1 /*
2 * Copyright (c) 1999, 2018, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 // no precompiled headers
26 #include "jvm.h"
27 #include "classfile/classLoader.hpp"
28 #include "classfile/systemDictionary.hpp"
29 #include "classfile/vmSymbols.hpp"
30 #include "code/icBuffer.hpp"
31 #include "code/vtableStubs.hpp"
32 #include "compiler/compileBroker.hpp"
33 #include "compiler/disassembler.hpp"
34 #include "interpreter/interpreter.hpp"
35 #include "logging/log.hpp"
36 #include "memory/allocation.inline.hpp"
37 #include "memory/filemap.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "os_linux.inline.hpp"
40 #include "os_share_linux.hpp"
41 #include "osContainer_linux.hpp"
42 #include "prims/jniFastGetField.hpp"
43 #include "prims/jvm_misc.hpp"
44 #include "runtime/arguments.hpp"
45 #include "runtime/atomic.hpp"
46 #include "runtime/extendedPC.hpp"
47 #include "runtime/globals.hpp"
48 #include "runtime/interfaceSupport.inline.hpp"
49 #include "runtime/init.hpp"
50 #include "runtime/java.hpp"
51 #include "runtime/javaCalls.hpp"
52 #include "runtime/mutexLocker.hpp"
53 #include "runtime/objectMonitor.hpp"
54 #include "runtime/orderAccess.hpp"
55 #include "runtime/osThread.hpp"
56 #include "runtime/perfMemory.hpp"
57 #include "runtime/sharedRuntime.hpp"
58 #include "runtime/statSampler.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "runtime/thread.inline.hpp"
61 #include "runtime/threadCritical.hpp"
62 #include "runtime/threadSMR.hpp"
63 #include "runtime/timer.hpp"
64 #include "semaphore_posix.hpp"
65 #include "services/attachListener.hpp"
66 #include "services/memTracker.hpp"
67 #include "services/runtimeService.hpp"
68 #include "utilities/align.hpp"
69 #include "utilities/decoder.hpp"
70 #include "utilities/defaultStream.hpp"
71 #include "utilities/events.hpp"
72 #include "utilities/elfFile.hpp"
73 #include "utilities/growableArray.hpp"
74 #include "utilities/macros.hpp"
75 #include "utilities/vmError.hpp"
76
77 // put OS-includes here
78 # include <sys/types.h>
79 # include <sys/mman.h>
80 # include <sys/stat.h>
81 # include <sys/select.h>
82 # include <pthread.h>
83 # include <signal.h>
84 # include <errno.h>
85 # include <dlfcn.h>
86 # include <stdio.h>
87 # include <unistd.h>
88 # include <sys/resource.h>
89 # include <pthread.h>
90 # include <sys/stat.h>
91 # include <sys/time.h>
92 # include <sys/times.h>
93 # include <sys/utsname.h>
94 # include <sys/socket.h>
95 # include <sys/wait.h>
96 # include <pwd.h>
97 # include <poll.h>
98 # include <fcntl.h>
99 # include <string.h>
100 # include <syscall.h>
101 # include <sys/sysinfo.h>
102 # include <gnu/libc-version.h>
103 # include <sys/ipc.h>
104 # include <sys/shm.h>
105 # include <link.h>
106 # include <stdint.h>
107 # include <inttypes.h>
108 # include <sys/ioctl.h>
109
110 #ifndef _GNU_SOURCE
111 #define _GNU_SOURCE
112 #include <sched.h>
113 #undef _GNU_SOURCE
114 #else
115 #include <sched.h>
116 #endif
117
118 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
119 // getrusage() is prepared to handle the associated failure.
120 #ifndef RUSAGE_THREAD
121 #define RUSAGE_THREAD (1) /* only the calling thread */
122 #endif
123
124 #define MAX_PATH (2 * K)
125
126 #define MAX_SECS 100000000
127
128 // for timer info max values which include all bits
129 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
130
131 #define LARGEPAGES_BIT (1 << 6)
132 #define DAX_SHARED_BIT (1 << 8)
133 ////////////////////////////////////////////////////////////////////////////////
134 // global variables
135 julong os::Linux::_physical_memory = 0;
136
137 address os::Linux::_initial_thread_stack_bottom = NULL;
138 uintptr_t os::Linux::_initial_thread_stack_size = 0;
139
140 int (*os::Linux::_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 // If the VM might have been created on the primordial thread, we need to resolve the
156 // primordial thread stack bounds and check if the current thread might be the
157 // primordial thread in places. If we know that the primordial thread is never used,
158 // such as when the VM was created by one of the standard java launchers, we can
159 // avoid this
160 static bool suppress_primordial_thread_resolution = false;
161
162 // For diagnostics to print a message once. see run_periodic_checks
163 static sigset_t check_signal_done;
164 static bool check_signals = true;
165
166 // Signal number used to suspend/resume a thread
167
168 // do not use any signal number less than SIGSEGV, see 4355769
169 static int SR_signum = SIGUSR2;
170 sigset_t SR_sigset;
171
172 // utility functions
173
174 static int SR_initialize();
175
176 julong os::available_memory() {
177 return Linux::available_memory();
178 }
179
180 julong os::Linux::available_memory() {
181 // values in struct sysinfo are "unsigned long"
182 struct sysinfo si;
183 julong avail_mem;
184
185 if (OSContainer::is_containerized()) {
186 jlong mem_limit, mem_usage;
187 if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
188 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
189 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
190 }
191 if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
192 log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
193 }
194 if (mem_limit > 0 && mem_usage > 0 ) {
195 avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
196 log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
197 return avail_mem;
198 }
199 }
200
201 sysinfo(&si);
202 avail_mem = (julong)si.freeram * si.mem_unit;
203 log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
204 return avail_mem;
205 }
206
207 julong os::physical_memory() {
208 jlong phys_mem = 0;
209 if (OSContainer::is_containerized()) {
210 jlong mem_limit;
211 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
212 log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
213 return mem_limit;
214 }
215 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
216 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
217 }
218
219 phys_mem = Linux::physical_memory();
220 log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
221 return phys_mem;
222 }
223
224 // Return true if user is running as root.
225
226 bool os::have_special_privileges() {
227 static bool init = false;
228 static bool privileges = false;
229 if (!init) {
230 privileges = (getuid() != geteuid()) || (getgid() != getegid());
231 init = true;
232 }
233 return privileges;
234 }
235
236
237 #ifndef SYS_gettid
238 // i386: 224, ia64: 1105, amd64: 186, sparc 143
239 #ifdef __ia64__
240 #define SYS_gettid 1105
241 #else
242 #ifdef __i386__
243 #define SYS_gettid 224
244 #else
245 #ifdef __amd64__
246 #define SYS_gettid 186
247 #else
248 #ifdef __sparc__
249 #define SYS_gettid 143
250 #else
251 #error define gettid for the arch
252 #endif
253 #endif
254 #endif
255 #endif
256 #endif
257
258
259 // pid_t gettid()
260 //
261 // Returns the kernel thread id of the currently running thread. Kernel
262 // thread id is used to access /proc.
263 pid_t os::Linux::gettid() {
264 int rslt = syscall(SYS_gettid);
265 assert(rslt != -1, "must be."); // old linuxthreads implementation?
266 return (pid_t)rslt;
267 }
268
269 // Most versions of linux have a bug where the number of processors are
270 // determined by looking at the /proc file system. In a chroot environment,
271 // the system call returns 1.
272 static bool unsafe_chroot_detected = false;
273 static const char *unstable_chroot_error = "/proc file system not found.\n"
274 "Java may be unstable running multithreaded in a chroot "
275 "environment on Linux when /proc filesystem is not mounted.";
276
277 void os::Linux::initialize_system_info() {
278 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
279 if (processor_count() == 1) {
280 pid_t pid = os::Linux::gettid();
281 char fname[32];
282 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
283 FILE *fp = fopen(fname, "r");
284 if (fp == NULL) {
285 unsafe_chroot_detected = true;
286 } else {
287 fclose(fp);
288 }
289 }
290 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
291 assert(processor_count() > 0, "linux error");
292 }
293
294 void os::init_system_properties_values() {
295 // The next steps are taken in the product version:
296 //
297 // Obtain the JAVA_HOME value from the location of libjvm.so.
298 // This library should be located at:
299 // <JAVA_HOME>/lib/{client|server}/libjvm.so.
300 //
301 // If "/jre/lib/" appears at the right place in the path, then we
302 // assume libjvm.so is installed in a JDK and we use this path.
303 //
304 // Otherwise exit with message: "Could not create the Java virtual machine."
305 //
306 // The following extra steps are taken in the debugging version:
307 //
308 // If "/jre/lib/" does NOT appear at the right place in the path
309 // instead of exit check for $JAVA_HOME environment variable.
310 //
311 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
312 // then we append a fake suffix "hotspot/libjvm.so" to this path so
313 // it looks like libjvm.so is installed there
314 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
315 //
316 // Otherwise exit.
317 //
318 // Important note: if the location of libjvm.so changes this
319 // code needs to be changed accordingly.
320
321 // See ld(1):
322 // The linker uses the following search paths to locate required
323 // shared libraries:
324 // 1: ...
325 // ...
326 // 7: The default directories, normally /lib and /usr/lib.
327 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
328 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
329 #else
330 #define DEFAULT_LIBPATH "/lib:/usr/lib"
331 #endif
332
333 // Base path of extensions installed on the system.
334 #define SYS_EXT_DIR "/usr/java/packages"
335 #define EXTENSIONS_DIR "/lib/ext"
336
337 // Buffer that fits several sprintfs.
338 // Note that the space for the colon and the trailing null are provided
339 // by the nulls included by the sizeof operator.
340 const size_t bufsize =
341 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
342 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
343 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
344
345 // sysclasspath, java_home, dll_dir
346 {
347 char *pslash;
348 os::jvm_path(buf, bufsize);
349
350 // Found the full path to libjvm.so.
351 // Now cut the path to <java_home>/jre if we can.
352 pslash = strrchr(buf, '/');
353 if (pslash != NULL) {
354 *pslash = '\0'; // Get rid of /libjvm.so.
355 }
356 pslash = strrchr(buf, '/');
357 if (pslash != NULL) {
358 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
359 }
360 Arguments::set_dll_dir(buf);
361
362 if (pslash != NULL) {
363 pslash = strrchr(buf, '/');
364 if (pslash != NULL) {
365 *pslash = '\0'; // Get rid of /lib.
366 }
367 }
368 Arguments::set_java_home(buf);
369 if (!set_boot_path('/', ':')) {
370 vm_exit_during_initialization("Failed setting boot class path.", NULL);
371 }
372 }
373
374 // Where to look for native libraries.
375 //
376 // Note: Due to a legacy implementation, most of the library path
377 // is set in the launcher. This was to accomodate linking restrictions
378 // on legacy Linux implementations (which are no longer supported).
379 // Eventually, all the library path setting will be done here.
380 //
381 // However, to prevent the proliferation of improperly built native
382 // libraries, the new path component /usr/java/packages is added here.
383 // Eventually, all the library path setting will be done here.
384 {
385 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
386 // should always exist (until the legacy problem cited above is
387 // addressed).
388 const char *v = ::getenv("LD_LIBRARY_PATH");
389 const char *v_colon = ":";
390 if (v == NULL) { v = ""; v_colon = ""; }
391 // That's +1 for the colon and +1 for the trailing '\0'.
392 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
393 strlen(v) + 1 +
394 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
395 mtInternal);
396 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
397 Arguments::set_library_path(ld_library_path);
398 FREE_C_HEAP_ARRAY(char, ld_library_path);
399 }
400
401 // Extensions directories.
402 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
403 Arguments::set_ext_dirs(buf);
404
405 FREE_C_HEAP_ARRAY(char, buf);
406
407 #undef DEFAULT_LIBPATH
408 #undef SYS_EXT_DIR
409 #undef EXTENSIONS_DIR
410 }
411
412 ////////////////////////////////////////////////////////////////////////////////
413 // breakpoint support
414
415 void os::breakpoint() {
416 BREAKPOINT;
417 }
418
419 extern "C" void breakpoint() {
420 // use debugger to set breakpoint here
421 }
422
423 ////////////////////////////////////////////////////////////////////////////////
424 // signal support
425
426 debug_only(static bool signal_sets_initialized = false);
427 static sigset_t unblocked_sigs, vm_sigs;
428
429 void os::Linux::signal_sets_init() {
430 // Should also have an assertion stating we are still single-threaded.
431 assert(!signal_sets_initialized, "Already initialized");
432 // Fill in signals that are necessarily unblocked for all threads in
433 // the VM. Currently, we unblock the following signals:
434 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
435 // by -Xrs (=ReduceSignalUsage));
436 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
437 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
438 // the dispositions or masks wrt these signals.
439 // Programs embedding the VM that want to use the above signals for their
440 // own purposes must, at this time, use the "-Xrs" option to prevent
441 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
442 // (See bug 4345157, and other related bugs).
443 // In reality, though, unblocking these signals is really a nop, since
444 // these signals are not blocked by default.
445 sigemptyset(&unblocked_sigs);
446 sigaddset(&unblocked_sigs, SIGILL);
447 sigaddset(&unblocked_sigs, SIGSEGV);
448 sigaddset(&unblocked_sigs, SIGBUS);
449 sigaddset(&unblocked_sigs, SIGFPE);
450 #if defined(PPC64)
451 sigaddset(&unblocked_sigs, SIGTRAP);
452 #endif
453 sigaddset(&unblocked_sigs, SR_signum);
454
455 if (!ReduceSignalUsage) {
456 if (!os::Posix::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
457 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
458 }
459 if (!os::Posix::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
460 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
461 }
462 if (!os::Posix::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
463 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
464 }
465 }
466 // Fill in signals that are blocked by all but the VM thread.
467 sigemptyset(&vm_sigs);
468 if (!ReduceSignalUsage) {
469 sigaddset(&vm_sigs, BREAK_SIGNAL);
470 }
471 debug_only(signal_sets_initialized = true);
472
473 }
474
475 // These are signals that are unblocked while a thread is running Java.
476 // (For some reason, they get blocked by default.)
477 sigset_t* os::Linux::unblocked_signals() {
478 assert(signal_sets_initialized, "Not initialized");
479 return &unblocked_sigs;
480 }
481
482 // These are the signals that are blocked while a (non-VM) thread is
483 // running Java. Only the VM thread handles these signals.
484 sigset_t* os::Linux::vm_signals() {
485 assert(signal_sets_initialized, "Not initialized");
486 return &vm_sigs;
487 }
488
489 void os::Linux::hotspot_sigmask(Thread* thread) {
490
491 //Save caller's signal mask before setting VM signal mask
492 sigset_t caller_sigmask;
493 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
494
495 OSThread* osthread = thread->osthread();
496 osthread->set_caller_sigmask(caller_sigmask);
497
498 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
499
500 if (!ReduceSignalUsage) {
501 if (thread->is_VM_thread()) {
502 // Only the VM thread handles BREAK_SIGNAL ...
503 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
504 } else {
505 // ... all other threads block BREAK_SIGNAL
506 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
507 }
508 }
509 }
510
511 //////////////////////////////////////////////////////////////////////////////
512 // detecting pthread library
513
514 void os::Linux::libpthread_init() {
515 // Save glibc and pthread version strings.
516 #if !defined(_CS_GNU_LIBC_VERSION) || \
517 !defined(_CS_GNU_LIBPTHREAD_VERSION)
518 #error "glibc too old (< 2.3.2)"
519 #endif
520
521 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
522 assert(n > 0, "cannot retrieve glibc version");
523 char *str = (char *)malloc(n, mtInternal);
524 confstr(_CS_GNU_LIBC_VERSION, str, n);
525 os::Linux::set_glibc_version(str);
526
527 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
528 assert(n > 0, "cannot retrieve pthread version");
529 str = (char *)malloc(n, mtInternal);
530 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
531 os::Linux::set_libpthread_version(str);
532 }
533
534 /////////////////////////////////////////////////////////////////////////////
535 // thread stack expansion
536
537 // os::Linux::manually_expand_stack() takes care of expanding the thread
538 // stack. Note that this is normally not needed: pthread stacks allocate
539 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
540 // committed. Therefore it is not necessary to expand the stack manually.
541 //
542 // Manually expanding the stack was historically needed on LinuxThreads
543 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
544 // it is kept to deal with very rare corner cases:
545 //
546 // For one, user may run the VM on an own implementation of threads
547 // whose stacks are - like the old LinuxThreads - implemented using
548 // mmap(MAP_GROWSDOWN).
549 //
550 // Also, this coding may be needed if the VM is running on the primordial
551 // thread. Normally we avoid running on the primordial thread; however,
552 // user may still invoke the VM on the primordial thread.
553 //
554 // The following historical comment describes the details about running
555 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
556
557
558 // Force Linux kernel to expand current thread stack. If "bottom" is close
559 // to the stack guard, caller should block all signals.
560 //
561 // MAP_GROWSDOWN:
562 // A special mmap() flag that is used to implement thread stacks. It tells
563 // kernel that the memory region should extend downwards when needed. This
564 // allows early versions of LinuxThreads to only mmap the first few pages
565 // when creating a new thread. Linux kernel will automatically expand thread
566 // stack as needed (on page faults).
567 //
568 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
569 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
570 // region, it's hard to tell if the fault is due to a legitimate stack
571 // access or because of reading/writing non-exist memory (e.g. buffer
572 // overrun). As a rule, if the fault happens below current stack pointer,
573 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
574 // application (see Linux kernel fault.c).
575 //
576 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
577 // stack overflow detection.
578 //
579 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
580 // not use MAP_GROWSDOWN.
581 //
582 // To get around the problem and allow stack banging on Linux, we need to
583 // manually expand thread stack after receiving the SIGSEGV.
584 //
585 // There are two ways to expand thread stack to address "bottom", we used
586 // both of them in JVM before 1.5:
587 // 1. adjust stack pointer first so that it is below "bottom", and then
588 // touch "bottom"
589 // 2. mmap() the page in question
590 //
591 // Now alternate signal stack is gone, it's harder to use 2. For instance,
592 // if current sp is already near the lower end of page 101, and we need to
593 // call mmap() to map page 100, it is possible that part of the mmap() frame
594 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
595 // That will destroy the mmap() frame and cause VM to crash.
596 //
597 // The following code works by adjusting sp first, then accessing the "bottom"
598 // page to force a page fault. Linux kernel will then automatically expand the
599 // stack mapping.
600 //
601 // _expand_stack_to() assumes its frame size is less than page size, which
602 // should always be true if the function is not inlined.
603
604 static void NOINLINE _expand_stack_to(address bottom) {
605 address sp;
606 size_t size;
607 volatile char *p;
608
609 // Adjust bottom to point to the largest address within the same page, it
610 // gives us a one-page buffer if alloca() allocates slightly more memory.
611 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
612 bottom += os::Linux::page_size() - 1;
613
614 // sp might be slightly above current stack pointer; if that's the case, we
615 // will alloca() a little more space than necessary, which is OK. Don't use
616 // os::current_stack_pointer(), as its result can be slightly below current
617 // stack pointer, causing us to not alloca enough to reach "bottom".
618 sp = (address)&sp;
619
620 if (sp > bottom) {
621 size = sp - bottom;
622 p = (volatile char *)alloca(size);
623 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
624 p[0] = '\0';
625 }
626 }
627
628 void os::Linux::expand_stack_to(address bottom) {
629 _expand_stack_to(bottom);
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 if (suppress_primordial_thread_resolution) {
917 return false;
918 }
919 char dummy;
920 // If called before init complete, thread stack bottom will be null.
921 // Can be called if fatal error occurs before initialization.
922 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
923 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
924 os::Linux::initial_thread_stack_size() != 0,
925 "os::init did not locate primordial thread's stack region");
926 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
927 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
928 os::Linux::initial_thread_stack_size()) {
929 return true;
930 } else {
931 return false;
932 }
933 }
934
935 // Find the virtual memory area that contains addr
936 static bool find_vma(address addr, address* vma_low, address* vma_high) {
937 FILE *fp = fopen("/proc/self/maps", "r");
938 if (fp) {
939 address low, high;
940 while (!feof(fp)) {
941 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
942 if (low <= addr && addr < high) {
943 if (vma_low) *vma_low = low;
944 if (vma_high) *vma_high = high;
945 fclose(fp);
946 return true;
947 }
948 }
949 for (;;) {
950 int ch = fgetc(fp);
951 if (ch == EOF || ch == (int)'\n') break;
952 }
953 }
954 fclose(fp);
955 }
956 return false;
957 }
958
959 // Locate primordial thread stack. This special handling of primordial thread stack
960 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
961 // bogus value for the primordial process thread. While the launcher has created
962 // the VM in a new thread since JDK 6, we still have to allow for the use of the
963 // JNI invocation API from a primordial thread.
964 void os::Linux::capture_initial_stack(size_t max_size) {
965
966 // max_size is either 0 (which means accept OS default for thread stacks) or
967 // a user-specified value known to be at least the minimum needed. If we
968 // are actually on the primordial thread we can make it appear that we have a
969 // smaller max_size stack by inserting the guard pages at that location. But we
970 // cannot do anything to emulate a larger stack than what has been provided by
971 // the OS or threading library. In fact if we try to use a stack greater than
972 // what is set by rlimit then we will crash the hosting process.
973
974 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
975 // If this is "unlimited" then it will be a huge value.
976 struct rlimit rlim;
977 getrlimit(RLIMIT_STACK, &rlim);
978 size_t stack_size = rlim.rlim_cur;
979
980 // 6308388: a bug in ld.so will relocate its own .data section to the
981 // lower end of primordial stack; reduce ulimit -s value a little bit
982 // so we won't install guard page on ld.so's data section.
983 // But ensure we don't underflow the stack size - allow 1 page spare
984 if (stack_size >= (size_t)(3 * page_size())) {
985 stack_size -= 2 * page_size();
986 }
987
988 // Try to figure out where the stack base (top) is. This is harder.
989 //
990 // When an application is started, glibc saves the initial stack pointer in
991 // a global variable "__libc_stack_end", which is then used by system
992 // libraries. __libc_stack_end should be pretty close to stack top. The
993 // variable is available since the very early days. However, because it is
994 // a private interface, it could disappear in the future.
