 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6101 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6102 if (p != NULL) {
6103 assert(is_aligned(p, alignment), "must be");
6104 small_page_write(p, size);
6105 os::Linux::release_memory_special_huge_tlbfs(p, size);
6106 }
6107 }
6108 }
6109
6110 // Case 2
6111 for (int i = 0; i < num_sizes; i++) {
6112 const size_t size = sizes[i];
6113 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6114 char* const req_addr = align_up(mapping1, alignment);
6115 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6116 if (p != NULL) {
6117 assert(p == req_addr, "must be");
6118 small_page_write(p, size);
6119 os::Linux::release_memory_special_huge_tlbfs(p, size);
6120 }
6121 }
6122 }
6123
6124 // Case 3
6125 for (int i = 0; i < num_sizes; i++) {
6126 const size_t size = sizes[i];
6127 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6128 char* const req_addr = align_up(mapping2, alignment);
6129 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6130 // as the area around req_addr contains already existing mappings, the API should always
6131 // return NULL (as per contract, it cannot return another address)
6132 assert(p == NULL, "must be");
6133 }
6134 }
6135
6136 ::munmap(mapping2, mapping_size);
6137
6138 }
6139
6140 static void test_reserve_memory_special_huge_tlbfs() {
6141 if (!UseHugeTLBFS) {
6142 return;
6143 }
6144
6145 test_reserve_memory_special_huge_tlbfs_only();
6146 test_reserve_memory_special_huge_tlbfs_mixed();
6147 }
6148
6149 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6150 if (!UseSHM) {
6151 return;
6152 }
6153
6154 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6155
6156 if (addr != NULL) {
6157 assert(is_aligned(addr, alignment), "Check");
6158 assert(is_aligned(addr, os::large_page_size()), "Check");
6159
6160 small_page_write(addr, size);
6161
6162 os::Linux::release_memory_special_shm(addr, size);
6163 }
6164 }
6165
6166 static void test_reserve_memory_special_shm() {
6167 size_t lp = os::large_page_size();
6168 size_t ag = os::vm_allocation_granularity();
6169
6170 for (size_t size = ag; size < lp * 3; size += ag) {
6171 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6172 test_reserve_memory_special_shm(size, alignment);
6173 }
6174 }
6175 }
6176
6177 static void test() {
6178 test_reserve_memory_special_huge_tlbfs();
6179 test_reserve_memory_special_shm();
6180 }
6181 };
6182
6183 void TestReserveMemorySpecial_test() {
6184 TestReserveMemorySpecial::test();
6185 }
6186
6187 #endif