1 /*
2 * Copyright (c) 1999, 2018, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 // no precompiled headers
26 #include "jvm.h"
27 #include "classfile/classLoader.hpp"
28 #include "classfile/systemDictionary.hpp"
29 #include "classfile/vmSymbols.hpp"
30 #include "code/icBuffer.hpp"
31 #include "code/vtableStubs.hpp"
32 #include "compiler/compileBroker.hpp"
33 #include "compiler/disassembler.hpp"
34 #include "interpreter/interpreter.hpp"
35 #include "logging/log.hpp"
36 #include "memory/allocation.inline.hpp"
37 #include "memory/filemap.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "os_linux.inline.hpp"
40 #include "os_share_linux.hpp"
41 #include "osContainer_linux.hpp"
42 #include "prims/jniFastGetField.hpp"
43 #include "prims/jvm_misc.hpp"
44 #include "runtime/arguments.hpp"
45 #include "runtime/atomic.hpp"
46 #include "runtime/extendedPC.hpp"
47 #include "runtime/globals.hpp"
48 #include "runtime/interfaceSupport.hpp"
49 #include "runtime/init.hpp"
50 #include "runtime/java.hpp"
51 #include "runtime/javaCalls.hpp"
52 #include "runtime/mutexLocker.hpp"
53 #include "runtime/objectMonitor.hpp"
54 #include "runtime/orderAccess.inline.hpp"
55 #include "runtime/osThread.hpp"
56 #include "runtime/perfMemory.hpp"
57 #include "runtime/sharedRuntime.hpp"
58 #include "runtime/statSampler.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "runtime/thread.inline.hpp"
61 #include "runtime/threadCritical.hpp"
62 #include "runtime/threadSMR.hpp"
63 #include "runtime/timer.hpp"
64 #include "semaphore_posix.hpp"
65 #include "services/attachListener.hpp"
66 #include "services/memTracker.hpp"
67 #include "services/runtimeService.hpp"
68 #include "utilities/align.hpp"
69 #include "utilities/decoder.hpp"
70 #include "utilities/defaultStream.hpp"
71 #include "utilities/events.hpp"
72 #include "utilities/elfFile.hpp"
73 #include "utilities/growableArray.hpp"
74 #include "utilities/macros.hpp"
75 #include "utilities/vmError.hpp"
76
77 // put OS-includes here
78 # include <sys/types.h>
79 # include <sys/mman.h>
80 # include <sys/stat.h>
81 # include <sys/select.h>
82 # include <pthread.h>
83 # include <signal.h>
84 # include <errno.h>
85 # include <dlfcn.h>
86 # include <stdio.h>
87 # include <unistd.h>
88 # include <sys/resource.h>
89 # include <pthread.h>
90 # include <sys/stat.h>
91 # include <sys/time.h>
92 # include <sys/times.h>
93 # include <sys/utsname.h>
94 # include <sys/socket.h>
95 # include <sys/wait.h>
96 # include <pwd.h>
97 # include <poll.h>
98 # include <fcntl.h>
99 # include <string.h>
100 # include <syscall.h>
101 # include <sys/sysinfo.h>
102 # include <gnu/libc-version.h>
103 # include <sys/ipc.h>
104 # include <sys/shm.h>
105 # include <link.h>
106 # include <stdint.h>
107 # include <inttypes.h>
108 # include <sys/ioctl.h>
109
110 #ifndef _GNU_SOURCE
111 #define _GNU_SOURCE
112 #include <sched.h>
113 #undef _GNU_SOURCE
114 #else
115 #include <sched.h>
116 #endif
117
118 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
119 // getrusage() is prepared to handle the associated failure.
120 #ifndef RUSAGE_THREAD
121 #define RUSAGE_THREAD (1) /* only the calling thread */
122 #endif
123
124 #define MAX_PATH (2 * K)
125
126 #define MAX_SECS 100000000
127
128 // for timer info max values which include all bits
129 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
130
131 #define LARGEPAGES_BIT (1 << 6)
132 #define DAX_SHARED_BIT (1 << 8)
133 ////////////////////////////////////////////////////////////////////////////////
134 // global variables
135 julong os::Linux::_physical_memory = 0;
136
137 address os::Linux::_initial_thread_stack_bottom = NULL;
138 uintptr_t os::Linux::_initial_thread_stack_size = 0;
139
140 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
141 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
142 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
143 Mutex* os::Linux::_createThread_lock = NULL;
144 pthread_t os::Linux::_main_thread;
145 int os::Linux::_page_size = -1;
146 bool os::Linux::_supports_fast_thread_cpu_time = false;
147 uint32_t os::Linux::_os_version = 0;
148 const char * os::Linux::_glibc_version = NULL;
149 const char * os::Linux::_libpthread_version = NULL;
150
151 static jlong initial_time_count=0;
152
153 static int clock_tics_per_sec = 100;
154
155 // For diagnostics to print a message once. see run_periodic_checks
156 static sigset_t check_signal_done;
157 static bool check_signals = true;
158
159 // Signal number used to suspend/resume a thread
160
161 // do not use any signal number less than SIGSEGV, see 4355769
162 static int SR_signum = SIGUSR2;
163 sigset_t SR_sigset;
164
165 // utility functions
166
167 static int SR_initialize();
168
169 julong os::available_memory() {
170 return Linux::available_memory();
171 }
172
173 julong os::Linux::available_memory() {
174 // values in struct sysinfo are "unsigned long"
175 struct sysinfo si;
176 julong avail_mem;
177
178 if (OSContainer::is_containerized()) {
179 jlong mem_limit, mem_usage;
180 if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
181 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
182 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
183 }
184 if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
185 log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
186 }
187 if (mem_limit > 0 && mem_usage > 0 ) {
188 avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
189 log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
190 return avail_mem;
191 }
192 }
193
194 sysinfo(&si);
195 avail_mem = (julong)si.freeram * si.mem_unit;
196 log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
197 return avail_mem;
198 }
199
200 julong os::physical_memory() {
201 jlong phys_mem = 0;
202 if (OSContainer::is_containerized()) {
203 jlong mem_limit;
204 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
205 log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
206 return mem_limit;
207 }
208 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
209 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
210 }
211
212 phys_mem = Linux::physical_memory();
213 log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
214 return phys_mem;
215 }
216
217 // Return true if user is running as root.
218
219 bool os::have_special_privileges() {
220 static bool init = false;
221 static bool privileges = false;
222 if (!init) {
223 privileges = (getuid() != geteuid()) || (getgid() != getegid());
224 init = true;
225 }
226 return privileges;
227 }
228
229
230 #ifndef SYS_gettid
231 // i386: 224, ia64: 1105, amd64: 186, sparc 143
232 #ifdef __ia64__
233 #define SYS_gettid 1105
234 #else
235 #ifdef __i386__
236 #define SYS_gettid 224
237 #else
238 #ifdef __amd64__
239 #define SYS_gettid 186
240 #else
241 #ifdef __sparc__
242 #define SYS_gettid 143
243 #else
244 #error define gettid for the arch
245 #endif
246 #endif
247 #endif
248 #endif
249 #endif
250
251
252 // pid_t gettid()
253 //
254 // Returns the kernel thread id of the currently running thread. Kernel
255 // thread id is used to access /proc.
256 pid_t os::Linux::gettid() {
257 int rslt = syscall(SYS_gettid);
258 assert(rslt != -1, "must be."); // old linuxthreads implementation?
259 return (pid_t)rslt;
260 }
261
262 // Most versions of linux have a bug where the number of processors are
263 // determined by looking at the /proc file system. In a chroot environment,
264 // the system call returns 1. This causes the VM to act as if it is
265 // a single processor and elide locking (see is_MP() call).
266 static bool unsafe_chroot_detected = false;
267 static const char *unstable_chroot_error = "/proc file system not found.\n"
268 "Java may be unstable running multithreaded in a chroot "
269 "environment on Linux when /proc filesystem is not mounted.";
270
271 void os::Linux::initialize_system_info() {
272 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
273 if (processor_count() == 1) {
274 pid_t pid = os::Linux::gettid();
275 char fname[32];
276 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
277 FILE *fp = fopen(fname, "r");
278 if (fp == NULL) {
279 unsafe_chroot_detected = true;
280 } else {
281 fclose(fp);
282 }
283 }
284 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
285 assert(processor_count() > 0, "linux error");
286 }
287
288 void os::init_system_properties_values() {
289 // The next steps are taken in the product version:
290 //
291 // Obtain the JAVA_HOME value from the location of libjvm.so.
292 // This library should be located at:
293 // <JAVA_HOME>/lib/{client|server}/libjvm.so.
294 //
295 // If "/jre/lib/" appears at the right place in the path, then we
296 // assume libjvm.so is installed in a JDK and we use this path.
297 //
298 // Otherwise exit with message: "Could not create the Java virtual machine."
299 //
300 // The following extra steps are taken in the debugging version:
301 //
302 // If "/jre/lib/" does NOT appear at the right place in the path
303 // instead of exit check for $JAVA_HOME environment variable.
304 //
305 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
306 // then we append a fake suffix "hotspot/libjvm.so" to this path so
307 // it looks like libjvm.so is installed there
308 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
309 //
310 // Otherwise exit.
311 //
312 // Important note: if the location of libjvm.so changes this
313 // code needs to be changed accordingly.
314
315 // See ld(1):
316 // The linker uses the following search paths to locate required
317 // shared libraries:
318 // 1: ...
319 // ...
320 // 7: The default directories, normally /lib and /usr/lib.
321 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
322 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
323 #else
324 #define DEFAULT_LIBPATH "/lib:/usr/lib"
325 #endif
326
327 // Base path of extensions installed on the system.
328 #define SYS_EXT_DIR "/usr/java/packages"
329 #define EXTENSIONS_DIR "/lib/ext"
330
331 // Buffer that fits several sprintfs.
332 // Note that the space for the colon and the trailing null are provided
333 // by the nulls included by the sizeof operator.
334 const size_t bufsize =
335 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
336 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
337 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
338
339 // sysclasspath, java_home, dll_dir
340 {
341 char *pslash;
342 os::jvm_path(buf, bufsize);
343
344 // Found the full path to libjvm.so.
345 // Now cut the path to <java_home>/jre if we can.
346 pslash = strrchr(buf, '/');
347 if (pslash != NULL) {
348 *pslash = '\0'; // Get rid of /libjvm.so.
349 }
350 pslash = strrchr(buf, '/');
351 if (pslash != NULL) {
352 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
353 }
354 Arguments::set_dll_dir(buf);
355
356 if (pslash != NULL) {
357 pslash = strrchr(buf, '/');
358 if (pslash != NULL) {
359 *pslash = '\0'; // Get rid of /lib.
360 }
361 }
362 Arguments::set_java_home(buf);
363 set_boot_path('/', ':');
364 }
365
366 // Where to look for native libraries.
367 //
368 // Note: Due to a legacy implementation, most of the library path
369 // is set in the launcher. This was to accomodate linking restrictions
370 // on legacy Linux implementations (which are no longer supported).
371 // Eventually, all the library path setting will be done here.
372 //
373 // However, to prevent the proliferation of improperly built native
374 // libraries, the new path component /usr/java/packages is added here.
375 // Eventually, all the library path setting will be done here.
376 {
377 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
378 // should always exist (until the legacy problem cited above is
379 // addressed).
380 const char *v = ::getenv("LD_LIBRARY_PATH");
381 const char *v_colon = ":";
382 if (v == NULL) { v = ""; v_colon = ""; }
383 // That's +1 for the colon and +1 for the trailing '\0'.
384 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
385 strlen(v) + 1 +
386 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
387 mtInternal);
388 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
389 Arguments::set_library_path(ld_library_path);
390 FREE_C_HEAP_ARRAY(char, ld_library_path);
391 }
392
393 // Extensions directories.
394 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
395 Arguments::set_ext_dirs(buf);
396
397 FREE_C_HEAP_ARRAY(char, buf);
398
399 #undef DEFAULT_LIBPATH
400 #undef SYS_EXT_DIR
401 #undef EXTENSIONS_DIR
402 }
403
404 ////////////////////////////////////////////////////////////////////////////////
405 // breakpoint support
406
407 void os::breakpoint() {
408 BREAKPOINT;
409 }
410
411 extern "C" void breakpoint() {
412 // use debugger to set breakpoint here
413 }
414
415 ////////////////////////////////////////////////////////////////////////////////
416 // signal support
417
418 debug_only(static bool signal_sets_initialized = false);
419 static sigset_t unblocked_sigs, vm_sigs;
420
421 bool os::Linux::is_sig_ignored(int sig) {
422 struct sigaction oact;
423 sigaction(sig, (struct sigaction*)NULL, &oact);
424 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
425 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
426 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
427 return true;
428 } else {
429 return false;
430 }
431 }
432
433 void os::Linux::signal_sets_init() {
434 // Should also have an assertion stating we are still single-threaded.
435 assert(!signal_sets_initialized, "Already initialized");
436 // Fill in signals that are necessarily unblocked for all threads in
437 // the VM. Currently, we unblock the following signals:
438 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
439 // by -Xrs (=ReduceSignalUsage));
440 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
441 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
442 // the dispositions or masks wrt these signals.
443 // Programs embedding the VM that want to use the above signals for their
444 // own purposes must, at this time, use the "-Xrs" option to prevent
445 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
446 // (See bug 4345157, and other related bugs).
447 // In reality, though, unblocking these signals is really a nop, since
448 // these signals are not blocked by default.
449 sigemptyset(&unblocked_sigs);
450 sigaddset(&unblocked_sigs, SIGILL);
451 sigaddset(&unblocked_sigs, SIGSEGV);
452 sigaddset(&unblocked_sigs, SIGBUS);
453 sigaddset(&unblocked_sigs, SIGFPE);
454 #if defined(PPC64)
455 sigaddset(&unblocked_sigs, SIGTRAP);
456 #endif
457 sigaddset(&unblocked_sigs, SR_signum);
458
459 if (!ReduceSignalUsage) {
460 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
461 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
462 }
463 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
464 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
465 }
466 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
467 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
468 }
469 }
470 // Fill in signals that are blocked by all but the VM thread.
471 sigemptyset(&vm_sigs);
472 if (!ReduceSignalUsage) {
473 sigaddset(&vm_sigs, BREAK_SIGNAL);
474 }
475 debug_only(signal_sets_initialized = true);
476
477 }
478
479 // These are signals that are unblocked while a thread is running Java.
480 // (For some reason, they get blocked by default.)
481 sigset_t* os::Linux::unblocked_signals() {
482 assert(signal_sets_initialized, "Not initialized");
483 return &unblocked_sigs;
484 }
485
486 // These are the signals that are blocked while a (non-VM) thread is
487 // running Java. Only the VM thread handles these signals.
488 sigset_t* os::Linux::vm_signals() {
489 assert(signal_sets_initialized, "Not initialized");
490 return &vm_sigs;
491 }
492
493 void os::Linux::hotspot_sigmask(Thread* thread) {
494
495 //Save caller's signal mask before setting VM signal mask
496 sigset_t caller_sigmask;
497 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
498
499 OSThread* osthread = thread->osthread();
500 osthread->set_caller_sigmask(caller_sigmask);
501
502 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
503
504 if (!ReduceSignalUsage) {
505 if (thread->is_VM_thread()) {
506 // Only the VM thread handles BREAK_SIGNAL ...
507 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
508 } else {
509 // ... all other threads block BREAK_SIGNAL
510 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
511 }
512 }
513 }
514
515 //////////////////////////////////////////////////////////////////////////////
516 // detecting pthread library
517
518 void os::Linux::libpthread_init() {
519 // Save glibc and pthread version strings.
520 #if !defined(_CS_GNU_LIBC_VERSION) || \
521 !defined(_CS_GNU_LIBPTHREAD_VERSION)
522 #error "glibc too old (< 2.3.2)"
523 #endif
524
525 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
526 assert(n > 0, "cannot retrieve glibc version");
527 char *str = (char *)malloc(n, mtInternal);
528 confstr(_CS_GNU_LIBC_VERSION, str, n);
529 os::Linux::set_glibc_version(str);
530
531 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
532 assert(n > 0, "cannot retrieve pthread version");
533 str = (char *)malloc(n, mtInternal);
534 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
535 os::Linux::set_libpthread_version(str);
536 }
537
538 /////////////////////////////////////////////////////////////////////////////
539 // thread stack expansion
540
541 // os::Linux::manually_expand_stack() takes care of expanding the thread
542 // stack. Note that this is normally not needed: pthread stacks allocate
543 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
544 // committed. Therefore it is not necessary to expand the stack manually.
545 //
546 // Manually expanding the stack was historically needed on LinuxThreads
547 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
548 // it is kept to deal with very rare corner cases:
549 //
550 // For one, user may run the VM on an own implementation of threads
551 // whose stacks are - like the old LinuxThreads - implemented using
552 // mmap(MAP_GROWSDOWN).
553 //
554 // Also, this coding may be needed if the VM is running on the primordial
555 // thread. Normally we avoid running on the primordial thread; however,
556 // user may still invoke the VM on the primordial thread.
557 //
558 // The following historical comment describes the details about running
559 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
560
561
562 // Force Linux kernel to expand current thread stack. If "bottom" is close
563 // to the stack guard, caller should block all signals.
564 //
565 // MAP_GROWSDOWN:
566 // A special mmap() flag that is used to implement thread stacks. It tells
567 // kernel that the memory region should extend downwards when needed. This
568 // allows early versions of LinuxThreads to only mmap the first few pages
569 // when creating a new thread. Linux kernel will automatically expand thread
570 // stack as needed (on page faults).
571 //
572 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
573 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
574 // region, it's hard to tell if the fault is due to a legitimate stack
575 // access or because of reading/writing non-exist memory (e.g. buffer
576 // overrun). As a rule, if the fault happens below current stack pointer,
577 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
578 // application (see Linux kernel fault.c).
579 //
580 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
581 // stack overflow detection.
582 //
583 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
584 // not use MAP_GROWSDOWN.
585 //
586 // To get around the problem and allow stack banging on Linux, we need to
587 // manually expand thread stack after receiving the SIGSEGV.
588 //
589 // There are two ways to expand thread stack to address "bottom", we used
590 // both of them in JVM before 1.5:
591 // 1. adjust stack pointer first so that it is below "bottom", and then
592 // touch "bottom"
593 // 2. mmap() the page in question
594 //
595 // Now alternate signal stack is gone, it's harder to use 2. For instance,
596 // if current sp is already near the lower end of page 101, and we need to
597 // call mmap() to map page 100, it is possible that part of the mmap() frame
598 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
599 // That will destroy the mmap() frame and cause VM to crash.
600 //
601 // The following code works by adjusting sp first, then accessing the "bottom"
602 // page to force a page fault. Linux kernel will then automatically expand the
603 // stack mapping.
604 //
605 // _expand_stack_to() assumes its frame size is less than page size, which
606 // should always be true if the function is not inlined.
607
608 static void NOINLINE _expand_stack_to(address bottom) {
609 address sp;
610 size_t size;
611 volatile char *p;
612
613 // Adjust bottom to point to the largest address within the same page, it
614 // gives us a one-page buffer if alloca() allocates slightly more memory.
615 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
616 bottom += os::Linux::page_size() - 1;
617
618 // sp might be slightly above current stack pointer; if that's the case, we
619 // will alloca() a little more space than necessary, which is OK. Don't use
620 // os::current_stack_pointer(), as its result can be slightly below current
621 // stack pointer, causing us to not alloca enough to reach "bottom".
622 sp = (address)&sp;
623
624 if (sp > bottom) {
625 size = sp - bottom;
626 p = (volatile char *)alloca(size);
627 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
628 p[0] = '\0';
629 }
630 }
631
632 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
633 assert(t!=NULL, "just checking");
634 assert(t->osthread()->expanding_stack(), "expand should be set");
635 assert(t->stack_base() != NULL, "stack_base was not initialized");
636
637 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
638 sigset_t mask_all, old_sigset;
639 sigfillset(&mask_all);
640 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
641 _expand_stack_to(addr);
642 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
643 return true;
644 }
645 return false;
646 }
647
648 //////////////////////////////////////////////////////////////////////////////
649 // create new thread
650
651 // Thread start routine for all newly created threads
652 static void *thread_native_entry(Thread *thread) {
653 // Try to randomize the cache line index of hot stack frames.
654 // This helps when threads of the same stack traces evict each other's
655 // cache lines. The threads can be either from the same JVM instance, or
656 // from different JVM instances. The benefit is especially true for
657 // processors with hyperthreading technology.
658 static int counter = 0;
659 int pid = os::current_process_id();
660 alloca(((pid ^ counter++) & 7) * 128);
661
662 thread->initialize_thread_current();
663
664 OSThread* osthread = thread->osthread();
665 Monitor* sync = osthread->startThread_lock();
666
667 osthread->set_thread_id(os::current_thread_id());
668
669 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
670 os::current_thread_id(), (uintx) pthread_self());
671
672 if (UseNUMA) {
673 int lgrp_id = os::numa_get_group_id();
674 if (lgrp_id != -1) {
675 thread->set_lgrp_id(lgrp_id);
676 }
677 }
678 // initialize signal mask for this thread
679 os::Linux::hotspot_sigmask(thread);
680
681 // initialize floating point control register
682 os::Linux::init_thread_fpu_state();
683
684 // handshaking with parent thread
685 {
686 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
687
688 // notify parent thread
689 osthread->set_state(INITIALIZED);
690 sync->notify_all();
691
692 // wait until os::start_thread()
693 while (osthread->get_state() == INITIALIZED) {
694 sync->wait(Mutex::_no_safepoint_check_flag);
695 }
696 }
697
698 // call one more level start routine
699 thread->run();
700
701 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
702 os::current_thread_id(), (uintx) pthread_self());
703
704 // If a thread has not deleted itself ("delete this") as part of its
705 // termination sequence, we have to ensure thread-local-storage is
706 // cleared before we actually terminate. No threads should ever be
707 // deleted asynchronously with respect to their termination.
708 if (Thread::current_or_null_safe() != NULL) {
709 assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
710 thread->clear_thread_current();
711 }
712
713 return 0;
714 }
715
716 bool os::create_thread(Thread* thread, ThreadType thr_type,
717 size_t req_stack_size) {
718 assert(thread->osthread() == NULL, "caller responsible");
719
720 // Allocate the OSThread object
721 OSThread* osthread = new OSThread(NULL, NULL);
722 if (osthread == NULL) {
723 return false;
724 }
725
726 // set the correct thread state
727 osthread->set_thread_type(thr_type);
728
729 // Initial state is ALLOCATED but not INITIALIZED
730 osthread->set_state(ALLOCATED);
731
732 thread->set_osthread(osthread);
733
734 // init thread attributes
735 pthread_attr_t attr;
736 pthread_attr_init(&attr);
737 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
738
739 // Calculate stack size if it's not specified by caller.
740 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
741 // In the Linux NPTL pthread implementation the guard size mechanism
742 // is not implemented properly. The posix standard requires adding
743 // the size of the guard pages to the stack size, instead Linux
744 // takes the space out of 'stacksize'. Thus we adapt the requested
745 // stack_size by the size of the guard pages to mimick proper
746 // behaviour. However, be careful not to end up with a size
747 // of zero due to overflow. Don't add the guard page in that case.
