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. This causes the VM to act as if it is
272 // a single processor and elide locking (see is_MP() call).
273 static bool unsafe_chroot_detected = false;
274 static const char *unstable_chroot_error = "/proc file system not found.\n"
275 "Java may be unstable running multithreaded in a chroot "
276 "environment on Linux when /proc filesystem is not mounted.";
277
278 void os::Linux::initialize_system_info() {
279 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
280 if (processor_count() == 1) {
281 pid_t pid = os::Linux::gettid();
282 char fname[32];
283 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
284 FILE *fp = fopen(fname, "r");
285 if (fp == NULL) {
286 unsafe_chroot_detected = true;
287 } else {
288 fclose(fp);
289 }
290 }
291 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
292 assert(processor_count() > 0, "linux error");
293 }
294
295 void os::init_system_properties_values() {
296 // The next steps are taken in the product version:
297 //
298 // Obtain the JAVA_HOME value from the location of libjvm.so.
299 // This library should be located at:
300 // <JAVA_HOME>/lib/{client|server}/libjvm.so.
301 //
302 // If "/jre/lib/" appears at the right place in the path, then we
303 // assume libjvm.so is installed in a JDK and we use this path.
304 //
305 // Otherwise exit with message: "Could not create the Java virtual machine."
306 //
307 // The following extra steps are taken in the debugging version:
308 //
309 // If "/jre/lib/" does NOT appear at the right place in the path
310 // instead of exit check for $JAVA_HOME environment variable.
311 //
312 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
313 // then we append a fake suffix "hotspot/libjvm.so" to this path so
314 // it looks like libjvm.so is installed there
315 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
316 //
317 // Otherwise exit.
318 //
319 // Important note: if the location of libjvm.so changes this
320 // code needs to be changed accordingly.
321
322 // See ld(1):
323 // The linker uses the following search paths to locate required
324 // shared libraries:
325 // 1: ...
326 // ...
327 // 7: The default directories, normally /lib and /usr/lib.
328 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
329 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
330 #else
331 #define DEFAULT_LIBPATH "/lib:/usr/lib"
332 #endif
333
334 // Base path of extensions installed on the system.
335 #define SYS_EXT_DIR "/usr/java/packages"
336 #define EXTENSIONS_DIR "/lib/ext"
337
338 // Buffer that fits several sprintfs.
339 // Note that the space for the colon and the trailing null are provided
340 // by the nulls included by the sizeof operator.
341 const size_t bufsize =
342 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
343 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
344 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
345
346 // sysclasspath, java_home, dll_dir
347 {
348 char *pslash;
349 os::jvm_path(buf, bufsize);
350
351 // Found the full path to libjvm.so.
352 // Now cut the path to <java_home>/jre if we can.
353 pslash = strrchr(buf, '/');
354 if (pslash != NULL) {
355 *pslash = '\0'; // Get rid of /libjvm.so.
356 }
357 pslash = strrchr(buf, '/');
358 if (pslash != NULL) {
359 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
360 }
361 Arguments::set_dll_dir(buf);
362
363 if (pslash != NULL) {
364 pslash = strrchr(buf, '/');
365 if (pslash != NULL) {
366 *pslash = '\0'; // Get rid of /lib.
367 }
368 }
369 Arguments::set_java_home(buf);
370 set_boot_path('/', ':');
371 }
372
373 // Where to look for native libraries.
374 //
375 // Note: Due to a legacy implementation, most of the library path
376 // is set in the launcher. This was to accomodate linking restrictions
377 // on legacy Linux implementations (which are no longer supported).
378 // Eventually, all the library path setting will be done here.
379 //
380 // However, to prevent the proliferation of improperly built native
381 // libraries, the new path component /usr/java/packages is added here.
382 // Eventually, all the library path setting will be done here.
383 {
384 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
385 // should always exist (until the legacy problem cited above is
386 // addressed).
387 const char *v = ::getenv("LD_LIBRARY_PATH");
388 const char *v_colon = ":";
389 if (v == NULL) { v = ""; v_colon = ""; }
390 // That's +1 for the colon and +1 for the trailing '\0'.
391 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
392 strlen(v) + 1 +
393 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
394 mtInternal);
395 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
396 Arguments::set_library_path(ld_library_path);
397 FREE_C_HEAP_ARRAY(char, ld_library_path);
398 }
399
400 // Extensions directories.