995 //
996 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
997 // to __libc_stack_end, it is very close to stack top, but isn't the real
998 // stack top. Note that /proc may not exist if VM is running as a chroot
999 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1000 // /proc/<pid>/stat could change in the future (though unlikely).
1001 //
1002 // We try __libc_stack_end first. If that doesn't work, look for
1003 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1004 // as a hint, which should work well in most cases.
1005
1006 uintptr_t stack_start;
1007
1008 // try __libc_stack_end first
1009 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1010 if (p && *p) {
1011 stack_start = *p;
1012 } else {
1013 // see if we can get the start_stack field from /proc/self/stat
1014 FILE *fp;
1015 int pid;
1016 char state;
1017 int ppid;
1018 int pgrp;
1019 int session;
1020 int nr;
1021 int tpgrp;
1022 unsigned long flags;
1023 unsigned long minflt;
1024 unsigned long cminflt;
1025 unsigned long majflt;
1026 unsigned long cmajflt;
1027 unsigned long utime;
1028 unsigned long stime;
1029 long cutime;
1030 long cstime;
1031 long prio;
1032 long nice;
1033 long junk;
1034 long it_real;
1035 uintptr_t start;
1036 uintptr_t vsize;
1037 intptr_t rss;
1038 uintptr_t rsslim;
1039 uintptr_t scodes;
1040 uintptr_t ecode;
1041 int i;
1042
1043 // Figure what the primordial thread stack base is. Code is inspired
1044 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1045 // followed by command name surrounded by parentheses, state, etc.
1046 char stat[2048];
1047 int statlen;
1048
1049 fp = fopen("/proc/self/stat", "r");
1050 if (fp) {
1051 statlen = fread(stat, 1, 2047, fp);
1052 stat[statlen] = '\0';
1053 fclose(fp);
1054
1055 // Skip pid and the command string. Note that we could be dealing with
1056 // weird command names, e.g. user could decide to rename java launcher
1057 // to "java 1.4.2 :)", then the stat file would look like
1058 // 1234 (java 1.4.2 :)) R ... ...
1059 // We don't really need to know the command string, just find the last
1060 // occurrence of ")" and then start parsing from there. See bug 4726580.
1061 char * s = strrchr(stat, ')');
1062
1063 i = 0;
1064 if (s) {
1065 // Skip blank chars
1066 do { s++; } while (s && isspace(*s));
1067
1068 #define _UFM UINTX_FORMAT
1069 #define _DFM INTX_FORMAT
1070
1071 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1072 // 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
1073 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,
1074 &state, // 3 %c
1075 &ppid, // 4 %d
1076 &pgrp, // 5 %d
1077 &session, // 6 %d
1078 &nr, // 7 %d
1079 &tpgrp, // 8 %d
1080 &flags, // 9 %lu
1081 &minflt, // 10 %lu
1082 &cminflt, // 11 %lu
1083 &majflt, // 12 %lu
1084 &cmajflt, // 13 %lu
1085 &utime, // 14 %lu
1086 &stime, // 15 %lu
1087 &cutime, // 16 %ld
1088 &cstime, // 17 %ld
1089 &prio, // 18 %ld
1090 &nice, // 19 %ld
1091 &junk, // 20 %ld
1092 &it_real, // 21 %ld
1093 &start, // 22 UINTX_FORMAT
1094 &vsize, // 23 UINTX_FORMAT
1095 &rss, // 24 INTX_FORMAT
1096 &rsslim, // 25 UINTX_FORMAT
1097 &scodes, // 26 UINTX_FORMAT
1098 &ecode, // 27 UINTX_FORMAT
1099 &stack_start); // 28 UINTX_FORMAT
1100 }
1101
1102 #undef _UFM
1103 #undef _DFM
1104
1105 if (i != 28 - 2) {
1106 assert(false, "Bad conversion from /proc/self/stat");
1107 // product mode - assume we are the primordial thread, good luck in the
1108 // embedded case.
1109 warning("Can't detect primordial thread stack location - bad conversion");
1110 stack_start = (uintptr_t) &rlim;
1111 }
1112 } else {
1113 // For some reason we can't open /proc/self/stat (for example, running on
1114 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1115 // most cases, so don't abort:
1116 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1117 stack_start = (uintptr_t) &rlim;
1118 }
1119 }
1120
1121 // Now we have a pointer (stack_start) very close to the stack top, the
1122 // next thing to do is to figure out the exact location of stack top. We
1123 // can find out the virtual memory area that contains stack_start by
1124 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1125 // and its upper limit is the real stack top. (again, this would fail if
1126 // running inside chroot, because /proc may not exist.)
1127
1128 uintptr_t stack_top;
1129 address low, high;
1130 if (find_vma((address)stack_start, &low, &high)) {
1131 // success, "high" is the true stack top. (ignore "low", because initial
1132 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1133 stack_top = (uintptr_t)high;
1134 } else {
1135 // failed, likely because /proc/self/maps does not exist
1136 warning("Can't detect primordial thread stack location - find_vma failed");
1137 // best effort: stack_start is normally within a few pages below the real
1138 // stack top, use it as stack top, and reduce stack size so we won't put
1139 // guard page outside stack.
1140 stack_top = stack_start;
1141 stack_size -= 16 * page_size();
1142 }
1143
1144 // stack_top could be partially down the page so align it
1145 stack_top = align_up(stack_top, page_size());
1146
1147 // Allowed stack value is minimum of max_size and what we derived from rlimit
1148 if (max_size > 0) {
1149 _initial_thread_stack_size = MIN2(max_size, stack_size);
1150 } else {
1151 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1152 // clamp it at 8MB as we do on Solaris
1153 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1154 }
1155 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1156 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1157
1158 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1159
1160 if (log_is_enabled(Info, os, thread)) {
1161 // See if we seem to be on primordial process thread
1162 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1163 uintptr_t(&rlim) < stack_top;
1164
1165 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1166 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1167 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1168 stack_top, intptr_t(_initial_thread_stack_bottom));
1169 }
1170 }
1171
1172 ////////////////////////////////////////////////////////////////////////////////
1173 // time support
1174
1175 // Time since start-up in seconds to a fine granularity.
1176 // Used by VMSelfDestructTimer and the MemProfiler.
1177 double os::elapsedTime() {
1178
1179 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1180 }
1181
1182 jlong os::elapsed_counter() {
1183 return javaTimeNanos() - initial_time_count;
1184 }
1185
1186 jlong os::elapsed_frequency() {
1187 return NANOSECS_PER_SEC; // nanosecond resolution
1188 }
1189
1190 bool os::supports_vtime() { return true; }
1191 bool os::enable_vtime() { return false; }
1192 bool os::vtime_enabled() { return false; }
1193
1194 double os::elapsedVTime() {
1195 struct rusage usage;
1196 int retval = getrusage(RUSAGE_THREAD, &usage);
1197 if (retval == 0) {
1198 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);
1199 } else {
1200 // better than nothing, but not much
1201 return elapsedTime();
1202 }
1203 }
1204
1205 jlong os::javaTimeMillis() {
1206 timeval time;
1207 int status = gettimeofday(&time, NULL);
1208 assert(status != -1, "linux error");
1209 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1210 }
1211
1212 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1213 timeval time;
1214 int status = gettimeofday(&time, NULL);
1215 assert(status != -1, "linux error");
1216 seconds = jlong(time.tv_sec);
1217 nanos = jlong(time.tv_usec) * 1000;
1218 }
1219
1220
1221 #ifndef CLOCK_MONOTONIC
1222 #define CLOCK_MONOTONIC (1)
1223 #endif
1224
1225 void os::Linux::clock_init() {
1226 // we do dlopen's in this particular order due to bug in linux
1227 // dynamical loader (see 6348968) leading to crash on exit
1228 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1229 if (handle == NULL) {
1230 handle = dlopen("librt.so", RTLD_LAZY);
1231 }
1232
1233 if (handle) {
1234 int (*clock_getres_func)(clockid_t, struct timespec*) =
1235 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1236 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1237 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1238 if (clock_getres_func && clock_gettime_func) {
1239 // See if monotonic clock is supported by the kernel. Note that some
1240 // early implementations simply return kernel jiffies (updated every
1241 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1242 // for nano time (though the monotonic property is still nice to have).
1243 // It's fixed in newer kernels, however clock_getres() still returns
1244 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1245 // resolution for now. Hopefully as people move to new kernels, this
1246 // won't be a problem.
1247 struct timespec res;
1248 struct timespec tp;
1249 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1250 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1251 // yes, monotonic clock is supported
1252 _clock_gettime = clock_gettime_func;
1253 return;
1254 } else {
1255 // close librt if there is no monotonic clock
1256 dlclose(handle);
1257 }
1258 }
1259 }
1260 warning("No monotonic clock was available - timed services may " \
1261 "be adversely affected if the time-of-day clock changes");
1262 }
1263
1264 #ifndef SYS_clock_getres
1265 #if defined(X86) || defined(PPC64) || defined(S390)
1266 #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261)
1267 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1268 #else
1269 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1270 #define sys_clock_getres(x,y) -1
1271 #endif
1272 #else
1273 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1274 #endif
1275
1276 void os::Linux::fast_thread_clock_init() {
1277 if (!UseLinuxPosixThreadCPUClocks) {
1278 return;
1279 }
1280 clockid_t clockid;
1281 struct timespec tp;
1282 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1283 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1284
1285 // Switch to using fast clocks for thread cpu time if
1286 // the sys_clock_getres() returns 0 error code.
1287 // Note, that some kernels may support the current thread
1288 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1289 // returned by the pthread_getcpuclockid().
1290 // If the fast Posix clocks are supported then the sys_clock_getres()
1291 // must return at least tp.tv_sec == 0 which means a resolution
1292 // better than 1 sec. This is extra check for reliability.
1293
1294 if (pthread_getcpuclockid_func &&
1295 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1296 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1297 _supports_fast_thread_cpu_time = true;
1298 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1299 }
1300 }
1301
1302 jlong os::javaTimeNanos() {
1303 if (os::supports_monotonic_clock()) {
1304 struct timespec tp;
1305 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1306 assert(status == 0, "gettime error");
1307 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1308 return result;
1309 } else {
1310 timeval time;
1311 int status = gettimeofday(&time, NULL);
1312 assert(status != -1, "linux error");
1313 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1314 return 1000 * usecs;
1315 }
1316 }
1317
1318 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1319 if (os::supports_monotonic_clock()) {
1320 info_ptr->max_value = ALL_64_BITS;
1321
1322 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1323 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1324 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1325 } else {
1326 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1327 info_ptr->max_value = ALL_64_BITS;
1328
1329 // gettimeofday is a real time clock so it skips
1330 info_ptr->may_skip_backward = true;
1331 info_ptr->may_skip_forward = true;
1332 }
1333
1334 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1335 }
1336
1337 // Return the real, user, and system times in seconds from an
1338 // arbitrary fixed point in the past.
1339 bool os::getTimesSecs(double* process_real_time,
1340 double* process_user_time,
1341 double* process_system_time) {
1342 struct tms ticks;
1343 clock_t real_ticks = times(&ticks);
1344
1345 if (real_ticks == (clock_t) (-1)) {
1346 return false;
1347 } else {
1348 double ticks_per_second = (double) clock_tics_per_sec;
1349 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1350 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1351 *process_real_time = ((double) real_ticks) / ticks_per_second;
1352
1353 return true;
1354 }
1355 }
1356
1357
1358 char * os::local_time_string(char *buf, size_t buflen) {
1359 struct tm t;
1360 time_t long_time;
1361 time(&long_time);
1362 localtime_r(&long_time, &t);
1363 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1364 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1365 t.tm_hour, t.tm_min, t.tm_sec);
1366 return buf;
1367 }
1368
1369 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1370 return localtime_r(clock, res);
1371 }
1372
1373 ////////////////////////////////////////////////////////////////////////////////
1374 // runtime exit support
1375
1376 // Note: os::shutdown() might be called very early during initialization, or
1377 // called from signal handler. Before adding something to os::shutdown(), make
1378 // sure it is async-safe and can handle partially initialized VM.
1379 void os::shutdown() {
1380
1381 // allow PerfMemory to attempt cleanup of any persistent resources
1382 perfMemory_exit();
1383
1384 // needs to remove object in file system
1385 AttachListener::abort();
1386
1387 // flush buffered output, finish log files
1388 ostream_abort();
1389
1390 // Check for abort hook
1391 abort_hook_t abort_hook = Arguments::abort_hook();
1392 if (abort_hook != NULL) {
1393 abort_hook();
1394 }
1395
1396 }
1397
1398 // Note: os::abort() might be called very early during initialization, or
1399 // called from signal handler. Before adding something to os::abort(), make
1400 // sure it is async-safe and can handle partially initialized VM.
1401 void os::abort(bool dump_core, void* siginfo, const void* context) {
1402 os::shutdown();
1403 if (dump_core) {
1404 #ifndef PRODUCT
1405 fdStream out(defaultStream::output_fd());
1406 out.print_raw("Current thread is ");
1407 char buf[16];
1408 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1409 out.print_raw_cr(buf);
1410 out.print_raw_cr("Dumping core ...");
1411 #endif
1412 ::abort(); // dump core
1413 }
1414
1415 ::exit(1);
1416 }
1417
1418 // Die immediately, no exit hook, no abort hook, no cleanup.
1419 void os::die() {
1420 ::abort();
1421 }
1422
1423 // thread_id is kernel thread id (similar to Solaris LWP id)
1424 intx os::current_thread_id() { return os::Linux::gettid(); }
1425 int os::current_process_id() {
1426 return ::getpid();
1427 }
1428
1429 // DLL functions
1430
1431 const char* os::dll_file_extension() { return ".so"; }
1432
1433 // This must be hard coded because it's the system's temporary
1434 // directory not the java application's temp directory, ala java.io.tmpdir.
1435 const char* os::get_temp_directory() { return "/tmp"; }
1436
1437 static bool file_exists(const char* filename) {
1438 struct stat statbuf;
1439 if (filename == NULL || strlen(filename) == 0) {
1440 return false;
1441 }
1442 return os::stat(filename, &statbuf) == 0;
1443 }
1444
1445 // check if addr is inside libjvm.so
1446 bool os::address_is_in_vm(address addr) {
1447 static address libjvm_base_addr;
1448 Dl_info dlinfo;
1449
1450 if (libjvm_base_addr == NULL) {
1451 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1452 libjvm_base_addr = (address)dlinfo.dli_fbase;
1453 }
1454 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1455 }
1456
1457 if (dladdr((void *)addr, &dlinfo) != 0) {
1458 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1459 }
1460
1461 return false;
1462 }
1463
1464 bool os::dll_address_to_function_name(address addr, char *buf,
1465 int buflen, int *offset,
1466 bool demangle) {
1467 // buf is not optional, but offset is optional
1468 assert(buf != NULL, "sanity check");
1469
1470 Dl_info dlinfo;
1471
1472 if (dladdr((void*)addr, &dlinfo) != 0) {
1473 // see if we have a matching symbol
1474 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1475 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1476 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1477 }
1478 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1479 return true;
1480 }
1481 // no matching symbol so try for just file info
1482 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1483 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1484 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1485 return true;
1486 }
1487 }
1488 }
1489
1490 buf[0] = '\0';
1491 if (offset != NULL) *offset = -1;
1492 return false;
1493 }
1494
1495 struct _address_to_library_name {
1496 address addr; // input : memory address
1497 size_t buflen; // size of fname
1498 char* fname; // output: library name
1499 address base; // library base addr
1500 };
1501
1502 static int address_to_library_name_callback(struct dl_phdr_info *info,
1503 size_t size, void *data) {
1504 int i;
1505 bool found = false;
1506 address libbase = NULL;
1507 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1508
1509 // iterate through all loadable segments
1510 for (i = 0; i < info->dlpi_phnum; i++) {
1511 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1512 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1513 // base address of a library is the lowest address of its loaded
1514 // segments.
1515 if (libbase == NULL || libbase > segbase) {
1516 libbase = segbase;
1517 }
1518 // see if 'addr' is within current segment
1519 if (segbase <= d->addr &&
1520 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1521 found = true;
1522 }
1523 }
1524 }
1525
1526 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1527 // so dll_address_to_library_name() can fall through to use dladdr() which
1528 // can figure out executable name from argv[0].
1529 if (found && info->dlpi_name && info->dlpi_name[0]) {
1530 d->base = libbase;
1531 if (d->fname) {
1532 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1533 }
1534 return 1;
1535 }
1536 return 0;
1537 }
1538
1539 bool os::dll_address_to_library_name(address addr, char* buf,
1540 int buflen, int* offset) {
1541 // buf is not optional, but offset is optional
1542 assert(buf != NULL, "sanity check");
1543
1544 Dl_info dlinfo;
1545 struct _address_to_library_name data;
1546
1547 // There is a bug in old glibc dladdr() implementation that it could resolve
1548 // to wrong library name if the .so file has a base address != NULL. Here
1549 // we iterate through the program headers of all loaded libraries to find
1550 // out which library 'addr' really belongs to. This workaround can be
1551 // removed once the minimum requirement for glibc is moved to 2.3.x.
1552 data.addr = addr;
1553 data.fname = buf;
1554 data.buflen = buflen;
1555 data.base = NULL;
1556 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1557
1558 if (rslt) {
1559 // buf already contains library name
1560 if (offset) *offset = addr - data.base;
1561 return true;
1562 }
1563 if (dladdr((void*)addr, &dlinfo) != 0) {
1564 if (dlinfo.dli_fname != NULL) {
1565 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1566 }
1567 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1568 *offset = addr - (address)dlinfo.dli_fbase;
1569 }
1570 return true;
1571 }
1572
1573 buf[0] = '\0';
1574 if (offset) *offset = -1;
1575 return false;
1576 }
1577
1578 // Loads .dll/.so and
1579 // in case of error it checks if .dll/.so was built for the
1580 // same architecture as Hotspot is running on
1581
1582
1583 // Remember the stack's state. The Linux dynamic linker will change
1584 // the stack to 'executable' at most once, so we must safepoint only once.
1585 bool os::Linux::_stack_is_executable = false;
1586
1587 // VM operation that loads a library. This is necessary if stack protection
1588 // of the Java stacks can be lost during loading the library. If we
1589 // do not stop the Java threads, they can stack overflow before the stacks
1590 // are protected again.
1591 class VM_LinuxDllLoad: public VM_Operation {
1592 private:
1593 const char *_filename;
1594 char *_ebuf;
1595 int _ebuflen;
1596 void *_lib;
1597 public:
1598 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1599 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1600 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1601 void doit() {
1602 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1603 os::Linux::_stack_is_executable = true;
1604 }
1605 void* loaded_library() { return _lib; }
1606 };
1607
1608 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1609 void * result = NULL;
1610 bool load_attempted = false;
1611
1612 // Check whether the library to load might change execution rights
1613 // of the stack. If they are changed, the protection of the stack
1614 // guard pages will be lost. We need a safepoint to fix this.
1615 //
1616 // See Linux man page execstack(8) for more info.
1617 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1618 if (!ElfFile::specifies_noexecstack(filename)) {
1619 if (!is_init_completed()) {
1620 os::Linux::_stack_is_executable = true;
1621 // This is OK - No Java threads have been created yet, and hence no
1622 // stack guard pages to fix.
1623 //
1624 // This should happen only when you are building JDK7 using a very
1625 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1626 //
1627 // Dynamic loader will make all stacks executable after
1628 // this function returns, and will not do that again.
1629 assert(Threads::number_of_threads() == 0, "no Java threads should exist yet.");
1630 } else {
1631 warning("You have loaded library %s which might have disabled stack guard. "
1632 "The VM will try to fix the stack guard now.\n"
1633 "It's highly recommended that you fix the library with "
1634 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1635 filename);
1636
1637 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1638 JavaThread *jt = JavaThread::current();
1639 if (jt->thread_state() != _thread_in_native) {
1640 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1641 // that requires ExecStack. Cannot enter safe point. Let's give up.
1642 warning("Unable to fix stack guard. Giving up.");
1643 } else {
1644 if (!LoadExecStackDllInVMThread) {
1645 // This is for the case where the DLL has an static
1646 // constructor function that executes JNI code. We cannot
1647 // load such DLLs in the VMThread.