748 size_t guard_size = os::Linux::default_guard_size(thr_type);
749 if (stack_size <= SIZE_MAX - guard_size) {
750 stack_size += guard_size;
751 }
752 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
753
754 int status = pthread_attr_setstacksize(&attr, stack_size);
755 assert_status(status == 0, status, "pthread_attr_setstacksize");
756
757 // Configure glibc guard page.
758 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
759
760 ThreadState state;
761
762 {
763 pthread_t tid;
764 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
765
766 char buf[64];
767 if (ret == 0) {
768 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
769 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
770 } else {
771 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
772 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
773 }
774
775 pthread_attr_destroy(&attr);
776
777 if (ret != 0) {
778 // Need to clean up stuff we've allocated so far
779 thread->set_osthread(NULL);
780 delete osthread;
781 return false;
782 }
783
784 // Store pthread info into the OSThread
785 osthread->set_pthread_id(tid);
786
787 // Wait until child thread is either initialized or aborted
788 {
789 Monitor* sync_with_child = osthread->startThread_lock();
790 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
791 while ((state = osthread->get_state()) == ALLOCATED) {
792 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
793 }
794 }
795 }
796
797 // Aborted due to thread limit being reached
798 if (state == ZOMBIE) {
799 thread->set_osthread(NULL);
800 delete osthread;
801 return false;
802 }
803
804 // The thread is returned suspended (in state INITIALIZED),
805 // and is started higher up in the call chain
806 assert(state == INITIALIZED, "race condition");
807 return true;
808 }
809
810 /////////////////////////////////////////////////////////////////////////////
811 // attach existing thread
812
813 // bootstrap the main thread
814 bool os::create_main_thread(JavaThread* thread) {
815 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
816 return create_attached_thread(thread);
817 }
818
819 bool os::create_attached_thread(JavaThread* thread) {
820 #ifdef ASSERT
821 thread->verify_not_published();
822 #endif
823
824 // Allocate the OSThread object
825 OSThread* osthread = new OSThread(NULL, NULL);
826
827 if (osthread == NULL) {
828 return false;
829 }
830
831 // Store pthread info into the OSThread
832 osthread->set_thread_id(os::Linux::gettid());
833 osthread->set_pthread_id(::pthread_self());
834
835 // initialize floating point control register
836 os::Linux::init_thread_fpu_state();
837
838 // Initial thread state is RUNNABLE
839 osthread->set_state(RUNNABLE);
840
841 thread->set_osthread(osthread);
842
843 if (UseNUMA) {
844 int lgrp_id = os::numa_get_group_id();
845 if (lgrp_id != -1) {
846 thread->set_lgrp_id(lgrp_id);
847 }
848 }
849
850 if (os::is_primordial_thread()) {
851 // If current thread is primordial thread, its stack is mapped on demand,
852 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
853 // the entire stack region to avoid SEGV in stack banging.
854 // It is also useful to get around the heap-stack-gap problem on SuSE
855 // kernel (see 4821821 for details). We first expand stack to the top
856 // of yellow zone, then enable stack yellow zone (order is significant,
857 // enabling yellow zone first will crash JVM on SuSE Linux), so there
858 // is no gap between the last two virtual memory regions.
859
860 JavaThread *jt = (JavaThread *)thread;
861 address addr = jt->stack_reserved_zone_base();
862 assert(addr != NULL, "initialization problem?");
863 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
864
865 osthread->set_expanding_stack();
866 os::Linux::manually_expand_stack(jt, addr);
867 osthread->clear_expanding_stack();
868 }
869
870 // initialize signal mask for this thread
871 // and save the caller's signal mask
872 os::Linux::hotspot_sigmask(thread);
873
874 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
875 os::current_thread_id(), (uintx) pthread_self());
876
877 return true;
878 }
879
880 void os::pd_start_thread(Thread* thread) {
881 OSThread * osthread = thread->osthread();
882 assert(osthread->get_state() != INITIALIZED, "just checking");
883 Monitor* sync_with_child = osthread->startThread_lock();
884 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
885 sync_with_child->notify();
886 }
887
888 // Free Linux resources related to the OSThread
889 void os::free_thread(OSThread* osthread) {
890 assert(osthread != NULL, "osthread not set");
891
892 // We are told to free resources of the argument thread,
893 // but we can only really operate on the current thread.
894 assert(Thread::current()->osthread() == osthread,
895 "os::free_thread but not current thread");
896
897 #ifdef ASSERT
898 sigset_t current;
899 sigemptyset(¤t);
900 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
901 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
902 #endif
903
904 // Restore caller's signal mask
905 sigset_t sigmask = osthread->caller_sigmask();
906 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
907
908 delete osthread;
909 }
910
911 //////////////////////////////////////////////////////////////////////////////
912 // primordial thread
913
914 // Check if current thread is the primordial thread, similar to Solaris thr_main.
915 bool os::is_primordial_thread(void) {
916 char dummy;
917 // If called before init complete, thread stack bottom will be null.
918 // Can be called if fatal error occurs before initialization.
919 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
920 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
921 os::Linux::initial_thread_stack_size() != 0,
922 "os::init did not locate primordial thread's stack region");
923 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
924 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
925 os::Linux::initial_thread_stack_size()) {
926 return true;
927 } else {
928 return false;
929 }
930 }
931
932 // Find the virtual memory area that contains addr
933 static bool find_vma(address addr, address* vma_low, address* vma_high) {
934 FILE *fp = fopen("/proc/self/maps", "r");
935 if (fp) {
936 address low, high;
937 while (!feof(fp)) {
938 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
939 if (low <= addr && addr < high) {
940 if (vma_low) *vma_low = low;
941 if (vma_high) *vma_high = high;
942 fclose(fp);
943 return true;
944 }
945 }
946 for (;;) {
947 int ch = fgetc(fp);
948 if (ch == EOF || ch == (int)'\n') break;
949 }
950 }
951 fclose(fp);
952 }
953 return false;
954 }
955
956 // Locate primordial thread stack. This special handling of primordial thread stack
957 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
958 // bogus value for the primordial process thread. While the launcher has created
959 // the VM in a new thread since JDK 6, we still have to allow for the use of the
960 // JNI invocation API from a primordial thread.
961 void os::Linux::capture_initial_stack(size_t max_size) {
962
963 // max_size is either 0 (which means accept OS default for thread stacks) or
964 // a user-specified value known to be at least the minimum needed. If we
965 // are actually on the primordial thread we can make it appear that we have a
966 // smaller max_size stack by inserting the guard pages at that location. But we
967 // cannot do anything to emulate a larger stack than what has been provided by
968 // the OS or threading library. In fact if we try to use a stack greater than
969 // what is set by rlimit then we will crash the hosting process.
970
971 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
972 // If this is "unlimited" then it will be a huge value.
973 struct rlimit rlim;
974 getrlimit(RLIMIT_STACK, &rlim);
975 size_t stack_size = rlim.rlim_cur;
976
977 // 6308388: a bug in ld.so will relocate its own .data section to the
978 // lower end of primordial stack; reduce ulimit -s value a little bit
979 // so we won't install guard page on ld.so's data section.
980 // But ensure we don't underflow the stack size - allow 1 page spare
981 if (stack_size >= (size_t)(3 * page_size())) {
982 stack_size -= 2 * page_size();
983 }
984
985 // Try to figure out where the stack base (top) is. This is harder.
986 //
987 // When an application is started, glibc saves the initial stack pointer in
988 // a global variable "__libc_stack_end", which is then used by system
989 // libraries. __libc_stack_end should be pretty close to stack top. The
990 // variable is available since the very early days. However, because it is
991 // a private interface, it could disappear in the future.
992 //
993 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
994 // to __libc_stack_end, it is very close to stack top, but isn't the real
995 // stack top. Note that /proc may not exist if VM is running as a chroot
996 // program, so reading /proc/<pid>/stat could fail. Also the contents of
997 // /proc/<pid>/stat could change in the future (though unlikely).
998 //
999 // We try __libc_stack_end first. If that doesn't work, look for
1000 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1001 // as a hint, which should work well in most cases.
1002
1003 uintptr_t stack_start;
1004
1005 // try __libc_stack_end first
1006 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1007 if (p && *p) {
1008 stack_start = *p;
1009 } else {
1010 // see if we can get the start_stack field from /proc/self/stat
1011 FILE *fp;
1012 int pid;
1013 char state;
1014 int ppid;
1015 int pgrp;
1016 int session;
1017 int nr;
1018 int tpgrp;
1019 unsigned long flags;
1020 unsigned long minflt;
1021 unsigned long cminflt;
1022 unsigned long majflt;
1023 unsigned long cmajflt;
1024 unsigned long utime;
1025 unsigned long stime;
1026 long cutime;
1027 long cstime;
1028 long prio;
1029 long nice;
1030 long junk;
1031 long it_real;
1032 uintptr_t start;
1033 uintptr_t vsize;
1034 intptr_t rss;
1035 uintptr_t rsslim;
1036 uintptr_t scodes;
1037 uintptr_t ecode;
1038 int i;
1039
1040 // Figure what the primordial thread stack base is. Code is inspired
1041 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1042 // followed by command name surrounded by parentheses, state, etc.
1043 char stat[2048];
1044 int statlen;
1045
1046 fp = fopen("/proc/self/stat", "r");
1047 if (fp) {
1048 statlen = fread(stat, 1, 2047, fp);
1049 stat[statlen] = '\0';
1050 fclose(fp);
1051
1052 // Skip pid and the command string. Note that we could be dealing with
1053 // weird command names, e.g. user could decide to rename java launcher
1054 // to "java 1.4.2 :)", then the stat file would look like
1055 // 1234 (java 1.4.2 :)) R ... ...
1056 // We don't really need to know the command string, just find the last
1057 // occurrence of ")" and then start parsing from there. See bug 4726580.
1058 char * s = strrchr(stat, ')');
1059
1060 i = 0;
1061 if (s) {
1062 // Skip blank chars
1063 do { s++; } while (s && isspace(*s));
1064
1065 #define _UFM UINTX_FORMAT
1066 #define _DFM INTX_FORMAT
1067
1068 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1069 // 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
1070 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,
1071 &state, // 3 %c
1072 &ppid, // 4 %d
1073 &pgrp, // 5 %d
1074 &session, // 6 %d
1075 &nr, // 7 %d
1076 &tpgrp, // 8 %d
1077 &flags, // 9 %lu
1078 &minflt, // 10 %lu
1079 &cminflt, // 11 %lu
1080 &majflt, // 12 %lu
1081 &cmajflt, // 13 %lu
1082 &utime, // 14 %lu
1083 &stime, // 15 %lu
1084 &cutime, // 16 %ld
1085 &cstime, // 17 %ld
1086 &prio, // 18 %ld
1087 &nice, // 19 %ld
1088 &junk, // 20 %ld
1089 &it_real, // 21 %ld
1090 &start, // 22 UINTX_FORMAT
1091 &vsize, // 23 UINTX_FORMAT
1092 &rss, // 24 INTX_FORMAT
1093 &rsslim, // 25 UINTX_FORMAT
1094 &scodes, // 26 UINTX_FORMAT
1095 &ecode, // 27 UINTX_FORMAT
1096 &stack_start); // 28 UINTX_FORMAT
1097 }
1098
1099 #undef _UFM
1100 #undef _DFM
1101
1102 if (i != 28 - 2) {
1103 assert(false, "Bad conversion from /proc/self/stat");
1104 // product mode - assume we are the primordial thread, good luck in the
1105 // embedded case.
1106 warning("Can't detect primordial thread stack location - bad conversion");
1107 stack_start = (uintptr_t) &rlim;
1108 }
1109 } else {
1110 // For some reason we can't open /proc/self/stat (for example, running on
1111 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1112 // most cases, so don't abort:
1113 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1114 stack_start = (uintptr_t) &rlim;
1115 }
1116 }
1117
1118 // Now we have a pointer (stack_start) very close to the stack top, the
1119 // next thing to do is to figure out the exact location of stack top. We
1120 // can find out the virtual memory area that contains stack_start by
1121 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1122 // and its upper limit is the real stack top. (again, this would fail if
1123 // running inside chroot, because /proc may not exist.)
1124
1125 uintptr_t stack_top;
1126 address low, high;
1127 if (find_vma((address)stack_start, &low, &high)) {
1128 // success, "high" is the true stack top. (ignore "low", because initial
1129 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1130 stack_top = (uintptr_t)high;
1131 } else {
1132 // failed, likely because /proc/self/maps does not exist
1133 warning("Can't detect primordial thread stack location - find_vma failed");
1134 // best effort: stack_start is normally within a few pages below the real
1135 // stack top, use it as stack top, and reduce stack size so we won't put
1136 // guard page outside stack.
1137 stack_top = stack_start;
1138 stack_size -= 16 * page_size();
1139 }
1140
1141 // stack_top could be partially down the page so align it
1142 stack_top = align_up(stack_top, page_size());
1143
1144 // Allowed stack value is minimum of max_size and what we derived from rlimit
1145 if (max_size > 0) {
1146 _initial_thread_stack_size = MIN2(max_size, stack_size);
1147 } else {
1148 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1149 // clamp it at 8MB as we do on Solaris
1150 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1151 }
1152 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1153 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1154
1155 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1156
1157 if (log_is_enabled(Info, os, thread)) {
1158 // See if we seem to be on primordial process thread
1159 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1160 uintptr_t(&rlim) < stack_top;
1161
1162 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1163 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1164 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1165 stack_top, intptr_t(_initial_thread_stack_bottom));
1166 }
1167 }
1168
1169 ////////////////////////////////////////////////////////////////////////////////
1170 // time support
1171
1172 // Time since start-up in seconds to a fine granularity.
1173 // Used by VMSelfDestructTimer and the MemProfiler.
1174 double os::elapsedTime() {
1175
1176 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1177 }
1178
1179 jlong os::elapsed_counter() {
1180 return javaTimeNanos() - initial_time_count;
1181 }
1182
1183 jlong os::elapsed_frequency() {
1184 return NANOSECS_PER_SEC; // nanosecond resolution
1185 }
1186
1187 bool os::supports_vtime() { return true; }
1188 bool os::enable_vtime() { return false; }
1189 bool os::vtime_enabled() { return false; }
1190
1191 double os::elapsedVTime() {
1192 struct rusage usage;
1193 int retval = getrusage(RUSAGE_THREAD, &usage);
1194 if (retval == 0) {
1195 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);
1196 } else {
1197 // better than nothing, but not much
1198 return elapsedTime();
1199 }
1200 }
1201
1202 jlong os::javaTimeMillis() {
1203 timeval time;
1204 int status = gettimeofday(&time, NULL);
1205 assert(status != -1, "linux error");
1206 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1207 }
1208
1209 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1210 timeval time;
1211 int status = gettimeofday(&time, NULL);
1212 assert(status != -1, "linux error");
1213 seconds = jlong(time.tv_sec);
1214 nanos = jlong(time.tv_usec) * 1000;
1215 }
1216
1217
1218 #ifndef CLOCK_MONOTONIC
1219 #define CLOCK_MONOTONIC (1)
1220 #endif
1221
1222 void os::Linux::clock_init() {
1223 // we do dlopen's in this particular order due to bug in linux
1224 // dynamical loader (see 6348968) leading to crash on exit
1225 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1226 if (handle == NULL) {
1227 handle = dlopen("librt.so", RTLD_LAZY);
1228 }
1229
1230 if (handle) {
1231 int (*clock_getres_func)(clockid_t, struct timespec*) =
1232 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1233 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1234 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1235 if (clock_getres_func && clock_gettime_func) {
1236 // See if monotonic clock is supported by the kernel. Note that some
1237 // early implementations simply return kernel jiffies (updated every
1238 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1239 // for nano time (though the monotonic property is still nice to have).
1240 // It's fixed in newer kernels, however clock_getres() still returns
1241 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1242 // resolution for now. Hopefully as people move to new kernels, this
1243 // won't be a problem.
1244 struct timespec res;
1245 struct timespec tp;
1246 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1247 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1248 // yes, monotonic clock is supported
1249 _clock_gettime = clock_gettime_func;
1250 return;
1251 } else {
1252 // close librt if there is no monotonic clock
1253 dlclose(handle);
1254 }
1255 }
1256 }
1257 warning("No monotonic clock was available - timed services may " \
1258 "be adversely affected if the time-of-day clock changes");
1259 }
1260
1261 #ifndef SYS_clock_getres
1262 #if defined(X86) || defined(PPC64) || defined(S390)
1263 #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261)
1264 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1265 #else
1266 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1267 #define sys_clock_getres(x,y) -1
1268 #endif
1269 #else
1270 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1271 #endif
1272
1273 void os::Linux::fast_thread_clock_init() {
1274 if (!UseLinuxPosixThreadCPUClocks) {
1275 return;
1276 }
1277 clockid_t clockid;
1278 struct timespec tp;
1279 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1280 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1281
1282 // Switch to using fast clocks for thread cpu time if
1283 // the sys_clock_getres() returns 0 error code.
1284 // Note, that some kernels may support the current thread
1285 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1286 // returned by the pthread_getcpuclockid().
1287 // If the fast Posix clocks are supported then the sys_clock_getres()
1288 // must return at least tp.tv_sec == 0 which means a resolution
1289 // better than 1 sec. This is extra check for reliability.
1290
1291 if (pthread_getcpuclockid_func &&
1292 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1293 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1294 _supports_fast_thread_cpu_time = true;
1295 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1296 }
1297 }
1298
1299 jlong os::javaTimeNanos() {
1300 if (os::supports_monotonic_clock()) {
1301 struct timespec tp;
1302 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1303 assert(status == 0, "gettime error");
1304 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1305 return result;
1306 } else {
1307 timeval time;
1308 int status = gettimeofday(&time, NULL);
1309 assert(status != -1, "linux error");
1310 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1311 return 1000 * usecs;
1312 }
1313 }
1314
1315 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1316 if (os::supports_monotonic_clock()) {
1317 info_ptr->max_value = ALL_64_BITS;
1318
1319 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1320 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1321 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1322 } else {
1323 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1324 info_ptr->max_value = ALL_64_BITS;
1325
1326 // gettimeofday is a real time clock so it skips
1327 info_ptr->may_skip_backward = true;
1328 info_ptr->may_skip_forward = true;
1329 }
1330
1331 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1332 }
1333
1334 // Return the real, user, and system times in seconds from an
1335 // arbitrary fixed point in the past.
1336 bool os::getTimesSecs(double* process_real_time,
1337 double* process_user_time,
1338 double* process_system_time) {
1339 struct tms ticks;
1340 clock_t real_ticks = times(&ticks);
1341
1342 if (real_ticks == (clock_t) (-1)) {
1343 return false;
1344 } else {
1345 double ticks_per_second = (double) clock_tics_per_sec;
1346 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1347 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1348 *process_real_time = ((double) real_ticks) / ticks_per_second;
1349
1350 return true;
1351 }
1352 }
1353
1354
1355 char * os::local_time_string(char *buf, size_t buflen) {
1356 struct tm t;
1357 time_t long_time;
1358 time(&long_time);
1359 localtime_r(&long_time, &t);
1360 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1361 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1362 t.tm_hour, t.tm_min, t.tm_sec);
1363 return buf;
1364 }
1365
1366 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1367 return localtime_r(clock, res);
1368 }
1369
1370 ////////////////////////////////////////////////////////////////////////////////
1371 // runtime exit support
1372
1373 // Note: os::shutdown() might be called very early during initialization, or
1374 // called from signal handler. Before adding something to os::shutdown(), make
1375 // sure it is async-safe and can handle partially initialized VM.
1376 void os::shutdown() {
1377
1378 // allow PerfMemory to attempt cleanup of any persistent resources
1379 perfMemory_exit();
1380
1381 // needs to remove object in file system
1382 AttachListener::abort();
1383
1384 // flush buffered output, finish log files
1385 ostream_abort();
1386
1387 // Check for abort hook
1388 abort_hook_t abort_hook = Arguments::abort_hook();
1389 if (abort_hook != NULL) {
1390 abort_hook();
1391 }
1392
1393 }
1394
1395 // Note: os::abort() might be called very early during initialization, or
1396 // called from signal handler. Before adding something to os::abort(), make
1397 // sure it is async-safe and can handle partially initialized VM.
1398 void os::abort(bool dump_core, void* siginfo, const void* context) {
1399 os::shutdown();
1400 if (dump_core) {
1401 #ifndef PRODUCT
1402 fdStream out(defaultStream::output_fd());
1403 out.print_raw("Current thread is ");
1404 char buf[16];
1405 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1406 out.print_raw_cr(buf);
1407 out.print_raw_cr("Dumping core ...");
1408 #endif
1409 ::abort(); // dump core
1410 }
1411
1412 ::exit(1);
1413 }
1414
1415 // Die immediately, no exit hook, no abort hook, no cleanup.
1416 void os::die() {
1417 ::abort();
1418 }
1419
1420
1421 // This method is a copy of JDK's sysGetLastErrorString
1422 // from src/solaris/hpi/src/system_md.c
1423
1424 size_t os::lasterror(char *buf, size_t len) {
1425 if (errno == 0) return 0;
1426
1427 const char *s = os::strerror(errno);
1428 size_t n = ::strlen(s);
1429 if (n >= len) {
1430 n = len - 1;
1431 }
1432 ::strncpy(buf, s, n);
1433 buf[n] = '\0';
1434 return n;
1435 }
1436
1437 // thread_id is kernel thread id (similar to Solaris LWP id)
1438 intx os::current_thread_id() { return os::Linux::gettid(); }
1439 int os::current_process_id() {
1440 return ::getpid();
1441 }
1442
1443 // DLL functions
1444
1445 const char* os::dll_file_extension() { return ".so"; }
1446
1447 // This must be hard coded because it's the system's temporary
1448 // directory not the java application's temp directory, ala java.io.tmpdir.
1449 const char* os::get_temp_directory() { return "/tmp"; }
1450
1451 static bool file_exists(const char* filename) {
1452 struct stat statbuf;
1453 if (filename == NULL || strlen(filename) == 0) {
1454 return false;
1455 }
1456 return os::stat(filename, &statbuf) == 0;
1457 }
1458
1459 // check if addr is inside libjvm.so
1460 bool os::address_is_in_vm(address addr) {
1461 static address libjvm_base_addr;
1462 Dl_info dlinfo;
1463
1464 if (libjvm_base_addr == NULL) {
1465 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1466 libjvm_base_addr = (address)dlinfo.dli_fbase;
1467 }
1468 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1469 }
1470
1471 if (dladdr((void *)addr, &dlinfo) != 0) {
1472 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1473 }
1474
1475 return false;
1476 }
1477
1478 bool os::dll_address_to_function_name(address addr, char *buf,
1479 int buflen, int *offset,
1480 bool demangle) {
1481 // buf is not optional, but offset is optional
1482 assert(buf != NULL, "sanity check");
1483
1484 Dl_info dlinfo;
1485
1486 if (dladdr((void*)addr, &dlinfo) != 0) {
1487 // see if we have a matching symbol
1488 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1489 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1490 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1491 }
1492 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1493 return true;
1494 }
1495 // no matching symbol so try for just file info
1496 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1497 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1498 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1499 return true;
1500 }
1501 }
1502 }
1503
1504 buf[0] = '\0';
1505 if (offset != NULL) *offset = -1;
1506 return false;
1507 }
1508
1509 struct _address_to_library_name {
1510 address addr; // input : memory address
1511 size_t buflen; // size of fname
1512 char* fname; // output: library name
1513 address base; // library base addr
1514 };
1515
1516 static int address_to_library_name_callback(struct dl_phdr_info *info,
1517 size_t size, void *data) {
1518 int i;
1519 bool found = false;
1520 address libbase = NULL;
1521 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1522
1523 // iterate through all loadable segments
1524 for (i = 0; i < info->dlpi_phnum; i++) {
1525 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1526 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1527 // base address of a library is the lowest address of its loaded
1528 // segments.