401 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
402 Arguments::set_ext_dirs(buf);
403
404 FREE_C_HEAP_ARRAY(char, buf);
405
406 #undef DEFAULT_LIBPATH
407 #undef SYS_EXT_DIR
408 #undef EXTENSIONS_DIR
409 }
410
411 ////////////////////////////////////////////////////////////////////////////////
412 // breakpoint support
413
414 void os::breakpoint() {
415 BREAKPOINT;
416 }
417
418 extern "C" void breakpoint() {
419 // use debugger to set breakpoint here
420 }
421
422 ////////////////////////////////////////////////////////////////////////////////
423 // signal support
424
425 debug_only(static bool signal_sets_initialized = false);
426 static sigset_t unblocked_sigs, vm_sigs;
427
428 void os::Linux::signal_sets_init() {
429 // Should also have an assertion stating we are still single-threaded.
430 assert(!signal_sets_initialized, "Already initialized");
431 // Fill in signals that are necessarily unblocked for all threads in
432 // the VM. Currently, we unblock the following signals:
433 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
434 // by -Xrs (=ReduceSignalUsage));
435 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
436 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
437 // the dispositions or masks wrt these signals.
438 // Programs embedding the VM that want to use the above signals for their
439 // own purposes must, at this time, use the "-Xrs" option to prevent
440 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
441 // (See bug 4345157, and other related bugs).
442 // In reality, though, unblocking these signals is really a nop, since
443 // these signals are not blocked by default.
444 sigemptyset(&unblocked_sigs);
445 sigaddset(&unblocked_sigs, SIGILL);
446 sigaddset(&unblocked_sigs, SIGSEGV);
447 sigaddset(&unblocked_sigs, SIGBUS);
448 sigaddset(&unblocked_sigs, SIGFPE);
449 #if defined(PPC64)
450 sigaddset(&unblocked_sigs, SIGTRAP);
451 #endif
452 sigaddset(&unblocked_sigs, SR_signum);
453
454 if (!ReduceSignalUsage) {
455 if (!os::Posix::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
456 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
457 }
458 if (!os::Posix::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
459 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
460 }
461 if (!os::Posix::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
462 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
463 }
464 }
465 // Fill in signals that are blocked by all but the VM thread.
466 sigemptyset(&vm_sigs);
467 if (!ReduceSignalUsage) {
468 sigaddset(&vm_sigs, BREAK_SIGNAL);
469 }
470 debug_only(signal_sets_initialized = true);
471
472 }
473
474 // These are signals that are unblocked while a thread is running Java.
475 // (For some reason, they get blocked by default.)
476 sigset_t* os::Linux::unblocked_signals() {
477 assert(signal_sets_initialized, "Not initialized");
478 return &unblocked_sigs;
479 }
480
481 // These are the signals that are blocked while a (non-VM) thread is
482 // running Java. Only the VM thread handles these signals.
483 sigset_t* os::Linux::vm_signals() {
484 assert(signal_sets_initialized, "Not initialized");
485 return &vm_sigs;
486 }
487
488 void os::Linux::hotspot_sigmask(Thread* thread) {
489
490 //Save caller's signal mask before setting VM signal mask
491 sigset_t caller_sigmask;
492 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
493
494 OSThread* osthread = thread->osthread();
495 osthread->set_caller_sigmask(caller_sigmask);
496
497 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
498
499 if (!ReduceSignalUsage) {
500 if (thread->is_VM_thread()) {
501 // Only the VM thread handles BREAK_SIGNAL ...
502 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
503 } else {
504 // ... all other threads block BREAK_SIGNAL
505 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
506 }
507 }
508 }
509
510 //////////////////////////////////////////////////////////////////////////////
511 // detecting pthread library
512
513 void os::Linux::libpthread_init() {
514 // Save glibc and pthread version strings.