1648 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1649 }
1650
1651 ThreadInVMfromNative tiv(jt);
1652 debug_only(VMNativeEntryWrapper vew;)
1653
1654 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1655 VMThread::execute(&op);
1656 if (LoadExecStackDllInVMThread) {
1657 result = op.loaded_library();
1658 }
1659 load_attempted = true;
1660 }
1661 }
1662 }
1663 }
1664
1665 if (!load_attempted) {
1666 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1667 }
1668
1669 if (result != NULL) {
1670 // Successful loading
1671 return result;
1672 }
1673
1674 Elf32_Ehdr elf_head;
1675 int diag_msg_max_length=ebuflen-strlen(ebuf);
1676 char* diag_msg_buf=ebuf+strlen(ebuf);
1677
1678 if (diag_msg_max_length==0) {
1679 // No more space in ebuf for additional diagnostics message
1680 return NULL;
1681 }
1682
1683
1684 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1685
1686 if (file_descriptor < 0) {
1687 // Can't open library, report dlerror() message
1688 return NULL;
1689 }
1690
1691 bool failed_to_read_elf_head=
1692 (sizeof(elf_head)!=
1693 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1694
1695 ::close(file_descriptor);
1696 if (failed_to_read_elf_head) {
1697 // file i/o error - report dlerror() msg
1698 return NULL;
1699 }
1700
1701 typedef struct {
1702 Elf32_Half code; // Actual value as defined in elf.h
1703 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1704 unsigned char elf_class; // 32 or 64 bit
1705 unsigned char endianess; // MSB or LSB
1706 char* name; // String representation
1707 } arch_t;
1708
1709 #ifndef EM_486
1710 #define EM_486 6 /* Intel 80486 */
1711 #endif
1712 #ifndef EM_AARCH64
1713 #define EM_AARCH64 183 /* ARM AARCH64 */
1714 #endif
1715
1716 static const arch_t arch_array[]={
1717 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1718 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1719 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1720 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1721 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1722 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1723 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1724 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1725 #if defined(VM_LITTLE_ENDIAN)
1726 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1727 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1728 #else
1729 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1730 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1731 #endif
1732 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1733 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1734 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1735 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1736 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1737 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1738 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1739 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1740 };
1741
1742 #if (defined IA32)
1743 static Elf32_Half running_arch_code=EM_386;
1744 #elif (defined AMD64) || (defined X32)
1745 static Elf32_Half running_arch_code=EM_X86_64;
1746 #elif (defined IA64)
1747 static Elf32_Half running_arch_code=EM_IA_64;
1748 #elif (defined __sparc) && (defined _LP64)
1749 static Elf32_Half running_arch_code=EM_SPARCV9;
1750 #elif (defined __sparc) && (!defined _LP64)
1751 static Elf32_Half running_arch_code=EM_SPARC;
1752 #elif (defined __powerpc64__)
1753 static Elf32_Half running_arch_code=EM_PPC64;
1754 #elif (defined __powerpc__)
1755 static Elf32_Half running_arch_code=EM_PPC;
1756 #elif (defined AARCH64)
1757 static Elf32_Half running_arch_code=EM_AARCH64;
1758 #elif (defined ARM)
1759 static Elf32_Half running_arch_code=EM_ARM;
1760 #elif (defined S390)
1761 static Elf32_Half running_arch_code=EM_S390;
1762 #elif (defined ALPHA)
1763 static Elf32_Half running_arch_code=EM_ALPHA;
1764 #elif (defined MIPSEL)
1765 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1766 #elif (defined PARISC)
1767 static Elf32_Half running_arch_code=EM_PARISC;
1768 #elif (defined MIPS)
1769 static Elf32_Half running_arch_code=EM_MIPS;
1770 #elif (defined M68K)
1771 static Elf32_Half running_arch_code=EM_68K;
1772 #elif (defined SH)
1773 static Elf32_Half running_arch_code=EM_SH;
1774 #else
1775 #error Method os::dll_load requires that one of following is defined:\
1776 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1777 #endif
1778
1779 // Identify compatability class for VM's architecture and library's architecture
1780 // Obtain string descriptions for architectures
1781
1782 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1783 int running_arch_index=-1;
1784
1785 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1786 if (running_arch_code == arch_array[i].code) {
1787 running_arch_index = i;
1788 }
1789 if (lib_arch.code == arch_array[i].code) {
1790 lib_arch.compat_class = arch_array[i].compat_class;
1791 lib_arch.name = arch_array[i].name;
1792 }
1793 }
1794
1795 assert(running_arch_index != -1,
1796 "Didn't find running architecture code (running_arch_code) in arch_array");
1797 if (running_arch_index == -1) {
1798 // Even though running architecture detection failed
1799 // we may still continue with reporting dlerror() message
1800 return NULL;
1801 }
1802
1803 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1804 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1805 return NULL;
1806 }
1807
1808 #ifndef S390
1809 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1810 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1811 return NULL;
1812 }
1813 #endif // !S390
1814
1815 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1816 if (lib_arch.name!=NULL) {
1817 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1818 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1819 lib_arch.name, arch_array[running_arch_index].name);
1820 } else {
1821 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1822 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1823 lib_arch.code,
1824 arch_array[running_arch_index].name);
1825 }
1826 }
1827
1828 return NULL;
1829 }
1830
1831 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1832 int ebuflen) {
1833 void * result = ::dlopen(filename, RTLD_LAZY);
1834 if (result == NULL) {
1835 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1836 ebuf[ebuflen-1] = '\0';
1837 }
1838 return result;
1839 }
1840
1841 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1842 int ebuflen) {
1843 void * result = NULL;
1844 if (LoadExecStackDllInVMThread) {
1845 result = dlopen_helper(filename, ebuf, ebuflen);
1846 }
1847
1848 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1849 // library that requires an executable stack, or which does not have this
1850 // stack attribute set, dlopen changes the stack attribute to executable. The
1851 // read protection of the guard pages gets lost.
1852 //
1853 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1854 // may have been queued at the same time.
1855
1856 if (!_stack_is_executable) {
1857 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
1858 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1859 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1860 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1861 warning("Attempt to reguard stack yellow zone failed.");
1862 }
1863 }
1864 }
1865 }
1866
1867 return result;
1868 }
1869
1870 void* os::dll_lookup(void* handle, const char* name) {
1871 void* res = dlsym(handle, name);
1872 return res;
1873 }
1874
1875 void* os::get_default_process_handle() {
1876 return (void*)::dlopen(NULL, RTLD_LAZY);
1877 }
1878
1879 static bool _print_ascii_file(const char* filename, outputStream* st) {
1880 int fd = ::open(filename, O_RDONLY);
1881 if (fd == -1) {
1882 return false;
1883 }
1884
1885 char buf[33];
1886 int bytes;
1887 buf[32] = '\0';
1888 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1889 st->print_raw(buf, bytes);
1890 }
1891
1892 ::close(fd);
1893
1894 return true;
1895 }
1896
1897 void os::print_dll_info(outputStream *st) {
1898 st->print_cr("Dynamic libraries:");
1899
1900 char fname[32];
1901 pid_t pid = os::Linux::gettid();
1902
1903 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1904
1905 if (!_print_ascii_file(fname, st)) {
1906 st->print("Can not get library information for pid = %d\n", pid);
1907 }
1908 }
1909
1910 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1911 FILE *procmapsFile = NULL;
1912
1913 // Open the procfs maps file for the current process
1914 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1915 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1916 char line[PATH_MAX + 100];
1917
1918 // Read line by line from 'file'
1919 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1920 u8 base, top, offset, inode;
1921 char permissions[5];
1922 char device[6];
1923 char name[PATH_MAX + 1];
1924
1925 // Parse fields from line
1926 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1927 &base, &top, permissions, &offset, device, &inode, name);
1928
1929 // Filter by device id '00:00' so that we only get file system mapped files.
1930 if (strcmp(device, "00:00") != 0) {
1931
1932 // Call callback with the fields of interest
1933 if(callback(name, (address)base, (address)top, param)) {
1934 // Oops abort, callback aborted
1935 fclose(procmapsFile);
1936 return 1;
1937 }
1938 }
1939 }
1940 fclose(procmapsFile);
1941 }
1942 return 0;
1943 }
1944
1945 void os::print_os_info_brief(outputStream* st) {
1946 os::Linux::print_distro_info(st);
1947
1948 os::Posix::print_uname_info(st);
1949
1950 os::Linux::print_libversion_info(st);
1951
1952 }
1953
1954 void os::print_os_info(outputStream* st) {
1955 st->print("OS:");
1956
1957 os::Linux::print_distro_info(st);
1958
1959 os::Posix::print_uname_info(st);
1960
1961 // Print warning if unsafe chroot environment detected
1962 if (unsafe_chroot_detected) {
1963 st->print("WARNING!! ");
1964 st->print_cr("%s", unstable_chroot_error);
1965 }
1966
1967 os::Linux::print_libversion_info(st);
1968
1969 os::Posix::print_rlimit_info(st);
1970
1971 os::Posix::print_load_average(st);
1972
1973 os::Linux::print_full_memory_info(st);
1974
1975 os::Linux::print_proc_sys_info(st);
1976
1977 os::Linux::print_container_info(st);
1978 }
1979
1980 // Try to identify popular distros.
1981 // Most Linux distributions have a /etc/XXX-release file, which contains
1982 // the OS version string. Newer Linux distributions have a /etc/lsb-release
1983 // file that also contains the OS version string. Some have more than one
1984 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
1985 // /etc/redhat-release.), so the order is important.
1986 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
1987 // their own specific XXX-release file as well as a redhat-release file.
1988 // Because of this the XXX-release file needs to be searched for before the
1989 // redhat-release file.
1990 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
1991 // search for redhat-release / SuSE-release needs to be before lsb-release.
1992 // Since the lsb-release file is the new standard it needs to be searched
1993 // before the older style release files.
1994 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
1995 // next to last resort. The os-release file is a new standard that contains
1996 // distribution information and the system-release file seems to be an old
1997 // standard that has been replaced by the lsb-release and os-release files.
1998 // Searching for the debian_version file is the last resort. It contains
1999 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2000 // "Debian " is printed before the contents of the debian_version file.
2001
2002 const char* distro_files[] = {
2003 "/etc/oracle-release",
2004 "/etc/mandriva-release",
2005 "/etc/mandrake-release",
2006 "/etc/sun-release",
2007 "/etc/redhat-release",
2008 "/etc/SuSE-release",
2009 "/etc/lsb-release",
2010 "/etc/turbolinux-release",
2011 "/etc/gentoo-release",
2012 "/etc/ltib-release",
2013 "/etc/angstrom-version",
2014 "/etc/system-release",
2015 "/etc/os-release",
2016 NULL };
2017
2018 void os::Linux::print_distro_info(outputStream* st) {
2019 for (int i = 0;; i++) {
2020 const char* file = distro_files[i];
2021 if (file == NULL) {
2022 break; // done
2023 }
2024 // If file prints, we found it.
2025 if (_print_ascii_file(file, st)) {
2026 return;
2027 }
2028 }
2029
2030 if (file_exists("/etc/debian_version")) {
2031 st->print("Debian ");
2032 _print_ascii_file("/etc/debian_version", st);
2033 } else {
2034 st->print("Linux");
2035 }
2036 st->cr();
2037 }
2038
2039 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2040 char buf[256];
2041 while (fgets(buf, sizeof(buf), fp)) {
2042 // Edit out extra stuff in expected format
2043 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2044 char* ptr = strstr(buf, "\""); // the name is in quotes
2045 if (ptr != NULL) {
2046 ptr++; // go beyond first quote
2047 char* nl = strchr(ptr, '\"');
2048 if (nl != NULL) *nl = '\0';
2049 strncpy(distro, ptr, length);
2050 } else {
2051 ptr = strstr(buf, "=");
2052 ptr++; // go beyond equals then
2053 char* nl = strchr(ptr, '\n');
2054 if (nl != NULL) *nl = '\0';
2055 strncpy(distro, ptr, length);
2056 }
2057 return;
2058 } else if (get_first_line) {
2059 char* nl = strchr(buf, '\n');
2060 if (nl != NULL) *nl = '\0';
2061 strncpy(distro, buf, length);
2062 return;
2063 }
2064 }
2065 // print last line and close
2066 char* nl = strchr(buf, '\n');
2067 if (nl != NULL) *nl = '\0';
2068 strncpy(distro, buf, length);
2069 }
2070
2071 static void parse_os_info(char* distro, size_t length, const char* file) {
2072 FILE* fp = fopen(file, "r");
2073 if (fp != NULL) {
2074 // if suse format, print out first line
2075 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2076 parse_os_info_helper(fp, distro, length, get_first_line);
2077 fclose(fp);
2078 }
2079 }
2080
2081 void os::get_summary_os_info(char* buf, size_t buflen) {
2082 for (int i = 0;; i++) {
2083 const char* file = distro_files[i];
2084 if (file == NULL) {
2085 break; // ran out of distro_files
2086 }
2087 if (file_exists(file)) {
2088 parse_os_info(buf, buflen, file);
2089 return;
2090 }
2091 }
2092 // special case for debian
2093 if (file_exists("/etc/debian_version")) {
2094 strncpy(buf, "Debian ", buflen);
2095 if (buflen > 7) {
2096 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2097 }
2098 } else {
2099 strncpy(buf, "Linux", buflen);
2100 }
2101 }
2102
2103 void os::Linux::print_libversion_info(outputStream* st) {
2104 // libc, pthread
2105 st->print("libc:");
2106 st->print("%s ", os::Linux::glibc_version());
2107 st->print("%s ", os::Linux::libpthread_version());
2108 st->cr();
2109 }
2110
2111 void os::Linux::print_proc_sys_info(outputStream* st) {
2112 st->cr();
2113 st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):");
2114 _print_ascii_file("/proc/sys/kernel/threads-max", st);
2115 st->cr();
2116 st->cr();
2117
2118 st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):");
2119 _print_ascii_file("/proc/sys/vm/max_map_count", st);
2120 st->cr();
2121 st->cr();
2122
2123 st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):");
2124 _print_ascii_file("/proc/sys/kernel/pid_max", st);
2125 st->cr();
2126 st->cr();
2127 }
2128
2129 void os::Linux::print_full_memory_info(outputStream* st) {
2130 st->print("\n/proc/meminfo:\n");
2131 _print_ascii_file("/proc/meminfo", st);
2132 st->cr();
2133 }
2134
2135 void os::Linux::print_container_info(outputStream* st) {
2136 if (!OSContainer::is_containerized()) {
2137 return;
2138 }
2139
2140 st->print("container (cgroup) information:\n");
2141
2142 const char *p_ct = OSContainer::container_type();
2143 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed");
2144
2145 char *p = OSContainer::cpu_cpuset_cpus();
2146 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed");
2147 free(p);
2148
2149 p = OSContainer::cpu_cpuset_memory_nodes();
2150 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed");
2151 free(p);
2152
2153 int i = OSContainer::active_processor_count();
2154 if (i > 0) {
2155 st->print("active_processor_count: %d\n", i);
2156 } else {
2157 st->print("active_processor_count: failed\n");
2158 }
2159
2160 i = OSContainer::cpu_quota();
2161 st->print("cpu_quota: %d\n", i);
2162
2163 i = OSContainer::cpu_period();
2164 st->print("cpu_period: %d\n", i);
2165
2166 i = OSContainer::cpu_shares();
2167 st->print("cpu_shares: %d\n", i);
2168
2169 jlong j = OSContainer::memory_limit_in_bytes();
2170 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2171
2172 j = OSContainer::memory_and_swap_limit_in_bytes();
2173 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2174
2175 j = OSContainer::memory_soft_limit_in_bytes();
2176 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2177
2178 j = OSContainer::OSContainer::memory_usage_in_bytes();
2179 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2180
2181 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2182 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2183 st->cr();
2184 }
2185
2186 void os::print_memory_info(outputStream* st) {
2187
2188 st->print("Memory:");
2189 st->print(" %dk page", os::vm_page_size()>>10);
2190
2191 // values in struct sysinfo are "unsigned long"
2192 struct sysinfo si;
2193 sysinfo(&si);
2194
2195 st->print(", physical " UINT64_FORMAT "k",
2196 os::physical_memory() >> 10);
2197 st->print("(" UINT64_FORMAT "k free)",
2198 os::available_memory() >> 10);
2199 st->print(", swap " UINT64_FORMAT "k",
2200 ((jlong)si.totalswap * si.mem_unit) >> 10);
2201 st->print("(" UINT64_FORMAT "k free)",
2202 ((jlong)si.freeswap * si.mem_unit) >> 10);
2203 st->cr();
2204 }
2205
2206 // Print the first "model name" line and the first "flags" line
2207 // that we find and nothing more. We assume "model name" comes
2208 // before "flags" so if we find a second "model name", then the
2209 // "flags" field is considered missing.
2210 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2211 #if defined(IA32) || defined(AMD64)
2212 // Other platforms have less repetitive cpuinfo files
2213 FILE *fp = fopen("/proc/cpuinfo", "r");
2214 if (fp) {
2215 while (!feof(fp)) {
2216 if (fgets(buf, buflen, fp)) {
2217 // Assume model name comes before flags
2218 bool model_name_printed = false;
2219 if (strstr(buf, "model name") != NULL) {
2220 if (!model_name_printed) {
2221 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2222 st->print_raw(buf);
2223 model_name_printed = true;
2224 } else {
2225 // model name printed but not flags? Odd, just return
2226 fclose(fp);
2227 return true;
2228 }
2229 }
2230 // print the flags line too
2231 if (strstr(buf, "flags") != NULL) {
2232 st->print_raw(buf);
2233 fclose(fp);
2234 return true;
2235 }
2236 }
2237 }
2238 fclose(fp);
2239 }
2240 #endif // x86 platforms
2241 return false;
2242 }
2243
2244 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2245 // Only print the model name if the platform provides this as a summary
2246 if (!print_model_name_and_flags(st, buf, buflen)) {
2247 st->print("\n/proc/cpuinfo:\n");
2248 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2249 st->print_cr(" <Not Available>");
2250 }
2251 }
2252 }
2253
2254 #if defined(AMD64) || defined(IA32) || defined(X32)
2255 const char* search_string = "model name";
2256 #elif defined(M68K)
2257 const char* search_string = "CPU";
2258 #elif defined(PPC64)
2259 const char* search_string = "cpu";
2260 #elif defined(S390)
2261 const char* search_string = "processor";
2262 #elif defined(SPARC)
2263 const char* search_string = "cpu";
2264 #else
2265 const char* search_string = "Processor";
2266 #endif
2267
2268 // Parses the cpuinfo file for string representing the model name.
2269 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2270 FILE* fp = fopen("/proc/cpuinfo", "r");
2271 if (fp != NULL) {
2272 while (!feof(fp)) {
2273 char buf[256];
2274 if (fgets(buf, sizeof(buf), fp)) {
2275 char* start = strstr(buf, search_string);
2276 if (start != NULL) {
2277 char *ptr = start + strlen(search_string);
2278 char *end = buf + strlen(buf);
2279 while (ptr != end) {
2280 // skip whitespace and colon for the rest of the name.
2281 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2282 break;
2283 }
2284 ptr++;
2285 }
2286 if (ptr != end) {
2287 // reasonable string, get rid of newline and keep the rest
2288 char* nl = strchr(buf, '\n');
2289 if (nl != NULL) *nl = '\0';
2290 strncpy(cpuinfo, ptr, length);
2291 fclose(fp);
2292 return;
2293 }
2294 }
2295 }
2296 }
2297 fclose(fp);
2298 }
2299 // cpuinfo not found or parsing failed, just print generic string. The entire
2300 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2301 #if defined(AARCH64)
2302 strncpy(cpuinfo, "AArch64", length);
2303 #elif defined(AMD64)
2304 strncpy(cpuinfo, "x86_64", length);
2305 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2306 strncpy(cpuinfo, "ARM", length);
2307 #elif defined(IA32)
2308 strncpy(cpuinfo, "x86_32", length);
2309 #elif defined(IA64)
2310 strncpy(cpuinfo, "IA64", length);
2311 #elif defined(PPC)
2312 strncpy(cpuinfo, "PPC64", length);
2313 #elif defined(S390)
2314 strncpy(cpuinfo, "S390", length);
2315 #elif defined(SPARC)
2316 strncpy(cpuinfo, "sparcv9", length);
2317 #elif defined(ZERO_LIBARCH)
2318 strncpy(cpuinfo, ZERO_LIBARCH, length);
2319 #else
2320 strncpy(cpuinfo, "unknown", length);
2321 #endif
2322 }
2323
2324 static void print_signal_handler(outputStream* st, int sig,
2325 char* buf, size_t buflen);
2326
2327 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2328 st->print_cr("Signal Handlers:");
2329 print_signal_handler(st, SIGSEGV, buf, buflen);
2330 print_signal_handler(st, SIGBUS , buf, buflen);
2331 print_signal_handler(st, SIGFPE , buf, buflen);
2332 print_signal_handler(st, SIGPIPE, buf, buflen);
2333 print_signal_handler(st, SIGXFSZ, buf, buflen);
2334 print_signal_handler(st, SIGILL , buf, buflen);
2335 print_signal_handler(st, SR_signum, buf, buflen);
2336 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2337 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2338 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2339 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2340 #if defined(PPC64)
2341 print_signal_handler(st, SIGTRAP, buf, buflen);
2342 #endif
2343 }
2344
2345 static char saved_jvm_path[MAXPATHLEN] = {0};
2346
2347 // Find the full path to the current module, libjvm.so
2348 void os::jvm_path(char *buf, jint buflen) {
2349 // Error checking.
2350 if (buflen < MAXPATHLEN) {
2351 assert(false, "must use a large-enough buffer");
2352 buf[0] = '\0';
2353 return;
2354 }
2355 // Lazy resolve the path to current module.
2356 if (saved_jvm_path[0] != 0) {
2357 strcpy(buf, saved_jvm_path);
2358 return;
2359 }
2360
2361 char dli_fname[MAXPATHLEN];
2362 bool ret = dll_address_to_library_name(
2363 CAST_FROM_FN_PTR(address, os::jvm_path),
2364 dli_fname, sizeof(dli_fname), NULL);
2365 assert(ret, "cannot locate libjvm");
2366 char *rp = NULL;
2367 if (ret && dli_fname[0] != '\0') {
2368 rp = os::Posix::realpath(dli_fname, buf, buflen);
2369 }
2370 if (rp == NULL) {
2371 return;
2372 }
2373
2374 if (Arguments::sun_java_launcher_is_altjvm()) {
2375 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2376 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2377 // If "/jre/lib/" appears at the right place in the string, then
2378 // assume we are installed in a JDK and we're done. Otherwise, check
2379 // for a JAVA_HOME environment variable and fix up the path so it
2380 // looks like libjvm.so is installed there (append a fake suffix
2381 // hotspot/libjvm.so).
2382 const char *p = buf + strlen(buf) - 1;
2383 for (int count = 0; p > buf && count < 5; ++count) {
2384 for (--p; p > buf && *p != '/'; --p)
2385 /* empty */ ;
2386 }
2387
2388 if (strncmp(p, "/jre/lib/", 9) != 0) {
2389 // Look for JAVA_HOME in the environment.
2390 char* java_home_var = ::getenv("JAVA_HOME");
2391 if (java_home_var != NULL && java_home_var[0] != 0) {
2392 char* jrelib_p;
2393 int len;
2394
2395 // Check the current module name "libjvm.so".
2396 p = strrchr(buf, '/');
2397 if (p == NULL) {
2398 return;
2399 }
2400 assert(strstr(p, "/libjvm") == p, "invalid library name");
2401
2402 rp = os::Posix::realpath(java_home_var, buf, buflen);
2403 if (rp == NULL) {
2404 return;
2405 }
2406
2407 // determine if this is a legacy image or modules image
2408 // modules image doesn't have "jre" subdirectory
2409 len = strlen(buf);
2410 assert(len < buflen, "Ran out of buffer room");
2411 jrelib_p = buf + len;
2412 snprintf(jrelib_p, buflen-len, "/jre/lib");
2413 if (0 != access(buf, F_OK)) {
2414 snprintf(jrelib_p, buflen-len, "/lib");
2415 }
2416
2417 if (0 == access(buf, F_OK)) {
2418 // Use current module name "libjvm.so"
2419 len = strlen(buf);
2420 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2421 } else {
2422 // Go back to path of .so
2423 rp = os::Posix::realpath(dli_fname, buf, buflen);
2424 if (rp == NULL) {
2425 return;
2426 }
2427 }
2428 }
2429 }
2430 }
2431
2432 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2433 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2434 }
2435
2436 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2437 // no prefix required, not even "_"
2438 }
2439
2440 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2441 // no suffix required
2442 }
2443
2444 ////////////////////////////////////////////////////////////////////////////////
2445 // sun.misc.Signal support
2446
2447 static volatile jint sigint_count = 0;
2448
2449 static void UserHandler(int sig, void *siginfo, void *context) {
2450 // 4511530 - sem_post is serialized and handled by the manager thread. When
2451 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2452 // don't want to flood the manager thread with sem_post requests.