1529 if (libbase == NULL || libbase > segbase) {
1530 libbase = segbase;
1531 }
1532 // see if 'addr' is within current segment
1533 if (segbase <= d->addr &&
1534 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1535 found = true;
1536 }
1537 }
1538 }
1539
1540 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1541 // so dll_address_to_library_name() can fall through to use dladdr() which
1542 // can figure out executable name from argv[0].
1543 if (found && info->dlpi_name && info->dlpi_name[0]) {
1544 d->base = libbase;
1545 if (d->fname) {
1546 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1547 }
1548 return 1;
1549 }
1550 return 0;
1551 }
1552
1553 bool os::dll_address_to_library_name(address addr, char* buf,
1554 int buflen, int* offset) {
1555 // buf is not optional, but offset is optional
1556 assert(buf != NULL, "sanity check");
1557
1558 Dl_info dlinfo;
1559 struct _address_to_library_name data;
1560
1561 // There is a bug in old glibc dladdr() implementation that it could resolve
1562 // to wrong library name if the .so file has a base address != NULL. Here
1563 // we iterate through the program headers of all loaded libraries to find
1564 // out which library 'addr' really belongs to. This workaround can be
1565 // removed once the minimum requirement for glibc is moved to 2.3.x.
1566 data.addr = addr;
1567 data.fname = buf;
1568 data.buflen = buflen;
1569 data.base = NULL;
1570 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1571
1572 if (rslt) {
1573 // buf already contains library name
1574 if (offset) *offset = addr - data.base;
1575 return true;
1576 }
1577 if (dladdr((void*)addr, &dlinfo) != 0) {
1578 if (dlinfo.dli_fname != NULL) {
1579 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1580 }
1581 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1582 *offset = addr - (address)dlinfo.dli_fbase;
1583 }
1584 return true;
1585 }
1586
1587 buf[0] = '\0';
1588 if (offset) *offset = -1;
1589 return false;
1590 }
1591
1592 // Loads .dll/.so and
1593 // in case of error it checks if .dll/.so was built for the
1594 // same architecture as Hotspot is running on
1595
1596
1597 // Remember the stack's state. The Linux dynamic linker will change
1598 // the stack to 'executable' at most once, so we must safepoint only once.
1599 bool os::Linux::_stack_is_executable = false;
1600
1601 // VM operation that loads a library. This is necessary if stack protection
1602 // of the Java stacks can be lost during loading the library. If we
1603 // do not stop the Java threads, they can stack overflow before the stacks
1604 // are protected again.
1605 class VM_LinuxDllLoad: public VM_Operation {
1606 private:
1607 const char *_filename;
1608 char *_ebuf;
1609 int _ebuflen;
1610 void *_lib;
1611 public:
1612 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1613 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1614 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1615 void doit() {
1616 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1617 os::Linux::_stack_is_executable = true;
1618 }
1619 void* loaded_library() { return _lib; }
1620 };
1621
1622 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1623 void * result = NULL;
1624 bool load_attempted = false;
1625
1626 // Check whether the library to load might change execution rights
1627 // of the stack. If they are changed, the protection of the stack
1628 // guard pages will be lost. We need a safepoint to fix this.
1629 //
1630 // See Linux man page execstack(8) for more info.
1631 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1632 if (!ElfFile::specifies_noexecstack(filename)) {
1633 if (!is_init_completed()) {
1634 os::Linux::_stack_is_executable = true;
1635 // This is OK - No Java threads have been created yet, and hence no
1636 // stack guard pages to fix.
1637 //
1638 // This should happen only when you are building JDK7 using a very
1639 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1640 //
1641 // Dynamic loader will make all stacks executable after
1642 // this function returns, and will not do that again.
1643 #ifdef ASSERT
1644 ThreadsListHandle tlh;
1645 assert(tlh.length() == 0, "no Java threads should exist yet.");
1646 #endif
1647 } else {
1648 warning("You have loaded library %s which might have disabled stack guard. "
1649 "The VM will try to fix the stack guard now.\n"
1650 "It's highly recommended that you fix the library with "
1651 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1652 filename);
1653
1654 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1655 JavaThread *jt = JavaThread::current();
1656 if (jt->thread_state() != _thread_in_native) {
1657 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1658 // that requires ExecStack. Cannot enter safe point. Let's give up.
1659 warning("Unable to fix stack guard. Giving up.");
1660 } else {
1661 if (!LoadExecStackDllInVMThread) {
1662 // This is for the case where the DLL has an static
1663 // constructor function that executes JNI code. We cannot
1664 // load such DLLs in the VMThread.
1665 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1666 }
1667
1668 ThreadInVMfromNative tiv(jt);
1669 debug_only(VMNativeEntryWrapper vew;)
1670
1671 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1672 VMThread::execute(&op);
1673 if (LoadExecStackDllInVMThread) {
1674 result = op.loaded_library();
1675 }
1676 load_attempted = true;
1677 }
1678 }
1679 }
1680 }
1681
1682 if (!load_attempted) {
1683 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1684 }
1685
1686 if (result != NULL) {
1687 // Successful loading
1688 return result;
1689 }
1690
1691 Elf32_Ehdr elf_head;
1692 int diag_msg_max_length=ebuflen-strlen(ebuf);
1693 char* diag_msg_buf=ebuf+strlen(ebuf);
1694
1695 if (diag_msg_max_length==0) {
1696 // No more space in ebuf for additional diagnostics message
1697 return NULL;
1698 }
1699
1700
1701 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1702
1703 if (file_descriptor < 0) {
1704 // Can't open library, report dlerror() message
1705 return NULL;
1706 }
1707
1708 bool failed_to_read_elf_head=
1709 (sizeof(elf_head)!=
1710 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1711
1712 ::close(file_descriptor);
1713 if (failed_to_read_elf_head) {
1714 // file i/o error - report dlerror() msg
1715 return NULL;
1716 }
1717
1718 typedef struct {
1719 Elf32_Half code; // Actual value as defined in elf.h
1720 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1721 unsigned char elf_class; // 32 or 64 bit
1722 unsigned char endianess; // MSB or LSB
1723 char* name; // String representation
1724 } arch_t;
1725
1726 #ifndef EM_486
1727 #define EM_486 6 /* Intel 80486 */
1728 #endif
1729 #ifndef EM_AARCH64
1730 #define EM_AARCH64 183 /* ARM AARCH64 */
1731 #endif
1732
1733 static const arch_t arch_array[]={
1734 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1735 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1736 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1737 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1738 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1739 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1740 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1741 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1742 #if defined(VM_LITTLE_ENDIAN)
1743 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1744 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1745 #else
1746 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1747 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1748 #endif
1749 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1750 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1751 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1752 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1753 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1754 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1755 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1756 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1757 };
1758
1759 #if (defined IA32)
1760 static Elf32_Half running_arch_code=EM_386;
1761 #elif (defined AMD64)
1762 static Elf32_Half running_arch_code=EM_X86_64;
1763 #elif (defined IA64)
1764 static Elf32_Half running_arch_code=EM_IA_64;
1765 #elif (defined __sparc) && (defined _LP64)
1766 static Elf32_Half running_arch_code=EM_SPARCV9;
1767 #elif (defined __sparc) && (!defined _LP64)
1768 static Elf32_Half running_arch_code=EM_SPARC;
1769 #elif (defined __powerpc64__)
1770 static Elf32_Half running_arch_code=EM_PPC64;
1771 #elif (defined __powerpc__)
1772 static Elf32_Half running_arch_code=EM_PPC;
1773 #elif (defined AARCH64)
1774 static Elf32_Half running_arch_code=EM_AARCH64;
1775 #elif (defined ARM)
1776 static Elf32_Half running_arch_code=EM_ARM;
1777 #elif (defined S390)
1778 static Elf32_Half running_arch_code=EM_S390;
1779 #elif (defined ALPHA)
1780 static Elf32_Half running_arch_code=EM_ALPHA;
1781 #elif (defined MIPSEL)
1782 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1783 #elif (defined PARISC)
1784 static Elf32_Half running_arch_code=EM_PARISC;
1785 #elif (defined MIPS)
1786 static Elf32_Half running_arch_code=EM_MIPS;
1787 #elif (defined M68K)
1788 static Elf32_Half running_arch_code=EM_68K;
1789 #elif (defined SH)
1790 static Elf32_Half running_arch_code=EM_SH;
1791 #else
1792 #error Method os::dll_load requires that one of following is defined:\
1793 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1794 #endif
1795
1796 // Identify compatability class for VM's architecture and library's architecture
1797 // Obtain string descriptions for architectures
1798
1799 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1800 int running_arch_index=-1;
1801
1802 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1803 if (running_arch_code == arch_array[i].code) {
1804 running_arch_index = i;
1805 }
1806 if (lib_arch.code == arch_array[i].code) {
1807 lib_arch.compat_class = arch_array[i].compat_class;
1808 lib_arch.name = arch_array[i].name;
1809 }
1810 }
1811
1812 assert(running_arch_index != -1,
1813 "Didn't find running architecture code (running_arch_code) in arch_array");
1814 if (running_arch_index == -1) {
1815 // Even though running architecture detection failed
1816 // we may still continue with reporting dlerror() message
1817 return NULL;
1818 }
1819
1820 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1821 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1822 return NULL;
1823 }
1824
1825 #ifndef S390
1826 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1827 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1828 return NULL;
1829 }
1830 #endif // !S390
1831
1832 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1833 if (lib_arch.name!=NULL) {
1834 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1835 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1836 lib_arch.name, arch_array[running_arch_index].name);
1837 } else {
1838 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1839 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1840 lib_arch.code,
1841 arch_array[running_arch_index].name);
1842 }
1843 }
1844
1845 return NULL;
1846 }
1847
1848 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1849 int ebuflen) {
1850 void * result = ::dlopen(filename, RTLD_LAZY);
1851 if (result == NULL) {
1852 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1853 ebuf[ebuflen-1] = '\0';
1854 }
1855 return result;
1856 }
1857
1858 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1859 int ebuflen) {
1860 void * result = NULL;
1861 if (LoadExecStackDllInVMThread) {
1862 result = dlopen_helper(filename, ebuf, ebuflen);
1863 }
1864
1865 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1866 // library that requires an executable stack, or which does not have this
1867 // stack attribute set, dlopen changes the stack attribute to executable. The
1868 // read protection of the guard pages gets lost.
1869 //
1870 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1871 // may have been queued at the same time.
1872
1873 if (!_stack_is_executable) {
1874 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
1875 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1876 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1877 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1878 warning("Attempt to reguard stack yellow zone failed.");
1879 }
1880 }
1881 }
1882 }
1883
1884 return result;
1885 }
1886
1887 void* os::dll_lookup(void* handle, const char* name) {
1888 void* res = dlsym(handle, name);
1889 return res;
1890 }
1891
1892 void* os::get_default_process_handle() {
1893 return (void*)::dlopen(NULL, RTLD_LAZY);
1894 }
1895
1896 static bool _print_ascii_file(const char* filename, outputStream* st) {
1897 int fd = ::open(filename, O_RDONLY);
1898 if (fd == -1) {
1899 return false;
1900 }
1901
1902 char buf[33];
1903 int bytes;
1904 buf[32] = '\0';
1905 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1906 st->print_raw(buf, bytes);
1907 }
1908
1909 ::close(fd);
1910
1911 return true;
1912 }
1913
1914 void os::print_dll_info(outputStream *st) {
1915 st->print_cr("Dynamic libraries:");
1916
1917 char fname[32];
1918 pid_t pid = os::Linux::gettid();
1919
1920 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1921
1922 if (!_print_ascii_file(fname, st)) {
1923 st->print("Can not get library information for pid = %d\n", pid);
1924 }
1925 }
1926
1927 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1928 FILE *procmapsFile = NULL;
1929
1930 // Open the procfs maps file for the current process
1931 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1932 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1933 char line[PATH_MAX + 100];
1934
1935 // Read line by line from 'file'
1936 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1937 u8 base, top, offset, inode;
1938 char permissions[5];
1939 char device[6];
1940 char name[PATH_MAX + 1];
1941
1942 // Parse fields from line
1943 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1944 &base, &top, permissions, &offset, device, &inode, name);
1945
1946 // Filter by device id '00:00' so that we only get file system mapped files.
1947 if (strcmp(device, "00:00") != 0) {
1948
1949 // Call callback with the fields of interest
1950 if(callback(name, (address)base, (address)top, param)) {
1951 // Oops abort, callback aborted
1952 fclose(procmapsFile);
1953 return 1;
1954 }
1955 }
1956 }
1957 fclose(procmapsFile);
1958 }
1959 return 0;
1960 }
1961
1962 void os::print_os_info_brief(outputStream* st) {
1963 os::Linux::print_distro_info(st);
1964
1965 os::Posix::print_uname_info(st);
1966
1967 os::Linux::print_libversion_info(st);
1968
1969 }
1970
1971 void os::print_os_info(outputStream* st) {
1972 st->print("OS:");
1973
1974 os::Linux::print_distro_info(st);
1975
1976 os::Posix::print_uname_info(st);
1977
1978 // Print warning if unsafe chroot environment detected
1979 if (unsafe_chroot_detected) {
1980 st->print("WARNING!! ");
1981 st->print_cr("%s", unstable_chroot_error);
1982 }
1983
1984 os::Linux::print_libversion_info(st);
1985
1986 os::Posix::print_rlimit_info(st);
1987
1988 os::Posix::print_load_average(st);
1989
1990 os::Linux::print_full_memory_info(st);
1991
1992 os::Linux::print_container_info(st);
1993 }
1994
1995 // Try to identify popular distros.
1996 // Most Linux distributions have a /etc/XXX-release file, which contains
1997 // the OS version string. Newer Linux distributions have a /etc/lsb-release
1998 // file that also contains the OS version string. Some have more than one
1999 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2000 // /etc/redhat-release.), so the order is important.
2001 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2002 // their own specific XXX-release file as well as a redhat-release file.
2003 // Because of this the XXX-release file needs to be searched for before the
2004 // redhat-release file.
2005 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2006 // search for redhat-release / SuSE-release needs to be before lsb-release.
2007 // Since the lsb-release file is the new standard it needs to be searched
2008 // before the older style release files.
2009 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2010 // next to last resort. The os-release file is a new standard that contains
2011 // distribution information and the system-release file seems to be an old
2012 // standard that has been replaced by the lsb-release and os-release files.
2013 // Searching for the debian_version file is the last resort. It contains
2014 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2015 // "Debian " is printed before the contents of the debian_version file.
2016
2017 const char* distro_files[] = {
2018 "/etc/oracle-release",
2019 "/etc/mandriva-release",
2020 "/etc/mandrake-release",
2021 "/etc/sun-release",
2022 "/etc/redhat-release",
2023 "/etc/SuSE-release",
2024 "/etc/lsb-release",
2025 "/etc/turbolinux-release",
2026 "/etc/gentoo-release",
2027 "/etc/ltib-release",
2028 "/etc/angstrom-version",
2029 "/etc/system-release",
2030 "/etc/os-release",
2031 NULL };
2032
2033 void os::Linux::print_distro_info(outputStream* st) {
2034 for (int i = 0;; i++) {
2035 const char* file = distro_files[i];
2036 if (file == NULL) {
2037 break; // done
2038 }
2039 // If file prints, we found it.
2040 if (_print_ascii_file(file, st)) {
2041 return;
2042 }
2043 }
2044
2045 if (file_exists("/etc/debian_version")) {
2046 st->print("Debian ");
2047 _print_ascii_file("/etc/debian_version", st);
2048 } else {
2049 st->print("Linux");
2050 }
2051 st->cr();
2052 }
2053
2054 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2055 char buf[256];
2056 while (fgets(buf, sizeof(buf), fp)) {
2057 // Edit out extra stuff in expected format
2058 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2059 char* ptr = strstr(buf, "\""); // the name is in quotes
2060 if (ptr != NULL) {
2061 ptr++; // go beyond first quote
2062 char* nl = strchr(ptr, '\"');
2063 if (nl != NULL) *nl = '\0';
2064 strncpy(distro, ptr, length);
2065 } else {
2066 ptr = strstr(buf, "=");
2067 ptr++; // go beyond equals then
2068 char* nl = strchr(ptr, '\n');
2069 if (nl != NULL) *nl = '\0';
2070 strncpy(distro, ptr, length);
2071 }
2072 return;
2073 } else if (get_first_line) {
2074 char* nl = strchr(buf, '\n');
2075 if (nl != NULL) *nl = '\0';
2076 strncpy(distro, buf, length);
2077 return;
2078 }
2079 }
2080 // print last line and close
2081 char* nl = strchr(buf, '\n');
2082 if (nl != NULL) *nl = '\0';
2083 strncpy(distro, buf, length);
2084 }
2085
2086 static void parse_os_info(char* distro, size_t length, const char* file) {
2087 FILE* fp = fopen(file, "r");
2088 if (fp != NULL) {
2089 // if suse format, print out first line
2090 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2091 parse_os_info_helper(fp, distro, length, get_first_line);
2092 fclose(fp);
2093 }
2094 }
2095
2096 void os::get_summary_os_info(char* buf, size_t buflen) {
2097 for (int i = 0;; i++) {
2098 const char* file = distro_files[i];
2099 if (file == NULL) {
2100 break; // ran out of distro_files
2101 }
2102 if (file_exists(file)) {
2103 parse_os_info(buf, buflen, file);
2104 return;
2105 }
2106 }
2107 // special case for debian
2108 if (file_exists("/etc/debian_version")) {
2109 strncpy(buf, "Debian ", buflen);
2110 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2111 } else {
2112 strncpy(buf, "Linux", buflen);
2113 }
2114 }
2115
2116 void os::Linux::print_libversion_info(outputStream* st) {
2117 // libc, pthread
2118 st->print("libc:");
2119 st->print("%s ", os::Linux::glibc_version());
2120 st->print("%s ", os::Linux::libpthread_version());
2121 st->cr();
2122 }
2123
2124 void os::Linux::print_full_memory_info(outputStream* st) {
2125 st->print("\n/proc/meminfo:\n");
2126 _print_ascii_file("/proc/meminfo", st);
2127 st->cr();
2128 }
2129
2130 void os::Linux::print_container_info(outputStream* st) {
2131 if (!OSContainer::is_containerized()) {
2132 return;
2133 }
2134
2135 st->print("container (cgroup) information:\n");
2136
2137 const char *p_ct = OSContainer::container_type();
2138 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed");
2139
2140 char *p = OSContainer::cpu_cpuset_cpus();
2141 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed");
2142 free(p);
2143
2144 p = OSContainer::cpu_cpuset_memory_nodes();
2145 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed");
2146 free(p);
2147
2148 int i = OSContainer::active_processor_count();
2149 if (i > 0) {
2150 st->print("active_processor_count: %d\n", i);
2151 } else {
2152 st->print("active_processor_count: failed\n");
2153 }
2154
2155 i = OSContainer::cpu_quota();
2156 st->print("cpu_quota: %d\n", i);
2157
2158 i = OSContainer::cpu_period();
2159 st->print("cpu_period: %d\n", i);
2160
2161 i = OSContainer::cpu_shares();
2162 st->print("cpu_shares: %d\n", i);
2163
2164 jlong j = OSContainer::memory_limit_in_bytes();
2165 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2166
2167 j = OSContainer::memory_and_swap_limit_in_bytes();
2168 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2169
2170 j = OSContainer::memory_soft_limit_in_bytes();
2171 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2172
2173 j = OSContainer::OSContainer::memory_usage_in_bytes();
2174 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2175
2176 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2177 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2178 st->cr();
2179 }
2180
2181 void os::print_memory_info(outputStream* st) {
2182
2183 st->print("Memory:");
2184 st->print(" %dk page", os::vm_page_size()>>10);
2185
2186 // values in struct sysinfo are "unsigned long"
2187 struct sysinfo si;
2188 sysinfo(&si);
2189
2190 st->print(", physical " UINT64_FORMAT "k",
2191 os::physical_memory() >> 10);
2192 st->print("(" UINT64_FORMAT "k free)",
2193 os::available_memory() >> 10);
2194 st->print(", swap " UINT64_FORMAT "k",
2195 ((jlong)si.totalswap * si.mem_unit) >> 10);
2196 st->print("(" UINT64_FORMAT "k free)",
2197 ((jlong)si.freeswap * si.mem_unit) >> 10);
2198 st->cr();
2199 }
2200
2201 // Print the first "model name" line and the first "flags" line
2202 // that we find and nothing more. We assume "model name" comes
2203 // before "flags" so if we find a second "model name", then the
2204 // "flags" field is considered missing.
2205 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2206 #if defined(IA32) || defined(AMD64)
2207 // Other platforms have less repetitive cpuinfo files
2208 FILE *fp = fopen("/proc/cpuinfo", "r");
2209 if (fp) {
2210 while (!feof(fp)) {
2211 if (fgets(buf, buflen, fp)) {
2212 // Assume model name comes before flags
2213 bool model_name_printed = false;
2214 if (strstr(buf, "model name") != NULL) {
2215 if (!model_name_printed) {
2216 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2217 st->print_raw(buf);
2218 model_name_printed = true;
2219 } else {
2220 // model name printed but not flags? Odd, just return
2221 fclose(fp);
2222 return true;
2223 }
2224 }
2225 // print the flags line too
2226 if (strstr(buf, "flags") != NULL) {
2227 st->print_raw(buf);
2228 fclose(fp);
2229 return true;
2230 }
2231 }
2232 }
2233 fclose(fp);
2234 }
2235 #endif // x86 platforms
2236 return false;
2237 }
2238
2239 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2240 // Only print the model name if the platform provides this as a summary
2241 if (!print_model_name_and_flags(st, buf, buflen)) {
2242 st->print("\n/proc/cpuinfo:\n");
2243 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2244 st->print_cr(" <Not Available>");
2245 }
2246 }
2247 }
2248
2249 #if defined(AMD64) || defined(IA32) || defined(X32)
2250 const char* search_string = "model name";
2251 #elif defined(M68K)
2252 const char* search_string = "CPU";
2253 #elif defined(PPC64)
2254 const char* search_string = "cpu";
2255 #elif defined(S390)
2256 const char* search_string = "processor";
2257 #elif defined(SPARC)
2258 const char* search_string = "cpu";
2259 #else
2260 const char* search_string = "Processor";
2261 #endif
2262
2263 // Parses the cpuinfo file for string representing the model name.