515 #if !defined(_CS_GNU_LIBC_VERSION) || \
516 !defined(_CS_GNU_LIBPTHREAD_VERSION)
517 #error "glibc too old (< 2.3.2)"
518 #endif
519
520 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
521 assert(n > 0, "cannot retrieve glibc version");
522 char *str = (char *)malloc(n, mtInternal);
523 confstr(_CS_GNU_LIBC_VERSION, str, n);
524 os::Linux::set_glibc_version(str);
525
526 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
527 assert(n > 0, "cannot retrieve pthread version");
528 str = (char *)malloc(n, mtInternal);
529 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
530 os::Linux::set_libpthread_version(str);
531 }
532
533 /////////////////////////////////////////////////////////////////////////////
534 // thread stack expansion
535
536 // os::Linux::manually_expand_stack() takes care of expanding the thread
537 // stack. Note that this is normally not needed: pthread stacks allocate
538 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
539 // committed. Therefore it is not necessary to expand the stack manually.
540 //
541 // Manually expanding the stack was historically needed on LinuxThreads
542 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
543 // it is kept to deal with very rare corner cases:
544 //
545 // For one, user may run the VM on an own implementation of threads
546 // whose stacks are - like the old LinuxThreads - implemented using
547 // mmap(MAP_GROWSDOWN).
548 //
549 // Also, this coding may be needed if the VM is running on the primordial
550 // thread. Normally we avoid running on the primordial thread; however,
551 // user may still invoke the VM on the primordial thread.
552 //
553 // The following historical comment describes the details about running
554 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
555
556
557 // Force Linux kernel to expand current thread stack. If "bottom" is close
558 // to the stack guard, caller should block all signals.
559 //
560 // MAP_GROWSDOWN:
561 // A special mmap() flag that is used to implement thread stacks. It tells
562 // kernel that the memory region should extend downwards when needed. This
563 // allows early versions of LinuxThreads to only mmap the first few pages
564 // when creating a new thread. Linux kernel will automatically expand thread
565 // stack as needed (on page faults).
566 //
567 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
568 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
569 // region, it's hard to tell if the fault is due to a legitimate stack
570 // access or because of reading/writing non-exist memory (e.g. buffer
571 // overrun). As a rule, if the fault happens below current stack pointer,
572 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
573 // application (see Linux kernel fault.c).
574 //
575 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
576 // stack overflow detection.
577 //
578 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
579 // not use MAP_GROWSDOWN.
580 //
581 // To get around the problem and allow stack banging on Linux, we need to
582 // manually expand thread stack after receiving the SIGSEGV.
583 //
584 // There are two ways to expand thread stack to address "bottom", we used
585 // both of them in JVM before 1.5:
586 // 1. adjust stack pointer first so that it is below "bottom", and then
587 // touch "bottom"
588 // 2. mmap() the page in question
589 //
590 // Now alternate signal stack is gone, it's harder to use 2. For instance,
591 // if current sp is already near the lower end of page 101, and we need to
592 // call mmap() to map page 100, it is possible that part of the mmap() frame
593 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
594 // That will destroy the mmap() frame and cause VM to crash.
595 //
596 // The following code works by adjusting sp first, then accessing the "bottom"
597 // page to force a page fault. Linux kernel will then automatically expand the
598 // stack mapping.
599 //
600 // _expand_stack_to() assumes its frame size is less than page size, which
601 // should always be true if the function is not inlined.
602
603 static void NOINLINE _expand_stack_to(address bottom) {
604 address sp;
605 size_t size;
606 volatile char *p;
607
608 // Adjust bottom to point to the largest address within the same page, it
609 // gives us a one-page buffer if alloca() allocates slightly more memory.
610 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
611 bottom += os::Linux::page_size() - 1;
612
613 // sp might be slightly above current stack pointer; if that's the case, we
614 // will alloca() a little more space than necessary, which is OK. Don't use
615 // os::current_stack_pointer(), as its result can be slightly below current
616 // stack pointer, causing us to not alloca enough to reach "bottom".
617 sp = (address)&sp;
618
619 if (sp > bottom) {
620 size = sp - bottom;
621 p = (volatile char *)alloca(size);
622 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
623 p[0] = '\0';
624 }
625 }
626
627 void os::Linux::expand_stack_to(address bottom) {
628 _expand_stack_to(bottom);
629 }
630
631 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
632 assert(t!=NULL, "just checking");
633 assert(t->osthread()->expanding_stack(), "expand should be set");
634 assert(t->stack_base() != NULL, "stack_base was not initialized");
635
636 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
637 sigset_t mask_all, old_sigset;
638 sigfillset(&mask_all);
639 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
640 _expand_stack_to(addr);
641 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
642 return true;
643 }
644 return false;
645 }
646
647 //////////////////////////////////////////////////////////////////////////////
648 // create new thread
649
650 // Thread start routine for all newly created threads
651 static void *thread_native_entry(Thread *thread) {
652 // Try to randomize the cache line index of hot stack frames.