2453 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2454 return;
2455 }
2456
2457 // Ctrl-C is pressed during error reporting, likely because the error
2458 // handler fails to abort. Let VM die immediately.
2459 if (sig == SIGINT && VMError::is_error_reported()) {
2460 os::die();
2461 }
2462
2463 os::signal_notify(sig);
2464 }
2465
2466 void* os::user_handler() {
2467 return CAST_FROM_FN_PTR(void*, UserHandler);
2468 }
2469
2470 static struct timespec create_semaphore_timespec(unsigned int sec, int nsec) {
2471 struct timespec ts;
2472 // Semaphore's are always associated with CLOCK_REALTIME
2473 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2474 // see os_posix.cpp for discussion on overflow checking
2475 if (sec >= MAX_SECS) {
2476 ts.tv_sec += MAX_SECS;
2477 ts.tv_nsec = 0;
2478 } else {
2479 ts.tv_sec += sec;
2480 ts.tv_nsec += nsec;
2481 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2482 ts.tv_nsec -= NANOSECS_PER_SEC;
2483 ++ts.tv_sec; // note: this must be <= max_secs
2484 }
2485 }
2486
2487 return ts;
2488 }
2489
2490 extern "C" {
2491 typedef void (*sa_handler_t)(int);
2492 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2493 }
2494
2495 void* os::signal(int signal_number, void* handler) {
2496 struct sigaction sigAct, oldSigAct;
2497
2498 sigfillset(&(sigAct.sa_mask));
2499 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2500 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2501
2502 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2503 // -1 means registration failed
2504 return (void *)-1;
2505 }
2506
2507 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2508 }
2509
2510 void os::signal_raise(int signal_number) {
2511 ::raise(signal_number);
2512 }
2513
2514 // The following code is moved from os.cpp for making this
2515 // code platform specific, which it is by its very nature.
2516
2517 // Will be modified when max signal is changed to be dynamic
2518 int os::sigexitnum_pd() {
2519 return NSIG;
2520 }
2521
2522 // a counter for each possible signal value
2523 static volatile jint pending_signals[NSIG+1] = { 0 };
2524
2525 // Linux(POSIX) specific hand shaking semaphore.
2526 static Semaphore* sig_sem = NULL;
2527 static PosixSemaphore sr_semaphore;
2528
2529 static void jdk_misc_signal_init() {
2530 // Initialize signal structures
2531 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2532
2533 // Initialize signal semaphore
2534 sig_sem = new Semaphore();
2535 }
2536
2537 void os::signal_notify(int sig) {
2538 if (sig_sem != NULL) {
2539 Atomic::inc(&pending_signals[sig]);
2540 sig_sem->signal();
2541 } else {
2542 // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2543 // initialization isn't called.
2544 assert(ReduceSignalUsage, "signal semaphore should be created");
2545 }
2546 }
2547
2548 static int check_pending_signals() {
2549 Atomic::store(0, &sigint_count);
2550 for (;;) {
2551 for (int i = 0; i < NSIG + 1; i++) {
2552 jint n = pending_signals[i];
2553 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2554 return i;
2555 }
2556 }
2557 JavaThread *thread = JavaThread::current();
2558 ThreadBlockInVM tbivm(thread);
2559
2560 bool threadIsSuspended;
2561 do {
2562 thread->set_suspend_equivalent();
2563 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2564 sig_sem->wait();
2565
2566 // were we externally suspended while we were waiting?
2567 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2568 if (threadIsSuspended) {
2569 // The semaphore has been incremented, but while we were waiting
2570 // another thread suspended us. We don't want to continue running
2571 // while suspended because that would surprise the thread that
2572 // suspended us.
2573 sig_sem->signal();
2574
2575 thread->java_suspend_self();
2576 }
2577 } while (threadIsSuspended);
2578 }
2579 }
2580
2581 int os::signal_wait() {
2582 return check_pending_signals();
2583 }
2584
2585 ////////////////////////////////////////////////////////////////////////////////
2586 // Virtual Memory
2587
2588 int os::vm_page_size() {
2589 // Seems redundant as all get out
2590 assert(os::Linux::page_size() != -1, "must call os::init");
2591 return os::Linux::page_size();
2592 }
2593
2594 // Solaris allocates memory by pages.
2595 int os::vm_allocation_granularity() {
2596 assert(os::Linux::page_size() != -1, "must call os::init");
2597 return os::Linux::page_size();
2598 }
2599
2600 // Rationale behind this function:
2601 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2602 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2603 // samples for JITted code. Here we create private executable mapping over the code cache
2604 // and then we can use standard (well, almost, as mapping can change) way to provide
2605 // info for the reporting script by storing timestamp and location of symbol
2606 void linux_wrap_code(char* base, size_t size) {
2607 static volatile jint cnt = 0;
2608
2609 if (!UseOprofile) {
2610 return;
2611 }
2612
2613 char buf[PATH_MAX+1];
2614 int num = Atomic::add(1, &cnt);
2615
2616 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2617 os::get_temp_directory(), os::current_process_id(), num);
2618 unlink(buf);
2619
2620 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2621
2622 if (fd != -1) {
2623 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2624 if (rv != (off_t)-1) {
2625 if (::write(fd, "", 1) == 1) {
2626 mmap(base, size,
2627 PROT_READ|PROT_WRITE|PROT_EXEC,
2628 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2629 }
2630 }
2631 ::close(fd);
2632 unlink(buf);
2633 }
2634 }
2635
2636 static bool recoverable_mmap_error(int err) {
2637 // See if the error is one we can let the caller handle. This
2638 // list of errno values comes from JBS-6843484. I can't find a
2639 // Linux man page that documents this specific set of errno
2640 // values so while this list currently matches Solaris, it may
2641 // change as we gain experience with this failure mode.
2642 switch (err) {
2643 case EBADF:
2644 case EINVAL:
2645 case ENOTSUP:
2646 // let the caller deal with these errors
2647 return true;
2648
2649 default:
2650 // Any remaining errors on this OS can cause our reserved mapping
2651 // to be lost. That can cause confusion where different data
2652 // structures think they have the same memory mapped. The worst
2653 // scenario is if both the VM and a library think they have the
2654 // same memory mapped.
2655 return false;
2656 }
2657 }
2658
2659 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2660 int err) {
2661 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2662 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2663 os::strerror(err), err);
2664 }
2665
2666 static void warn_fail_commit_memory(char* addr, size_t size,
2667 size_t alignment_hint, bool exec,
2668 int err) {
2669 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2670 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2671 alignment_hint, exec, os::strerror(err), err);
2672 }
2673
2674 // NOTE: Linux kernel does not really reserve the pages for us.
2675 // All it does is to check if there are enough free pages
2676 // left at the time of mmap(). This could be a potential
2677 // problem.
2678 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2679 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2680 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2681 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2682 if (res != (uintptr_t) MAP_FAILED) {
2683 if (UseNUMAInterleaving) {
2684 numa_make_global(addr, size);
2685 }
2686 return 0;
2687 }
2688
2689 int err = errno; // save errno from mmap() call above
2690
2691 if (!recoverable_mmap_error(err)) {
2692 warn_fail_commit_memory(addr, size, exec, err);
2693 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2694 }
2695
2696 return err;
2697 }
2698
2699 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2700 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2701 }
2702
2703 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2704 const char* mesg) {
2705 assert(mesg != NULL, "mesg must be specified");
2706 int err = os::Linux::commit_memory_impl(addr, size, exec);
2707 if (err != 0) {
2708 // the caller wants all commit errors to exit with the specified mesg:
2709 warn_fail_commit_memory(addr, size, exec, err);
2710 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2711 }
2712 }
2713
2714 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2715 #ifndef MAP_HUGETLB
2716 #define MAP_HUGETLB 0x40000
2717 #endif
2718
2719 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2720 #ifndef MADV_HUGEPAGE
2721 #define MADV_HUGEPAGE 14
2722 #endif
2723
2724 int os::Linux::commit_memory_impl(char* addr, size_t size,
2725 size_t alignment_hint, bool exec) {
2726 int err = os::Linux::commit_memory_impl(addr, size, exec);
2727 if (err == 0) {
2728 realign_memory(addr, size, alignment_hint);
2729 }
2730 return err;
2731 }
2732
2733 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2734 bool exec) {
2735 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2736 }
2737
2738 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2739 size_t alignment_hint, bool exec,
2740 const char* mesg) {
2741 assert(mesg != NULL, "mesg must be specified");
2742 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2743 if (err != 0) {
2744 // the caller wants all commit errors to exit with the specified mesg:
2745 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2746 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2747 }
2748 }
2749
2750 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2751 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2752 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2753 // be supported or the memory may already be backed by huge pages.
2754 ::madvise(addr, bytes, MADV_HUGEPAGE);
2755 }
2756 }
2757
2758 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2759 // This method works by doing an mmap over an existing mmaping and effectively discarding
2760 // the existing pages. However it won't work for SHM-based large pages that cannot be
2761 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2762 // small pages on top of the SHM segment. This method always works for small pages, so we
2763 // allow that in any case.
2764 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2765 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2766 }
2767 }
2768
2769 void os::numa_make_global(char *addr, size_t bytes) {
2770 Linux::numa_interleave_memory(addr, bytes);
2771 }
2772
2773 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2774 // bind policy to MPOL_PREFERRED for the current thread.
2775 #define USE_MPOL_PREFERRED 0
2776
2777 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2778 // To make NUMA and large pages more robust when both enabled, we need to ease
2779 // the requirements on where the memory should be allocated. MPOL_BIND is the
2780 // default policy and it will force memory to be allocated on the specified
2781 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2782 // the specified node, but will not force it. Using this policy will prevent
2783 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2784 // free large pages.
2785 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2786 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2787 }
2788
2789 bool os::numa_topology_changed() { return false; }
2790
2791 size_t os::numa_get_groups_num() {
2792 // Return just the number of nodes in which it's possible to allocate memory
2793 // (in numa terminology, configured nodes).
2794 return Linux::numa_num_configured_nodes();
2795 }
2796
2797 int os::numa_get_group_id() {
2798 int cpu_id = Linux::sched_getcpu();
2799 if (cpu_id != -1) {
2800 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2801 if (lgrp_id != -1) {
2802 return lgrp_id;
2803 }
2804 }
2805 return 0;
2806 }
2807
2808 int os::Linux::get_existing_num_nodes() {
2809 int node;
2810 int highest_node_number = Linux::numa_max_node();
2811 int num_nodes = 0;
2812
2813 // Get the total number of nodes in the system including nodes without memory.
2814 for (node = 0; node <= highest_node_number; node++) {
2815 if (isnode_in_existing_nodes(node)) {
2816 num_nodes++;
2817 }
2818 }
2819 return num_nodes;
2820 }
2821
2822 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2823 int highest_node_number = Linux::numa_max_node();
2824 size_t i = 0;
2825
2826 // Map all node ids in which it is possible to allocate memory. Also nodes are
2827 // not always consecutively available, i.e. available from 0 to the highest
2828 // node number. If the nodes have been bound explicitly using numactl membind,
2829 // then allocate memory from those nodes only.
2830 for (int node = 0; node <= highest_node_number; node++) {
2831 if (Linux::isnode_in_bound_nodes((unsigned int)node)) {
2832 ids[i++] = node;
2833 }
2834 }
2835 return i;
2836 }
2837
2838 bool os::get_page_info(char *start, page_info* info) {
2839 return false;
2840 }
2841
2842 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2843 page_info* page_found) {
2844 return end;
2845 }
2846
2847
2848 int os::Linux::sched_getcpu_syscall(void) {
2849 unsigned int cpu = 0;
2850 int retval = -1;
2851
2852 #if defined(IA32)
2853 #ifndef SYS_getcpu
2854 #define SYS_getcpu 318
2855 #endif
2856 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2857 #elif defined(AMD64)
2858 // Unfortunately we have to bring all these macros here from vsyscall.h
2859 // to be able to compile on old linuxes.
2860 #define __NR_vgetcpu 2
2861 #define VSYSCALL_START (-10UL << 20)
2862 #define VSYSCALL_SIZE 1024
2863 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2864 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2865 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2866 retval = vgetcpu(&cpu, NULL, NULL);
2867 #endif
2868
2869 return (retval == -1) ? retval : cpu;
2870 }
2871
2872 void os::Linux::sched_getcpu_init() {
2873 // sched_getcpu() should be in libc.
2874 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2875 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2876
2877 // If it's not, try a direct syscall.
2878 if (sched_getcpu() == -1) {
2879 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2880 (void*)&sched_getcpu_syscall));
2881 }
2882
2883 if (sched_getcpu() == -1) {
2884 vm_exit_during_initialization("getcpu(2) system call not supported by kernel");
2885 }
2886 }
2887
2888 // Something to do with the numa-aware allocator needs these symbols
2889 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2890 extern "C" JNIEXPORT void numa_error(char *where) { }
2891
2892 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2893 // load symbol from base version instead.
2894 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2895 void *f = dlvsym(handle, name, "libnuma_1.1");
2896 if (f == NULL) {
2897 f = dlsym(handle, name);
2898 }
2899 return f;
2900 }
2901
2902 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2903 // Return NULL if the symbol is not defined in this particular version.
2904 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2905 return dlvsym(handle, name, "libnuma_1.2");
2906 }
2907
2908 bool os::Linux::libnuma_init() {
2909 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
2910 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2911 if (handle != NULL) {
2912 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2913 libnuma_dlsym(handle, "numa_node_to_cpus")));
2914 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2915 libnuma_dlsym(handle, "numa_max_node")));
2916 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2917 libnuma_dlsym(handle, "numa_num_configured_nodes")));
2918 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2919 libnuma_dlsym(handle, "numa_available")));
2920 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2921 libnuma_dlsym(handle, "numa_tonode_memory")));
2922 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2923 libnuma_dlsym(handle, "numa_interleave_memory")));
2924 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2925 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2926 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2927 libnuma_dlsym(handle, "numa_set_bind_policy")));
2928 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2929 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2930 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2931 libnuma_dlsym(handle, "numa_distance")));
2932 set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t,
2933 libnuma_v2_dlsym(handle, "numa_get_membind")));
2934
2935 if (numa_available() != -1) {
2936 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2937 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2938 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2939 // Create an index -> node mapping, since nodes are not always consecutive
2940 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2941 rebuild_nindex_to_node_map();
2942 // Create a cpu -> node mapping
2943 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2944 rebuild_cpu_to_node_map();
2945 return true;
2946 }
2947 }
2948 }
2949 return false;
2950 }
2951
2952 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2953 // Creating guard page is very expensive. Java thread has HotSpot
2954 // guard pages, only enable glibc guard page for non-Java threads.
2955 // (Remember: compiler thread is a Java thread, too!)
2956 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2957 }
2958
2959 void os::Linux::rebuild_nindex_to_node_map() {
2960 int highest_node_number = Linux::numa_max_node();
2961
2962 nindex_to_node()->clear();
2963 for (int node = 0; node <= highest_node_number; node++) {
2964 if (Linux::isnode_in_existing_nodes(node)) {
2965 nindex_to_node()->append(node);
2966 }
2967 }
2968 }
2969
2970 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2971 // The table is later used in get_node_by_cpu().
2972 void os::Linux::rebuild_cpu_to_node_map() {
2973 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2974 // in libnuma (possible values are starting from 16,
2975 // and continuing up with every other power of 2, but less
2976 // than the maximum number of CPUs supported by kernel), and
2977 // is a subject to change (in libnuma version 2 the requirements
2978 // are more reasonable) we'll just hardcode the number they use
2979 // in the library.
2980 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2981
2982 size_t cpu_num = processor_count();
2983 size_t cpu_map_size = NCPUS / BitsPerCLong;
2984 size_t cpu_map_valid_size =
2985 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2986
2987 cpu_to_node()->clear();
2988 cpu_to_node()->at_grow(cpu_num - 1);
2989
2990 size_t node_num = get_existing_num_nodes();
2991
2992 int distance = 0;
2993 int closest_distance = INT_MAX;
2994 int closest_node = 0;
2995 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2996 for (size_t i = 0; i < node_num; i++) {
2997 // Check if node is configured (not a memory-less node). If it is not, find
2998 // the closest configured node. Check also if node is bound, i.e. it's allowed
2999 // to allocate memory from the node. If it's not allowed, map cpus in that node
3000 // to the closest node from which memory allocation is allowed.
3001 if (!isnode_in_configured_nodes(nindex_to_node()->at(i)) ||
3002 !isnode_in_bound_nodes(nindex_to_node()->at(i))) {
3003 closest_distance = INT_MAX;
3004 // Check distance from all remaining nodes in the system. Ignore distance
3005 // from itself, from another non-configured node, and from another non-bound
3006 // node.
3007 for (size_t m = 0; m < node_num; m++) {
3008 if (m != i &&
3009 isnode_in_configured_nodes(nindex_to_node()->at(m)) &&
3010 isnode_in_bound_nodes(nindex_to_node()->at(m))) {
3011 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3012 // If a closest node is found, update. There is always at least one
3013 // configured and bound node in the system so there is always at least
3014 // one node close.
3015 if (distance != 0 && distance < closest_distance) {
3016 closest_distance = distance;
3017 closest_node = nindex_to_node()->at(m);
3018 }
3019 }
3020 }
3021 } else {
3022 // Current node is already a configured node.
3023 closest_node = nindex_to_node()->at(i);
3024 }
3025
3026 // Get cpus from the original node and map them to the closest node. If node
3027 // is a configured node (not a memory-less node), then original node and
3028 // closest node are the same.
3029 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3030 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3031 if (cpu_map[j] != 0) {
3032 for (size_t k = 0; k < BitsPerCLong; k++) {
3033 if (cpu_map[j] & (1UL << k)) {
3034 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3035 }
3036 }
3037 }
3038 }
3039 }
3040 }
3041 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3042 }
3043
3044 int os::Linux::get_node_by_cpu(int cpu_id) {
3045 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3046 return cpu_to_node()->at(cpu_id);
3047 }
3048 return -1;
3049 }
3050
3051 GrowableArray<int>* os::Linux::_cpu_to_node;
3052 GrowableArray<int>* os::Linux::_nindex_to_node;
3053 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3054 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3055 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3056 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3057 os::Linux::numa_available_func_t os::Linux::_numa_available;
3058 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3059 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3060 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3061 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3062 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3063 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3064 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind;
3065 unsigned long* os::Linux::_numa_all_nodes;
3066 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3067 struct bitmask* os::Linux::_numa_nodes_ptr;
3068
3069 bool os::pd_uncommit_memory(char* addr, size_t size) {
3070 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3071 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3072 return res != (uintptr_t) MAP_FAILED;
3073 }
3074
3075 static address get_stack_commited_bottom(address bottom, size_t size) {
3076 address nbot = bottom;
3077 address ntop = bottom + size;
3078
3079 size_t page_sz = os::vm_page_size();
3080 unsigned pages = size / page_sz;
3081
3082 unsigned char vec[1];
3083 unsigned imin = 1, imax = pages + 1, imid;
3084 int mincore_return_value = 0;
3085
3086 assert(imin <= imax, "Unexpected page size");
3087
3088 while (imin < imax) {
3089 imid = (imax + imin) / 2;
3090 nbot = ntop - (imid * page_sz);
3091
3092 // Use a trick with mincore to check whether the page is mapped or not.
3093 // mincore sets vec to 1 if page resides in memory and to 0 if page
3094 // is swapped output but if page we are asking for is unmapped
3095 // it returns -1,ENOMEM
3096 mincore_return_value = mincore(nbot, page_sz, vec);
3097
3098 if (mincore_return_value == -1) {
3099 // Page is not mapped go up
3100 // to find first mapped page
3101 if (errno != EAGAIN) {
3102 assert(errno == ENOMEM, "Unexpected mincore errno");
3103 imax = imid;
3104 }
3105 } else {
3106 // Page is mapped go down
3107 // to find first not mapped page
3108 imin = imid + 1;
3109 }
3110 }
3111
3112 nbot = nbot + page_sz;
3113
3114 // Adjust stack bottom one page up if last checked page is not mapped
3115 if (mincore_return_value == -1) {
3116 nbot = nbot + page_sz;
3117 }
3118
3119 return nbot;
3120 }
3121
3122 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3123 int mincore_return_value;
3124 const size_t stripe = 1024; // query this many pages each time
3125 unsigned char vec[stripe + 1];
3126 // set a guard
3127 vec[stripe] = 'X';
3128
3129 const size_t page_sz = os::vm_page_size();
3130 size_t pages = size / page_sz;
3131
3132 assert(is_aligned(start, page_sz), "Start address must be page aligned");
3133 assert(is_aligned(size, page_sz), "Size must be page aligned");
3134
3135 committed_start = NULL;
3136
3137 int loops = (pages + stripe - 1) / stripe;
3138 int committed_pages = 0;
3139 address loop_base = start;
3140 bool found_range = false;
3141
3142 for (int index = 0; index < loops && !found_range; index ++) {
3143 assert(pages > 0, "Nothing to do");
3144 int pages_to_query = (pages >= stripe) ? stripe : pages;
3145 pages -= pages_to_query;
3146
3147 // Get stable read
3148 while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3149
3150 // During shutdown, some memory goes away without properly notifying NMT,
3151 // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3152 // Bailout and return as not committed for now.