2264 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2265 FILE* fp = fopen("/proc/cpuinfo", "r");
2266 if (fp != NULL) {
2267 while (!feof(fp)) {
2268 char buf[256];
2269 if (fgets(buf, sizeof(buf), fp)) {
2270 char* start = strstr(buf, search_string);
2271 if (start != NULL) {
2272 char *ptr = start + strlen(search_string);
2273 char *end = buf + strlen(buf);
2274 while (ptr != end) {
2275 // skip whitespace and colon for the rest of the name.
2276 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2277 break;
2278 }
2279 ptr++;
2280 }
2281 if (ptr != end) {
2282 // reasonable string, get rid of newline and keep the rest
2283 char* nl = strchr(buf, '\n');
2284 if (nl != NULL) *nl = '\0';
2285 strncpy(cpuinfo, ptr, length);
2286 fclose(fp);
2287 return;
2288 }
2289 }
2290 }
2291 }
2292 fclose(fp);
2293 }
2294 // cpuinfo not found or parsing failed, just print generic string. The entire
2295 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2296 #if defined(AARCH64)
2297 strncpy(cpuinfo, "AArch64", length);
2298 #elif defined(AMD64)
2299 strncpy(cpuinfo, "x86_64", length);
2300 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2301 strncpy(cpuinfo, "ARM", length);
2302 #elif defined(IA32)
2303 strncpy(cpuinfo, "x86_32", length);
2304 #elif defined(IA64)
2305 strncpy(cpuinfo, "IA64", length);
2306 #elif defined(PPC)
2307 strncpy(cpuinfo, "PPC64", length);
2308 #elif defined(S390)
2309 strncpy(cpuinfo, "S390", length);
2310 #elif defined(SPARC)
2311 strncpy(cpuinfo, "sparcv9", length);
2312 #elif defined(ZERO_LIBARCH)
2313 strncpy(cpuinfo, ZERO_LIBARCH, length);
2314 #else
2315 strncpy(cpuinfo, "unknown", length);
2316 #endif
2317 }
2318
2319 static void print_signal_handler(outputStream* st, int sig,
2320 char* buf, size_t buflen);
2321
2322 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2323 st->print_cr("Signal Handlers:");
2324 print_signal_handler(st, SIGSEGV, buf, buflen);
2325 print_signal_handler(st, SIGBUS , buf, buflen);
2326 print_signal_handler(st, SIGFPE , buf, buflen);
2327 print_signal_handler(st, SIGPIPE, buf, buflen);
2328 print_signal_handler(st, SIGXFSZ, buf, buflen);
2329 print_signal_handler(st, SIGILL , buf, buflen);
2330 print_signal_handler(st, SR_signum, buf, buflen);
2331 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2332 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2333 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2334 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2335 #if defined(PPC64)
2336 print_signal_handler(st, SIGTRAP, buf, buflen);
2337 #endif
2338 }
2339
2340 static char saved_jvm_path[MAXPATHLEN] = {0};
2341
2342 // Find the full path to the current module, libjvm.so
2343 void os::jvm_path(char *buf, jint buflen) {
2344 // Error checking.
2345 if (buflen < MAXPATHLEN) {
2346 assert(false, "must use a large-enough buffer");
2347 buf[0] = '\0';
2348 return;
2349 }
2350 // Lazy resolve the path to current module.
2351 if (saved_jvm_path[0] != 0) {
2352 strcpy(buf, saved_jvm_path);
2353 return;
2354 }
2355
2356 char dli_fname[MAXPATHLEN];
2357 bool ret = dll_address_to_library_name(
2358 CAST_FROM_FN_PTR(address, os::jvm_path),
2359 dli_fname, sizeof(dli_fname), NULL);
2360 assert(ret, "cannot locate libjvm");
2361 char *rp = NULL;
2362 if (ret && dli_fname[0] != '\0') {
2363 rp = os::Posix::realpath(dli_fname, buf, buflen);
2364 }
2365 if (rp == NULL) {
2366 return;
2367 }
2368
2369 if (Arguments::sun_java_launcher_is_altjvm()) {
2370 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2371 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2372 // If "/jre/lib/" appears at the right place in the string, then
2373 // assume we are installed in a JDK and we're done. Otherwise, check
2374 // for a JAVA_HOME environment variable and fix up the path so it
2375 // looks like libjvm.so is installed there (append a fake suffix
2376 // hotspot/libjvm.so).
2377 const char *p = buf + strlen(buf) - 1;
2378 for (int count = 0; p > buf && count < 5; ++count) {
2379 for (--p; p > buf && *p != '/'; --p)
2380 /* empty */ ;
2381 }
2382
2383 if (strncmp(p, "/jre/lib/", 9) != 0) {
2384 // Look for JAVA_HOME in the environment.
2385 char* java_home_var = ::getenv("JAVA_HOME");
2386 if (java_home_var != NULL && java_home_var[0] != 0) {
2387 char* jrelib_p;
2388 int len;
2389
2390 // Check the current module name "libjvm.so".
2391 p = strrchr(buf, '/');
2392 if (p == NULL) {
2393 return;
2394 }
2395 assert(strstr(p, "/libjvm") == p, "invalid library name");
2396
2397 rp = os::Posix::realpath(java_home_var, buf, buflen);
2398 if (rp == NULL) {
2399 return;
2400 }
2401
2402 // determine if this is a legacy image or modules image
2403 // modules image doesn't have "jre" subdirectory
2404 len = strlen(buf);
2405 assert(len < buflen, "Ran out of buffer room");
2406 jrelib_p = buf + len;
2407 snprintf(jrelib_p, buflen-len, "/jre/lib");
2408 if (0 != access(buf, F_OK)) {
2409 snprintf(jrelib_p, buflen-len, "/lib");
2410 }
2411
2412 if (0 == access(buf, F_OK)) {
2413 // Use current module name "libjvm.so"
2414 len = strlen(buf);
2415 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2416 } else {
2417 // Go back to path of .so
2418 rp = os::Posix::realpath(dli_fname, buf, buflen);
2419 if (rp == NULL) {
2420 return;
2421 }
2422 }
2423 }
2424 }
2425 }
2426
2427 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2428 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2429 }
2430
2431 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2432 // no prefix required, not even "_"
2433 }
2434
2435 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2436 // no suffix required
2437 }
2438
2439 ////////////////////////////////////////////////////////////////////////////////
2440 // sun.misc.Signal support
2441
2442 static volatile jint sigint_count = 0;
2443
2444 static void UserHandler(int sig, void *siginfo, void *context) {
2445 // 4511530 - sem_post is serialized and handled by the manager thread. When
2446 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2447 // don't want to flood the manager thread with sem_post requests.
2448 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2449 return;
2450 }
2451
2452 // Ctrl-C is pressed during error reporting, likely because the error
2453 // handler fails to abort. Let VM die immediately.
2454 if (sig == SIGINT && VMError::is_error_reported()) {
2455 os::die();
2456 }
2457
2458 os::signal_notify(sig);
2459 }
2460
2461 void* os::user_handler() {
2462 return CAST_FROM_FN_PTR(void*, UserHandler);
2463 }
2464
2465 static struct timespec create_semaphore_timespec(unsigned int sec, int nsec) {
2466 struct timespec ts;
2467 // Semaphore's are always associated with CLOCK_REALTIME
2468 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2469 // see os_posix.cpp for discussion on overflow checking
2470 if (sec >= MAX_SECS) {
2471 ts.tv_sec += MAX_SECS;
2472 ts.tv_nsec = 0;
2473 } else {
2474 ts.tv_sec += sec;
2475 ts.tv_nsec += nsec;
2476 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2477 ts.tv_nsec -= NANOSECS_PER_SEC;
2478 ++ts.tv_sec; // note: this must be <= max_secs
2479 }
2480 }
2481
2482 return ts;
2483 }
2484
2485 extern "C" {
2486 typedef void (*sa_handler_t)(int);
2487 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2488 }
2489
2490 void* os::signal(int signal_number, void* handler) {
2491 struct sigaction sigAct, oldSigAct;
2492
2493 sigfillset(&(sigAct.sa_mask));
2494 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2495 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2496
2497 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2498 // -1 means registration failed
2499 return (void *)-1;
2500 }
2501
2502 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2503 }
2504
2505 void os::signal_raise(int signal_number) {
2506 ::raise(signal_number);
2507 }
2508
2509 // The following code is moved from os.cpp for making this
2510 // code platform specific, which it is by its very nature.
2511
2512 // Will be modified when max signal is changed to be dynamic
2513 int os::sigexitnum_pd() {
2514 return NSIG;
2515 }
2516
2517 // a counter for each possible signal value
2518 static volatile jint pending_signals[NSIG+1] = { 0 };
2519
2520 // Linux(POSIX) specific hand shaking semaphore.
2521 static Semaphore* sig_sem = NULL;
2522 static PosixSemaphore sr_semaphore;
2523
2524 void os::signal_init_pd() {
2525 // Initialize signal structures
2526 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2527
2528 // Initialize signal semaphore
2529 sig_sem = new Semaphore();
2530 }
2531
2532 void os::signal_notify(int sig) {
2533 if (sig_sem != NULL) {
2534 Atomic::inc(&pending_signals[sig]);
2535 sig_sem->signal();
2536 } else {
2537 // Signal thread is not created with ReduceSignalUsage and signal_init_pd
2538 // initialization isn't called.
2539 assert(ReduceSignalUsage, "signal semaphore should be created");
2540 }
2541 }
2542
2543 static int check_pending_signals() {
2544 Atomic::store(0, &sigint_count);
2545 for (;;) {
2546 for (int i = 0; i < NSIG + 1; i++) {
2547 jint n = pending_signals[i];
2548 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2549 return i;
2550 }
2551 }
2552 JavaThread *thread = JavaThread::current();
2553 ThreadBlockInVM tbivm(thread);
2554
2555 bool threadIsSuspended;
2556 do {
2557 thread->set_suspend_equivalent();
2558 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2559 sig_sem->wait();
2560
2561 // were we externally suspended while we were waiting?
2562 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2563 if (threadIsSuspended) {
2564 // The semaphore has been incremented, but while we were waiting
2565 // another thread suspended us. We don't want to continue running
2566 // while suspended because that would surprise the thread that
2567 // suspended us.
2568 sig_sem->signal();
2569
2570 thread->java_suspend_self();
2571 }
2572 } while (threadIsSuspended);
2573 }
2574 }
2575
2576 int os::signal_wait() {
2577 return check_pending_signals();
2578 }
2579
2580 ////////////////////////////////////////////////////////////////////////////////
2581 // Virtual Memory
2582
2583 int os::vm_page_size() {
2584 // Seems redundant as all get out
2585 assert(os::Linux::page_size() != -1, "must call os::init");
2586 return os::Linux::page_size();
2587 }
2588
2589 // Solaris allocates memory by pages.
2590 int os::vm_allocation_granularity() {
2591 assert(os::Linux::page_size() != -1, "must call os::init");
2592 return os::Linux::page_size();
2593 }
2594
2595 // Rationale behind this function:
2596 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2597 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2598 // samples for JITted code. Here we create private executable mapping over the code cache
2599 // and then we can use standard (well, almost, as mapping can change) way to provide
2600 // info for the reporting script by storing timestamp and location of symbol
2601 void linux_wrap_code(char* base, size_t size) {
2602 static volatile jint cnt = 0;
2603
2604 if (!UseOprofile) {
2605 return;
2606 }
2607
2608 char buf[PATH_MAX+1];
2609 int num = Atomic::add(1, &cnt);
2610
2611 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2612 os::get_temp_directory(), os::current_process_id(), num);
2613 unlink(buf);
2614
2615 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2616
2617 if (fd != -1) {
2618 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2619 if (rv != (off_t)-1) {
2620 if (::write(fd, "", 1) == 1) {
2621 mmap(base, size,
2622 PROT_READ|PROT_WRITE|PROT_EXEC,
2623 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2624 }
2625 }
2626 ::close(fd);
2627 unlink(buf);
2628 }
2629 }
2630
2631 static bool recoverable_mmap_error(int err) {
2632 // See if the error is one we can let the caller handle. This
2633 // list of errno values comes from JBS-6843484. I can't find a
2634 // Linux man page that documents this specific set of errno
2635 // values so while this list currently matches Solaris, it may
2636 // change as we gain experience with this failure mode.
2637 switch (err) {
2638 case EBADF:
2639 case EINVAL:
2640 case ENOTSUP:
2641 // let the caller deal with these errors
2642 return true;
2643
2644 default:
2645 // Any remaining errors on this OS can cause our reserved mapping
2646 // to be lost. That can cause confusion where different data
2647 // structures think they have the same memory mapped. The worst
2648 // scenario is if both the VM and a library think they have the
2649 // same memory mapped.
2650 return false;
2651 }
2652 }
2653
2654 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2655 int err) {
2656 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2657 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2658 os::strerror(err), err);
2659 }
2660
2661 static void warn_fail_commit_memory(char* addr, size_t size,
2662 size_t alignment_hint, bool exec,
2663 int err) {
2664 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2665 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2666 alignment_hint, exec, os::strerror(err), err);
2667 }
2668
2669 // NOTE: Linux kernel does not really reserve the pages for us.
2670 // All it does is to check if there are enough free pages
2671 // left at the time of mmap(). This could be a potential
2672 // problem.
2673 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2674 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2675 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2676 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2677 if (res != (uintptr_t) MAP_FAILED) {
2678 if (UseNUMAInterleaving) {
2679 numa_make_global(addr, size);
2680 }
2681 return 0;
2682 }
2683
2684 int err = errno; // save errno from mmap() call above
2685
2686 if (!recoverable_mmap_error(err)) {
2687 warn_fail_commit_memory(addr, size, exec, err);
2688 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2689 }
2690
2691 return err;
2692 }
2693
2694 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2695 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2696 }
2697
2698 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2699 const char* mesg) {
2700 assert(mesg != NULL, "mesg must be specified");
2701 int err = os::Linux::commit_memory_impl(addr, size, exec);
2702 if (err != 0) {
2703 // the caller wants all commit errors to exit with the specified mesg:
2704 warn_fail_commit_memory(addr, size, exec, err);
2705 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2706 }
2707 }
2708
2709 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2710 #ifndef MAP_HUGETLB
2711 #define MAP_HUGETLB 0x40000
2712 #endif
2713
2714 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2715 #ifndef MADV_HUGEPAGE
2716 #define MADV_HUGEPAGE 14
2717 #endif
2718
2719 int os::Linux::commit_memory_impl(char* addr, size_t size,
2720 size_t alignment_hint, bool exec) {
2721 int err = os::Linux::commit_memory_impl(addr, size, exec);
2722 if (err == 0) {
2723 realign_memory(addr, size, alignment_hint);
2724 }
2725 return err;
2726 }
2727
2728 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2729 bool exec) {
2730 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2731 }
2732
2733 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2734 size_t alignment_hint, bool exec,
2735 const char* mesg) {
2736 assert(mesg != NULL, "mesg must be specified");
2737 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2738 if (err != 0) {
2739 // the caller wants all commit errors to exit with the specified mesg:
2740 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2741 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2742 }
2743 }
2744
2745 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2746 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2747 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2748 // be supported or the memory may already be backed by huge pages.
2749 ::madvise(addr, bytes, MADV_HUGEPAGE);
2750 }
2751 }
2752
2753 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2754 // This method works by doing an mmap over an existing mmaping and effectively discarding
2755 // the existing pages. However it won't work for SHM-based large pages that cannot be
2756 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2757 // small pages on top of the SHM segment. This method always works for small pages, so we
2758 // allow that in any case.
2759 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2760 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2761 }
2762 }
2763
2764 void os::numa_make_global(char *addr, size_t bytes) {
2765 Linux::numa_interleave_memory(addr, bytes);
2766 }
2767
2768 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2769 // bind policy to MPOL_PREFERRED for the current thread.
2770 #define USE_MPOL_PREFERRED 0
2771
2772 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2773 // To make NUMA and large pages more robust when both enabled, we need to ease
2774 // the requirements on where the memory should be allocated. MPOL_BIND is the
2775 // default policy and it will force memory to be allocated on the specified
2776 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2777 // the specified node, but will not force it. Using this policy will prevent
2778 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2779 // free large pages.
2780 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2781 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2782 }
2783
2784 bool os::numa_topology_changed() { return false; }
2785
2786 size_t os::numa_get_groups_num() {
2787 // Return just the number of nodes in which it's possible to allocate memory
2788 // (in numa terminology, configured nodes).
2789 return Linux::numa_num_configured_nodes();
2790 }
2791
2792 int os::numa_get_group_id() {
2793 int cpu_id = Linux::sched_getcpu();
2794 if (cpu_id != -1) {
2795 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2796 if (lgrp_id != -1) {
2797 return lgrp_id;
2798 }
2799 }
2800 return 0;
2801 }
2802
2803 int os::Linux::get_existing_num_nodes() {
2804 size_t node;
2805 size_t highest_node_number = Linux::numa_max_node();
2806 int num_nodes = 0;
2807
2808 // Get the total number of nodes in the system including nodes without memory.
2809 for (node = 0; node <= highest_node_number; node++) {
2810 if (isnode_in_existing_nodes(node)) {
2811 num_nodes++;
2812 }
2813 }
2814 return num_nodes;
2815 }
2816
2817 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2818 size_t highest_node_number = Linux::numa_max_node();
2819 size_t i = 0;
2820
2821 // Map all node ids in which is possible to allocate memory. Also nodes are
2822 // not always consecutively available, i.e. available from 0 to the highest
2823 // node number.
2824 for (size_t node = 0; node <= highest_node_number; node++) {
2825 if (Linux::isnode_in_configured_nodes(node)) {
2826 ids[i++] = node;
2827 }
2828 }
2829 return i;
2830 }
2831
2832 bool os::get_page_info(char *start, page_info* info) {
2833 return false;
2834 }
2835
2836 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2837 page_info* page_found) {
2838 return end;
2839 }
2840
2841
2842 int os::Linux::sched_getcpu_syscall(void) {
2843 unsigned int cpu = 0;
2844 int retval = -1;
2845
2846 #if defined(IA32)
2847 #ifndef SYS_getcpu
2848 #define SYS_getcpu 318
2849 #endif
2850 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2851 #elif defined(AMD64)
2852 // Unfortunately we have to bring all these macros here from vsyscall.h
2853 // to be able to compile on old linuxes.
2854 #define __NR_vgetcpu 2
2855 #define VSYSCALL_START (-10UL << 20)
2856 #define VSYSCALL_SIZE 1024
2857 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2858 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2859 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2860 retval = vgetcpu(&cpu, NULL, NULL);
2861 #endif
2862
2863 return (retval == -1) ? retval : cpu;
2864 }
2865
2866 void os::Linux::sched_getcpu_init() {
2867 // sched_getcpu() should be in libc.
2868 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2869 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2870
2871 // If it's not, try a direct syscall.
2872 if (sched_getcpu() == -1) {
2873 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2874 (void*)&sched_getcpu_syscall));
2875 }
2876 }
2877
2878 // Something to do with the numa-aware allocator needs these symbols
2879 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2880 extern "C" JNIEXPORT void numa_error(char *where) { }
2881
2882 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2883 // load symbol from base version instead.
2884 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2885 void *f = dlvsym(handle, name, "libnuma_1.1");
2886 if (f == NULL) {
2887 f = dlsym(handle, name);
2888 }
2889 return f;
2890 }
2891
2892 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2893 // Return NULL if the symbol is not defined in this particular version.
2894 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2895 return dlvsym(handle, name, "libnuma_1.2");
2896 }
2897
2898 bool os::Linux::libnuma_init() {
2899 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
2900 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2901 if (handle != NULL) {
2902 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2903 libnuma_dlsym(handle, "numa_node_to_cpus")));
2904 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2905 libnuma_dlsym(handle, "numa_max_node")));
2906 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2907 libnuma_dlsym(handle, "numa_num_configured_nodes")));
2908 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2909 libnuma_dlsym(handle, "numa_available")));
2910 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2911 libnuma_dlsym(handle, "numa_tonode_memory")));
2912 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2913 libnuma_dlsym(handle, "numa_interleave_memory")));
2914 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2915 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2916 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2917 libnuma_dlsym(handle, "numa_set_bind_policy")));
2918 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2919 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2920 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2921 libnuma_dlsym(handle, "numa_distance")));
2922
2923 if (numa_available() != -1) {
2924 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2925 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2926 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2927 // Create an index -> node mapping, since nodes are not always consecutive
2928 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2929 rebuild_nindex_to_node_map();
2930 // Create a cpu -> node mapping
2931 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2932 rebuild_cpu_to_node_map();
2933 return true;
2934 }
2935 }
2936 }
2937 return false;
2938 }
2939
2940 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2941 // Creating guard page is very expensive. Java thread has HotSpot
2942 // guard pages, only enable glibc guard page for non-Java threads.
2943 // (Remember: compiler thread is a Java thread, too!)
2944 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2945 }
2946
2947 void os::Linux::rebuild_nindex_to_node_map() {
2948 int highest_node_number = Linux::numa_max_node();
2949
2950 nindex_to_node()->clear();
2951 for (int node = 0; node <= highest_node_number; node++) {
2952 if (Linux::isnode_in_existing_nodes(node)) {
2953 nindex_to_node()->append(node);
2954 }
2955 }
2956 }
2957
2958 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2959 // The table is later used in get_node_by_cpu().
2960 void os::Linux::rebuild_cpu_to_node_map() {
2961 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2962 // in libnuma (possible values are starting from 16,
2963 // and continuing up with every other power of 2, but less
2964 // than the maximum number of CPUs supported by kernel), and
2965 // is a subject to change (in libnuma version 2 the requirements
2966 // are more reasonable) we'll just hardcode the number they use
2967 // in the library.
2968 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2969
2970 size_t cpu_num = processor_count();
2971 size_t cpu_map_size = NCPUS / BitsPerCLong;
2972 size_t cpu_map_valid_size =
2973 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2974
2975 cpu_to_node()->clear();
2976 cpu_to_node()->at_grow(cpu_num - 1);
2977
2978 size_t node_num = get_existing_num_nodes();
2979
2980 int distance = 0;
2981 int closest_distance = INT_MAX;
2982 int closest_node = 0;
2983 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2984 for (size_t i = 0; i < node_num; i++) {
2985 // Check if node is configured (not a memory-less node). If it is not, find
2986 // the closest configured node.
2987 if (!isnode_in_configured_nodes(nindex_to_node()->at(i))) {
2988 closest_distance = INT_MAX;
2989 // Check distance from all remaining nodes in the system. Ignore distance
2990 // from itself and from another non-configured node.
2991 for (size_t m = 0; m < node_num; m++) {
2992 if (m != i && isnode_in_configured_nodes(nindex_to_node()->at(m))) {
2993 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
2994 // If a closest node is found, update. There is always at least one
2995 // configured node in the system so there is always at least one node
2996 // close.