653 // This helps when threads of the same stack traces evict each other's
654 // cache lines. The threads can be either from the same JVM instance, or
655 // from different JVM instances. The benefit is especially true for
656 // processors with hyperthreading technology.
657 static int counter = 0;
658 int pid = os::current_process_id();
659 alloca(((pid ^ counter++) & 7) * 128);
660
661 thread->initialize_thread_current();
662
663 OSThread* osthread = thread->osthread();
664 Monitor* sync = osthread->startThread_lock();
665
666 osthread->set_thread_id(os::current_thread_id());
667
668 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
669 os::current_thread_id(), (uintx) pthread_self());
670
671 if (UseNUMA) {
672 int lgrp_id = os::numa_get_group_id();
673 if (lgrp_id != -1) {
674 thread->set_lgrp_id(lgrp_id);
675 }
676 }
677 // initialize signal mask for this thread
678 os::Linux::hotspot_sigmask(thread);
679
680 // initialize floating point control register
681 os::Linux::init_thread_fpu_state();
682
683 // handshaking with parent thread
684 {
685 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
686
687 // notify parent thread
688 osthread->set_state(INITIALIZED);
689 sync->notify_all();
690
691 // wait until os::start_thread()
692 while (osthread->get_state() == INITIALIZED) {
693 sync->wait(Mutex::_no_safepoint_check_flag);
694 }
695 }
696
697 // call one more level start routine
698 thread->run();
699
700 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
701 os::current_thread_id(), (uintx) pthread_self());
702
703 // If a thread has not deleted itself ("delete this") as part of its
704 // termination sequence, we have to ensure thread-local-storage is
705 // cleared before we actually terminate. No threads should ever be
706 // deleted asynchronously with respect to their termination.
707 if (Thread::current_or_null_safe() != NULL) {
708 assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
709 thread->clear_thread_current();
710 }
711
712 return 0;
713 }
714
715 bool os::create_thread(Thread* thread, ThreadType thr_type,
716 size_t req_stack_size) {
717 assert(thread->osthread() == NULL, "caller responsible");
718
719 // Allocate the OSThread object
720 OSThread* osthread = new OSThread(NULL, NULL);
721 if (osthread == NULL) {
722 return false;
723 }
724
725 // set the correct thread state
726 osthread->set_thread_type(thr_type);
727
728 // Initial state is ALLOCATED but not INITIALIZED
729 osthread->set_state(ALLOCATED);
730
731 thread->set_osthread(osthread);
732
733 // init thread attributes
734 pthread_attr_t attr;
735 pthread_attr_init(&attr);
736 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
737
738 // Calculate stack size if it's not specified by caller.
739 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
740 // In the Linux NPTL pthread implementation the guard size mechanism
741 // is not implemented properly. The posix standard requires adding
742 // the size of the guard pages to the stack size, instead Linux
743 // takes the space out of 'stacksize'. Thus we adapt the requested
744 // stack_size by the size of the guard pages to mimick proper
745 // behaviour. However, be careful not to end up with a size
746 // of zero due to overflow. Don't add the guard page in that case.
747 size_t guard_size = os::Linux::default_guard_size(thr_type);
748 if (stack_size <= SIZE_MAX - guard_size) {
749 stack_size += guard_size;
750 }
751 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
752
753 int status = pthread_attr_setstacksize(&attr, stack_size);
754 assert_status(status == 0, status, "pthread_attr_setstacksize");
755
756 // Configure glibc guard page.