3153 if (mincore_return_value == -1 && errno == ENOMEM) {
3154 return false;
3155 }
3156
3157 assert(vec[stripe] == 'X', "overflow guard");
3158 assert(mincore_return_value == 0, "Range must be valid");
3159 // Process this stripe
3160 for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3161 if ((vec[vecIdx] & 0x01) == 0) { // not committed
3162 // End of current contiguous region
3163 if (committed_start != NULL) {
3164 found_range = true;
3165 break;
3166 }
3167 } else { // committed
3168 // Start of region
3169 if (committed_start == NULL) {
3170 committed_start = loop_base + page_sz * vecIdx;
3171 }
3172 committed_pages ++;
3173 }
3174 }
3175
3176 loop_base += pages_to_query * page_sz;
3177 }
3178
3179 if (committed_start != NULL) {
3180 assert(committed_pages > 0, "Must have committed region");
3181 assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3182 assert(committed_start >= start && committed_start < start + size, "Out of range");
3183 committed_size = page_sz * committed_pages;
3184 return true;
3185 } else {
3186 assert(committed_pages == 0, "Should not have committed region");
3187 return false;
3188 }
3189 }
3190
3191
3192 // Linux uses a growable mapping for the stack, and if the mapping for
3193 // the stack guard pages is not removed when we detach a thread the
3194 // stack cannot grow beyond the pages where the stack guard was
3195 // mapped. If at some point later in the process the stack expands to
3196 // that point, the Linux kernel cannot expand the stack any further
3197 // because the guard pages are in the way, and a segfault occurs.
3198 //
3199 // However, it's essential not to split the stack region by unmapping
3200 // a region (leaving a hole) that's already part of the stack mapping,
3201 // so if the stack mapping has already grown beyond the guard pages at
3202 // the time we create them, we have to truncate the stack mapping.
3203 // So, we need to know the extent of the stack mapping when
3204 // create_stack_guard_pages() is called.
3205
3206 // We only need this for stacks that are growable: at the time of
3207 // writing thread stacks don't use growable mappings (i.e. those
3208 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3209 // only applies to the main thread.
3210
3211 // If the (growable) stack mapping already extends beyond the point
3212 // where we're going to put our guard pages, truncate the mapping at
3213 // that point by munmap()ping it. This ensures that when we later
3214 // munmap() the guard pages we don't leave a hole in the stack
3215 // mapping. This only affects the main/primordial thread
3216
3217 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3218 if (os::is_primordial_thread()) {
3219 // As we manually grow stack up to bottom inside create_attached_thread(),
3220 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3221 // we don't need to do anything special.
3222 // Check it first, before calling heavy function.
3223 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3224 unsigned char vec[1];
3225
3226 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3227 // Fallback to slow path on all errors, including EAGAIN
3228 stack_extent = (uintptr_t) get_stack_commited_bottom(
3229 os::Linux::initial_thread_stack_bottom(),
3230 (size_t)addr - stack_extent);
3231 }
3232
3233 if (stack_extent < (uintptr_t)addr) {
3234 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3235 }
3236 }
3237
3238 return os::commit_memory(addr, size, !ExecMem);
3239 }
3240
3241 // If this is a growable mapping, remove the guard pages entirely by
3242 // munmap()ping them. If not, just call uncommit_memory(). This only
3243 // affects the main/primordial thread, but guard against future OS changes.
3244 // It's safe to always unmap guard pages for primordial thread because we
3245 // always place it right after end of the mapped region.
3246
3247 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3248 uintptr_t stack_extent, stack_base;
3249
3250 if (os::is_primordial_thread()) {
3251 return ::munmap(addr, size) == 0;
3252 }
3253
3254 return os::uncommit_memory(addr, size);
3255 }
3256
3257 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3258 // at 'requested_addr'. If there are existing memory mappings at the same
3259 // location, however, they will be overwritten. If 'fixed' is false,
3260 // 'requested_addr' is only treated as a hint, the return value may or
3261 // may not start from the requested address. Unlike Linux mmap(), this
3262 // function returns NULL to indicate failure.
3263 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3264 char * addr;
3265 int flags;
3266
3267 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3268 if (fixed) {
3269 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3270 flags |= MAP_FIXED;
3271 }
3272
3273 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3274 // touch an uncommitted page. Otherwise, the read/write might
3275 // succeed if we have enough swap space to back the physical page.
3276 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3277 flags, -1, 0);
3278
3279 return addr == MAP_FAILED ? NULL : addr;
3280 }
3281
3282 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3283 // (req_addr != NULL) or with a given alignment.
3284 // - bytes shall be a multiple of alignment.
3285 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3286 // - alignment sets the alignment at which memory shall be allocated.
3287 // It must be a multiple of allocation granularity.
3288 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3289 // req_addr or NULL.
3290 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3291
3292 size_t extra_size = bytes;
3293 if (req_addr == NULL && alignment > 0) {
3294 extra_size += alignment;
3295 }
3296
3297 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3298 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3299 -1, 0);
3300 if (start == MAP_FAILED) {
3301 start = NULL;
3302 } else {
3303 if (req_addr != NULL) {
3304 if (start != req_addr) {
3305 ::munmap(start, extra_size);
3306 start = NULL;
3307 }
3308 } else {
3309 char* const start_aligned = align_up(start, alignment);
3310 char* const end_aligned = start_aligned + bytes;
3311 char* const end = start + extra_size;
3312 if (start_aligned > start) {
3313 ::munmap(start, start_aligned - start);
3314 }
3315 if (end_aligned < end) {
3316 ::munmap(end_aligned, end - end_aligned);
3317 }
3318 start = start_aligned;
3319 }
3320 }
3321 return start;
3322 }
3323
3324 static int anon_munmap(char * addr, size_t size) {
3325 return ::munmap(addr, size) == 0;
3326 }
3327
3328 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3329 size_t alignment_hint) {
3330 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3331 }
3332
3333 bool os::pd_release_memory(char* addr, size_t size) {
3334 return anon_munmap(addr, size);
3335 }
3336
3337 static bool linux_mprotect(char* addr, size_t size, int prot) {
3338 // Linux wants the mprotect address argument to be page aligned.
3339 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3340
3341 // According to SUSv3, mprotect() should only be used with mappings
3342 // established by mmap(), and mmap() always maps whole pages. Unaligned
3343 // 'addr' likely indicates problem in the VM (e.g. trying to change
3344 // protection of malloc'ed or statically allocated memory). Check the
3345 // caller if you hit this assert.
3346 assert(addr == bottom, "sanity check");
3347
3348 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3349 return ::mprotect(bottom, size, prot) == 0;
3350 }
3351
3352 // Set protections specified
3353 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3354 bool is_committed) {
3355 unsigned int p = 0;
3356 switch (prot) {
3357 case MEM_PROT_NONE: p = PROT_NONE; break;
3358 case MEM_PROT_READ: p = PROT_READ; break;
3359 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3360 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3361 default:
3362 ShouldNotReachHere();
3363 }
3364 // is_committed is unused.
3365 return linux_mprotect(addr, bytes, p);
3366 }
3367
3368 bool os::guard_memory(char* addr, size_t size) {
3369 return linux_mprotect(addr, size, PROT_NONE);
3370 }
3371
3372 bool os::unguard_memory(char* addr, size_t size) {
3373 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3374 }
3375
3376 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3377 size_t page_size) {
3378 bool result = false;
3379 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3380 MAP_ANONYMOUS|MAP_PRIVATE,
3381 -1, 0);
3382 if (p != MAP_FAILED) {
3383 void *aligned_p = align_up(p, page_size);
3384
3385 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3386
3387 munmap(p, page_size * 2);
3388 }
3389
3390 if (warn && !result) {
3391 warning("TransparentHugePages is not supported by the operating system.");
3392 }
3393
3394 return result;
3395 }
3396
3397 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3398 bool result = false;
3399 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3400 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3401 -1, 0);
3402
3403 if (p != MAP_FAILED) {
3404 // We don't know if this really is a huge page or not.
3405 FILE *fp = fopen("/proc/self/maps", "r");
3406 if (fp) {
3407 while (!feof(fp)) {
3408 char chars[257];
3409 long x = 0;
3410 if (fgets(chars, sizeof(chars), fp)) {
3411 if (sscanf(chars, "%lx-%*x", &x) == 1
3412 && x == (long)p) {
3413 if (strstr (chars, "hugepage")) {
3414 result = true;
3415 break;
3416 }
3417 }
3418 }
3419 }
3420 fclose(fp);
3421 }
3422 munmap(p, page_size);
3423 }
3424
3425 if (warn && !result) {
3426 warning("HugeTLBFS is not supported by the operating system.");
3427 }
3428
3429 return result;
3430 }
3431
3432 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3433 //
3434 // From the coredump_filter documentation:
3435 //
3436 // - (bit 0) anonymous private memory
3437 // - (bit 1) anonymous shared memory
3438 // - (bit 2) file-backed private memory
3439 // - (bit 3) file-backed shared memory
3440 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3441 // effective only if the bit 2 is cleared)
3442 // - (bit 5) hugetlb private memory
3443 // - (bit 6) hugetlb shared memory
3444 // - (bit 7) dax private memory
3445 // - (bit 8) dax shared memory
3446 //
3447 static void set_coredump_filter(bool largepages, bool dax_shared) {
3448 FILE *f;
3449 long cdm;
3450 bool filter_changed = false;
3451
3452 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3453 return;
3454 }
3455
3456 if (fscanf(f, "%lx", &cdm) != 1) {
3457 fclose(f);
3458 return;
3459 }
3460
3461 rewind(f);
3462
3463 if (largepages && (cdm & LARGEPAGES_BIT) == 0) {
3464 cdm |= LARGEPAGES_BIT;
3465 filter_changed = true;
3466 }
3467 if (dax_shared && (cdm & DAX_SHARED_BIT) == 0) {
3468 cdm |= DAX_SHARED_BIT;
3469 filter_changed = true;
3470 }
3471 if (filter_changed) {
3472 fprintf(f, "%#lx", cdm);
3473 }
3474
3475 fclose(f);
3476 }
3477
3478 // Large page support
3479
3480 static size_t _large_page_size = 0;
3481
3482 size_t os::Linux::find_large_page_size() {
3483 size_t large_page_size = 0;
3484
3485 // large_page_size on Linux is used to round up heap size. x86 uses either
3486 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3487 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3488 // page as large as 256M.
3489 //
3490 // Here we try to figure out page size by parsing /proc/meminfo and looking
3491 // for a line with the following format:
3492 // Hugepagesize: 2048 kB
3493 //
3494 // If we can't determine the value (e.g. /proc is not mounted, or the text
3495 // format has been changed), we'll use the largest page size supported by
3496 // the processor.
3497
3498 #ifndef ZERO
3499 large_page_size =
3500 AARCH64_ONLY(2 * M)
3501 AMD64_ONLY(2 * M)
3502 ARM32_ONLY(2 * M)
3503 IA32_ONLY(4 * M)
3504 IA64_ONLY(256 * M)
3505 PPC_ONLY(4 * M)
3506 S390_ONLY(1 * M)
3507 SPARC_ONLY(4 * M);
3508 #endif // ZERO
3509
3510 FILE *fp = fopen("/proc/meminfo", "r");
3511 if (fp) {
3512 while (!feof(fp)) {
3513 int x = 0;
3514 char buf[16];
3515 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3516 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3517 large_page_size = x * K;
3518 break;
3519 }
3520 } else {
3521 // skip to next line
3522 for (;;) {
3523 int ch = fgetc(fp);
3524 if (ch == EOF || ch == (int)'\n') break;
3525 }
3526 }
3527 }
3528 fclose(fp);
3529 }
3530
3531 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3532 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3533 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3534 proper_unit_for_byte_size(large_page_size));
3535 }
3536
3537 return large_page_size;
3538 }
3539
3540 size_t os::Linux::setup_large_page_size() {
3541 _large_page_size = Linux::find_large_page_size();
3542 const size_t default_page_size = (size_t)Linux::page_size();
3543 if (_large_page_size > default_page_size) {
3544 _page_sizes[0] = _large_page_size;
3545 _page_sizes[1] = default_page_size;
3546 _page_sizes[2] = 0;
3547 }
3548
3549 return _large_page_size;
3550 }
3551
3552 bool os::Linux::setup_large_page_type(size_t page_size) {
3553 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3554 FLAG_IS_DEFAULT(UseSHM) &&
3555 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3556
3557 // The type of large pages has not been specified by the user.
3558
3559 // Try UseHugeTLBFS and then UseSHM.
3560 UseHugeTLBFS = UseSHM = true;
3561
3562 // Don't try UseTransparentHugePages since there are known
3563 // performance issues with it turned on. This might change in the future.
3564 UseTransparentHugePages = false;
3565 }
3566
3567 if (UseTransparentHugePages) {
3568 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3569 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3570 UseHugeTLBFS = false;
3571 UseSHM = false;
3572 return true;
3573 }
3574 UseTransparentHugePages = false;
3575 }
3576
3577 if (UseHugeTLBFS) {
3578 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3579 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3580 UseSHM = false;
3581 return true;
3582 }
3583 UseHugeTLBFS = false;
3584 }
3585
3586 return UseSHM;
3587 }
3588
3589 void os::large_page_init() {
3590 if (!UseLargePages &&
3591 !UseTransparentHugePages &&
3592 !UseHugeTLBFS &&
3593 !UseSHM) {
3594 // Not using large pages.
3595 return;
3596 }
3597
3598 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3599 // The user explicitly turned off large pages.
3600 // Ignore the rest of the large pages flags.
3601 UseTransparentHugePages = false;
3602 UseHugeTLBFS = false;
3603 UseSHM = false;
3604 return;
3605 }
3606
3607 size_t large_page_size = Linux::setup_large_page_size();
3608 UseLargePages = Linux::setup_large_page_type(large_page_size);
3609
3610 set_coredump_filter(true /*largepages*/, false /*dax_shared*/);
3611 }
3612
3613 #ifndef SHM_HUGETLB
3614 #define SHM_HUGETLB 04000
3615 #endif
3616
3617 #define shm_warning_format(format, ...) \
3618 do { \
3619 if (UseLargePages && \
3620 (!FLAG_IS_DEFAULT(UseLargePages) || \
3621 !FLAG_IS_DEFAULT(UseSHM) || \
3622 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3623 warning(format, __VA_ARGS__); \
3624 } \
3625 } while (0)
3626
3627 #define shm_warning(str) shm_warning_format("%s", str)
3628
3629 #define shm_warning_with_errno(str) \
3630 do { \
3631 int err = errno; \
3632 shm_warning_format(str " (error = %d)", err); \
3633 } while (0)
3634
3635 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3636 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3637
3638 if (!is_aligned(alignment, SHMLBA)) {
3639 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3640 return NULL;
3641 }
3642
3643 // To ensure that we get 'alignment' aligned memory from shmat,
3644 // we pre-reserve aligned virtual memory and then attach to that.
3645
3646 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3647 if (pre_reserved_addr == NULL) {
3648 // Couldn't pre-reserve aligned memory.
3649 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3650 return NULL;
3651 }
3652
3653 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3654 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3655
3656 if ((intptr_t)addr == -1) {
3657 int err = errno;
3658 shm_warning_with_errno("Failed to attach shared memory.");
3659
3660 assert(err != EACCES, "Unexpected error");
3661 assert(err != EIDRM, "Unexpected error");
3662 assert(err != EINVAL, "Unexpected error");
3663
3664 // Since we don't know if the kernel unmapped the pre-reserved memory area
3665 // we can't unmap it, since that would potentially unmap memory that was
3666 // mapped from other threads.
3667 return NULL;
3668 }
3669
3670 return addr;
3671 }
3672
3673 static char* shmat_at_address(int shmid, char* req_addr) {
3674 if (!is_aligned(req_addr, SHMLBA)) {
3675 assert(false, "Requested address needs to be SHMLBA aligned");
3676 return NULL;
3677 }
3678
3679 char* addr = (char*)shmat(shmid, req_addr, 0);
3680
3681 if ((intptr_t)addr == -1) {
3682 shm_warning_with_errno("Failed to attach shared memory.");
3683 return NULL;
3684 }
3685
3686 return addr;
3687 }
3688
3689 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3690 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3691 if (req_addr != NULL) {
3692 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3693 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3694 return shmat_at_address(shmid, req_addr);
3695 }
3696
3697 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3698 // return large page size aligned memory addresses when req_addr == NULL.
3699 // However, if the alignment is larger than the large page size, we have
3700 // to manually ensure that the memory returned is 'alignment' aligned.
3701 if (alignment > os::large_page_size()) {
3702 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3703 return shmat_with_alignment(shmid, bytes, alignment);
3704 } else {
3705 return shmat_at_address(shmid, NULL);
3706 }
3707 }
3708
3709 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3710 char* req_addr, bool exec) {
3711 // "exec" is passed in but not used. Creating the shared image for
3712 // the code cache doesn't have an SHM_X executable permission to check.
3713 assert(UseLargePages && UseSHM, "only for SHM large pages");
3714 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3715 assert(is_aligned(req_addr, alignment), "Unaligned address");
3716
3717 if (!is_aligned(bytes, os::large_page_size())) {
3718 return NULL; // Fallback to small pages.
3719 }
3720
3721 // Create a large shared memory region to attach to based on size.
3722 // Currently, size is the total size of the heap.
3723 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3724 if (shmid == -1) {
3725 // Possible reasons for shmget failure:
3726 // 1. shmmax is too small for Java heap.
3727 // > check shmmax value: cat /proc/sys/kernel/shmmax
3728 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3729 // 2. not enough large page memory.
3730 // > check available large pages: cat /proc/meminfo
3731 // > increase amount of large pages:
3732 // echo new_value > /proc/sys/vm/nr_hugepages
3733 // Note 1: different Linux may use different name for this property,
3734 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3735 // Note 2: it's possible there's enough physical memory available but
3736 // they are so fragmented after a long run that they can't
3737 // coalesce into large pages. Try to reserve large pages when
3738 // the system is still "fresh".
3739 shm_warning_with_errno("Failed to reserve shared memory.");
3740 return NULL;
3741 }
3742
3743 // Attach to the region.
3744 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3745
3746 // Remove shmid. If shmat() is successful, the actual shared memory segment
3747 // will be deleted when it's detached by shmdt() or when the process
3748 // terminates. If shmat() is not successful this will remove the shared
3749 // segment immediately.
3750 shmctl(shmid, IPC_RMID, NULL);
3751
3752 return addr;
3753 }
3754
3755 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3756 int error) {
3757 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3758
3759 bool warn_on_failure = UseLargePages &&
3760 (!FLAG_IS_DEFAULT(UseLargePages) ||
3761 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3762 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3763
3764 if (warn_on_failure) {
3765 char msg[128];
3766 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3767 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3768 warning("%s", msg);
3769 }
3770 }
3771
3772 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3773 char* req_addr,
3774 bool exec) {
3775 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3776 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3777 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3778
3779 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3780 char* addr = (char*)::mmap(req_addr, bytes, prot,
3781 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3782 -1, 0);
3783
3784 if (addr == MAP_FAILED) {
3785 warn_on_large_pages_failure(req_addr, bytes, errno);
3786 return NULL;
3787 }
3788
3789 assert(is_aligned(addr, os::large_page_size()), "Must be");
3790
3791 return addr;
3792 }
3793
3794 // Reserve memory using mmap(MAP_HUGETLB).
3795 // - bytes shall be a multiple of alignment.
3796 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3797 // - alignment sets the alignment at which memory shall be allocated.
3798 // It must be a multiple of allocation granularity.
3799 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3800 // req_addr or NULL.
3801 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3802 size_t alignment,
3803 char* req_addr,
3804 bool exec) {
3805 size_t large_page_size = os::large_page_size();
3806 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3807
3808 assert(is_aligned(req_addr, alignment), "Must be");
3809 assert(is_aligned(bytes, alignment), "Must be");
3810
3811 // First reserve - but not commit - the address range in small pages.
3812 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3813
3814 if (start == NULL) {
3815 return NULL;
3816 }
3817
3818 assert(is_aligned(start, alignment), "Must be");
3819
3820 char* end = start + bytes;
3821
3822 // Find the regions of the allocated chunk that can be promoted to large pages.
3823 char* lp_start = align_up(start, large_page_size);
3824 char* lp_end = align_down(end, large_page_size);
3825
3826 size_t lp_bytes = lp_end - lp_start;
3827
3828 assert(is_aligned(lp_bytes, large_page_size), "Must be");
3829
3830 if (lp_bytes == 0) {
3831 // The mapped region doesn't even span the start and the end of a large page.
3832 // Fall back to allocate a non-special area.
3833 ::munmap(start, end - start);
3834 return NULL;
3835 }
3836
3837 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3838
3839 void* result;
3840
3841 // Commit small-paged leading area.
3842 if (start != lp_start) {
3843 result = ::mmap(start, lp_start - start, prot,
3844 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3845 -1, 0);
3846 if (result == MAP_FAILED) {
3847 ::munmap(lp_start, end - lp_start);
3848 return NULL;
3849 }
3850 }
3851
3852 // Commit large-paged area.
3853 result = ::mmap(lp_start, lp_bytes, prot,
3854 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3855 -1, 0);
3856 if (result == MAP_FAILED) {
3857 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3858 // If the mmap above fails, the large pages region will be unmapped and we
3859 // have regions before and after with small pages. Release these regions.
3860 //
3861 // | mapped | unmapped | mapped |
3862 // ^ ^ ^ ^
3863 // start lp_start lp_end end
3864 //
3865 ::munmap(start, lp_start - start);
3866 ::munmap(lp_end, end - lp_end);
3867 return NULL;
3868 }
3869
3870 // Commit small-paged trailing area.