2997 if (distance != 0 && distance < closest_distance) {
2998 closest_distance = distance;
2999 closest_node = nindex_to_node()->at(m);
3000 }
3001 }
3002 }
3003 } else {
3004 // Current node is already a configured node.
3005 closest_node = nindex_to_node()->at(i);
3006 }
3007
3008 // Get cpus from the original node and map them to the closest node. If node
3009 // is a configured node (not a memory-less node), then original node and
3010 // closest node are the same.
3011 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3012 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3013 if (cpu_map[j] != 0) {
3014 for (size_t k = 0; k < BitsPerCLong; k++) {
3015 if (cpu_map[j] & (1UL << k)) {
3016 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3017 }
3018 }
3019 }
3020 }
3021 }
3022 }
3023 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3024 }
3025
3026 int os::Linux::get_node_by_cpu(int cpu_id) {
3027 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3028 return cpu_to_node()->at(cpu_id);
3029 }
3030 return -1;
3031 }
3032
3033 GrowableArray<int>* os::Linux::_cpu_to_node;
3034 GrowableArray<int>* os::Linux::_nindex_to_node;
3035 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3036 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3037 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3038 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3039 os::Linux::numa_available_func_t os::Linux::_numa_available;
3040 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3041 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3042 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3043 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3044 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3045 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3046 unsigned long* os::Linux::_numa_all_nodes;
3047 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3048 struct bitmask* os::Linux::_numa_nodes_ptr;
3049
3050 bool os::pd_uncommit_memory(char* addr, size_t size) {
3051 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3052 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3053 return res != (uintptr_t) MAP_FAILED;
3054 }
3055
3056 static address get_stack_commited_bottom(address bottom, size_t size) {
3057 address nbot = bottom;
3058 address ntop = bottom + size;
3059
3060 size_t page_sz = os::vm_page_size();
3061 unsigned pages = size / page_sz;
3062
3063 unsigned char vec[1];
3064 unsigned imin = 1, imax = pages + 1, imid;
3065 int mincore_return_value = 0;
3066
3067 assert(imin <= imax, "Unexpected page size");
3068
3069 while (imin < imax) {
3070 imid = (imax + imin) / 2;
3071 nbot = ntop - (imid * page_sz);
3072
3073 // Use a trick with mincore to check whether the page is mapped or not.
3074 // mincore sets vec to 1 if page resides in memory and to 0 if page
3075 // is swapped output but if page we are asking for is unmapped
3076 // it returns -1,ENOMEM
3077 mincore_return_value = mincore(nbot, page_sz, vec);
3078
3079 if (mincore_return_value == -1) {
3080 // Page is not mapped go up
3081 // to find first mapped page
3082 if (errno != EAGAIN) {
3083 assert(errno == ENOMEM, "Unexpected mincore errno");
3084 imax = imid;
3085 }
3086 } else {
3087 // Page is mapped go down
3088 // to find first not mapped page
3089 imin = imid + 1;
3090 }
3091 }
3092
3093 nbot = nbot + page_sz;
3094
3095 // Adjust stack bottom one page up if last checked page is not mapped
3096 if (mincore_return_value == -1) {
3097 nbot = nbot + page_sz;
3098 }
3099
3100 return nbot;
3101 }
3102
3103 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3104 int mincore_return_value;
3105 const size_t stripe = 1024; // query this many pages each time
3106 unsigned char vec[stripe];
3107 const size_t page_sz = os::vm_page_size();
3108 size_t pages = size / page_sz;
3109
3110 assert(is_aligned(start, page_sz), "Start address must be page aligned");
3111 assert(is_aligned(size, page_sz), "Size must be page aligned");
3112
3113 committed_start = NULL;
3114
3115 int loops = (pages + stripe - 1) / stripe;
3116 int committed_pages = 0;
3117 address loop_base = start;
3118 for (int index = 0; index < loops; index ++) {
3119 assert(pages > 0, "Nothing to do");
3120 int pages_to_query = (pages >= stripe) ? stripe : pages;
3121 pages -= pages_to_query;
3122
3123 // Get stable read
3124 while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3125
3126 // During shutdown, some memory goes away without properly notifying NMT,
3127 // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3128 // Bailout and return as not committed for now.
3129 if (mincore_return_value == -1 && errno == ENOMEM) {
3130 return false;
3131 }
3132
3133 assert(mincore_return_value == 0, "Range must be valid");
3134 // Process this stripe
3135 for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3136 if ((vec[vecIdx] & 0x01) == 0) { // not committed
3137 // End of current contiguous region
3138 if (committed_start != NULL) {
3139 break;
3140 }
3141 } else { // committed
3142 // Start of region
3143 if (committed_start == NULL) {
3144 committed_start = loop_base + page_sz * vecIdx;
3145 }
3146 committed_pages ++;
3147 }
3148 }
3149
3150 loop_base += pages_to_query * page_sz;
3151 }
3152
3153 if (committed_start != NULL) {
3154 assert(committed_pages > 0, "Must have committed region");
3155 assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3156 assert(committed_start >= start && committed_start < start + size, "Out of range");
3157 committed_size = page_sz * committed_pages;
3158 return true;
3159 } else {
3160 assert(committed_pages == 0, "Should not have committed region");
3161 return false;
3162 }
3163 }
3164
3165
3166 // Linux uses a growable mapping for the stack, and if the mapping for
3167 // the stack guard pages is not removed when we detach a thread the
3168 // stack cannot grow beyond the pages where the stack guard was
3169 // mapped. If at some point later in the process the stack expands to
3170 // that point, the Linux kernel cannot expand the stack any further
3171 // because the guard pages are in the way, and a segfault occurs.
3172 //
3173 // However, it's essential not to split the stack region by unmapping
3174 // a region (leaving a hole) that's already part of the stack mapping,
3175 // so if the stack mapping has already grown beyond the guard pages at
3176 // the time we create them, we have to truncate the stack mapping.
3177 // So, we need to know the extent of the stack mapping when
3178 // create_stack_guard_pages() is called.
3179
3180 // We only need this for stacks that are growable: at the time of
3181 // writing thread stacks don't use growable mappings (i.e. those
3182 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3183 // only applies to the main thread.
3184
3185 // If the (growable) stack mapping already extends beyond the point
3186 // where we're going to put our guard pages, truncate the mapping at
3187 // that point by munmap()ping it. This ensures that when we later
3188 // munmap() the guard pages we don't leave a hole in the stack
3189 // mapping. This only affects the main/primordial thread
3190
3191 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3192 if (os::is_primordial_thread()) {
3193 // As we manually grow stack up to bottom inside create_attached_thread(),
3194 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3195 // we don't need to do anything special.
3196 // Check it first, before calling heavy function.
3197 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3198 unsigned char vec[1];
3199
3200 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3201 // Fallback to slow path on all errors, including EAGAIN
3202 stack_extent = (uintptr_t) get_stack_commited_bottom(
3203 os::Linux::initial_thread_stack_bottom(),
3204 (size_t)addr - stack_extent);
3205 }
3206
3207 if (stack_extent < (uintptr_t)addr) {
3208 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3209 }
3210 }
3211
3212 return os::commit_memory(addr, size, !ExecMem);
3213 }
3214
3215 // If this is a growable mapping, remove the guard pages entirely by
3216 // munmap()ping them. If not, just call uncommit_memory(). This only
3217 // affects the main/primordial thread, but guard against future OS changes.
3218 // It's safe to always unmap guard pages for primordial thread because we
3219 // always place it right after end of the mapped region.
3220
3221 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3222 uintptr_t stack_extent, stack_base;
3223
3224 if (os::is_primordial_thread()) {
3225 return ::munmap(addr, size) == 0;
3226 }
3227
3228 return os::uncommit_memory(addr, size);
3229 }
3230
3231 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3232 // at 'requested_addr'. If there are existing memory mappings at the same
3233 // location, however, they will be overwritten. If 'fixed' is false,
3234 // 'requested_addr' is only treated as a hint, the return value may or
3235 // may not start from the requested address. Unlike Linux mmap(), this
3236 // function returns NULL to indicate failure.
3237 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3238 char * addr;
3239 int flags;
3240
3241 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3242 if (fixed) {
3243 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3244 flags |= MAP_FIXED;
3245 }
3246
3247 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3248 // touch an uncommitted page. Otherwise, the read/write might
3249 // succeed if we have enough swap space to back the physical page.
3250 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3251 flags, -1, 0);
3252
3253 return addr == MAP_FAILED ? NULL : addr;
3254 }
3255
3256 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3257 // (req_addr != NULL) or with a given alignment.
3258 // - bytes shall be a multiple of alignment.
3259 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3260 // - alignment sets the alignment at which memory shall be allocated.
3261 // It must be a multiple of allocation granularity.
3262 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3263 // req_addr or NULL.
3264 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3265
3266 size_t extra_size = bytes;
3267 if (req_addr == NULL && alignment > 0) {
3268 extra_size += alignment;
3269 }
3270
3271 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3272 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3273 -1, 0);
3274 if (start == MAP_FAILED) {
3275 start = NULL;
3276 } else {
3277 if (req_addr != NULL) {
3278 if (start != req_addr) {
3279 ::munmap(start, extra_size);
3280 start = NULL;
3281 }
3282 } else {
3283 char* const start_aligned = align_up(start, alignment);
3284 char* const end_aligned = start_aligned + bytes;
3285 char* const end = start + extra_size;
3286 if (start_aligned > start) {
3287 ::munmap(start, start_aligned - start);
3288 }
3289 if (end_aligned < end) {
3290 ::munmap(end_aligned, end - end_aligned);
3291 }
3292 start = start_aligned;
3293 }
3294 }
3295 return start;
3296 }
3297
3298 static int anon_munmap(char * addr, size_t size) {
3299 return ::munmap(addr, size) == 0;
3300 }
3301
3302 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3303 size_t alignment_hint) {
3304 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3305 }
3306
3307 bool os::pd_release_memory(char* addr, size_t size) {
3308 return anon_munmap(addr, size);
3309 }
3310
3311 static bool linux_mprotect(char* addr, size_t size, int prot) {
3312 // Linux wants the mprotect address argument to be page aligned.
3313 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3314
3315 // According to SUSv3, mprotect() should only be used with mappings
3316 // established by mmap(), and mmap() always maps whole pages. Unaligned
3317 // 'addr' likely indicates problem in the VM (e.g. trying to change
3318 // protection of malloc'ed or statically allocated memory). Check the
3319 // caller if you hit this assert.
3320 assert(addr == bottom, "sanity check");
3321
3322 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3323 return ::mprotect(bottom, size, prot) == 0;
3324 }
3325
3326 // Set protections specified
3327 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3328 bool is_committed) {
3329 unsigned int p = 0;
3330 switch (prot) {
3331 case MEM_PROT_NONE: p = PROT_NONE; break;
3332 case MEM_PROT_READ: p = PROT_READ; break;
3333 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3334 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3335 default:
3336 ShouldNotReachHere();
3337 }
3338 // is_committed is unused.
3339 return linux_mprotect(addr, bytes, p);
3340 }
3341
3342 bool os::guard_memory(char* addr, size_t size) {
3343 return linux_mprotect(addr, size, PROT_NONE);
3344 }
3345
3346 bool os::unguard_memory(char* addr, size_t size) {
3347 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3348 }
3349
3350 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3351 size_t page_size) {
3352 bool result = false;
3353 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3354 MAP_ANONYMOUS|MAP_PRIVATE,
3355 -1, 0);
3356 if (p != MAP_FAILED) {
3357 void *aligned_p = align_up(p, page_size);
3358
3359 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3360
3361 munmap(p, page_size * 2);
3362 }
3363
3364 if (warn && !result) {
3365 warning("TransparentHugePages is not supported by the operating system.");
3366 }
3367
3368 return result;
3369 }
3370
3371 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3372 bool result = false;
3373 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3374 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3375 -1, 0);
3376
3377 if (p != MAP_FAILED) {
3378 // We don't know if this really is a huge page or not.
3379 FILE *fp = fopen("/proc/self/maps", "r");
3380 if (fp) {
3381 while (!feof(fp)) {
3382 char chars[257];
3383 long x = 0;
3384 if (fgets(chars, sizeof(chars), fp)) {
3385 if (sscanf(chars, "%lx-%*x", &x) == 1
3386 && x == (long)p) {
3387 if (strstr (chars, "hugepage")) {
3388 result = true;
3389 break;
3390 }
3391 }
3392 }
3393 }
3394 fclose(fp);
3395 }
3396 munmap(p, page_size);
3397 }
3398
3399 if (warn && !result) {
3400 warning("HugeTLBFS is not supported by the operating system.");
3401 }
3402
3403 return result;
3404 }
3405
3406 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3407 //
3408 // From the coredump_filter documentation:
3409 //
3410 // - (bit 0) anonymous private memory
3411 // - (bit 1) anonymous shared memory
3412 // - (bit 2) file-backed private memory
3413 // - (bit 3) file-backed shared memory
3414 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3415 // effective only if the bit 2 is cleared)
3416 // - (bit 5) hugetlb private memory
3417 // - (bit 6) hugetlb shared memory
3418 // - (bit 7) dax private memory
3419 // - (bit 8) dax shared memory
3420 //
3421 static void set_coredump_filter(bool largepages, bool dax_shared) {
3422 FILE *f;
3423 long cdm;
3424 bool filter_changed = false;
3425
3426 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3427 return;
3428 }
3429
3430 if (fscanf(f, "%lx", &cdm) != 1) {
3431 fclose(f);
3432 return;
3433 }
3434
3435 rewind(f);
3436
3437 if (largepages && (cdm & LARGEPAGES_BIT) == 0) {
3438 cdm |= LARGEPAGES_BIT;
3439 filter_changed = true;
3440 }
3441 if (dax_shared && (cdm & DAX_SHARED_BIT) == 0) {
3442 cdm |= DAX_SHARED_BIT;
3443 filter_changed = true;
3444 }
3445 if (filter_changed) {
3446 fprintf(f, "%#lx", cdm);
3447 }
3448
3449 fclose(f);
3450 }
3451
3452 // Large page support
3453
3454 static size_t _large_page_size = 0;
3455
3456 size_t os::Linux::find_large_page_size() {
3457 size_t large_page_size = 0;
3458
3459 // large_page_size on Linux is used to round up heap size. x86 uses either
3460 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3461 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3462 // page as large as 256M.
3463 //
3464 // Here we try to figure out page size by parsing /proc/meminfo and looking
3465 // for a line with the following format:
3466 // Hugepagesize: 2048 kB
3467 //
3468 // If we can't determine the value (e.g. /proc is not mounted, or the text
3469 // format has been changed), we'll use the largest page size supported by
3470 // the processor.
3471
3472 #ifndef ZERO
3473 large_page_size =
3474 AARCH64_ONLY(2 * M)
3475 AMD64_ONLY(2 * M)
3476 ARM32_ONLY(2 * M)
3477 IA32_ONLY(4 * M)
3478 IA64_ONLY(256 * M)
3479 PPC_ONLY(4 * M)
3480 S390_ONLY(1 * M)
3481 SPARC_ONLY(4 * M);
3482 #endif // ZERO
3483
3484 FILE *fp = fopen("/proc/meminfo", "r");
3485 if (fp) {
3486 while (!feof(fp)) {
3487 int x = 0;
3488 char buf[16];
3489 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3490 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3491 large_page_size = x * K;
3492 break;
3493 }
3494 } else {
3495 // skip to next line
3496 for (;;) {
3497 int ch = fgetc(fp);
3498 if (ch == EOF || ch == (int)'\n') break;
3499 }
3500 }
3501 }
3502 fclose(fp);
3503 }
3504
3505 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3506 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3507 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3508 proper_unit_for_byte_size(large_page_size));
3509 }
3510
3511 return large_page_size;
3512 }
3513
3514 size_t os::Linux::setup_large_page_size() {
3515 _large_page_size = Linux::find_large_page_size();
3516 const size_t default_page_size = (size_t)Linux::page_size();
3517 if (_large_page_size > default_page_size) {
3518 _page_sizes[0] = _large_page_size;
3519 _page_sizes[1] = default_page_size;
3520 _page_sizes[2] = 0;
3521 }
3522
3523 return _large_page_size;
3524 }
3525
3526 bool os::Linux::setup_large_page_type(size_t page_size) {
3527 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3528 FLAG_IS_DEFAULT(UseSHM) &&
3529 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3530
3531 // The type of large pages has not been specified by the user.
3532
3533 // Try UseHugeTLBFS and then UseSHM.
3534 UseHugeTLBFS = UseSHM = true;
3535
3536 // Don't try UseTransparentHugePages since there are known
3537 // performance issues with it turned on. This might change in the future.
3538 UseTransparentHugePages = false;
3539 }
3540
3541 if (UseTransparentHugePages) {
3542 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3543 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3544 UseHugeTLBFS = false;
3545 UseSHM = false;
3546 return true;
3547 }
3548 UseTransparentHugePages = false;
3549 }
3550
3551 if (UseHugeTLBFS) {
3552 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3553 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3554 UseSHM = false;
3555 return true;
3556 }
3557 UseHugeTLBFS = false;
3558 }
3559
3560 return UseSHM;
3561 }
3562
3563 void os::large_page_init() {
3564 if (!UseLargePages &&
3565 !UseTransparentHugePages &&
3566 !UseHugeTLBFS &&
3567 !UseSHM) {
3568 // Not using large pages.
3569 return;
3570 }
3571
3572 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3573 // The user explicitly turned off large pages.
3574 // Ignore the rest of the large pages flags.
3575 UseTransparentHugePages = false;
3576 UseHugeTLBFS = false;
3577 UseSHM = false;
3578 return;
3579 }
3580
3581 size_t large_page_size = Linux::setup_large_page_size();
3582 UseLargePages = Linux::setup_large_page_type(large_page_size);
3583
3584 set_coredump_filter(true /*largepages*/, false /*dax_shared*/);
3585 }
3586
3587 #ifndef SHM_HUGETLB
3588 #define SHM_HUGETLB 04000
3589 #endif
3590
3591 #define shm_warning_format(format, ...) \
3592 do { \
3593 if (UseLargePages && \
3594 (!FLAG_IS_DEFAULT(UseLargePages) || \
3595 !FLAG_IS_DEFAULT(UseSHM) || \
3596 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3597 warning(format, __VA_ARGS__); \
3598 } \
3599 } while (0)
3600
3601 #define shm_warning(str) shm_warning_format("%s", str)
3602
3603 #define shm_warning_with_errno(str) \
3604 do { \
3605 int err = errno; \
3606 shm_warning_format(str " (error = %d)", err); \
3607 } while (0)
3608
3609 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3610 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3611
3612 if (!is_aligned(alignment, SHMLBA)) {
3613 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3614 return NULL;
3615 }
3616
3617 // To ensure that we get 'alignment' aligned memory from shmat,
3618 // we pre-reserve aligned virtual memory and then attach to that.
3619
3620 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3621 if (pre_reserved_addr == NULL) {
3622 // Couldn't pre-reserve aligned memory.
3623 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3624 return NULL;
3625 }
3626
3627 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3628 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3629
3630 if ((intptr_t)addr == -1) {
3631 int err = errno;
3632 shm_warning_with_errno("Failed to attach shared memory.");
3633
3634 assert(err != EACCES, "Unexpected error");
3635 assert(err != EIDRM, "Unexpected error");
3636 assert(err != EINVAL, "Unexpected error");
3637
3638 // Since we don't know if the kernel unmapped the pre-reserved memory area
3639 // we can't unmap it, since that would potentially unmap memory that was
3640 // mapped from other threads.
3641 return NULL;
3642 }
3643
3644 return addr;
3645 }
3646
3647 static char* shmat_at_address(int shmid, char* req_addr) {
3648 if (!is_aligned(req_addr, SHMLBA)) {
3649 assert(false, "Requested address needs to be SHMLBA aligned");
3650 return NULL;
3651 }
3652
3653 char* addr = (char*)shmat(shmid, req_addr, 0);
3654
3655 if ((intptr_t)addr == -1) {
3656 shm_warning_with_errno("Failed to attach shared memory.");
3657 return NULL;
3658 }
3659
3660 return addr;
3661 }
3662
3663 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3664 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3665 if (req_addr != NULL) {
3666 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3667 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3668 return shmat_at_address(shmid, req_addr);
3669 }
3670
3671 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3672 // return large page size aligned memory addresses when req_addr == NULL.
3673 // However, if the alignment is larger than the large page size, we have
3674 // to manually ensure that the memory returned is 'alignment' aligned.
3675 if (alignment > os::large_page_size()) {
3676 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3677 return shmat_with_alignment(shmid, bytes, alignment);
3678 } else {
3679 return shmat_at_address(shmid, NULL);
3680 }
3681 }
3682
3683 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3684 char* req_addr, bool exec) {
3685 // "exec" is passed in but not used. Creating the shared image for
3686 // the code cache doesn't have an SHM_X executable permission to check.
3687 assert(UseLargePages && UseSHM, "only for SHM large pages");
3688 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3689 assert(is_aligned(req_addr, alignment), "Unaligned address");
3690
3691 if (!is_aligned(bytes, os::large_page_size())) {
3692 return NULL; // Fallback to small pages.
3693 }
3694
3695 // Create a large shared memory region to attach to based on size.
3696 // Currently, size is the total size of the heap.
3697 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3698 if (shmid == -1) {
3699 // Possible reasons for shmget failure:
3700 // 1. shmmax is too small for Java heap.
3701 // > check shmmax value: cat /proc/sys/kernel/shmmax
3702 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3703 // 2. not enough large page memory.
3704 // > check available large pages: cat /proc/meminfo
3705 // > increase amount of large pages:
3706 // echo new_value > /proc/sys/vm/nr_hugepages
3707 // Note 1: different Linux may use different name for this property,
3708 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3709 // Note 2: it's possible there's enough physical memory available but
3710 // they are so fragmented after a long run that they can't
3711 // coalesce into large pages. Try to reserve large pages when
3712 // the system is still "fresh".
3713 shm_warning_with_errno("Failed to reserve shared memory.");
3714 return NULL;
3715 }
3716
3717 // Attach to the region.
3718 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3719
3720 // Remove shmid. If shmat() is successful, the actual shared memory segment
3721 // will be deleted when it's detached by shmdt() or when the process
3722 // terminates. If shmat() is not successful this will remove the shared
3723 // segment immediately.
3724 shmctl(shmid, IPC_RMID, NULL);
3725
3726 return addr;
3727 }
3728
3729 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3730 int error) {
3731 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3732
3733 bool warn_on_failure = UseLargePages &&
3734 (!FLAG_IS_DEFAULT(UseLargePages) ||
3735 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3736 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3737
3738 if (warn_on_failure) {
3739 char msg[128];
3740 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3741 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3742 warning("%s", msg);
3743 }
3744 }
3745
3746 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3747 char* req_addr,
3748 bool exec) {
3749 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3750 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3751 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3752
3753 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3754 char* addr = (char*)::mmap(req_addr, bytes, prot,
3755 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3756 -1, 0);
3757
3758 if (addr == MAP_FAILED) {
3759 warn_on_large_pages_failure(req_addr, bytes, errno);
3760 return NULL;
3761 }
3762
3763 assert(is_aligned(addr, os::large_page_size()), "Must be");
3764
3765 return addr;
3766 }
3767
3768 // Reserve memory using mmap(MAP_HUGETLB).