757 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
758
759 ThreadState state;
760
761 {
762 pthread_t tid;
763 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
764
765 char buf[64];
766 if (ret == 0) {
767 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
768 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
769 } else {
770 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
771 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
772 }
773
774 pthread_attr_destroy(&attr);
775
776 if (ret != 0) {
777 // Need to clean up stuff we've allocated so far
778 thread->set_osthread(NULL);
779 delete osthread;
780 return false;
781 }
782
783 // Store pthread info into the OSThread
784 osthread->set_pthread_id(tid);
785
786 // Wait until child thread is either initialized or aborted
787 {
788 Monitor* sync_with_child = osthread->startThread_lock();
789 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
790 while ((state = osthread->get_state()) == ALLOCATED) {
791 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
792 }
793 }
794 }
795
796 // Aborted due to thread limit being reached
797 if (state == ZOMBIE) {
798 thread->set_osthread(NULL);
799 delete osthread;
800 return false;
801 }
802
803 // The thread is returned suspended (in state INITIALIZED),
804 // and is started higher up in the call chain
805 assert(state == INITIALIZED, "race condition");
806 return true;
807 }
808
809 /////////////////////////////////////////////////////////////////////////////
810 // attach existing thread
811
812 // bootstrap the main thread
813 bool os::create_main_thread(JavaThread* thread) {
814 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
815 return create_attached_thread(thread);
816 }
817
818 bool os::create_attached_thread(JavaThread* thread) {
819 #ifdef ASSERT
820 thread->verify_not_published();
821 #endif
822
823 // Allocate the OSThread object
824 OSThread* osthread = new OSThread(NULL, NULL);
825
826 if (osthread == NULL) {
827 return false;
828 }
829
830 // Store pthread info into the OSThread
831 osthread->set_thread_id(os::Linux::gettid());
832 osthread->set_pthread_id(::pthread_self());
833
834 // initialize floating point control register
835 os::Linux::init_thread_fpu_state();
836
837 // Initial thread state is RUNNABLE
838 osthread->set_state(RUNNABLE);
839
840 thread->set_osthread(osthread);
841
842 if (UseNUMA) {
843 int lgrp_id = os::numa_get_group_id();
844 if (lgrp_id != -1) {
845 thread->set_lgrp_id(lgrp_id);
846 }
847 }
848
849 if (os::is_primordial_thread()) {
850 // If current thread is primordial thread, its stack is mapped on demand,
851 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
852 // the entire stack region to avoid SEGV in stack banging.
853 // It is also useful to get around the heap-stack-gap problem on SuSE
854 // kernel (see 4821821 for details). We first expand stack to the top
855 // of yellow zone, then enable stack yellow zone (order is significant,
856 // enabling yellow zone first will crash JVM on SuSE Linux), so there
857 // is no gap between the last two virtual memory regions.
858
859 JavaThread *jt = (JavaThread *)thread;
860 address addr = jt->stack_reserved_zone_base();
861 assert(addr != NULL, "initialization problem?");
862 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
863
864 osthread->set_expanding_stack();
865 os::Linux::manually_expand_stack(jt, addr);
866 osthread->clear_expanding_stack();
867 }
868
869 // initialize signal mask for this thread
870 // and save the caller's signal mask
871 os::Linux::hotspot_sigmask(thread);
872
873 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
874 os::current_thread_id(), (uintx) pthread_self());
875
876 return true;
877 }
878
879 void os::pd_start_thread(Thread* thread) {
880 OSThread * osthread = thread->osthread();
881 assert(osthread->get_state() != INITIALIZED, "just checking");
882 Monitor* sync_with_child = osthread->startThread_lock();
883 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
884 sync_with_child->notify();
885 }
886
887 // Free Linux resources related to the OSThread
888 void os::free_thread(OSThread* osthread) {
889 assert(osthread != NULL, "osthread not set");
890
891 // We are told to free resources of the argument thread,
892 // but we can only really operate on the current thread.
893 assert(Thread::current()->osthread() == osthread,
894 "os::free_thread but not current thread");
895
896 #ifdef ASSERT
897 sigset_t current;
898 sigemptyset(¤t);
899 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
900 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
901 #endif
902
903 // Restore caller's signal mask
904 sigset_t sigmask = osthread->caller_sigmask();
905 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
906
907 delete osthread;
908 }
909
910 //////////////////////////////////////////////////////////////////////////////
911 // primordial thread
912
913 // Check if current thread is the primordial thread, similar to Solaris thr_main.
914 bool os::is_primordial_thread(void) {
915 if (suppress_primordial_thread_resolution) {
916 return false;
917 }
918 char dummy;
919 // If called before init complete, thread stack bottom will be null.
920 // Can be called if fatal error occurs before initialization.