3871 if (lp_end != end) {
3872 result = ::mmap(lp_end, end - lp_end, prot,
3873 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3874 -1, 0);
3875 if (result == MAP_FAILED) {
3876 ::munmap(start, lp_end - start);
3877 return NULL;
3878 }
3879 }
3880
3881 return start;
3882 }
3883
3884 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3885 size_t alignment,
3886 char* req_addr,
3887 bool exec) {
3888 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3889 assert(is_aligned(req_addr, alignment), "Must be");
3890 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3891 assert(is_power_of_2(os::large_page_size()), "Must be");
3892 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3893
3894 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3895 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3896 } else {
3897 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3898 }
3899 }
3900
3901 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3902 char* req_addr, bool exec) {
3903 assert(UseLargePages, "only for large pages");
3904
3905 char* addr;
3906 if (UseSHM) {
3907 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3908 } else {
3909 assert(UseHugeTLBFS, "must be");
3910 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3911 }
3912
3913 if (addr != NULL) {
3914 if (UseNUMAInterleaving) {
3915 numa_make_global(addr, bytes);
3916 }
3917
3918 // The memory is committed
3919 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3920 }
3921
3922 return addr;
3923 }
3924
3925 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3926 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3927 return shmdt(base) == 0;
3928 }
3929
3930 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3931 return pd_release_memory(base, bytes);
3932 }
3933
3934 bool os::release_memory_special(char* base, size_t bytes) {
3935 bool res;
3936 if (MemTracker::tracking_level() > NMT_minimal) {
3937 Tracker tkr(Tracker::release);
3938 res = os::Linux::release_memory_special_impl(base, bytes);
3939 if (res) {
3940 tkr.record((address)base, bytes);
3941 }
3942
3943 } else {
3944 res = os::Linux::release_memory_special_impl(base, bytes);
3945 }
3946 return res;
3947 }
3948
3949 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3950 assert(UseLargePages, "only for large pages");
3951 bool res;
3952
3953 if (UseSHM) {
3954 res = os::Linux::release_memory_special_shm(base, bytes);
3955 } else {
3956 assert(UseHugeTLBFS, "must be");
3957 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3958 }
3959 return res;
3960 }
3961
3962 size_t os::large_page_size() {
3963 return _large_page_size;
3964 }
3965
3966 // With SysV SHM the entire memory region must be allocated as shared
3967 // memory.
3968 // HugeTLBFS allows application to commit large page memory on demand.
3969 // However, when committing memory with HugeTLBFS fails, the region
3970 // that was supposed to be committed will lose the old reservation
3971 // and allow other threads to steal that memory region. Because of this
3972 // behavior we can't commit HugeTLBFS memory.
3973 bool os::can_commit_large_page_memory() {
3974 return UseTransparentHugePages;
3975 }
3976
3977 bool os::can_execute_large_page_memory() {
3978 return UseTransparentHugePages || UseHugeTLBFS;
3979 }
3980
3981 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
3982 assert(file_desc >= 0, "file_desc is not valid");
3983 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
3984 if (result != NULL) {
3985 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
3986 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
3987 }
3988 }
3989 return result;
3990 }
3991
3992 // Reserve memory at an arbitrary address, only if that area is
3993 // available (and not reserved for something else).
3994
3995 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3996 const int max_tries = 10;
3997 char* base[max_tries];
3998 size_t size[max_tries];
3999 const size_t gap = 0x000000;
4000
4001 // Assert only that the size is a multiple of the page size, since
4002 // that's all that mmap requires, and since that's all we really know
4003 // about at this low abstraction level. If we need higher alignment,
4004 // we can either pass an alignment to this method or verify alignment
4005 // in one of the methods further up the call chain. See bug 5044738.
4006 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
4007
4008 // Repeatedly allocate blocks until the block is allocated at the
4009 // right spot.
4010
4011 // Linux mmap allows caller to pass an address as hint; give it a try first,
4012 // if kernel honors the hint then we can return immediately.
4013 char * addr = anon_mmap(requested_addr, bytes, false);
4014 if (addr == requested_addr) {
4015 return requested_addr;
4016 }
4017
4018 if (addr != NULL) {
4019 // mmap() is successful but it fails to reserve at the requested address
4020 anon_munmap(addr, bytes);
4021 }
4022
4023 int i;
4024 for (i = 0; i < max_tries; ++i) {
4025 base[i] = reserve_memory(bytes);
4026
4027 if (base[i] != NULL) {
4028 // Is this the block we wanted?
4029 if (base[i] == requested_addr) {
4030 size[i] = bytes;
4031 break;
4032 }
4033
4034 // Does this overlap the block we wanted? Give back the overlapped
4035 // parts and try again.
4036
4037 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
4038 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
4039 unmap_memory(base[i], top_overlap);
4040 base[i] += top_overlap;
4041 size[i] = bytes - top_overlap;
4042 } else {
4043 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
4044 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
4045 unmap_memory(requested_addr, bottom_overlap);
4046 size[i] = bytes - bottom_overlap;
4047 } else {
4048 size[i] = bytes;
4049 }
4050 }
4051 }
4052 }
4053
4054 // Give back the unused reserved pieces.
4055
4056 for (int j = 0; j < i; ++j) {
4057 if (base[j] != NULL) {
4058 unmap_memory(base[j], size[j]);
4059 }
4060 }
4061
4062 if (i < max_tries) {
4063 return requested_addr;
4064 } else {
4065 return NULL;
4066 }
4067 }
4068
4069 size_t os::read(int fd, void *buf, unsigned int nBytes) {
4070 return ::read(fd, buf, nBytes);
4071 }
4072
4073 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
4074 return ::pread(fd, buf, nBytes, offset);
4075 }
4076
4077 // Short sleep, direct OS call.
4078 //
4079 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
4080 // sched_yield(2) will actually give up the CPU:
4081 //
4082 // * Alone on this pariticular CPU, keeps running.
4083 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
4084 // (pre 2.6.39).
4085 //
4086 // So calling this with 0 is an alternative.
4087 //
4088 void os::naked_short_sleep(jlong ms) {
4089 struct timespec req;
4090
4091 assert(ms < 1000, "Un-interruptable sleep, short time use only");
4092 req.tv_sec = 0;
4093 if (ms > 0) {
4094 req.tv_nsec = (ms % 1000) * 1000000;
4095 } else {
4096 req.tv_nsec = 1;
4097 }
4098
4099 nanosleep(&req, NULL);
4100
4101 return;
4102 }
4103
4104 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4105 void os::infinite_sleep() {
4106 while (true) { // sleep forever ...
4107 ::sleep(100); // ... 100 seconds at a time
4108 }
4109 }
4110
4111 // Used to convert frequent JVM_Yield() to nops
4112 bool os::dont_yield() {
4113 return DontYieldALot;
4114 }
4115
4116 void os::naked_yield() {
4117 sched_yield();
4118 }
4119
4120 ////////////////////////////////////////////////////////////////////////////////
4121 // thread priority support
4122
4123 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4124 // only supports dynamic priority, static priority must be zero. For real-time
4125 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4126 // However, for large multi-threaded applications, SCHED_RR is not only slower
4127 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4128 // of 5 runs - Sep 2005).
4129 //
4130 // The following code actually changes the niceness of kernel-thread/LWP. It
4131 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4132 // not the entire user process, and user level threads are 1:1 mapped to kernel
4133 // threads. It has always been the case, but could change in the future. For
4134 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4135 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4136
4137 int os::java_to_os_priority[CriticalPriority + 1] = {
4138 19, // 0 Entry should never be used
4139
4140 4, // 1 MinPriority
4141 3, // 2
4142 2, // 3
4143
4144 1, // 4
4145 0, // 5 NormPriority
4146 -1, // 6
4147
4148 -2, // 7
4149 -3, // 8
4150 -4, // 9 NearMaxPriority
4151
4152 -5, // 10 MaxPriority
4153
4154 -5 // 11 CriticalPriority
4155 };
4156
4157 static int prio_init() {
4158 if (ThreadPriorityPolicy == 1) {
4159 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4160 // if effective uid is not root. Perhaps, a more elegant way of doing
4161 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4162 if (geteuid() != 0) {
4163 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4164 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4165 }
4166 ThreadPriorityPolicy = 0;
4167 }
4168 }
4169 if (UseCriticalJavaThreadPriority) {
4170 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4171 }
4172 return 0;
4173 }
4174
4175 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4176 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4177
4178 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4179 return (ret == 0) ? OS_OK : OS_ERR;
4180 }
4181
4182 OSReturn os::get_native_priority(const Thread* const thread,
4183 int *priority_ptr) {
4184 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4185 *priority_ptr = java_to_os_priority[NormPriority];
4186 return OS_OK;
4187 }
4188
4189 errno = 0;
4190 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4191 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4192 }
4193
4194 ////////////////////////////////////////////////////////////////////////////////
4195 // suspend/resume support
4196
4197 // The low-level signal-based suspend/resume support is a remnant from the
4198 // old VM-suspension that used to be for java-suspension, safepoints etc,
4199 // within hotspot. Currently used by JFR's OSThreadSampler
4200 //
4201 // The remaining code is greatly simplified from the more general suspension
4202 // code that used to be used.
4203 //
4204 // The protocol is quite simple:
4205 // - suspend:
4206 // - sends a signal to the target thread
4207 // - polls the suspend state of the osthread using a yield loop
4208 // - target thread signal handler (SR_handler) sets suspend state
4209 // and blocks in sigsuspend until continued
4210 // - resume:
4211 // - sets target osthread state to continue
4212 // - sends signal to end the sigsuspend loop in the SR_handler
4213 //
4214 // Note that the SR_lock plays no role in this suspend/resume protocol,
4215 // but is checked for NULL in SR_handler as a thread termination indicator.
4216 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4217 //
4218 // Note that resume_clear_context() and suspend_save_context() are needed
4219 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4220 // which in part is used by:
4221 // - Forte Analyzer: AsyncGetCallTrace()
4222 // - StackBanging: get_frame_at_stack_banging_point()
4223
4224 static void resume_clear_context(OSThread *osthread) {
4225 osthread->set_ucontext(NULL);
4226 osthread->set_siginfo(NULL);
4227 }
4228
4229 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4230 ucontext_t* context) {
4231 osthread->set_ucontext(context);
4232 osthread->set_siginfo(siginfo);
4233 }
4234
4235 // Handler function invoked when a thread's execution is suspended or
4236 // resumed. We have to be careful that only async-safe functions are
4237 // called here (Note: most pthread functions are not async safe and
4238 // should be avoided.)
4239 //
4240 // Note: sigwait() is a more natural fit than sigsuspend() from an
4241 // interface point of view, but sigwait() prevents the signal hander
4242 // from being run. libpthread would get very confused by not having
4243 // its signal handlers run and prevents sigwait()'s use with the
4244 // mutex granting granting signal.
4245 //
4246 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4247 //
4248 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4249 // Save and restore errno to avoid confusing native code with EINTR
4250 // after sigsuspend.
4251 int old_errno = errno;
4252
4253 Thread* thread = Thread::current_or_null_safe();
4254 assert(thread != NULL, "Missing current thread in SR_handler");
4255
4256 // On some systems we have seen signal delivery get "stuck" until the signal
4257 // mask is changed as part of thread termination. Check that the current thread
4258 // has not already terminated (via SR_lock()) - else the following assertion
4259 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4260 // destructor has completed.
4261
4262 if (thread->SR_lock() == NULL) {
4263 return;
4264 }
4265
4266 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4267
4268 OSThread* osthread = thread->osthread();
4269
4270 os::SuspendResume::State current = osthread->sr.state();
4271 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4272 suspend_save_context(osthread, siginfo, context);
4273
4274 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4275 os::SuspendResume::State state = osthread->sr.suspended();
4276 if (state == os::SuspendResume::SR_SUSPENDED) {
4277 sigset_t suspend_set; // signals for sigsuspend()
4278 sigemptyset(&suspend_set);
4279 // get current set of blocked signals and unblock resume signal
4280 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4281 sigdelset(&suspend_set, SR_signum);
4282
4283 sr_semaphore.signal();
4284 // wait here until we are resumed
4285 while (1) {
4286 sigsuspend(&suspend_set);
4287
4288 os::SuspendResume::State result = osthread->sr.running();
4289 if (result == os::SuspendResume::SR_RUNNING) {
4290 sr_semaphore.signal();
4291 break;
4292 }
4293 }
4294
4295 } else if (state == os::SuspendResume::SR_RUNNING) {
4296 // request was cancelled, continue
4297 } else {
4298 ShouldNotReachHere();
4299 }
4300
4301 resume_clear_context(osthread);
4302 } else if (current == os::SuspendResume::SR_RUNNING) {
4303 // request was cancelled, continue
4304 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4305 // ignore
4306 } else {
4307 // ignore
4308 }
4309
4310 errno = old_errno;
4311 }
4312
4313 static int SR_initialize() {
4314 struct sigaction act;
4315 char *s;
4316
4317 // Get signal number to use for suspend/resume
4318 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4319 int sig = ::strtol(s, 0, 10);
4320 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4321 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4322 SR_signum = sig;
4323 } else {
4324 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4325 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4326 }
4327 }
4328
4329 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4330 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4331
4332 sigemptyset(&SR_sigset);
4333 sigaddset(&SR_sigset, SR_signum);
4334
4335 // Set up signal handler for suspend/resume
4336 act.sa_flags = SA_RESTART|SA_SIGINFO;
4337 act.sa_handler = (void (*)(int)) SR_handler;
4338
4339 // SR_signum is blocked by default.
4340 // 4528190 - We also need to block pthread restart signal (32 on all
4341 // supported Linux platforms). Note that LinuxThreads need to block
4342 // this signal for all threads to work properly. So we don't have
4343 // to use hard-coded signal number when setting up the mask.
4344 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4345
4346 if (sigaction(SR_signum, &act, 0) == -1) {
4347 return -1;
4348 }
4349
4350 // Save signal flag
4351 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4352 return 0;
4353 }
4354
4355 static int sr_notify(OSThread* osthread) {
4356 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4357 assert_status(status == 0, status, "pthread_kill");
4358 return status;
4359 }
4360
4361 // "Randomly" selected value for how long we want to spin
4362 // before bailing out on suspending a thread, also how often
4363 // we send a signal to a thread we want to resume
4364 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4365 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4366
4367 // returns true on success and false on error - really an error is fatal
4368 // but this seems the normal response to library errors
4369 static bool do_suspend(OSThread* osthread) {
4370 assert(osthread->sr.is_running(), "thread should be running");
4371 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4372
4373 // mark as suspended and send signal
4374 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4375 // failed to switch, state wasn't running?
4376 ShouldNotReachHere();
4377 return false;
4378 }
4379
4380 if (sr_notify(osthread) != 0) {
4381 ShouldNotReachHere();
4382 }
4383
4384 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4385 while (true) {
4386 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4387 break;
4388 } else {
4389 // timeout
4390 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4391 if (cancelled == os::SuspendResume::SR_RUNNING) {
4392 return false;
4393 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4394 // make sure that we consume the signal on the semaphore as well
4395 sr_semaphore.wait();
4396 break;
4397 } else {
4398 ShouldNotReachHere();
4399 return false;
4400 }
4401 }
4402 }
4403
4404 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4405 return true;
4406 }
4407
4408 static void do_resume(OSThread* osthread) {
4409 assert(osthread->sr.is_suspended(), "thread should be suspended");
4410 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4411
4412 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4413 // failed to switch to WAKEUP_REQUEST
4414 ShouldNotReachHere();
4415 return;
4416 }
4417
4418 while (true) {
4419 if (sr_notify(osthread) == 0) {
4420 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4421 if (osthread->sr.is_running()) {
4422 return;
4423 }
4424 }
4425 } else {
4426 ShouldNotReachHere();
4427 }
4428 }
4429
4430 guarantee(osthread->sr.is_running(), "Must be running!");
4431 }
4432
4433 ///////////////////////////////////////////////////////////////////////////////////
4434 // signal handling (except suspend/resume)
4435
4436 // This routine may be used by user applications as a "hook" to catch signals.
4437 // The user-defined signal handler must pass unrecognized signals to this
4438 // routine, and if it returns true (non-zero), then the signal handler must
4439 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4440 // routine will never retun false (zero), but instead will execute a VM panic
4441 // routine kill the process.
4442 //
4443 // If this routine returns false, it is OK to call it again. This allows
4444 // the user-defined signal handler to perform checks either before or after
4445 // the VM performs its own checks. Naturally, the user code would be making
4446 // a serious error if it tried to handle an exception (such as a null check
4447 // or breakpoint) that the VM was generating for its own correct operation.
4448 //
4449 // This routine may recognize any of the following kinds of signals:
4450 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4451 // It should be consulted by handlers for any of those signals.
4452 //
4453 // The caller of this routine must pass in the three arguments supplied
4454 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4455 // field of the structure passed to sigaction(). This routine assumes that
4456 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4457 //
4458 // Note that the VM will print warnings if it detects conflicting signal
4459 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4460 //
4461 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4462 siginfo_t* siginfo,
4463 void* ucontext,
4464 int abort_if_unrecognized);
4465
4466 static void signalHandler(int sig, siginfo_t* info, void* uc) {
4467 assert(info != NULL && uc != NULL, "it must be old kernel");
4468 int orig_errno = errno; // Preserve errno value over signal handler.
4469 JVM_handle_linux_signal(sig, info, uc, true);
4470 errno = orig_errno;
4471 }
4472
4473
4474 // This boolean allows users to forward their own non-matching signals
4475 // to JVM_handle_linux_signal, harmlessly.
4476 bool os::Linux::signal_handlers_are_installed = false;
4477
4478 // For signal-chaining
4479 struct sigaction sigact[NSIG];
4480 uint64_t sigs = 0;
4481 #if (64 < NSIG-1)
4482 #error "Not all signals can be encoded in sigs. Adapt its type!"
4483 #endif
4484 bool os::Linux::libjsig_is_loaded = false;
4485 typedef struct sigaction *(*get_signal_t)(int);
4486 get_signal_t os::Linux::get_signal_action = NULL;
4487
4488 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4489 struct sigaction *actp = NULL;
4490
4491 if (libjsig_is_loaded) {
4492 // Retrieve the old signal handler from libjsig
4493 actp = (*get_signal_action)(sig);
4494 }
4495 if (actp == NULL) {
4496 // Retrieve the preinstalled signal handler from jvm
4497 actp = get_preinstalled_handler(sig);
4498 }
4499
4500 return actp;
4501 }
4502
4503 static bool call_chained_handler(struct sigaction *actp, int sig,
4504 siginfo_t *siginfo, void *context) {
4505 // Call the old signal handler
4506 if (actp->sa_handler == SIG_DFL) {
4507 // It's more reasonable to let jvm treat it as an unexpected exception
4508 // instead of taking the default action.
4509 return false;
4510 } else if (actp->sa_handler != SIG_IGN) {
4511 if ((actp->sa_flags & SA_NODEFER) == 0) {
4512 // automaticlly block the signal
4513 sigaddset(&(actp->sa_mask), sig);
4514 }
4515
4516 sa_handler_t hand = NULL;
4517 sa_sigaction_t sa = NULL;
4518 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4519 // retrieve the chained handler
4520 if (siginfo_flag_set) {
4521 sa = actp->sa_sigaction;
4522 } else {
4523 hand = actp->sa_handler;
4524 }
4525
4526 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4527 actp->sa_handler = SIG_DFL;
4528 }
4529
4530 // try to honor the signal mask
4531 sigset_t oset;
4532 sigemptyset(&oset);
4533 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4534
4535 // call into the chained handler
4536 if (siginfo_flag_set) {
4537 (*sa)(sig, siginfo, context);
4538 } else {
4539 (*hand)(sig);
4540 }
4541
4542 // restore the signal mask
4543 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4544 }
4545 // Tell jvm's signal handler the signal is taken care of.
4546 return true;
4547 }
4548
4549 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4550 bool chained = false;
4551 // signal-chaining
4552 if (UseSignalChaining) {
4553 struct sigaction *actp = get_chained_signal_action(sig);
4554 if (actp != NULL) {
4555 chained = call_chained_handler(actp, sig, siginfo, context);
4556 }
4557 }
4558 return chained;
4559 }
4560
4561 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4562 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4563 return &sigact[sig];
4564 }
4565 return NULL;
4566 }
4567
4568 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4569 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4570 sigact[sig] = oldAct;
4571 sigs |= (uint64_t)1 << (sig-1);
4572 }
4573
4574 // for diagnostic
4575 int sigflags[NSIG];
4576
4577 int os::Linux::get_our_sigflags(int sig) {
4578 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4579 return sigflags[sig];
4580 }
4581
4582 void os::Linux::set_our_sigflags(int sig, int flags) {
4583 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4584 if (sig > 0 && sig < NSIG) {
4585 sigflags[sig] = flags;
4586 }
4587 }
4588
4589 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4590 // Check for overwrite.
4591 struct sigaction oldAct;
4592 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4593
4594 void* oldhand = oldAct.sa_sigaction
4595 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4596 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4597 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4598 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4599 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4600 if (AllowUserSignalHandlers || !set_installed) {
4601 // Do not overwrite; user takes responsibility to forward to us.
4602 return;
4603 } else if (UseSignalChaining) {
4604 // save the old handler in jvm
4605 save_preinstalled_handler(sig, oldAct);
4606 // libjsig also interposes the sigaction() call below and saves the
4607 // old sigaction on it own.
4608 } else {
4609 fatal("Encountered unexpected pre-existing sigaction handler "
4610 "%#lx for signal %d.", (long)oldhand, sig);
4611 }
4612 }
4613
4614 struct sigaction sigAct;
4615 sigfillset(&(sigAct.sa_mask));
4616 sigAct.sa_handler = SIG_DFL;
4617 if (!set_installed) {
4618 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4619 } else {
4620 sigAct.sa_sigaction = signalHandler;
4621 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4622 }
4623 // Save flags, which are set by ours
4624 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4625 sigflags[sig] = sigAct.sa_flags;
4626
4627 int ret = sigaction(sig, &sigAct, &oldAct);
4628 assert(ret == 0, "check");
4629
4630 void* oldhand2 = oldAct.sa_sigaction
4631 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4632 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4633 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4634 }
4635
4636 // install signal handlers for signals that HotSpot needs to
4637 // handle in order to support Java-level exception handling.