3769 // - bytes shall be a multiple of alignment.
3770 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3771 // - alignment sets the alignment at which memory shall be allocated.
3772 // It must be a multiple of allocation granularity.
3773 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3774 // req_addr or NULL.
3775 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3776 size_t alignment,
3777 char* req_addr,
3778 bool exec) {
3779 size_t large_page_size = os::large_page_size();
3780 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3781
3782 assert(is_aligned(req_addr, alignment), "Must be");
3783 assert(is_aligned(bytes, alignment), "Must be");
3784
3785 // First reserve - but not commit - the address range in small pages.
3786 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3787
3788 if (start == NULL) {
3789 return NULL;
3790 }
3791
3792 assert(is_aligned(start, alignment), "Must be");
3793
3794 char* end = start + bytes;
3795
3796 // Find the regions of the allocated chunk that can be promoted to large pages.
3797 char* lp_start = align_up(start, large_page_size);
3798 char* lp_end = align_down(end, large_page_size);
3799
3800 size_t lp_bytes = lp_end - lp_start;
3801
3802 assert(is_aligned(lp_bytes, large_page_size), "Must be");
3803
3804 if (lp_bytes == 0) {
3805 // The mapped region doesn't even span the start and the end of a large page.
3806 // Fall back to allocate a non-special area.
3807 ::munmap(start, end - start);
3808 return NULL;
3809 }
3810
3811 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3812
3813 void* result;
3814
3815 // Commit small-paged leading area.
3816 if (start != lp_start) {
3817 result = ::mmap(start, lp_start - start, prot,
3818 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3819 -1, 0);
3820 if (result == MAP_FAILED) {
3821 ::munmap(lp_start, end - lp_start);
3822 return NULL;
3823 }
3824 }
3825
3826 // Commit large-paged area.
3827 result = ::mmap(lp_start, lp_bytes, prot,
3828 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3829 -1, 0);
3830 if (result == MAP_FAILED) {
3831 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3832 // If the mmap above fails, the large pages region will be unmapped and we
3833 // have regions before and after with small pages. Release these regions.
3834 //
3835 // | mapped | unmapped | mapped |
3836 // ^ ^ ^ ^
3837 // start lp_start lp_end end
3838 //
3839 ::munmap(start, lp_start - start);
3840 ::munmap(lp_end, end - lp_end);
3841 return NULL;
3842 }
3843
3844 // Commit small-paged trailing area.
3845 if (lp_end != end) {
3846 result = ::mmap(lp_end, end - lp_end, prot,
3847 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3848 -1, 0);
3849 if (result == MAP_FAILED) {
3850 ::munmap(start, lp_end - start);
3851 return NULL;
3852 }
3853 }
3854
3855 return start;
3856 }
3857
3858 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3859 size_t alignment,
3860 char* req_addr,
3861 bool exec) {
3862 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3863 assert(is_aligned(req_addr, alignment), "Must be");
3864 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3865 assert(is_power_of_2(os::large_page_size()), "Must be");
3866 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3867
3868 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3869 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3870 } else {
3871 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3872 }
3873 }
3874
3875 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3876 char* req_addr, bool exec) {
3877 assert(UseLargePages, "only for large pages");
3878
3879 char* addr;
3880 if (UseSHM) {
3881 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3882 } else {
3883 assert(UseHugeTLBFS, "must be");
3884 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3885 }
3886
3887 if (addr != NULL) {
3888 if (UseNUMAInterleaving) {
3889 numa_make_global(addr, bytes);
3890 }
3891
3892 // The memory is committed
3893 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3894 }
3895
3896 return addr;
3897 }
3898
3899 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3900 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3901 return shmdt(base) == 0;
3902 }
3903
3904 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3905 return pd_release_memory(base, bytes);
3906 }
3907
3908 bool os::release_memory_special(char* base, size_t bytes) {
3909 bool res;
3910 if (MemTracker::tracking_level() > NMT_minimal) {
3911 Tracker tkr(Tracker::release);
3912 res = os::Linux::release_memory_special_impl(base, bytes);
3913 if (res) {
3914 tkr.record((address)base, bytes);
3915 }
3916
3917 } else {
3918 res = os::Linux::release_memory_special_impl(base, bytes);
3919 }
3920 return res;
3921 }
3922
3923 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3924 assert(UseLargePages, "only for large pages");
3925 bool res;
3926
3927 if (UseSHM) {
3928 res = os::Linux::release_memory_special_shm(base, bytes);
3929 } else {
3930 assert(UseHugeTLBFS, "must be");
3931 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3932 }
3933 return res;
3934 }
3935
3936 size_t os::large_page_size() {
3937 return _large_page_size;
3938 }
3939
3940 // With SysV SHM the entire memory region must be allocated as shared
3941 // memory.
3942 // HugeTLBFS allows application to commit large page memory on demand.
3943 // However, when committing memory with HugeTLBFS fails, the region
3944 // that was supposed to be committed will lose the old reservation
3945 // and allow other threads to steal that memory region. Because of this
3946 // behavior we can't commit HugeTLBFS memory.
3947 bool os::can_commit_large_page_memory() {
3948 return UseTransparentHugePages;
3949 }
3950
3951 bool os::can_execute_large_page_memory() {
3952 return UseTransparentHugePages || UseHugeTLBFS;
3953 }
3954
3955 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
3956 assert(file_desc >= 0, "file_desc is not valid");
3957 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
3958 if (result != NULL) {
3959 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
3960 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
3961 }
3962 }
3963 return result;
3964 }
3965
3966 // Reserve memory at an arbitrary address, only if that area is
3967 // available (and not reserved for something else).
3968
3969 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3970 const int max_tries = 10;
3971 char* base[max_tries];
3972 size_t size[max_tries];
3973 const size_t gap = 0x000000;
3974
3975 // Assert only that the size is a multiple of the page size, since
3976 // that's all that mmap requires, and since that's all we really know
3977 // about at this low abstraction level. If we need higher alignment,
3978 // we can either pass an alignment to this method or verify alignment
3979 // in one of the methods further up the call chain. See bug 5044738.
3980 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3981
3982 // Repeatedly allocate blocks until the block is allocated at the
3983 // right spot.
3984
3985 // Linux mmap allows caller to pass an address as hint; give it a try first,
3986 // if kernel honors the hint then we can return immediately.
3987 char * addr = anon_mmap(requested_addr, bytes, false);
3988 if (addr == requested_addr) {
3989 return requested_addr;
3990 }
3991
3992 if (addr != NULL) {
3993 // mmap() is successful but it fails to reserve at the requested address
3994 anon_munmap(addr, bytes);
3995 }
3996
3997 int i;
3998 for (i = 0; i < max_tries; ++i) {
3999 base[i] = reserve_memory(bytes);
4000
4001 if (base[i] != NULL) {
4002 // Is this the block we wanted?
4003 if (base[i] == requested_addr) {
4004 size[i] = bytes;
4005 break;
4006 }
4007
4008 // Does this overlap the block we wanted? Give back the overlapped
4009 // parts and try again.
4010
4011 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
4012 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
4013 unmap_memory(base[i], top_overlap);
4014 base[i] += top_overlap;
4015 size[i] = bytes - top_overlap;
4016 } else {
4017 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
4018 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
4019 unmap_memory(requested_addr, bottom_overlap);
4020 size[i] = bytes - bottom_overlap;
4021 } else {
4022 size[i] = bytes;
4023 }
4024 }
4025 }
4026 }
4027
4028 // Give back the unused reserved pieces.
4029
4030 for (int j = 0; j < i; ++j) {
4031 if (base[j] != NULL) {
4032 unmap_memory(base[j], size[j]);
4033 }
4034 }
4035
4036 if (i < max_tries) {
4037 return requested_addr;
4038 } else {
4039 return NULL;
4040 }
4041 }
4042
4043 size_t os::read(int fd, void *buf, unsigned int nBytes) {
4044 return ::read(fd, buf, nBytes);
4045 }
4046
4047 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
4048 return ::pread(fd, buf, nBytes, offset);
4049 }
4050
4051 // Short sleep, direct OS call.
4052 //
4053 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
4054 // sched_yield(2) will actually give up the CPU:
4055 //
4056 // * Alone on this pariticular CPU, keeps running.
4057 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
4058 // (pre 2.6.39).
4059 //
4060 // So calling this with 0 is an alternative.
4061 //
4062 void os::naked_short_sleep(jlong ms) {
4063 struct timespec req;
4064
4065 assert(ms < 1000, "Un-interruptable sleep, short time use only");
4066 req.tv_sec = 0;
4067 if (ms > 0) {
4068 req.tv_nsec = (ms % 1000) * 1000000;
4069 } else {
4070 req.tv_nsec = 1;
4071 }
4072
4073 nanosleep(&req, NULL);
4074
4075 return;
4076 }
4077
4078 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4079 void os::infinite_sleep() {
4080 while (true) { // sleep forever ...
4081 ::sleep(100); // ... 100 seconds at a time
4082 }
4083 }
4084
4085 // Used to convert frequent JVM_Yield() to nops
4086 bool os::dont_yield() {
4087 return DontYieldALot;
4088 }
4089
4090 void os::naked_yield() {
4091 sched_yield();
4092 }
4093
4094 ////////////////////////////////////////////////////////////////////////////////
4095 // thread priority support
4096
4097 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4098 // only supports dynamic priority, static priority must be zero. For real-time
4099 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4100 // However, for large multi-threaded applications, SCHED_RR is not only slower
4101 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4102 // of 5 runs - Sep 2005).
4103 //
4104 // The following code actually changes the niceness of kernel-thread/LWP. It
4105 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4106 // not the entire user process, and user level threads are 1:1 mapped to kernel
4107 // threads. It has always been the case, but could change in the future. For
4108 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4109 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4110
4111 int os::java_to_os_priority[CriticalPriority + 1] = {
4112 19, // 0 Entry should never be used
4113
4114 4, // 1 MinPriority
4115 3, // 2
4116 2, // 3
4117
4118 1, // 4
4119 0, // 5 NormPriority
4120 -1, // 6
4121
4122 -2, // 7
4123 -3, // 8
4124 -4, // 9 NearMaxPriority
4125
4126 -5, // 10 MaxPriority
4127
4128 -5 // 11 CriticalPriority
4129 };
4130
4131 static int prio_init() {
4132 if (ThreadPriorityPolicy == 1) {
4133 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4134 // if effective uid is not root. Perhaps, a more elegant way of doing
4135 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4136 if (geteuid() != 0) {
4137 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4138 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4139 }
4140 ThreadPriorityPolicy = 0;
4141 }
4142 }
4143 if (UseCriticalJavaThreadPriority) {
4144 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4145 }
4146 return 0;
4147 }
4148
4149 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4150 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4151
4152 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4153 return (ret == 0) ? OS_OK : OS_ERR;
4154 }
4155
4156 OSReturn os::get_native_priority(const Thread* const thread,
4157 int *priority_ptr) {
4158 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4159 *priority_ptr = java_to_os_priority[NormPriority];
4160 return OS_OK;
4161 }
4162
4163 errno = 0;
4164 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4165 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4166 }
4167
4168 // Hint to the underlying OS that a task switch would not be good.
4169 // Void return because it's a hint and can fail.
4170 void os::hint_no_preempt() {}
4171
4172 ////////////////////////////////////////////////////////////////////////////////
4173 // suspend/resume support
4174
4175 // The low-level signal-based suspend/resume support is a remnant from the
4176 // old VM-suspension that used to be for java-suspension, safepoints etc,
4177 // within hotspot. Currently used by JFR's OSThreadSampler
4178 //
4179 // The remaining code is greatly simplified from the more general suspension
4180 // code that used to be used.
4181 //
4182 // The protocol is quite simple:
4183 // - suspend:
4184 // - sends a signal to the target thread
4185 // - polls the suspend state of the osthread using a yield loop
4186 // - target thread signal handler (SR_handler) sets suspend state
4187 // and blocks in sigsuspend until continued
4188 // - resume:
4189 // - sets target osthread state to continue
4190 // - sends signal to end the sigsuspend loop in the SR_handler
4191 //
4192 // Note that the SR_lock plays no role in this suspend/resume protocol,
4193 // but is checked for NULL in SR_handler as a thread termination indicator.
4194 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4195 //
4196 // Note that resume_clear_context() and suspend_save_context() are needed
4197 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4198 // which in part is used by:
4199 // - Forte Analyzer: AsyncGetCallTrace()
4200 // - StackBanging: get_frame_at_stack_banging_point()
4201
4202 static void resume_clear_context(OSThread *osthread) {
4203 osthread->set_ucontext(NULL);
4204 osthread->set_siginfo(NULL);
4205 }
4206
4207 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4208 ucontext_t* context) {
4209 osthread->set_ucontext(context);
4210 osthread->set_siginfo(siginfo);
4211 }
4212
4213 // Handler function invoked when a thread's execution is suspended or
4214 // resumed. We have to be careful that only async-safe functions are
4215 // called here (Note: most pthread functions are not async safe and
4216 // should be avoided.)
4217 //
4218 // Note: sigwait() is a more natural fit than sigsuspend() from an
4219 // interface point of view, but sigwait() prevents the signal hander
4220 // from being run. libpthread would get very confused by not having
4221 // its signal handlers run and prevents sigwait()'s use with the
4222 // mutex granting granting signal.
4223 //
4224 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4225 //
4226 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4227 // Save and restore errno to avoid confusing native code with EINTR
4228 // after sigsuspend.
4229 int old_errno = errno;
4230
4231 Thread* thread = Thread::current_or_null_safe();
4232 assert(thread != NULL, "Missing current thread in SR_handler");
4233
4234 // On some systems we have seen signal delivery get "stuck" until the signal
4235 // mask is changed as part of thread termination. Check that the current thread
4236 // has not already terminated (via SR_lock()) - else the following assertion
4237 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4238 // destructor has completed.
4239
4240 if (thread->SR_lock() == NULL) {
4241 return;
4242 }
4243
4244 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4245
4246 OSThread* osthread = thread->osthread();
4247
4248 os::SuspendResume::State current = osthread->sr.state();
4249 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4250 suspend_save_context(osthread, siginfo, context);
4251
4252 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4253 os::SuspendResume::State state = osthread->sr.suspended();
4254 if (state == os::SuspendResume::SR_SUSPENDED) {
4255 sigset_t suspend_set; // signals for sigsuspend()
4256 sigemptyset(&suspend_set);
4257 // get current set of blocked signals and unblock resume signal
4258 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4259 sigdelset(&suspend_set, SR_signum);
4260
4261 sr_semaphore.signal();
4262 // wait here until we are resumed
4263 while (1) {
4264 sigsuspend(&suspend_set);
4265
4266 os::SuspendResume::State result = osthread->sr.running();
4267 if (result == os::SuspendResume::SR_RUNNING) {
4268 sr_semaphore.signal();
4269 break;
4270 }
4271 }
4272
4273 } else if (state == os::SuspendResume::SR_RUNNING) {
4274 // request was cancelled, continue
4275 } else {
4276 ShouldNotReachHere();
4277 }
4278
4279 resume_clear_context(osthread);
4280 } else if (current == os::SuspendResume::SR_RUNNING) {
4281 // request was cancelled, continue
4282 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4283 // ignore
4284 } else {
4285 // ignore
4286 }
4287
4288 errno = old_errno;
4289 }
4290
4291 static int SR_initialize() {
4292 struct sigaction act;
4293 char *s;
4294
4295 // Get signal number to use for suspend/resume
4296 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4297 int sig = ::strtol(s, 0, 10);
4298 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4299 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4300 SR_signum = sig;
4301 } else {
4302 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4303 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4304 }
4305 }
4306
4307 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4308 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4309
4310 sigemptyset(&SR_sigset);
4311 sigaddset(&SR_sigset, SR_signum);
4312
4313 // Set up signal handler for suspend/resume
4314 act.sa_flags = SA_RESTART|SA_SIGINFO;
4315 act.sa_handler = (void (*)(int)) SR_handler;
4316
4317 // SR_signum is blocked by default.
4318 // 4528190 - We also need to block pthread restart signal (32 on all
4319 // supported Linux platforms). Note that LinuxThreads need to block
4320 // this signal for all threads to work properly. So we don't have
4321 // to use hard-coded signal number when setting up the mask.
4322 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4323
4324 if (sigaction(SR_signum, &act, 0) == -1) {
4325 return -1;
4326 }
4327
4328 // Save signal flag
4329 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4330 return 0;
4331 }
4332
4333 static int sr_notify(OSThread* osthread) {
4334 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4335 assert_status(status == 0, status, "pthread_kill");
4336 return status;
4337 }
4338
4339 // "Randomly" selected value for how long we want to spin
4340 // before bailing out on suspending a thread, also how often
4341 // we send a signal to a thread we want to resume
4342 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4343 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4344
4345 // returns true on success and false on error - really an error is fatal
4346 // but this seems the normal response to library errors
4347 static bool do_suspend(OSThread* osthread) {
4348 assert(osthread->sr.is_running(), "thread should be running");
4349 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4350
4351 // mark as suspended and send signal
4352 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4353 // failed to switch, state wasn't running?
4354 ShouldNotReachHere();
4355 return false;
4356 }
4357
4358 if (sr_notify(osthread) != 0) {
4359 ShouldNotReachHere();
4360 }
4361
4362 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4363 while (true) {
4364 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4365 break;
4366 } else {
4367 // timeout
4368 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4369 if (cancelled == os::SuspendResume::SR_RUNNING) {
4370 return false;
4371 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4372 // make sure that we consume the signal on the semaphore as well
4373 sr_semaphore.wait();
4374 break;
4375 } else {
4376 ShouldNotReachHere();
4377 return false;
4378 }
4379 }
4380 }
4381
4382 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4383 return true;
4384 }
4385
4386 static void do_resume(OSThread* osthread) {
4387 assert(osthread->sr.is_suspended(), "thread should be suspended");
4388 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4389
4390 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4391 // failed to switch to WAKEUP_REQUEST
4392 ShouldNotReachHere();
4393 return;
4394 }
4395
4396 while (true) {
4397 if (sr_notify(osthread) == 0) {
4398 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4399 if (osthread->sr.is_running()) {
4400 return;
4401 }
4402 }
4403 } else {
4404 ShouldNotReachHere();
4405 }
4406 }
4407
4408 guarantee(osthread->sr.is_running(), "Must be running!");
4409 }
4410
4411 ///////////////////////////////////////////////////////////////////////////////////
4412 // signal handling (except suspend/resume)
4413
4414 // This routine may be used by user applications as a "hook" to catch signals.
4415 // The user-defined signal handler must pass unrecognized signals to this
4416 // routine, and if it returns true (non-zero), then the signal handler must
4417 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4418 // routine will never retun false (zero), but instead will execute a VM panic
4419 // routine kill the process.
4420 //
4421 // If this routine returns false, it is OK to call it again. This allows
4422 // the user-defined signal handler to perform checks either before or after
4423 // the VM performs its own checks. Naturally, the user code would be making
4424 // a serious error if it tried to handle an exception (such as a null check
4425 // or breakpoint) that the VM was generating for its own correct operation.
4426 //
4427 // This routine may recognize any of the following kinds of signals:
4428 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4429 // It should be consulted by handlers for any of those signals.
4430 //
4431 // The caller of this routine must pass in the three arguments supplied
4432 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4433 // field of the structure passed to sigaction(). This routine assumes that
4434 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4435 //
4436 // Note that the VM will print warnings if it detects conflicting signal
4437 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4438 //
4439 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4440 siginfo_t* siginfo,
4441 void* ucontext,
4442 int abort_if_unrecognized);
4443
4444 void signalHandler(int sig, siginfo_t* info, void* uc) {
4445 assert(info != NULL && uc != NULL, "it must be old kernel");
4446 int orig_errno = errno; // Preserve errno value over signal handler.
4447 JVM_handle_linux_signal(sig, info, uc, true);
4448 errno = orig_errno;
4449 }
4450
4451
4452 // This boolean allows users to forward their own non-matching signals
4453 // to JVM_handle_linux_signal, harmlessly.
4454 bool os::Linux::signal_handlers_are_installed = false;
4455
4456 // For signal-chaining
4457 struct sigaction sigact[NSIG];
4458 uint64_t sigs = 0;
4459 #if (64 < NSIG-1)
4460 #error "Not all signals can be encoded in sigs. Adapt its type!"
4461 #endif
4462 bool os::Linux::libjsig_is_loaded = false;
4463 typedef struct sigaction *(*get_signal_t)(int);
4464 get_signal_t os::Linux::get_signal_action = NULL;
4465
4466 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4467 struct sigaction *actp = NULL;
4468
4469 if (libjsig_is_loaded) {
4470 // Retrieve the old signal handler from libjsig
4471 actp = (*get_signal_action)(sig);
4472 }
4473 if (actp == NULL) {
4474 // Retrieve the preinstalled signal handler from jvm
4475 actp = get_preinstalled_handler(sig);
4476 }
4477
4478 return actp;
4479 }
4480
4481 static bool call_chained_handler(struct sigaction *actp, int sig,
4482 siginfo_t *siginfo, void *context) {
4483 // Call the old signal handler
4484 if (actp->sa_handler == SIG_DFL) {
4485 // It's more reasonable to let jvm treat it as an unexpected exception
4486 // instead of taking the default action.
4487 return false;
4488 } else if (actp->sa_handler != SIG_IGN) {
4489 if ((actp->sa_flags & SA_NODEFER) == 0) {
4490 // automaticlly block the signal
4491 sigaddset(&(actp->sa_mask), sig);
4492 }
4493
4494 sa_handler_t hand = NULL;
4495 sa_sigaction_t sa = NULL;
4496 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4497 // retrieve the chained handler
4498 if (siginfo_flag_set) {
4499 sa = actp->sa_sigaction;
4500 } else {
4501 hand = actp->sa_handler;
4502 }
4503
4504 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4505 actp->sa_handler = SIG_DFL;
4506 }
4507
4508 // try to honor the signal mask
4509 sigset_t oset;
4510 sigemptyset(&oset);
4511 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4512
4513 // call into the chained handler
4514 if (siginfo_flag_set) {
4515 (*sa)(sig, siginfo, context);
4516 } else {
4517 (*hand)(sig);
4518 }
4519
4520 // restore the signal mask
4521 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4522 }
4523 // Tell jvm's signal handler the signal is taken care of.
4524 return true;
4525 }
4526
4527 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4528 bool chained = false;
4529 // signal-chaining
4530 if (UseSignalChaining) {
4531 struct sigaction *actp = get_chained_signal_action(sig);
4532 if (actp != NULL) {
4533 chained = call_chained_handler(actp, sig, siginfo, context);
4534 }
4535 }
4536 return chained;
4537 }
4538
4539 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4540 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4541 return &sigact[sig];
4542 }
4543 return NULL;
4544 }
4545
4546 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4547 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4548 sigact[sig] = oldAct;
4549 sigs |= (uint64_t)1 << (sig-1);
4550 }
4551
4552 // for diagnostic
4553 int sigflags[NSIG];
4554
4555 int os::Linux::get_our_sigflags(int sig) {
4556 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4557 return sigflags[sig];
4558 }
4559
4560 void os::Linux::set_our_sigflags(int sig, int flags) {
4561 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4562 if (sig > 0 && sig < NSIG) {
4563 sigflags[sig] = flags;
4564 }
4565 }
4566
4567 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4568 // Check for overwrite.