921 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
922 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
923 os::Linux::initial_thread_stack_size() != 0,
924 "os::init did not locate primordial thread's stack region");
925 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
926 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
927 os::Linux::initial_thread_stack_size()) {
928 return true;
929 } else {
930 return false;
931 }
932 }
933
934 // Find the virtual memory area that contains addr
935 static bool find_vma(address addr, address* vma_low, address* vma_high) {
936 FILE *fp = fopen("/proc/self/maps", "r");
937 if (fp) {
938 address low, high;
939 while (!feof(fp)) {
940 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
941 if (low <= addr && addr < high) {
942 if (vma_low) *vma_low = low;
943 if (vma_high) *vma_high = high;
944 fclose(fp);
945 return true;
946 }
947 }
948 for (;;) {
949 int ch = fgetc(fp);
950 if (ch == EOF || ch == (int)'\n') break;
951 }
952 }
953 fclose(fp);
954 }
955 return false;
956 }
957
958 // Locate primordial thread stack. This special handling of primordial thread stack
959 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
960 // bogus value for the primordial process thread. While the launcher has created
961 // the VM in a new thread since JDK 6, we still have to allow for the use of the
962 // JNI invocation API from a primordial thread.
963 void os::Linux::capture_initial_stack(size_t max_size) {
964
965 // max_size is either 0 (which means accept OS default for thread stacks) or
966 // a user-specified value known to be at least the minimum needed. If we
967 // are actually on the primordial thread we can make it appear that we have a
968 // smaller max_size stack by inserting the guard pages at that location. But we
969 // cannot do anything to emulate a larger stack than what has been provided by
970 // the OS or threading library. In fact if we try to use a stack greater than
971 // what is set by rlimit then we will crash the hosting process.
972
973 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
974 // If this is "unlimited" then it will be a huge value.
975 struct rlimit rlim;
976 getrlimit(RLIMIT_STACK, &rlim);
977 size_t stack_size = rlim.rlim_cur;
978
979 // 6308388: a bug in ld.so will relocate its own .data section to the
980 // lower end of primordial stack; reduce ulimit -s value a little bit
981 // so we won't install guard page on ld.so's data section.
982 // But ensure we don't underflow the stack size - allow 1 page spare
983 if (stack_size >= (size_t)(3 * page_size())) {
984 stack_size -= 2 * page_size();
985 }
986
987 // Try to figure out where the stack base (top) is. This is harder.
988 //
989 // When an application is started, glibc saves the initial stack pointer in
990 // a global variable "__libc_stack_end", which is then used by system
991 // libraries. __libc_stack_end should be pretty close to stack top. The
992 // variable is available since the very early days. However, because it is
993 // a private interface, it could disappear in the future.
994 //
995 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
996 // to __libc_stack_end, it is very close to stack top, but isn't the real
997 // stack top. Note that /proc may not exist if VM is running as a chroot
998 // program, so reading /proc/<pid>/stat could fail. Also the contents of
999 // /proc/<pid>/stat could change in the future (though unlikely).
1000 //
1001 // We try __libc_stack_end first. If that doesn't work, look for
1002 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1003 // as a hint, which should work well in most cases.
1004
1005 uintptr_t stack_start;
1006
1007 // try __libc_stack_end first
1008 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1009 if (p && *p) {
1010 stack_start = *p;
1011 } else {
1012 // see if we can get the start_stack field from /proc/self/stat
1013 FILE *fp;
1014 int pid;
1015 char state;
1016 int ppid;
1017 int pgrp;
1018 int session;
1019 int nr;
1020 int tpgrp;
1021 unsigned long flags;
1022 unsigned long minflt;
1023 unsigned long cminflt;
1024 unsigned long majflt;
1025 unsigned long cmajflt;
1026 unsigned long utime;
1027 unsigned long stime;
1028 long cutime;
1029 long cstime;
1030 long prio;
1031 long nice;
1032 long junk;
1033 long it_real;
1034 uintptr_t start;
1035 uintptr_t vsize;
1036 intptr_t rss;
1037 uintptr_t rsslim;
1038 uintptr_t scodes;
1039 uintptr_t ecode;
1040 int i;
1041
1042 // Figure what the primordial thread stack base is. Code is inspired
1043 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1044 // followed by command name surrounded by parentheses, state, etc.