4638
4639 void os::Linux::install_signal_handlers() {
4640 if (!signal_handlers_are_installed) {
4641 signal_handlers_are_installed = true;
4642
4643 // signal-chaining
4644 typedef void (*signal_setting_t)();
4645 signal_setting_t begin_signal_setting = NULL;
4646 signal_setting_t end_signal_setting = NULL;
4647 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4648 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4649 if (begin_signal_setting != NULL) {
4650 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4651 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4652 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4653 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4654 libjsig_is_loaded = true;
4655 assert(UseSignalChaining, "should enable signal-chaining");
4656 }
4657 if (libjsig_is_loaded) {
4658 // Tell libjsig jvm is setting signal handlers
4659 (*begin_signal_setting)();
4660 }
4661
4662 set_signal_handler(SIGSEGV, true);
4663 set_signal_handler(SIGPIPE, true);
4664 set_signal_handler(SIGBUS, true);
4665 set_signal_handler(SIGILL, true);
4666 set_signal_handler(SIGFPE, true);
4667 #if defined(PPC64)
4668 set_signal_handler(SIGTRAP, true);
4669 #endif
4670 set_signal_handler(SIGXFSZ, true);
4671
4672 if (libjsig_is_loaded) {
4673 // Tell libjsig jvm finishes setting signal handlers
4674 (*end_signal_setting)();
4675 }
4676
4677 // We don't activate signal checker if libjsig is in place, we trust ourselves
4678 // and if UserSignalHandler is installed all bets are off.
4679 // Log that signal checking is off only if -verbose:jni is specified.
4680 if (CheckJNICalls) {
4681 if (libjsig_is_loaded) {
4682 if (PrintJNIResolving) {
4683 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4684 }
4685 check_signals = false;
4686 }
4687 if (AllowUserSignalHandlers) {
4688 if (PrintJNIResolving) {
4689 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4690 }
4691 check_signals = false;
4692 }
4693 }
4694 }
4695 }
4696
4697 // This is the fastest way to get thread cpu time on Linux.
4698 // Returns cpu time (user+sys) for any thread, not only for current.
4699 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4700 // It might work on 2.6.10+ with a special kernel/glibc patch.
4701 // For reference, please, see IEEE Std 1003.1-2004:
4702 // http://www.unix.org/single_unix_specification
4703
4704 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4705 struct timespec tp;
4706 int rc = os::Linux::clock_gettime(clockid, &tp);
4707 assert(rc == 0, "clock_gettime is expected to return 0 code");
4708
4709 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4710 }
4711
4712 void os::Linux::initialize_os_info() {
4713 assert(_os_version == 0, "OS info already initialized");
4714
4715 struct utsname _uname;
4716
4717 uint32_t major;
4718 uint32_t minor;
4719 uint32_t fix;
4720
4721 int rc;
4722
4723 // Kernel version is unknown if
4724 // verification below fails.
4725 _os_version = 0x01000000;
4726
4727 rc = uname(&_uname);
4728 if (rc != -1) {
4729
4730 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4731 if (rc == 3) {
4732
4733 if (major < 256 && minor < 256 && fix < 256) {
4734 // Kernel version format is as expected,
4735 // set it overriding unknown state.
4736 _os_version = (major << 16) |
4737 (minor << 8 ) |
4738 (fix << 0 ) ;
4739 }
4740 }
4741 }
4742 }
4743
4744 uint32_t os::Linux::os_version() {
4745 assert(_os_version != 0, "not initialized");
4746 return _os_version & 0x00FFFFFF;
4747 }
4748
4749 bool os::Linux::os_version_is_known() {
4750 assert(_os_version != 0, "not initialized");
4751 return _os_version & 0x01000000 ? false : true;
4752 }
4753
4754 /////
4755 // glibc on Linux platform uses non-documented flag
4756 // to indicate, that some special sort of signal
4757 // trampoline is used.
4758 // We will never set this flag, and we should
4759 // ignore this flag in our diagnostic
4760 #ifdef SIGNIFICANT_SIGNAL_MASK
4761 #undef SIGNIFICANT_SIGNAL_MASK
4762 #endif
4763 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4764
4765 static const char* get_signal_handler_name(address handler,
4766 char* buf, int buflen) {
4767 int offset = 0;
4768 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4769 if (found) {
4770 // skip directory names
4771 const char *p1, *p2;
4772 p1 = buf;
4773 size_t len = strlen(os::file_separator());
4774 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4775 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4776 } else {
4777 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4778 }
4779 return buf;
4780 }
4781
4782 static void print_signal_handler(outputStream* st, int sig,
4783 char* buf, size_t buflen) {
4784 struct sigaction sa;
4785
4786 sigaction(sig, NULL, &sa);
4787
4788 // See comment for SIGNIFICANT_SIGNAL_MASK define
4789 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4790
4791 st->print("%s: ", os::exception_name(sig, buf, buflen));
4792
4793 address handler = (sa.sa_flags & SA_SIGINFO)
4794 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4795 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4796
4797 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4798 st->print("SIG_DFL");
4799 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4800 st->print("SIG_IGN");
4801 } else {
4802 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4803 }
4804
4805 st->print(", sa_mask[0]=");
4806 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4807
4808 address rh = VMError::get_resetted_sighandler(sig);
4809 // May be, handler was resetted by VMError?
4810 if (rh != NULL) {
4811 handler = rh;
4812 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4813 }
4814
4815 st->print(", sa_flags=");
4816 os::Posix::print_sa_flags(st, sa.sa_flags);
4817
4818 // Check: is it our handler?
4819 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4820 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4821 // It is our signal handler
4822 // check for flags, reset system-used one!
4823 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4824 st->print(
4825 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4826 os::Linux::get_our_sigflags(sig));
4827 }
4828 }
4829 st->cr();
4830 }
4831
4832
4833 #define DO_SIGNAL_CHECK(sig) \
4834 do { \
4835 if (!sigismember(&check_signal_done, sig)) { \
4836 os::Linux::check_signal_handler(sig); \
4837 } \
4838 } while (0)
4839
4840 // This method is a periodic task to check for misbehaving JNI applications
4841 // under CheckJNI, we can add any periodic checks here
4842
4843 void os::run_periodic_checks() {
4844 if (check_signals == false) return;
4845
4846 // SEGV and BUS if overridden could potentially prevent
4847 // generation of hs*.log in the event of a crash, debugging
4848 // such a case can be very challenging, so we absolutely
4849 // check the following for a good measure:
4850 DO_SIGNAL_CHECK(SIGSEGV);
4851 DO_SIGNAL_CHECK(SIGILL);
4852 DO_SIGNAL_CHECK(SIGFPE);
4853 DO_SIGNAL_CHECK(SIGBUS);
4854 DO_SIGNAL_CHECK(SIGPIPE);
4855 DO_SIGNAL_CHECK(SIGXFSZ);
4856 #if defined(PPC64)
4857 DO_SIGNAL_CHECK(SIGTRAP);
4858 #endif
4859
4860 // ReduceSignalUsage allows the user to override these handlers
4861 // see comments at the very top and jvm_md.h
4862 if (!ReduceSignalUsage) {
4863 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4864 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4865 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4866 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4867 }
4868
4869 DO_SIGNAL_CHECK(SR_signum);
4870 }
4871
4872 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4873
4874 static os_sigaction_t os_sigaction = NULL;
4875
4876 void os::Linux::check_signal_handler(int sig) {
4877 char buf[O_BUFLEN];
4878 address jvmHandler = NULL;
4879
4880
4881 struct sigaction act;
4882 if (os_sigaction == NULL) {
4883 // only trust the default sigaction, in case it has been interposed
4884 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4885 if (os_sigaction == NULL) return;
4886 }
4887
4888 os_sigaction(sig, (struct sigaction*)NULL, &act);
4889
4890
4891 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4892
4893 address thisHandler = (act.sa_flags & SA_SIGINFO)
4894 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4895 : CAST_FROM_FN_PTR(address, act.sa_handler);
4896
4897
4898 switch (sig) {
4899 case SIGSEGV:
4900 case SIGBUS:
4901 case SIGFPE:
4902 case SIGPIPE:
4903 case SIGILL:
4904 case SIGXFSZ:
4905 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4906 break;
4907
4908 case SHUTDOWN1_SIGNAL:
4909 case SHUTDOWN2_SIGNAL:
4910 case SHUTDOWN3_SIGNAL:
4911 case BREAK_SIGNAL:
4912 jvmHandler = (address)user_handler();
4913 break;
4914
4915 default:
4916 if (sig == SR_signum) {
4917 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4918 } else {
4919 return;
4920 }
4921 break;
4922 }
4923
4924 if (thisHandler != jvmHandler) {
4925 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4926 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4927 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4928 // No need to check this sig any longer
4929 sigaddset(&check_signal_done, sig);
4930 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4931 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4932 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4933 exception_name(sig, buf, O_BUFLEN));
4934 }
4935 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4936 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4937 tty->print("expected:");
4938 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4939 tty->cr();
4940 tty->print(" found:");
4941 os::Posix::print_sa_flags(tty, act.sa_flags);
4942 tty->cr();
4943 // No need to check this sig any longer
4944 sigaddset(&check_signal_done, sig);
4945 }
4946
4947 // Dump all the signal
4948 if (sigismember(&check_signal_done, sig)) {
4949 print_signal_handlers(tty, buf, O_BUFLEN);
4950 }
4951 }
4952
4953 extern void report_error(char* file_name, int line_no, char* title,
4954 char* format, ...);
4955
4956 // this is called _before_ most of the global arguments have been parsed
4957 void os::init(void) {
4958 char dummy; // used to get a guess on initial stack address
4959
4960 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4961
4962 init_random(1234567);
4963
4964 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4965 if (Linux::page_size() == -1) {
4966 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4967 os::strerror(errno));
4968 }
4969 init_page_sizes((size_t) Linux::page_size());
4970
4971 Linux::initialize_system_info();
4972
4973 Linux::initialize_os_info();
4974
4975 // _main_thread points to the thread that created/loaded the JVM.
4976 Linux::_main_thread = pthread_self();
4977
4978 Linux::clock_init();
4979 initial_time_count = javaTimeNanos();
4980
4981 // retrieve entry point for pthread_setname_np
4982 Linux::_pthread_setname_np =
4983 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4984
4985 os::Posix::init();
4986 }
4987
4988 // To install functions for atexit system call
4989 extern "C" {
4990 static void perfMemory_exit_helper() {
4991 perfMemory_exit();
4992 }
4993 }
4994
4995 void os::pd_init_container_support() {
4996 OSContainer::init();
4997 }
4998
4999 // this is called _after_ the global arguments have been parsed
5000 jint os::init_2(void) {
5001
5002 os::Posix::init_2();
5003
5004 Linux::fast_thread_clock_init();
5005
5006 // initialize suspend/resume support - must do this before signal_sets_init()
5007 if (SR_initialize() != 0) {
5008 perror("SR_initialize failed");
5009 return JNI_ERR;
5010 }
5011
5012 Linux::signal_sets_init();
5013 Linux::install_signal_handlers();
5014 // Initialize data for jdk.internal.misc.Signal
5015 if (!ReduceSignalUsage) {
5016 jdk_misc_signal_init();
5017 }
5018
5019 // Check and sets minimum stack sizes against command line options
5020 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
5021 return JNI_ERR;
5022 }
5023
5024 suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
5025 if (!suppress_primordial_thread_resolution) {
5026 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5027 }
5028
5029 #if defined(IA32)
5030 workaround_expand_exec_shield_cs_limit();
5031 #endif
5032
5033 Linux::libpthread_init();
5034 Linux::sched_getcpu_init();
5035 log_info(os)("HotSpot is running with %s, %s",
5036 Linux::glibc_version(), Linux::libpthread_version());
5037
5038 if (UseNUMA) {
5039 if (!Linux::libnuma_init()) {
5040 UseNUMA = false;
5041 } else {
5042 if ((Linux::numa_max_node() < 1) || Linux::isbound_to_single_node()) {
5043 // If there's only one node (they start from 0) or if the process
5044 // is bound explicitly to a single node using membind, disable NUMA.
5045 UseNUMA = false;
5046 }
5047 }
5048
5049 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5050 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5051 // we can make the adaptive lgrp chunk resizing work. If the user specified both
5052 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5053 // and disable adaptive resizing.
5054 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5055 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5056 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5057 UseAdaptiveSizePolicy = false;
5058 UseAdaptiveNUMAChunkSizing = false;
5059 }
5060 }
5061
5062 if (!UseNUMA && ForceNUMA) {
5063 UseNUMA = true;
5064 }
5065 }
5066
5067 if (MaxFDLimit) {
5068 // set the number of file descriptors to max. print out error
5069 // if getrlimit/setrlimit fails but continue regardless.
5070 struct rlimit nbr_files;
5071 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5072 if (status != 0) {
5073 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5074 } else {
5075 nbr_files.rlim_cur = nbr_files.rlim_max;
5076 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5077 if (status != 0) {
5078 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5079 }
5080 }
5081 }
5082
5083 // Initialize lock used to serialize thread creation (see os::create_thread)
5084 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5085
5086 // at-exit methods are called in the reverse order of their registration.
5087 // atexit functions are called on return from main or as a result of a
5088 // call to exit(3C). There can be only 32 of these functions registered
5089 // and atexit() does not set errno.
5090
5091 if (PerfAllowAtExitRegistration) {
5092 // only register atexit functions if PerfAllowAtExitRegistration is set.
5093 // atexit functions can be delayed until process exit time, which
5094 // can be problematic for embedded VM situations. Embedded VMs should
5095 // call DestroyJavaVM() to assure that VM resources are released.
5096
5097 // note: perfMemory_exit_helper atexit function may be removed in
5098 // the future if the appropriate cleanup code can be added to the
5099 // VM_Exit VMOperation's doit method.
5100 if (atexit(perfMemory_exit_helper) != 0) {
5101 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5102 }
5103 }
5104
5105 // initialize thread priority policy
5106 prio_init();
5107
5108 if (!FLAG_IS_DEFAULT(AllocateHeapAt)) {
5109 set_coredump_filter(false /*largepages*/, true /*dax_shared*/);
5110 }
5111 return JNI_OK;
5112 }
5113
5114 // Mark the polling page as unreadable
5115 void os::make_polling_page_unreadable(void) {
5116 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5117 fatal("Could not disable polling page");
5118 }
5119 }
5120
5121 // Mark the polling page as readable
5122 void os::make_polling_page_readable(void) {
5123 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5124 fatal("Could not enable polling page");
5125 }
5126 }
5127
5128 // older glibc versions don't have this macro (which expands to
5129 // an optimized bit-counting function) so we have to roll our own
5130 #ifndef CPU_COUNT
5131
5132 static int _cpu_count(const cpu_set_t* cpus) {
5133 int count = 0;
5134 // only look up to the number of configured processors
5135 for (int i = 0; i < os::processor_count(); i++) {
5136 if (CPU_ISSET(i, cpus)) {
5137 count++;
5138 }
5139 }
5140 return count;
5141 }
5142
5143 #define CPU_COUNT(cpus) _cpu_count(cpus)
5144
5145 #endif // CPU_COUNT
5146
5147 // Get the current number of available processors for this process.
5148 // This value can change at any time during a process's lifetime.
5149 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5150 // If it appears there may be more than 1024 processors then we do a
5151 // dynamic check - see 6515172 for details.
5152 // If anything goes wrong we fallback to returning the number of online
5153 // processors - which can be greater than the number available to the process.
5154 int os::Linux::active_processor_count() {
5155 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5156 cpu_set_t* cpus_p = &cpus;
5157 int cpus_size = sizeof(cpu_set_t);
5158
5159 int configured_cpus = os::processor_count(); // upper bound on available cpus
5160 int cpu_count = 0;
5161
5162 // old build platforms may not support dynamic cpu sets
5163 #ifdef CPU_ALLOC
5164
5165 // To enable easy testing of the dynamic path on different platforms we
5166 // introduce a diagnostic flag: UseCpuAllocPath
5167 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5168 // kernel may use a mask bigger than cpu_set_t
5169 log_trace(os)("active_processor_count: using dynamic path %s"
5170 "- configured processors: %d",
5171 UseCpuAllocPath ? "(forced) " : "",
5172 configured_cpus);
5173 cpus_p = CPU_ALLOC(configured_cpus);
5174 if (cpus_p != NULL) {
5175 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5176 // zero it just to be safe
5177 CPU_ZERO_S(cpus_size, cpus_p);
5178 }
5179 else {
5180 // failed to allocate so fallback to online cpus
5181 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5182 log_trace(os)("active_processor_count: "
5183 "CPU_ALLOC failed (%s) - using "
5184 "online processor count: %d",
5185 os::strerror(errno), online_cpus);
5186 return online_cpus;
5187 }
5188 }
5189 else {
5190 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5191 configured_cpus);
5192 }
5193 #else // CPU_ALLOC
5194 // these stubs won't be executed
5195 #define CPU_COUNT_S(size, cpus) -1
5196 #define CPU_FREE(cpus)
5197
5198 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5199 configured_cpus);
5200 #endif // CPU_ALLOC
5201
5202 // pid 0 means the current thread - which we have to assume represents the process
5203 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5204 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5205 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5206 }
5207 else {
5208 cpu_count = CPU_COUNT(cpus_p);
5209 }
5210 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5211 }
5212 else {
5213 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5214 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5215 "which may exceed available processors", os::strerror(errno), cpu_count);
5216 }
5217
5218 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5219 CPU_FREE(cpus_p);
5220 }
5221
5222 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5223 return cpu_count;
5224 }
5225
5226 // Determine the active processor count from one of
5227 // three different sources:
5228 //
5229 // 1. User option -XX:ActiveProcessorCount
5230 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5231 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5232 //
5233 // Option 1, if specified, will always override.
5234 // If the cgroup subsystem is active and configured, we
5235 // will return the min of the cgroup and option 2 results.
5236 // This is required since tools, such as numactl, that
5237 // alter cpu affinity do not update cgroup subsystem
5238 // cpuset configuration files.
5239 int os::active_processor_count() {
5240 // User has overridden the number of active processors
5241 if (ActiveProcessorCount > 0) {
5242 log_trace(os)("active_processor_count: "
5243 "active processor count set by user : %d",
5244 ActiveProcessorCount);
5245 return ActiveProcessorCount;
5246 }
5247
5248 int active_cpus;
5249 if (OSContainer::is_containerized()) {
5250 active_cpus = OSContainer::active_processor_count();
5251 log_trace(os)("active_processor_count: determined by OSContainer: %d",
5252 active_cpus);
5253 } else {
5254 active_cpus = os::Linux::active_processor_count();
5255 }
5256
5257 return active_cpus;
5258 }
5259
5260 uint os::processor_id() {
5261 const int id = Linux::sched_getcpu();
5262 assert(id >= 0 && id < _processor_count, "Invalid processor id");
5263 return (uint)id;
5264 }
5265
5266 void os::set_native_thread_name(const char *name) {
5267 if (Linux::_pthread_setname_np) {
5268 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5269 snprintf(buf, sizeof(buf), "%s", name);
5270 buf[sizeof(buf) - 1] = '\0';
5271 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5272 // ERANGE should not happen; all other errors should just be ignored.
5273 assert(rc != ERANGE, "pthread_setname_np failed");
5274 }
5275 }
5276
5277 bool os::distribute_processes(uint length, uint* distribution) {
5278 // Not yet implemented.
5279 return false;
5280 }
5281
5282 bool os::bind_to_processor(uint processor_id) {
5283 // Not yet implemented.
5284 return false;
5285 }
5286
5287 ///
5288
5289 void os::SuspendedThreadTask::internal_do_task() {
5290 if (do_suspend(_thread->osthread())) {
5291 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5292 do_task(context);
5293 do_resume(_thread->osthread());
5294 }
5295 }
5296
5297 ////////////////////////////////////////////////////////////////////////////////
5298 // debug support
5299
5300 bool os::find(address addr, outputStream* st) {
5301 Dl_info dlinfo;
5302 memset(&dlinfo, 0, sizeof(dlinfo));
5303 if (dladdr(addr, &dlinfo) != 0) {
5304 st->print(PTR_FORMAT ": ", p2i(addr));
5305 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5306 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5307 p2i(addr) - p2i(dlinfo.dli_saddr));
5308 } else if (dlinfo.dli_fbase != NULL) {
5309 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5310 } else {
5311 st->print("<absolute address>");
5312 }
5313 if (dlinfo.dli_fname != NULL) {
5314 st->print(" in %s", dlinfo.dli_fname);
5315 }
5316 if (dlinfo.dli_fbase != NULL) {
5317 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5318 }
5319 st->cr();
5320
5321 if (Verbose) {
5322 // decode some bytes around the PC
5323 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5324 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5325 address lowest = (address) dlinfo.dli_sname;
5326 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5327 if (begin < lowest) begin = lowest;
5328 Dl_info dlinfo2;
5329 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5330 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5331 end = (address) dlinfo2.dli_saddr;
5332 }
5333 Disassembler::decode(begin, end, st);
5334 }
5335 return true;
5336 }
5337 return false;
5338 }
5339
5340 ////////////////////////////////////////////////////////////////////////////////
5341 // misc
5342
5343 // This does not do anything on Linux. This is basically a hook for being
5344 // able to use structured exception handling (thread-local exception filters)
5345 // on, e.g., Win32.
5346 void
5347 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5348 JavaCallArguments* args, Thread* thread) {
5349 f(value, method, args, thread);
5350 }
5351
5352 void os::print_statistics() {
5353 }
5354
5355 bool os::message_box(const char* title, const char* message) {
5356 int i;
5357 fdStream err(defaultStream::error_fd());
5358 for (i = 0; i < 78; i++) err.print_raw("=");
5359 err.cr();
5360 err.print_raw_cr(title);
5361 for (i = 0; i < 78; i++) err.print_raw("-");
5362 err.cr();
5363 err.print_raw_cr(message);
5364 for (i = 0; i < 78; i++) err.print_raw("=");
5365 err.cr();
5366
5367 char buf[16];
5368 // Prevent process from exiting upon "read error" without consuming all CPU
5369 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5370
5371 return buf[0] == 'y' || buf[0] == 'Y';
5372 }
5373
5374 // Is a (classpath) directory empty?
5375 bool os::dir_is_empty(const char* path) {
5376 DIR *dir = NULL;
5377 struct dirent *ptr;
5378
5379 dir = opendir(path);
5380 if (dir == NULL) return true;
5381
5382 // Scan the directory
5383 bool result = true;
5384 while (result && (ptr = readdir(dir)) != NULL) {
5385 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5386 result = false;
5387 }
5388 }
5389 closedir(dir);
5390 return result;
5391 }
5392
5393 // This code originates from JDK's sysOpen and open64_w
5394 // from src/solaris/hpi/src/system_md.c
5395
5396 int os::open(const char *path, int oflag, int mode) {
5397 if (strlen(path) > MAX_PATH - 1) {
5398 errno = ENAMETOOLONG;
5399 return -1;
5400 }
5401
5402 // All file descriptors that are opened in the Java process and not
5403 // specifically destined for a subprocess should have the close-on-exec
5404 // flag set. If we don't set it, then careless 3rd party native code
5405 // might fork and exec without closing all appropriate file descriptors
5406 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5407 // turn might:
5408 //
5409 // - cause end-of-file to fail to be detected on some file
5410 // descriptors, resulting in mysterious hangs, or
5411 //
5412 // - might cause an fopen in the subprocess to fail on a system
5413 // suffering from bug 1085341.