4569 struct sigaction oldAct;
4570 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4571
4572 void* oldhand = oldAct.sa_sigaction
4573 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4574 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4575 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4576 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4577 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4578 if (AllowUserSignalHandlers || !set_installed) {
4579 // Do not overwrite; user takes responsibility to forward to us.
4580 return;
4581 } else if (UseSignalChaining) {
4582 // save the old handler in jvm
4583 save_preinstalled_handler(sig, oldAct);
4584 // libjsig also interposes the sigaction() call below and saves the
4585 // old sigaction on it own.
4586 } else {
4587 fatal("Encountered unexpected pre-existing sigaction handler "
4588 "%#lx for signal %d.", (long)oldhand, sig);
4589 }
4590 }
4591
4592 struct sigaction sigAct;
4593 sigfillset(&(sigAct.sa_mask));
4594 sigAct.sa_handler = SIG_DFL;
4595 if (!set_installed) {
4596 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4597 } else {
4598 sigAct.sa_sigaction = signalHandler;
4599 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4600 }
4601 // Save flags, which are set by ours
4602 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4603 sigflags[sig] = sigAct.sa_flags;
4604
4605 int ret = sigaction(sig, &sigAct, &oldAct);
4606 assert(ret == 0, "check");
4607
4608 void* oldhand2 = oldAct.sa_sigaction
4609 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4610 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4611 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4612 }
4613
4614 // install signal handlers for signals that HotSpot needs to
4615 // handle in order to support Java-level exception handling.
4616
4617 void os::Linux::install_signal_handlers() {
4618 if (!signal_handlers_are_installed) {
4619 signal_handlers_are_installed = true;
4620
4621 // signal-chaining
4622 typedef void (*signal_setting_t)();
4623 signal_setting_t begin_signal_setting = NULL;
4624 signal_setting_t end_signal_setting = NULL;
4625 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4626 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4627 if (begin_signal_setting != NULL) {
4628 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4629 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4630 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4631 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4632 libjsig_is_loaded = true;
4633 assert(UseSignalChaining, "should enable signal-chaining");
4634 }
4635 if (libjsig_is_loaded) {
4636 // Tell libjsig jvm is setting signal handlers
4637 (*begin_signal_setting)();
4638 }
4639
4640 set_signal_handler(SIGSEGV, true);
4641 set_signal_handler(SIGPIPE, true);
4642 set_signal_handler(SIGBUS, true);
4643 set_signal_handler(SIGILL, true);
4644 set_signal_handler(SIGFPE, true);
4645 #if defined(PPC64)
4646 set_signal_handler(SIGTRAP, true);
4647 #endif
4648 set_signal_handler(SIGXFSZ, true);
4649
4650 if (libjsig_is_loaded) {
4651 // Tell libjsig jvm finishes setting signal handlers
4652 (*end_signal_setting)();
4653 }
4654
4655 // We don't activate signal checker if libjsig is in place, we trust ourselves
4656 // and if UserSignalHandler is installed all bets are off.
4657 // Log that signal checking is off only if -verbose:jni is specified.
4658 if (CheckJNICalls) {
4659 if (libjsig_is_loaded) {
4660 if (PrintJNIResolving) {
4661 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4662 }
4663 check_signals = false;
4664 }
4665 if (AllowUserSignalHandlers) {
4666 if (PrintJNIResolving) {
4667 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4668 }
4669 check_signals = false;
4670 }
4671 }
4672 }
4673 }
4674
4675 // This is the fastest way to get thread cpu time on Linux.
4676 // Returns cpu time (user+sys) for any thread, not only for current.
4677 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4678 // It might work on 2.6.10+ with a special kernel/glibc patch.
4679 // For reference, please, see IEEE Std 1003.1-2004:
4680 // http://www.unix.org/single_unix_specification
4681
4682 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4683 struct timespec tp;
4684 int rc = os::Linux::clock_gettime(clockid, &tp);
4685 assert(rc == 0, "clock_gettime is expected to return 0 code");
4686
4687 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4688 }
4689
4690 void os::Linux::initialize_os_info() {
4691 assert(_os_version == 0, "OS info already initialized");
4692
4693 struct utsname _uname;
4694
4695 uint32_t major;
4696 uint32_t minor;
4697 uint32_t fix;
4698
4699 int rc;
4700
4701 // Kernel version is unknown if
4702 // verification below fails.
4703 _os_version = 0x01000000;
4704
4705 rc = uname(&_uname);
4706 if (rc != -1) {
4707
4708 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4709 if (rc == 3) {
4710
4711 if (major < 256 && minor < 256 && fix < 256) {
4712 // Kernel version format is as expected,
4713 // set it overriding unknown state.
4714 _os_version = (major << 16) |
4715 (minor << 8 ) |
4716 (fix << 0 ) ;
4717 }
4718 }
4719 }
4720 }
4721
4722 uint32_t os::Linux::os_version() {
4723 assert(_os_version != 0, "not initialized");
4724 return _os_version & 0x00FFFFFF;
4725 }
4726
4727 bool os::Linux::os_version_is_known() {
4728 assert(_os_version != 0, "not initialized");
4729 return _os_version & 0x01000000 ? false : true;
4730 }
4731
4732 /////
4733 // glibc on Linux platform uses non-documented flag
4734 // to indicate, that some special sort of signal
4735 // trampoline is used.
4736 // We will never set this flag, and we should
4737 // ignore this flag in our diagnostic
4738 #ifdef SIGNIFICANT_SIGNAL_MASK
4739 #undef SIGNIFICANT_SIGNAL_MASK
4740 #endif
4741 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4742
4743 static const char* get_signal_handler_name(address handler,
4744 char* buf, int buflen) {
4745 int offset = 0;
4746 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4747 if (found) {
4748 // skip directory names
4749 const char *p1, *p2;
4750 p1 = buf;
4751 size_t len = strlen(os::file_separator());
4752 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4753 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4754 } else {
4755 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4756 }
4757 return buf;
4758 }
4759
4760 static void print_signal_handler(outputStream* st, int sig,
4761 char* buf, size_t buflen) {
4762 struct sigaction sa;
4763
4764 sigaction(sig, NULL, &sa);
4765
4766 // See comment for SIGNIFICANT_SIGNAL_MASK define
4767 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4768
4769 st->print("%s: ", os::exception_name(sig, buf, buflen));
4770
4771 address handler = (sa.sa_flags & SA_SIGINFO)
4772 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4773 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4774
4775 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4776 st->print("SIG_DFL");
4777 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4778 st->print("SIG_IGN");
4779 } else {
4780 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4781 }
4782
4783 st->print(", sa_mask[0]=");
4784 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4785
4786 address rh = VMError::get_resetted_sighandler(sig);
4787 // May be, handler was resetted by VMError?
4788 if (rh != NULL) {
4789 handler = rh;
4790 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4791 }
4792
4793 st->print(", sa_flags=");
4794 os::Posix::print_sa_flags(st, sa.sa_flags);
4795
4796 // Check: is it our handler?
4797 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4798 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4799 // It is our signal handler
4800 // check for flags, reset system-used one!
4801 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4802 st->print(
4803 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4804 os::Linux::get_our_sigflags(sig));
4805 }
4806 }
4807 st->cr();
4808 }
4809
4810
4811 #define DO_SIGNAL_CHECK(sig) \
4812 do { \
4813 if (!sigismember(&check_signal_done, sig)) { \
4814 os::Linux::check_signal_handler(sig); \
4815 } \
4816 } while (0)
4817
4818 // This method is a periodic task to check for misbehaving JNI applications
4819 // under CheckJNI, we can add any periodic checks here
4820
4821 void os::run_periodic_checks() {
4822 if (check_signals == false) return;
4823
4824 // SEGV and BUS if overridden could potentially prevent
4825 // generation of hs*.log in the event of a crash, debugging
4826 // such a case can be very challenging, so we absolutely
4827 // check the following for a good measure:
4828 DO_SIGNAL_CHECK(SIGSEGV);
4829 DO_SIGNAL_CHECK(SIGILL);
4830 DO_SIGNAL_CHECK(SIGFPE);
4831 DO_SIGNAL_CHECK(SIGBUS);
4832 DO_SIGNAL_CHECK(SIGPIPE);
4833 DO_SIGNAL_CHECK(SIGXFSZ);
4834 #if defined(PPC64)
4835 DO_SIGNAL_CHECK(SIGTRAP);
4836 #endif
4837
4838 // ReduceSignalUsage allows the user to override these handlers
4839 // see comments at the very top and jvm_md.h
4840 if (!ReduceSignalUsage) {
4841 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4842 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4843 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4844 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4845 }
4846
4847 DO_SIGNAL_CHECK(SR_signum);
4848 }
4849
4850 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4851
4852 static os_sigaction_t os_sigaction = NULL;
4853
4854 void os::Linux::check_signal_handler(int sig) {
4855 char buf[O_BUFLEN];
4856 address jvmHandler = NULL;
4857
4858
4859 struct sigaction act;
4860 if (os_sigaction == NULL) {
4861 // only trust the default sigaction, in case it has been interposed
4862 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4863 if (os_sigaction == NULL) return;
4864 }
4865
4866 os_sigaction(sig, (struct sigaction*)NULL, &act);
4867
4868
4869 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4870
4871 address thisHandler = (act.sa_flags & SA_SIGINFO)
4872 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4873 : CAST_FROM_FN_PTR(address, act.sa_handler);
4874
4875
4876 switch (sig) {
4877 case SIGSEGV:
4878 case SIGBUS:
4879 case SIGFPE:
4880 case SIGPIPE:
4881 case SIGILL:
4882 case SIGXFSZ:
4883 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4884 break;
4885
4886 case SHUTDOWN1_SIGNAL:
4887 case SHUTDOWN2_SIGNAL:
4888 case SHUTDOWN3_SIGNAL:
4889 case BREAK_SIGNAL:
4890 jvmHandler = (address)user_handler();
4891 break;
4892
4893 default:
4894 if (sig == SR_signum) {
4895 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4896 } else {
4897 return;
4898 }
4899 break;
4900 }
4901
4902 if (thisHandler != jvmHandler) {
4903 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4904 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4905 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4906 // No need to check this sig any longer
4907 sigaddset(&check_signal_done, sig);
4908 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4909 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4910 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4911 exception_name(sig, buf, O_BUFLEN));
4912 }
4913 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4914 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4915 tty->print("expected:");
4916 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4917 tty->cr();
4918 tty->print(" found:");
4919 os::Posix::print_sa_flags(tty, act.sa_flags);
4920 tty->cr();
4921 // No need to check this sig any longer
4922 sigaddset(&check_signal_done, sig);
4923 }
4924
4925 // Dump all the signal
4926 if (sigismember(&check_signal_done, sig)) {
4927 print_signal_handlers(tty, buf, O_BUFLEN);
4928 }
4929 }
4930
4931 extern void report_error(char* file_name, int line_no, char* title,
4932 char* format, ...);
4933
4934 // this is called _before_ most of the global arguments have been parsed
4935 void os::init(void) {
4936 char dummy; // used to get a guess on initial stack address
4937
4938 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4939
4940 init_random(1234567);
4941
4942 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4943 if (Linux::page_size() == -1) {
4944 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4945 os::strerror(errno));
4946 }
4947 init_page_sizes((size_t) Linux::page_size());
4948
4949 Linux::initialize_system_info();
4950
4951 Linux::initialize_os_info();
4952
4953 // _main_thread points to the thread that created/loaded the JVM.
4954 Linux::_main_thread = pthread_self();
4955
4956 Linux::clock_init();
4957 initial_time_count = javaTimeNanos();
4958
4959 // retrieve entry point for pthread_setname_np
4960 Linux::_pthread_setname_np =
4961 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4962
4963 os::Posix::init();
4964 }
4965
4966 // To install functions for atexit system call
4967 extern "C" {
4968 static void perfMemory_exit_helper() {
4969 perfMemory_exit();
4970 }
4971 }
4972
4973 void os::pd_init_container_support() {
4974 OSContainer::init();
4975 }
4976
4977 // this is called _after_ the global arguments have been parsed
4978 jint os::init_2(void) {
4979
4980 os::Posix::init_2();
4981
4982 Linux::fast_thread_clock_init();
4983
4984 // initialize suspend/resume support - must do this before signal_sets_init()
4985 if (SR_initialize() != 0) {
4986 perror("SR_initialize failed");
4987 return JNI_ERR;
4988 }
4989
4990 Linux::signal_sets_init();
4991 Linux::install_signal_handlers();
4992
4993 // Check and sets minimum stack sizes against command line options
4994 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4995 return JNI_ERR;
4996 }
4997 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4998
4999 #if defined(IA32)
5000 workaround_expand_exec_shield_cs_limit();
5001 #endif
5002
5003 Linux::libpthread_init();
5004 Linux::sched_getcpu_init();
5005 log_info(os)("HotSpot is running with %s, %s",
5006 Linux::glibc_version(), Linux::libpthread_version());
5007
5008 if (UseNUMA) {
5009 if (!Linux::libnuma_init()) {
5010 UseNUMA = false;
5011 } else {
5012 if ((Linux::numa_max_node() < 1)) {
5013 // There's only one node(they start from 0), disable NUMA.
5014 UseNUMA = false;
5015 }
5016 }
5017
5018 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5019 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5020 // we can make the adaptive lgrp chunk resizing work. If the user specified both
5021 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5022 // and disable adaptive resizing.
5023 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5024 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5025 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5026 UseAdaptiveSizePolicy = false;
5027 UseAdaptiveNUMAChunkSizing = false;
5028 }
5029 }
5030
5031 if (!UseNUMA && ForceNUMA) {
5032 UseNUMA = true;
5033 }
5034 }
5035
5036 if (MaxFDLimit) {
5037 // set the number of file descriptors to max. print out error
5038 // if getrlimit/setrlimit fails but continue regardless.
5039 struct rlimit nbr_files;
5040 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5041 if (status != 0) {
5042 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5043 } else {
5044 nbr_files.rlim_cur = nbr_files.rlim_max;
5045 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5046 if (status != 0) {
5047 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5048 }
5049 }
5050 }
5051
5052 // Initialize lock used to serialize thread creation (see os::create_thread)
5053 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5054
5055 // at-exit methods are called in the reverse order of their registration.
5056 // atexit functions are called on return from main or as a result of a
5057 // call to exit(3C). There can be only 32 of these functions registered
5058 // and atexit() does not set errno.
5059
5060 if (PerfAllowAtExitRegistration) {
5061 // only register atexit functions if PerfAllowAtExitRegistration is set.
5062 // atexit functions can be delayed until process exit time, which
5063 // can be problematic for embedded VM situations. Embedded VMs should
5064 // call DestroyJavaVM() to assure that VM resources are released.
5065
5066 // note: perfMemory_exit_helper atexit function may be removed in
5067 // the future if the appropriate cleanup code can be added to the
5068 // VM_Exit VMOperation's doit method.
5069 if (atexit(perfMemory_exit_helper) != 0) {
5070 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5071 }
5072 }
5073
5074 // initialize thread priority policy
5075 prio_init();
5076
5077 if (!FLAG_IS_DEFAULT(AllocateHeapAt)) {
5078 set_coredump_filter(false /*largepages*/, true /*dax_shared*/);
5079 }
5080 return JNI_OK;
5081 }
5082
5083 // Mark the polling page as unreadable
5084 void os::make_polling_page_unreadable(void) {
5085 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5086 fatal("Could not disable polling page");
5087 }
5088 }
5089
5090 // Mark the polling page as readable
5091 void os::make_polling_page_readable(void) {
5092 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5093 fatal("Could not enable polling page");
5094 }
5095 }
5096
5097 // older glibc versions don't have this macro (which expands to
5098 // an optimized bit-counting function) so we have to roll our own
5099 #ifndef CPU_COUNT
5100
5101 static int _cpu_count(const cpu_set_t* cpus) {
5102 int count = 0;
5103 // only look up to the number of configured processors
5104 for (int i = 0; i < os::processor_count(); i++) {
5105 if (CPU_ISSET(i, cpus)) {
5106 count++;
5107 }
5108 }
5109 return count;
5110 }
5111
5112 #define CPU_COUNT(cpus) _cpu_count(cpus)
5113
5114 #endif // CPU_COUNT
5115
5116 // Get the current number of available processors for this process.
5117 // This value can change at any time during a process's lifetime.
5118 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5119 // If it appears there may be more than 1024 processors then we do a
5120 // dynamic check - see 6515172 for details.
5121 // If anything goes wrong we fallback to returning the number of online
5122 // processors - which can be greater than the number available to the process.
5123 int os::Linux::active_processor_count() {
5124 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5125 cpu_set_t* cpus_p = &cpus;
5126 int cpus_size = sizeof(cpu_set_t);
5127
5128 int configured_cpus = os::processor_count(); // upper bound on available cpus
5129 int cpu_count = 0;
5130
5131 // old build platforms may not support dynamic cpu sets
5132 #ifdef CPU_ALLOC
5133
5134 // To enable easy testing of the dynamic path on different platforms we
5135 // introduce a diagnostic flag: UseCpuAllocPath
5136 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5137 // kernel may use a mask bigger than cpu_set_t
5138 log_trace(os)("active_processor_count: using dynamic path %s"
5139 "- configured processors: %d",
5140 UseCpuAllocPath ? "(forced) " : "",
5141 configured_cpus);
5142 cpus_p = CPU_ALLOC(configured_cpus);
5143 if (cpus_p != NULL) {
5144 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5145 // zero it just to be safe
5146 CPU_ZERO_S(cpus_size, cpus_p);
5147 }
5148 else {
5149 // failed to allocate so fallback to online cpus
5150 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5151 log_trace(os)("active_processor_count: "
5152 "CPU_ALLOC failed (%s) - using "
5153 "online processor count: %d",
5154 os::strerror(errno), online_cpus);
5155 return online_cpus;
5156 }
5157 }
5158 else {
5159 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5160 configured_cpus);
5161 }
5162 #else // CPU_ALLOC
5163 // these stubs won't be executed
5164 #define CPU_COUNT_S(size, cpus) -1
5165 #define CPU_FREE(cpus)
5166
5167 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5168 configured_cpus);
5169 #endif // CPU_ALLOC
5170
5171 // pid 0 means the current thread - which we have to assume represents the process
5172 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5173 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5174 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5175 }
5176 else {
5177 cpu_count = CPU_COUNT(cpus_p);
5178 }
5179 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5180 }
5181 else {
5182 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5183 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5184 "which may exceed available processors", os::strerror(errno), cpu_count);
5185 }
5186
5187 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5188 CPU_FREE(cpus_p);
5189 }
5190
5191 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5192 return cpu_count;
5193 }
5194
5195 // Determine the active processor count from one of
5196 // three different sources:
5197 //
5198 // 1. User option -XX:ActiveProcessorCount
5199 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5200 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5201 //
5202 // Option 1, if specified, will always override.
5203 // If the cgroup subsystem is active and configured, we
5204 // will return the min of the cgroup and option 2 results.
5205 // This is required since tools, such as numactl, that
5206 // alter cpu affinity do not update cgroup subsystem
5207 // cpuset configuration files.
5208 int os::active_processor_count() {
5209 // User has overridden the number of active processors
5210 if (ActiveProcessorCount > 0) {
5211 log_trace(os)("active_processor_count: "
5212 "active processor count set by user : %d",
5213 ActiveProcessorCount);
5214 return ActiveProcessorCount;
5215 }
5216
5217 int active_cpus;
5218 if (OSContainer::is_containerized()) {
5219 active_cpus = OSContainer::active_processor_count();
5220 log_trace(os)("active_processor_count: determined by OSContainer: %d",
5221 active_cpus);
5222 } else {
5223 active_cpus = os::Linux::active_processor_count();
5224 }
5225
5226 return active_cpus;
5227 }
5228
5229 void os::set_native_thread_name(const char *name) {
5230 if (Linux::_pthread_setname_np) {
5231 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5232 snprintf(buf, sizeof(buf), "%s", name);
5233 buf[sizeof(buf) - 1] = '\0';
5234 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5235 // ERANGE should not happen; all other errors should just be ignored.
5236 assert(rc != ERANGE, "pthread_setname_np failed");
5237 }
5238 }
5239
5240 bool os::distribute_processes(uint length, uint* distribution) {
5241 // Not yet implemented.
5242 return false;
5243 }
5244
5245 bool os::bind_to_processor(uint processor_id) {
5246 // Not yet implemented.
5247 return false;
5248 }
5249
5250 ///
5251
5252 void os::SuspendedThreadTask::internal_do_task() {
5253 if (do_suspend(_thread->osthread())) {
5254 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5255 do_task(context);
5256 do_resume(_thread->osthread());
5257 }
5258 }
5259
5260 ////////////////////////////////////////////////////////////////////////////////
5261 // debug support
5262
5263 bool os::find(address addr, outputStream* st) {
5264 Dl_info dlinfo;
5265 memset(&dlinfo, 0, sizeof(dlinfo));
5266 if (dladdr(addr, &dlinfo) != 0) {
5267 st->print(PTR_FORMAT ": ", p2i(addr));
5268 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5269 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5270 p2i(addr) - p2i(dlinfo.dli_saddr));
5271 } else if (dlinfo.dli_fbase != NULL) {
5272 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5273 } else {
5274 st->print("<absolute address>");
5275 }
5276 if (dlinfo.dli_fname != NULL) {
5277 st->print(" in %s", dlinfo.dli_fname);
5278 }
5279 if (dlinfo.dli_fbase != NULL) {
5280 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5281 }
5282 st->cr();
5283
5284 if (Verbose) {
5285 // decode some bytes around the PC
5286 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5287 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5288 address lowest = (address) dlinfo.dli_sname;
5289 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5290 if (begin < lowest) begin = lowest;
5291 Dl_info dlinfo2;
5292 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5293 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5294 end = (address) dlinfo2.dli_saddr;
5295 }
5296 Disassembler::decode(begin, end, st);
5297 }
5298 return true;
5299 }
5300 return false;
5301 }
5302
5303 ////////////////////////////////////////////////////////////////////////////////
5304 // misc
5305
5306 // This does not do anything on Linux. This is basically a hook for being
5307 // able to use structured exception handling (thread-local exception filters)
5308 // on, e.g., Win32.