1045 char stat[2048];
1046 int statlen;
1047
1048 fp = fopen("/proc/self/stat", "r");
1049 if (fp) {
1050 statlen = fread(stat, 1, 2047, fp);
1051 stat[statlen] = '\0';
1052 fclose(fp);
1053
1054 // Skip pid and the command string. Note that we could be dealing with
1055 // weird command names, e.g. user could decide to rename java launcher
1056 // to "java 1.4.2 :)", then the stat file would look like
1057 // 1234 (java 1.4.2 :)) R ... ...
1058 // We don't really need to know the command string, just find the last
1059 // occurrence of ")" and then start parsing from there. See bug 4726580.
1060 char * s = strrchr(stat, ')');
1061
1062 i = 0;
1063 if (s) {
1064 // Skip blank chars
1065 do { s++; } while (s && isspace(*s));
1066
1067 #define _UFM UINTX_FORMAT
1068 #define _DFM INTX_FORMAT
1069
1070 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1071 // 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
1072 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,
1073 &state, // 3 %c
1074 &ppid, // 4 %d
1075 &pgrp, // 5 %d
1076 &session, // 6 %d
1077 &nr, // 7 %d
1078 &tpgrp, // 8 %d
1079 &flags, // 9 %lu
1080 &minflt, // 10 %lu
1081 &cminflt, // 11 %lu
1082 &majflt, // 12 %lu
1083 &cmajflt, // 13 %lu
1084 &utime, // 14 %lu
1085 &stime, // 15 %lu
1086 &cutime, // 16 %ld
1087 &cstime, // 17 %ld
1088 &prio, // 18 %ld
1089 &nice, // 19 %ld
1090 &junk, // 20 %ld
1091 &it_real, // 21 %ld
1092 &start, // 22 UINTX_FORMAT
1093 &vsize, // 23 UINTX_FORMAT
1094 &rss, // 24 INTX_FORMAT
1095 &rsslim, // 25 UINTX_FORMAT
1096 &scodes, // 26 UINTX_FORMAT
1097 &ecode, // 27 UINTX_FORMAT
1098 &stack_start); // 28 UINTX_FORMAT
1099 }
1100
1101 #undef _UFM
1102 #undef _DFM
1103
1104 if (i != 28 - 2) {
1105 assert(false, "Bad conversion from /proc/self/stat");
1106 // product mode - assume we are the primordial thread, good luck in the
1107 // embedded case.
1108 warning("Can't detect primordial thread stack location - bad conversion");
1109 stack_start = (uintptr_t) &rlim;
1110 }
1111 } else {
1112 // For some reason we can't open /proc/self/stat (for example, running on
1113 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1114 // most cases, so don't abort:
1115 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1116 stack_start = (uintptr_t) &rlim;
1117 }
1118 }
1119
1120 // Now we have a pointer (stack_start) very close to the stack top, the
1121 // next thing to do is to figure out the exact location of stack top. We
1122 // can find out the virtual memory area that contains stack_start by
1123 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1124 // and its upper limit is the real stack top. (again, this would fail if
1125 // running inside chroot, because /proc may not exist.)
1126
1127 uintptr_t stack_top;
1128 address low, high;
1129 if (find_vma((address)stack_start, &low, &high)) {
1130 // success, "high" is the true stack top. (ignore "low", because initial
1131 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1132 stack_top = (uintptr_t)high;
1133 } else {
1134 // failed, likely because /proc/self/maps does not exist
1135 warning("Can't detect primordial thread stack location - find_vma failed");
1136 // best effort: stack_start is normally within a few pages below the real
1137 // stack top, use it as stack top, and reduce stack size so we won't put
1138 // guard page outside stack.
1139 stack_top = stack_start;
1140 stack_size -= 16 * page_size();
1141 }
1142
1143 // stack_top could be partially down the page so align it
1144 stack_top = align_up(stack_top, page_size());
1145
1146 // Allowed stack value is minimum of max_size and what we derived from rlimit
1147 if (max_size > 0) {
1148 _initial_thread_stack_size = MIN2(max_size, stack_size);
1149 } else {
1150 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1151 // clamp it at 8MB as we do on Solaris
1152 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1153 }
1154 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1155 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1156
1157 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1158
1159 if (log_is_enabled(Info, os, thread)) {
1160 // See if we seem to be on primordial process thread
1161 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1162 uintptr_t(&rlim) < stack_top;
1163
1164 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1165 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1166 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1167 stack_top, intptr_t(_initial_thread_stack_bottom));
1168 }
1169 }
1170
1171 ////////////////////////////////////////////////////////////////////////////////
1172 // time support
1173
1174 // Time since start-up in seconds to a fine granularity.