5414 //
5415 // (Yes, the default setting of the close-on-exec flag is a Unix
5416 // design flaw)
5417 //
5418 // See:
5419 // 1085341: 32-bit stdio routines should support file descriptors >255
5420 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5421 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5422 //
5423 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5424 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5425 // because it saves a system call and removes a small window where the flag
5426 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5427 // and we fall back to using FD_CLOEXEC (see below).
5428 #ifdef O_CLOEXEC
5429 oflag |= O_CLOEXEC;
5430 #endif
5431
5432 int fd = ::open64(path, oflag, mode);
5433 if (fd == -1) return -1;
5434
5435 //If the open succeeded, the file might still be a directory
5436 {
5437 struct stat64 buf64;
5438 int ret = ::fstat64(fd, &buf64);
5439 int st_mode = buf64.st_mode;
5440
5441 if (ret != -1) {
5442 if ((st_mode & S_IFMT) == S_IFDIR) {
5443 errno = EISDIR;
5444 ::close(fd);
5445 return -1;
5446 }
5447 } else {
5448 ::close(fd);
5449 return -1;
5450 }
5451 }
5452
5453 #ifdef FD_CLOEXEC
5454 // Validate that the use of the O_CLOEXEC flag on open above worked.
5455 // With recent kernels, we will perform this check exactly once.
5456 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5457 if (!O_CLOEXEC_is_known_to_work) {
5458 int flags = ::fcntl(fd, F_GETFD);
5459 if (flags != -1) {
5460 if ((flags & FD_CLOEXEC) != 0)
5461 O_CLOEXEC_is_known_to_work = 1;
5462 else
5463 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5464 }
5465 }
5466 #endif
5467
5468 return fd;
5469 }
5470
5471
5472 // create binary file, rewriting existing file if required
5473 int os::create_binary_file(const char* path, bool rewrite_existing) {
5474 int oflags = O_WRONLY | O_CREAT;
5475 if (!rewrite_existing) {
5476 oflags |= O_EXCL;
5477 }
5478 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5479 }
5480
5481 // return current position of file pointer
5482 jlong os::current_file_offset(int fd) {
5483 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5484 }
5485
5486 // move file pointer to the specified offset
5487 jlong os::seek_to_file_offset(int fd, jlong offset) {
5488 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5489 }
5490
5491 // This code originates from JDK's sysAvailable
5492 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5493
5494 int os::available(int fd, jlong *bytes) {
5495 jlong cur, end;
5496 int mode;
5497 struct stat64 buf64;
5498
5499 if (::fstat64(fd, &buf64) >= 0) {
5500 mode = buf64.st_mode;
5501 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5502 int n;
5503 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5504 *bytes = n;
5505 return 1;
5506 }
5507 }
5508 }
5509 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5510 return 0;
5511 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5512 return 0;
5513 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5514 return 0;
5515 }
5516 *bytes = end - cur;
5517 return 1;
5518 }
5519
5520 // Map a block of memory.
5521 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5522 char *addr, size_t bytes, bool read_only,
5523 bool allow_exec) {
5524 int prot;
5525 int flags = MAP_PRIVATE;
5526
5527 if (read_only) {
5528 prot = PROT_READ;
5529 } else {
5530 prot = PROT_READ | PROT_WRITE;
5531 }
5532
5533 if (allow_exec) {
5534 prot |= PROT_EXEC;
5535 }
5536
5537 if (addr != NULL) {
5538 flags |= MAP_FIXED;
5539 }
5540
5541 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5542 fd, file_offset);
5543 if (mapped_address == MAP_FAILED) {
5544 return NULL;
5545 }
5546 return mapped_address;
5547 }
5548
5549
5550 // Remap a block of memory.
5551 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5552 char *addr, size_t bytes, bool read_only,
5553 bool allow_exec) {
5554 // same as map_memory() on this OS
5555 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5556 allow_exec);
5557 }
5558
5559
5560 // Unmap a block of memory.
5561 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5562 return munmap(addr, bytes) == 0;
5563 }
5564
5565 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5566
5567 static jlong fast_cpu_time(Thread *thread) {
5568 clockid_t clockid;
5569 int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(),
5570 &clockid);
5571 if (rc == 0) {
5572 return os::Linux::fast_thread_cpu_time(clockid);
5573 } else {
5574 // It's possible to encounter a terminated native thread that failed
5575 // to detach itself from the VM - which should result in ESRCH.
5576 assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed");
5577 return -1;
5578 }
5579 }
5580
5581 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5582 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5583 // of a thread.
5584 //
5585 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5586 // the fast estimate available on the platform.
5587
5588 jlong os::current_thread_cpu_time() {
5589 if (os::Linux::supports_fast_thread_cpu_time()) {
5590 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5591 } else {
5592 // return user + sys since the cost is the same
5593 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5594 }
5595 }
5596
5597 jlong os::thread_cpu_time(Thread* thread) {
5598 // consistent with what current_thread_cpu_time() returns
5599 if (os::Linux::supports_fast_thread_cpu_time()) {
5600 return fast_cpu_time(thread);
5601 } else {
5602 return slow_thread_cpu_time(thread, true /* user + sys */);
5603 }
5604 }
5605
5606 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5607 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5608 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5609 } else {
5610 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5611 }
5612 }
5613
5614 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5615 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5616 return fast_cpu_time(thread);
5617 } else {
5618 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5619 }
5620 }
5621
5622 // -1 on error.
5623 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5624 pid_t tid = thread->osthread()->thread_id();
5625 char *s;
5626 char stat[2048];
5627 int statlen;
5628 char proc_name[64];
5629 int count;
5630 long sys_time, user_time;
5631 char cdummy;
5632 int idummy;
5633 long ldummy;
5634 FILE *fp;
5635
5636 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5637 fp = fopen(proc_name, "r");
5638 if (fp == NULL) return -1;
5639 statlen = fread(stat, 1, 2047, fp);
5640 stat[statlen] = '\0';
5641 fclose(fp);
5642
5643 // Skip pid and the command string. Note that we could be dealing with
5644 // weird command names, e.g. user could decide to rename java launcher
5645 // to "java 1.4.2 :)", then the stat file would look like
5646 // 1234 (java 1.4.2 :)) R ... ...
5647 // We don't really need to know the command string, just find the last
5648 // occurrence of ")" and then start parsing from there. See bug 4726580.
5649 s = strrchr(stat, ')');
5650 if (s == NULL) return -1;
5651
5652 // Skip blank chars
5653 do { s++; } while (s && isspace(*s));
5654
5655 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5656 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5657 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5658 &user_time, &sys_time);
5659 if (count != 13) return -1;
5660 if (user_sys_cpu_time) {
5661 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5662 } else {
5663 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5664 }
5665 }
5666
5667 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5668 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5669 info_ptr->may_skip_backward = false; // elapsed time not wall time
5670 info_ptr->may_skip_forward = false; // elapsed time not wall time
5671 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5672 }
5673
5674 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5675 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5676 info_ptr->may_skip_backward = false; // elapsed time not wall time
5677 info_ptr->may_skip_forward = false; // elapsed time not wall time
5678 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5679 }
5680
5681 bool os::is_thread_cpu_time_supported() {
5682 return true;
5683 }
5684
5685 // System loadavg support. Returns -1 if load average cannot be obtained.
5686 // Linux doesn't yet have a (official) notion of processor sets,
5687 // so just return the system wide load average.
5688 int os::loadavg(double loadavg[], int nelem) {
5689 return ::getloadavg(loadavg, nelem);
5690 }
5691
5692 void os::pause() {
5693 char filename[MAX_PATH];
5694 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5695 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5696 } else {
5697 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5698 }
5699
5700 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5701 if (fd != -1) {
5702 struct stat buf;
5703 ::close(fd);
5704 while (::stat(filename, &buf) == 0) {
5705 (void)::poll(NULL, 0, 100);
5706 }
5707 } else {
5708 jio_fprintf(stderr,
5709 "Could not open pause file '%s', continuing immediately.\n", filename);
5710 }
5711 }
5712
5713 extern char** environ;
5714
5715 // Run the specified command in a separate process. Return its exit value,
5716 // or -1 on failure (e.g. can't fork a new process).
5717 // Unlike system(), this function can be called from signal handler. It
5718 // doesn't block SIGINT et al.
5719 int os::fork_and_exec(char* cmd) {
5720 const char * argv[4] = {"sh", "-c", cmd, NULL};
5721
5722 pid_t pid = fork();
5723
5724 if (pid < 0) {
5725 // fork failed
5726 return -1;
5727
5728 } else if (pid == 0) {
5729 // child process
5730
5731 execve("/bin/sh", (char* const*)argv, environ);
5732
5733 // execve failed
5734 _exit(-1);
5735
5736 } else {
5737 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5738 // care about the actual exit code, for now.
5739
5740 int status;
5741
5742 // Wait for the child process to exit. This returns immediately if
5743 // the child has already exited. */
5744 while (waitpid(pid, &status, 0) < 0) {
5745 switch (errno) {
5746 case ECHILD: return 0;
5747 case EINTR: break;
5748 default: return -1;
5749 }
5750 }
5751
5752 if (WIFEXITED(status)) {
5753 // The child exited normally; get its exit code.
5754 return WEXITSTATUS(status);
5755 } else if (WIFSIGNALED(status)) {
5756 // The child exited because of a signal
5757 // The best value to return is 0x80 + signal number,
5758 // because that is what all Unix shells do, and because
5759 // it allows callers to distinguish between process exit and
5760 // process death by signal.
5761 return 0x80 + WTERMSIG(status);
5762 } else {
5763 // Unknown exit code; pass it through
5764 return status;
5765 }
5766 }
5767 }
5768
5769 // Get the default path to the core file
5770 // Returns the length of the string
5771 int os::get_core_path(char* buffer, size_t bufferSize) {
5772 /*
5773 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5774 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5775 */
5776 const int core_pattern_len = 129;
5777 char core_pattern[core_pattern_len] = {0};
5778
5779 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5780 if (core_pattern_file == -1) {
5781 return -1;
5782 }
5783
5784 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5785 ::close(core_pattern_file);
5786 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5787 return -1;
5788 }
5789 if (core_pattern[ret-1] == '\n') {
5790 core_pattern[ret-1] = '\0';
5791 } else {
5792 core_pattern[ret] = '\0';
5793 }
5794
5795 // Replace the %p in the core pattern with the process id. NOTE: we do this
5796 // only if the pattern doesn't start with "|", and we support only one %p in
5797 // the pattern.
5798 char *pid_pos = strstr(core_pattern, "%p");
5799 const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : ""; // skip over the "%p"
5800 int written;
5801
5802 if (core_pattern[0] == '/') {
5803 if (pid_pos != NULL) {
5804 *pid_pos = '\0';
5805 written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern,
5806 current_process_id(), tail);
5807 } else {
5808 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5809 }
5810 } else {
5811 char cwd[PATH_MAX];
5812
5813 const char* p = get_current_directory(cwd, PATH_MAX);
5814 if (p == NULL) {
5815 return -1;
5816 }
5817
5818 if (core_pattern[0] == '|') {
5819 written = jio_snprintf(buffer, bufferSize,
5820 "\"%s\" (or dumping to %s/core.%d)",
5821 &core_pattern[1], p, current_process_id());
5822 } else if (pid_pos != NULL) {
5823 *pid_pos = '\0';
5824 written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern,
5825 current_process_id(), tail);
5826 } else {
5827 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5828 }
5829 }
5830
5831 if (written < 0) {
5832 return -1;
5833 }
5834
5835 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5836 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5837
5838 if (core_uses_pid_file != -1) {
5839 char core_uses_pid = 0;
5840 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5841 ::close(core_uses_pid_file);
5842
5843 if (core_uses_pid == '1') {
5844 jio_snprintf(buffer + written, bufferSize - written,
5845 ".%d", current_process_id());
5846 }
5847 }
5848 }
5849
5850 return strlen(buffer);
5851 }
5852
5853 bool os::start_debugging(char *buf, int buflen) {
5854 int len = (int)strlen(buf);
5855 char *p = &buf[len];
5856
5857 jio_snprintf(p, buflen-len,
5858 "\n\n"
5859 "Do you want to debug the problem?\n\n"
5860 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5861 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5862 "Otherwise, press RETURN to abort...",
5863 os::current_process_id(), os::current_process_id(),
5864 os::current_thread_id(), os::current_thread_id());
5865
5866 bool yes = os::message_box("Unexpected Error", buf);
5867
5868 if (yes) {
5869 // yes, user asked VM to launch debugger
5870 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5871 os::current_process_id(), os::current_process_id());
5872
5873 os::fork_and_exec(buf);
5874 yes = false;
5875 }
5876 return yes;
5877 }
5878
5879
5880 // Java/Compiler thread:
5881 //
5882 // Low memory addresses
5883 // P0 +------------------------+
5884 // | |\ Java thread created by VM does not have glibc
5885 // | glibc guard page | - guard page, attached Java thread usually has
5886 // | |/ 1 glibc guard page.
5887 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5888 // | |\
5889 // | HotSpot Guard Pages | - red, yellow and reserved pages
5890 // | |/
5891 // +------------------------+ JavaThread::stack_reserved_zone_base()
5892 // | |\
5893 // | Normal Stack | -
5894 // | |/
5895 // P2 +------------------------+ Thread::stack_base()
5896 //
5897 // Non-Java thread:
5898 //
5899 // Low memory addresses
5900 // P0 +------------------------+
5901 // | |\
5902 // | glibc guard page | - usually 1 page
5903 // | |/
5904 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5905 // | |\
5906 // | Normal Stack | -
5907 // | |/
5908 // P2 +------------------------+ Thread::stack_base()
5909 //
5910 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
5911 // returned from pthread_attr_getstack().
5912 // ** Due to NPTL implementation error, linux takes the glibc guard page out
5913 // of the stack size given in pthread_attr. We work around this for
5914 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
5915 //
5916 #ifndef ZERO
5917 static void current_stack_region(address * bottom, size_t * size) {
5918 if (os::is_primordial_thread()) {
5919 // primordial thread needs special handling because pthread_getattr_np()
5920 // may return bogus value.
5921 *bottom = os::Linux::initial_thread_stack_bottom();
5922 *size = os::Linux::initial_thread_stack_size();
5923 } else {
5924 pthread_attr_t attr;
5925
5926 int rslt = pthread_getattr_np(pthread_self(), &attr);
5927
5928 // JVM needs to know exact stack location, abort if it fails
5929 if (rslt != 0) {
5930 if (rslt == ENOMEM) {
5931 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
5932 } else {
5933 fatal("pthread_getattr_np failed with error = %d", rslt);
5934 }
5935 }
5936
5937 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
5938 fatal("Cannot locate current stack attributes!");
5939 }
5940
5941 // Work around NPTL stack guard error.
5942 size_t guard_size = 0;
5943 rslt = pthread_attr_getguardsize(&attr, &guard_size);
5944 if (rslt != 0) {
5945 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
5946 }
5947 *bottom += guard_size;
5948 *size -= guard_size;
5949
5950 pthread_attr_destroy(&attr);
5951
5952 }
5953 assert(os::current_stack_pointer() >= *bottom &&
5954 os::current_stack_pointer() < *bottom + *size, "just checking");
5955 }
5956
5957 address os::current_stack_base() {
5958 address bottom;
5959 size_t size;
5960 current_stack_region(&bottom, &size);
5961 return (bottom + size);
5962 }
5963
5964 size_t os::current_stack_size() {
5965 // This stack size includes the usable stack and HotSpot guard pages
5966 // (for the threads that have Hotspot guard pages).
5967 address bottom;
5968 size_t size;
5969 current_stack_region(&bottom, &size);
5970 return size;
5971 }
5972 #endif
5973
5974 static inline struct timespec get_mtime(const char* filename) {
5975 struct stat st;
5976 int ret = os::stat(filename, &st);
5977 assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
5978 return st.st_mtim;
5979 }
5980
5981 int os::compare_file_modified_times(const char* file1, const char* file2) {
5982 struct timespec filetime1 = get_mtime(file1);
5983 struct timespec filetime2 = get_mtime(file2);
5984 int diff = filetime1.tv_sec - filetime2.tv_sec;
5985 if (diff == 0) {
5986 return filetime1.tv_nsec - filetime2.tv_nsec;
5987 }
5988 return diff;
5989 }
5990
5991 /////////////// Unit tests ///////////////
5992
5993 #ifndef PRODUCT
5994
5995 #define test_log(...) \
5996 do { \
5997 if (VerboseInternalVMTests) { \
5998 tty->print_cr(__VA_ARGS__); \
5999 tty->flush(); \
6000 } \
6001 } while (false)
6002
6003 class TestReserveMemorySpecial : AllStatic {
6004 public:
6005 static void small_page_write(void* addr, size_t size) {
6006 size_t page_size = os::vm_page_size();
6007
6008 char* end = (char*)addr + size;
6009 for (char* p = (char*)addr; p < end; p += page_size) {
6010 *p = 1;
6011 }
6012 }
6013
6014 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6015 if (!UseHugeTLBFS) {
6016 return;
6017 }
6018
6019 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6020
6021 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6022
6023 if (addr != NULL) {
6024 small_page_write(addr, size);
6025
6026 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6027 }
6028 }
6029
6030 static void test_reserve_memory_special_huge_tlbfs_only() {
6031 if (!UseHugeTLBFS) {
6032 return;
6033 }
6034
6035 size_t lp = os::large_page_size();
6036
6037 for (size_t size = lp; size <= lp * 10; size += lp) {
6038 test_reserve_memory_special_huge_tlbfs_only(size);
6039 }
6040 }
6041
6042 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6043 size_t lp = os::large_page_size();
6044 size_t ag = os::vm_allocation_granularity();
6045
6046 // sizes to test
6047 const size_t sizes[] = {
6048 lp, lp + ag, lp + lp / 2, lp * 2,
6049 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6050 lp * 10, lp * 10 + lp / 2
6051 };
6052 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6053
6054 // For each size/alignment combination, we test three scenarios:
6055 // 1) with req_addr == NULL
6056 // 2) with a non-null req_addr at which we expect to successfully allocate
6057 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6058 // expect the allocation to either fail or to ignore req_addr
6059
6060 // Pre-allocate two areas; they shall be as large as the largest allocation
6061 // and aligned to the largest alignment we will be testing.
6062 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6063 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6064 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6065 -1, 0);
6066 assert(mapping1 != MAP_FAILED, "should work");
6067
6068 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6069 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6070 -1, 0);
6071 assert(mapping2 != MAP_FAILED, "should work");
6072
6073 // Unmap the first mapping, but leave the second mapping intact: the first
6074 // mapping will serve as a value for a "good" req_addr (case 2). The second
6075 // mapping, still intact, as "bad" req_addr (case 3).
6076 ::munmap(mapping1, mapping_size);
6077
6078 // Case 1
6079 test_log("%s, req_addr NULL:", __FUNCTION__);
6080 test_log("size align result");
6081
6082 for (int i = 0; i < num_sizes; i++) {
6083 const size_t size = sizes[i];
6084 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6085 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6086 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6087 size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6088 if (p != NULL) {
6089 assert(is_aligned(p, alignment), "must be");
6090 small_page_write(p, size);
6091 os::Linux::release_memory_special_huge_tlbfs(p, size);
6092 }
6093 }
6094 }
6095
6096 // Case 2
6097 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6098 test_log("size align req_addr result");
6099
6100 for (int i = 0; i < num_sizes; i++) {
6101 const size_t size = sizes[i];
6102 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6103 char* const req_addr = align_up(mapping1, alignment);
6104 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6105 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6106 size, alignment, p2i(req_addr), p2i(p),
6107 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6108 if (p != NULL) {
6109 assert(p == req_addr, "must be");
6110 small_page_write(p, size);
6111 os::Linux::release_memory_special_huge_tlbfs(p, size);
6112 }
6113 }
6114 }
6115
6116 // Case 3
6117 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6118 test_log("size align req_addr result");
6119
6120 for (int i = 0; i < num_sizes; i++) {
6121 const size_t size = sizes[i];
6122 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6123 char* const req_addr = align_up(mapping2, alignment);
6124 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6125 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6126 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6127 // as the area around req_addr contains already existing mappings, the API should always
6128 // return NULL (as per contract, it cannot return another address)
6129 assert(p == NULL, "must be");
6130 }
6131 }
6132
6133 ::munmap(mapping2, mapping_size);
6134
6135 }
6136
6137 static void test_reserve_memory_special_huge_tlbfs() {
6138 if (!UseHugeTLBFS) {
6139 return;
6140 }
6141
6142 test_reserve_memory_special_huge_tlbfs_only();
6143 test_reserve_memory_special_huge_tlbfs_mixed();
6144 }
6145
6146 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6147 if (!UseSHM) {
6148 return;
6149 }
6150
6151 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6152
6153 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6154
6155 if (addr != NULL) {
6156 assert(is_aligned(addr, alignment), "Check");
6157 assert(is_aligned(addr, os::large_page_size()), "Check");
6158
6159 small_page_write(addr, size);
6160
6161 os::Linux::release_memory_special_shm(addr, size);
6162 }
6163 }
6164
6165 static void test_reserve_memory_special_shm() {
6166 size_t lp = os::large_page_size();
6167 size_t ag = os::vm_allocation_granularity();
6168
6169 for (size_t size = ag; size < lp * 3; size += ag) {
6170 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6171 test_reserve_memory_special_shm(size, alignment);
6172 }
6173 }
6174 }
6175
6176 static void test() {
6177 test_reserve_memory_special_huge_tlbfs();
6178 test_reserve_memory_special_shm();
6179 }
6180 };
6181
6182 void TestReserveMemorySpecial_test() {
6183 TestReserveMemorySpecial::test();
6184 }
6185
6186 #endif