5309 void
5310 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5311 JavaCallArguments* args, Thread* thread) {
5312 f(value, method, args, thread);
5313 }
5314
5315 void os::print_statistics() {
5316 }
5317
5318 bool os::message_box(const char* title, const char* message) {
5319 int i;
5320 fdStream err(defaultStream::error_fd());
5321 for (i = 0; i < 78; i++) err.print_raw("=");
5322 err.cr();
5323 err.print_raw_cr(title);
5324 for (i = 0; i < 78; i++) err.print_raw("-");
5325 err.cr();
5326 err.print_raw_cr(message);
5327 for (i = 0; i < 78; i++) err.print_raw("=");
5328 err.cr();
5329
5330 char buf[16];
5331 // Prevent process from exiting upon "read error" without consuming all CPU
5332 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5333
5334 return buf[0] == 'y' || buf[0] == 'Y';
5335 }
5336
5337 int os::stat(const char *path, struct stat *sbuf) {
5338 char pathbuf[MAX_PATH];
5339 if (strlen(path) > MAX_PATH - 1) {
5340 errno = ENAMETOOLONG;
5341 return -1;
5342 }
5343 os::native_path(strcpy(pathbuf, path));
5344 return ::stat(pathbuf, sbuf);
5345 }
5346
5347 // Is a (classpath) directory empty?
5348 bool os::dir_is_empty(const char* path) {
5349 DIR *dir = NULL;
5350 struct dirent *ptr;
5351
5352 dir = opendir(path);
5353 if (dir == NULL) return true;
5354
5355 // Scan the directory
5356 bool result = true;
5357 char buf[sizeof(struct dirent) + MAX_PATH];
5358 while (result && (ptr = ::readdir(dir)) != NULL) {
5359 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5360 result = false;
5361 }
5362 }
5363 closedir(dir);
5364 return result;
5365 }
5366
5367 // This code originates from JDK's sysOpen and open64_w
5368 // from src/solaris/hpi/src/system_md.c
5369
5370 int os::open(const char *path, int oflag, int mode) {
5371 if (strlen(path) > MAX_PATH - 1) {
5372 errno = ENAMETOOLONG;
5373 return -1;
5374 }
5375
5376 // All file descriptors that are opened in the Java process and not
5377 // specifically destined for a subprocess should have the close-on-exec
5378 // flag set. If we don't set it, then careless 3rd party native code
5379 // might fork and exec without closing all appropriate file descriptors
5380 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5381 // turn might:
5382 //
5383 // - cause end-of-file to fail to be detected on some file
5384 // descriptors, resulting in mysterious hangs, or
5385 //
5386 // - might cause an fopen in the subprocess to fail on a system
5387 // suffering from bug 1085341.
5388 //
5389 // (Yes, the default setting of the close-on-exec flag is a Unix
5390 // design flaw)
5391 //
5392 // See:
5393 // 1085341: 32-bit stdio routines should support file descriptors >255
5394 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5395 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5396 //
5397 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5398 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5399 // because it saves a system call and removes a small window where the flag
5400 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5401 // and we fall back to using FD_CLOEXEC (see below).
5402 #ifdef O_CLOEXEC
5403 oflag |= O_CLOEXEC;
5404 #endif
5405
5406 int fd = ::open64(path, oflag, mode);
5407 if (fd == -1) return -1;
5408
5409 //If the open succeeded, the file might still be a directory
5410 {
5411 struct stat64 buf64;
5412 int ret = ::fstat64(fd, &buf64);
5413 int st_mode = buf64.st_mode;
5414
5415 if (ret != -1) {
5416 if ((st_mode & S_IFMT) == S_IFDIR) {
5417 errno = EISDIR;
5418 ::close(fd);
5419 return -1;
5420 }
5421 } else {
5422 ::close(fd);
5423 return -1;
5424 }
5425 }
5426
5427 #ifdef FD_CLOEXEC
5428 // Validate that the use of the O_CLOEXEC flag on open above worked.
5429 // With recent kernels, we will perform this check exactly once.
5430 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5431 if (!O_CLOEXEC_is_known_to_work) {
5432 int flags = ::fcntl(fd, F_GETFD);
5433 if (flags != -1) {
5434 if ((flags & FD_CLOEXEC) != 0)
5435 O_CLOEXEC_is_known_to_work = 1;
5436 else
5437 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5438 }
5439 }
5440 #endif
5441
5442 return fd;
5443 }
5444
5445
5446 // create binary file, rewriting existing file if required
5447 int os::create_binary_file(const char* path, bool rewrite_existing) {
5448 int oflags = O_WRONLY | O_CREAT;
5449 if (!rewrite_existing) {
5450 oflags |= O_EXCL;
5451 }
5452 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5453 }
5454
5455 // return current position of file pointer
5456 jlong os::current_file_offset(int fd) {
5457 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5458 }
5459
5460 // move file pointer to the specified offset
5461 jlong os::seek_to_file_offset(int fd, jlong offset) {
5462 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5463 }
5464
5465 // This code originates from JDK's sysAvailable
5466 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5467
5468 int os::available(int fd, jlong *bytes) {
5469 jlong cur, end;
5470 int mode;
5471 struct stat64 buf64;
5472
5473 if (::fstat64(fd, &buf64) >= 0) {
5474 mode = buf64.st_mode;
5475 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5476 int n;
5477 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5478 *bytes = n;
5479 return 1;
5480 }
5481 }
5482 }
5483 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5484 return 0;
5485 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5486 return 0;
5487 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5488 return 0;
5489 }
5490 *bytes = end - cur;
5491 return 1;
5492 }
5493
5494 // Map a block of memory.
5495 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5496 char *addr, size_t bytes, bool read_only,
5497 bool allow_exec) {
5498 int prot;
5499 int flags = MAP_PRIVATE;
5500
5501 if (read_only) {
5502 prot = PROT_READ;
5503 } else {
5504 prot = PROT_READ | PROT_WRITE;
5505 }
5506
5507 if (allow_exec) {
5508 prot |= PROT_EXEC;
5509 }
5510
5511 if (addr != NULL) {
5512 flags |= MAP_FIXED;
5513 }
5514
5515 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5516 fd, file_offset);
5517 if (mapped_address == MAP_FAILED) {
5518 return NULL;
5519 }
5520 return mapped_address;
5521 }
5522
5523
5524 // Remap a block of memory.
5525 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5526 char *addr, size_t bytes, bool read_only,
5527 bool allow_exec) {
5528 // same as map_memory() on this OS
5529 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5530 allow_exec);
5531 }
5532
5533
5534 // Unmap a block of memory.
5535 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5536 return munmap(addr, bytes) == 0;
5537 }
5538
5539 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5540
5541 static clockid_t thread_cpu_clockid(Thread* thread) {
5542 pthread_t tid = thread->osthread()->pthread_id();
5543 clockid_t clockid;
5544
5545 // Get thread clockid
5546 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5547 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5548 return clockid;
5549 }
5550
5551 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5552 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5553 // of a thread.
5554 //
5555 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5556 // the fast estimate available on the platform.
5557
5558 jlong os::current_thread_cpu_time() {
5559 if (os::Linux::supports_fast_thread_cpu_time()) {
5560 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5561 } else {
5562 // return user + sys since the cost is the same
5563 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5564 }
5565 }
5566
5567 jlong os::thread_cpu_time(Thread* thread) {
5568 // consistent with what current_thread_cpu_time() returns
5569 if (os::Linux::supports_fast_thread_cpu_time()) {
5570 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5571 } else {
5572 return slow_thread_cpu_time(thread, true /* user + sys */);
5573 }
5574 }
5575
5576 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5577 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5578 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5579 } else {
5580 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5581 }
5582 }
5583
5584 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5585 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5586 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5587 } else {
5588 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5589 }
5590 }
5591
5592 // -1 on error.
5593 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5594 pid_t tid = thread->osthread()->thread_id();
5595 char *s;
5596 char stat[2048];
5597 int statlen;
5598 char proc_name[64];
5599 int count;
5600 long sys_time, user_time;
5601 char cdummy;
5602 int idummy;
5603 long ldummy;
5604 FILE *fp;
5605
5606 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5607 fp = fopen(proc_name, "r");
5608 if (fp == NULL) return -1;
5609 statlen = fread(stat, 1, 2047, fp);
5610 stat[statlen] = '\0';
5611 fclose(fp);
5612
5613 // Skip pid and the command string. Note that we could be dealing with
5614 // weird command names, e.g. user could decide to rename java launcher
5615 // to "java 1.4.2 :)", then the stat file would look like
5616 // 1234 (java 1.4.2 :)) R ... ...
5617 // We don't really need to know the command string, just find the last
5618 // occurrence of ")" and then start parsing from there. See bug 4726580.
5619 s = strrchr(stat, ')');
5620 if (s == NULL) return -1;
5621
5622 // Skip blank chars
5623 do { s++; } while (s && isspace(*s));
5624
5625 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5626 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5627 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5628 &user_time, &sys_time);
5629 if (count != 13) return -1;
5630 if (user_sys_cpu_time) {
5631 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5632 } else {
5633 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5634 }
5635 }
5636
5637 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5638 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5639 info_ptr->may_skip_backward = false; // elapsed time not wall time
5640 info_ptr->may_skip_forward = false; // elapsed time not wall time
5641 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5642 }
5643
5644 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5645 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5646 info_ptr->may_skip_backward = false; // elapsed time not wall time
5647 info_ptr->may_skip_forward = false; // elapsed time not wall time
5648 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5649 }
5650
5651 bool os::is_thread_cpu_time_supported() {
5652 return true;
5653 }
5654
5655 // System loadavg support. Returns -1 if load average cannot be obtained.
5656 // Linux doesn't yet have a (official) notion of processor sets,
5657 // so just return the system wide load average.
5658 int os::loadavg(double loadavg[], int nelem) {
5659 return ::getloadavg(loadavg, nelem);
5660 }
5661
5662 void os::pause() {
5663 char filename[MAX_PATH];
5664 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5665 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5666 } else {
5667 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5668 }
5669
5670 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5671 if (fd != -1) {
5672 struct stat buf;
5673 ::close(fd);
5674 while (::stat(filename, &buf) == 0) {
5675 (void)::poll(NULL, 0, 100);
5676 }
5677 } else {
5678 jio_fprintf(stderr,
5679 "Could not open pause file '%s', continuing immediately.\n", filename);
5680 }
5681 }
5682
5683 extern char** environ;
5684
5685 // Run the specified command in a separate process. Return its exit value,
5686 // or -1 on failure (e.g. can't fork a new process).
5687 // Unlike system(), this function can be called from signal handler. It
5688 // doesn't block SIGINT et al.
5689 int os::fork_and_exec(char* cmd) {
5690 const char * argv[4] = {"sh", "-c", cmd, NULL};
5691
5692 pid_t pid = fork();
5693
5694 if (pid < 0) {
5695 // fork failed
5696 return -1;
5697
5698 } else if (pid == 0) {
5699 // child process
5700
5701 execve("/bin/sh", (char* const*)argv, environ);
5702
5703 // execve failed
5704 _exit(-1);
5705
5706 } else {
5707 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5708 // care about the actual exit code, for now.
5709
5710 int status;
5711
5712 // Wait for the child process to exit. This returns immediately if
5713 // the child has already exited. */
5714 while (waitpid(pid, &status, 0) < 0) {
5715 switch (errno) {
5716 case ECHILD: return 0;
5717 case EINTR: break;
5718 default: return -1;
5719 }
5720 }
5721
5722 if (WIFEXITED(status)) {
5723 // The child exited normally; get its exit code.
5724 return WEXITSTATUS(status);
5725 } else if (WIFSIGNALED(status)) {
5726 // The child exited because of a signal
5727 // The best value to return is 0x80 + signal number,
5728 // because that is what all Unix shells do, and because
5729 // it allows callers to distinguish between process exit and
5730 // process death by signal.
5731 return 0x80 + WTERMSIG(status);
5732 } else {
5733 // Unknown exit code; pass it through
5734 return status;
5735 }
5736 }
5737 }
5738
5739 // Get the default path to the core file
5740 // Returns the length of the string
5741 int os::get_core_path(char* buffer, size_t bufferSize) {
5742 /*
5743 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5744 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5745 */
5746 const int core_pattern_len = 129;
5747 char core_pattern[core_pattern_len] = {0};
5748
5749 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5750 if (core_pattern_file == -1) {
5751 return -1;
5752 }
5753
5754 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5755 ::close(core_pattern_file);
5756 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5757 return -1;
5758 }
5759 if (core_pattern[ret-1] == '\n') {
5760 core_pattern[ret-1] = '\0';
5761 } else {
5762 core_pattern[ret] = '\0';
5763 }
5764
5765 char *pid_pos = strstr(core_pattern, "%p");
5766 int written;
5767
5768 if (core_pattern[0] == '/') {
5769 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5770 } else {
5771 char cwd[PATH_MAX];
5772
5773 const char* p = get_current_directory(cwd, PATH_MAX);
5774 if (p == NULL) {
5775 return -1;
5776 }
5777
5778 if (core_pattern[0] == '|') {
5779 written = jio_snprintf(buffer, bufferSize,
5780 "\"%s\" (or dumping to %s/core.%d)",
5781 &core_pattern[1], p, current_process_id());
5782 } else {
5783 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5784 }
5785 }
5786
5787 if (written < 0) {
5788 return -1;
5789 }
5790
5791 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5792 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5793
5794 if (core_uses_pid_file != -1) {
5795 char core_uses_pid = 0;
5796 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5797 ::close(core_uses_pid_file);
5798
5799 if (core_uses_pid == '1') {
5800 jio_snprintf(buffer + written, bufferSize - written,
5801 ".%d", current_process_id());
5802 }
5803 }
5804 }
5805
5806 return strlen(buffer);
5807 }
5808
5809 bool os::start_debugging(char *buf, int buflen) {
5810 int len = (int)strlen(buf);
5811 char *p = &buf[len];
5812
5813 jio_snprintf(p, buflen-len,
5814 "\n\n"
5815 "Do you want to debug the problem?\n\n"
5816 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5817 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5818 "Otherwise, press RETURN to abort...",
5819 os::current_process_id(), os::current_process_id(),
5820 os::current_thread_id(), os::current_thread_id());
5821
5822 bool yes = os::message_box("Unexpected Error", buf);
5823
5824 if (yes) {
5825 // yes, user asked VM to launch debugger
5826 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5827 os::current_process_id(), os::current_process_id());
5828
5829 os::fork_and_exec(buf);
5830 yes = false;
5831 }
5832 return yes;
5833 }
5834
5835
5836 // Java/Compiler thread:
5837 //
5838 // Low memory addresses
5839 // P0 +------------------------+
5840 // | |\ Java thread created by VM does not have glibc
5841 // | glibc guard page | - guard page, attached Java thread usually has
5842 // | |/ 1 glibc guard page.
5843 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5844 // | |\
5845 // | HotSpot Guard Pages | - red, yellow and reserved pages
5846 // | |/
5847 // +------------------------+ JavaThread::stack_reserved_zone_base()
5848 // | |\
5849 // | Normal Stack | -
5850 // | |/
5851 // P2 +------------------------+ Thread::stack_base()
5852 //
5853 // Non-Java thread:
5854 //
5855 // Low memory addresses
5856 // P0 +------------------------+
5857 // | |\
5858 // | glibc guard page | - usually 1 page
5859 // | |/
5860 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5861 // | |\
5862 // | Normal Stack | -
5863 // | |/
5864 // P2 +------------------------+ Thread::stack_base()
5865 //
5866 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
5867 // returned from pthread_attr_getstack().
5868 // ** Due to NPTL implementation error, linux takes the glibc guard page out
5869 // of the stack size given in pthread_attr. We work around this for
5870 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
5871 //
5872 #ifndef ZERO
5873 static void current_stack_region(address * bottom, size_t * size) {
5874 if (os::is_primordial_thread()) {
5875 // primordial thread needs special handling because pthread_getattr_np()
5876 // may return bogus value.
5877 *bottom = os::Linux::initial_thread_stack_bottom();
5878 *size = os::Linux::initial_thread_stack_size();
5879 } else {
5880 pthread_attr_t attr;
5881
5882 int rslt = pthread_getattr_np(pthread_self(), &attr);
5883
5884 // JVM needs to know exact stack location, abort if it fails
5885 if (rslt != 0) {
5886 if (rslt == ENOMEM) {
5887 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
5888 } else {
5889 fatal("pthread_getattr_np failed with error = %d", rslt);
5890 }
5891 }
5892
5893 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
5894 fatal("Cannot locate current stack attributes!");
5895 }
5896
5897 // Work around NPTL stack guard error.
5898 size_t guard_size = 0;
5899 rslt = pthread_attr_getguardsize(&attr, &guard_size);
5900 if (rslt != 0) {
5901 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
5902 }
5903 *bottom += guard_size;
5904 *size -= guard_size;
5905
5906 pthread_attr_destroy(&attr);
5907
5908 }
5909 assert(os::current_stack_pointer() >= *bottom &&
5910 os::current_stack_pointer() < *bottom + *size, "just checking");
5911 }
5912
5913 address os::current_stack_base() {
5914 address bottom;
5915 size_t size;
5916 current_stack_region(&bottom, &size);
5917 return (bottom + size);
5918 }
5919
5920 size_t os::current_stack_size() {
5921 // This stack size includes the usable stack and HotSpot guard pages
5922 // (for the threads that have Hotspot guard pages).
5923 address bottom;
5924 size_t size;
5925 current_stack_region(&bottom, &size);
5926 return size;
5927 }
5928 #endif
5929
5930 static inline struct timespec get_mtime(const char* filename) {
5931 struct stat st;
5932 int ret = os::stat(filename, &st);
5933 assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
5934 return st.st_mtim;
5935 }
5936
5937 int os::compare_file_modified_times(const char* file1, const char* file2) {
5938 struct timespec filetime1 = get_mtime(file1);
5939 struct timespec filetime2 = get_mtime(file2);
5940 int diff = filetime1.tv_sec - filetime2.tv_sec;
5941 if (diff == 0) {
5942 return filetime1.tv_nsec - filetime2.tv_nsec;
5943 }
5944 return diff;
5945 }
5946
5947 /////////////// Unit tests ///////////////
5948
5949 #ifndef PRODUCT
5950
5951 #define test_log(...) \
5952 do { \
5953 if (VerboseInternalVMTests) { \
5954 tty->print_cr(__VA_ARGS__); \
5955 tty->flush(); \
5956 } \
5957 } while (false)
5958
5959 class TestReserveMemorySpecial : AllStatic {
5960 public:
5961 static void small_page_write(void* addr, size_t size) {
5962 size_t page_size = os::vm_page_size();
5963
5964 char* end = (char*)addr + size;
5965 for (char* p = (char*)addr; p < end; p += page_size) {
5966 *p = 1;
5967 }
5968 }
5969
5970 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
5971 if (!UseHugeTLBFS) {
5972 return;
5973 }
5974
5975 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
5976
5977 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
5978
5979 if (addr != NULL) {
5980 small_page_write(addr, size);
5981
5982 os::Linux::release_memory_special_huge_tlbfs(addr, size);
5983 }
5984 }
5985
5986 static void test_reserve_memory_special_huge_tlbfs_only() {
5987 if (!UseHugeTLBFS) {
5988 return;
5989 }
5990
5991 size_t lp = os::large_page_size();
5992
5993 for (size_t size = lp; size <= lp * 10; size += lp) {
5994 test_reserve_memory_special_huge_tlbfs_only(size);
5995 }
5996 }
5997
5998 static void test_reserve_memory_special_huge_tlbfs_mixed() {
5999 size_t lp = os::large_page_size();
6000 size_t ag = os::vm_allocation_granularity();
6001
6002 // sizes to test
6003 const size_t sizes[] = {
6004 lp, lp + ag, lp + lp / 2, lp * 2,
6005 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6006 lp * 10, lp * 10 + lp / 2
6007 };
6008 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6009
6010 // For each size/alignment combination, we test three scenarios:
6011 // 1) with req_addr == NULL
6012 // 2) with a non-null req_addr at which we expect to successfully allocate
6013 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6014 // expect the allocation to either fail or to ignore req_addr
6015
6016 // Pre-allocate two areas; they shall be as large as the largest allocation
6017 // and aligned to the largest alignment we will be testing.
6018 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6019 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6020 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6021 -1, 0);
6022 assert(mapping1 != MAP_FAILED, "should work");
6023
6024 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6025 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6026 -1, 0);
6027 assert(mapping2 != MAP_FAILED, "should work");
6028
6029 // Unmap the first mapping, but leave the second mapping intact: the first
6030 // mapping will serve as a value for a "good" req_addr (case 2). The second
6031 // mapping, still intact, as "bad" req_addr (case 3).
6032 ::munmap(mapping1, mapping_size);
6033
6034 // Case 1
6035 test_log("%s, req_addr NULL:", __FUNCTION__);
6036 test_log("size align result");
6037
6038 for (int i = 0; i < num_sizes; i++) {
6039 const size_t size = sizes[i];
6040 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6041 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6042 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6043 size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6044 if (p != NULL) {
6045 assert(is_aligned(p, alignment), "must be");
6046 small_page_write(p, size);
6047 os::Linux::release_memory_special_huge_tlbfs(p, size);
6048 }
6049 }
6050 }
6051
6052 // Case 2
6053 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6054 test_log("size align req_addr result");
6055
6056 for (int i = 0; i < num_sizes; i++) {
6057 const size_t size = sizes[i];
6058 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6059 char* const req_addr = align_up(mapping1, alignment);
6060 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6061 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6062 size, alignment, p2i(req_addr), p2i(p),
6063 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6064 if (p != NULL) {
6065 assert(p == req_addr, "must be");
6066 small_page_write(p, size);
6067 os::Linux::release_memory_special_huge_tlbfs(p, size);
6068 }
6069 }
6070 }
6071
6072 // Case 3
6073 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6074 test_log("size align req_addr result");
6075
6076 for (int i = 0; i < num_sizes; i++) {
6077 const size_t size = sizes[i];
6078 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6079 char* const req_addr = align_up(mapping2, alignment);
6080 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6081 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6082 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6083 // as the area around req_addr contains already existing mappings, the API should always
6084 // return NULL (as per contract, it cannot return another address)
6085 assert(p == NULL, "must be");
6086 }
6087 }
6088
6089 ::munmap(mapping2, mapping_size);
6090
6091 }
6092
6093 static void test_reserve_memory_special_huge_tlbfs() {
6094 if (!UseHugeTLBFS) {
6095 return;
6096 }
6097
6098 test_reserve_memory_special_huge_tlbfs_only();
6099 test_reserve_memory_special_huge_tlbfs_mixed();
6100 }
6101
6102 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6103 if (!UseSHM) {
6104 return;
6105 }
6106
6107 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6108
6109 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6110
6111 if (addr != NULL) {
6112 assert(is_aligned(addr, alignment), "Check");
6113 assert(is_aligned(addr, os::large_page_size()), "Check");
6114
6115 small_page_write(addr, size);
6116
6117 os::Linux::release_memory_special_shm(addr, size);
6118 }
6119 }
6120
6121 static void test_reserve_memory_special_shm() {
6122 size_t lp = os::large_page_size();
6123 size_t ag = os::vm_allocation_granularity();
6124
6125 for (size_t size = ag; size < lp * 3; size += ag) {
6126 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6127 test_reserve_memory_special_shm(size, alignment);
6128 }
6129 }
6130 }
6131
6132 static void test() {
6133 test_reserve_memory_special_huge_tlbfs();
6134 test_reserve_memory_special_shm();
6135 }
6136 };
6137
6138 void TestReserveMemorySpecial_test() {
6139 TestReserveMemorySpecial::test();
6140 }
6141
6142 #endif