1175 // Used by VMSelfDestructTimer and the MemProfiler.
1176 double os::elapsedTime() {
1177
1178 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1179 }
1180
1181 jlong os::elapsed_counter() {
1182 return javaTimeNanos() - initial_time_count;
1183 }
1184
1185 jlong os::elapsed_frequency() {
1186 return NANOSECS_PER_SEC; // nanosecond resolution
1187 }
1188
1189 bool os::supports_vtime() { return true; }
1190 bool os::enable_vtime() { return false; }
1191 bool os::vtime_enabled() { return false; }
1192
1193 double os::elapsedVTime() {
1194 struct rusage usage;
1195 int retval = getrusage(RUSAGE_THREAD, &usage);
1196 if (retval == 0) {
1197 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);
1198 } else {
1199 // better than nothing, but not much
1200 return elapsedTime();
1201 }
1202 }
1203
1204 jlong os::javaTimeMillis() {
1205 timeval time;
1206 int status = gettimeofday(&time, NULL);
1207 assert(status != -1, "linux error");
1208 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1209 }
1210
1211 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1212 timeval time;
1213 int status = gettimeofday(&time, NULL);
1214 assert(status != -1, "linux error");
1215 seconds = jlong(time.tv_sec);
1216 nanos = jlong(time.tv_usec) * 1000;
1217 }
1218
1219
1220 #ifndef CLOCK_MONOTONIC
1221 #define CLOCK_MONOTONIC (1)
1222 #endif
1223
1224 void os::Linux::clock_init() {
1225 // we do dlopen's in this particular order due to bug in linux
1226 // dynamical loader (see 6348968) leading to crash on exit
1227 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1228 if (handle == NULL) {
1229 handle = dlopen("librt.so", RTLD_LAZY);
1230 }
1231
1232 if (handle) {
1233 int (*clock_getres_func)(clockid_t, struct timespec*) =
1234 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1235 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1236 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1237 if (clock_getres_func && clock_gettime_func) {
1238 // See if monotonic clock is supported by the kernel. Note that some
1239 // early implementations simply return kernel jiffies (updated every
1240 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1241 // for nano time (though the monotonic property is still nice to have).
1242 // It's fixed in newer kernels, however clock_getres() still returns
1243 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1244 // resolution for now. Hopefully as people move to new kernels, this
1245 // won't be a problem.
1246 struct timespec res;
1247 struct timespec tp;
1248 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1249 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1250 // yes, monotonic clock is supported
1251 _clock_gettime = clock_gettime_func;
1252 return;
1253 } else {
1254 // close librt if there is no monotonic clock
1255 dlclose(handle);
1256 }
1257 }
1258 }
1259 warning("No monotonic clock was available - timed services may " \
1260 "be adversely affected if the time-of-day clock changes");
1261 }
1262
1263 #ifndef SYS_clock_getres
1264 #if defined(X86) || defined(PPC64) || defined(S390)
1265 #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261)
1266 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1267 #else
1268 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1269 #define sys_clock_getres(x,y) -1
1270 #endif
1271 #else
1272 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1273 #endif
1274
1275 void os::Linux::fast_thread_clock_init() {
1276 if (!UseLinuxPosixThreadCPUClocks) {
1277 return;
1278 }
1279 clockid_t clockid;
1280 struct timespec tp;
1281 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1282 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1283
1284 // Switch to using fast clocks for thread cpu time if
1285 // the sys_clock_getres() returns 0 error code.
1286 // Note, that some kernels may support the current thread
1287 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1288 // returned by the pthread_getcpuclockid().
1289 // If the fast Posix clocks are supported then the sys_clock_getres()
1290 // must return at least tp.tv_sec == 0 which means a resolution
1291 // better than 1 sec. This is extra check for reliability.
1292
1293 if (pthread_getcpuclockid_func &&
1294 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1295 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1296 _supports_fast_thread_cpu_time = true;
1297 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1298 }
1299 }
1300
1301 jlong os::javaTimeNanos() {
1302 if (os::supports_monotonic_clock()) {
1303 struct timespec tp;
1304 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1305 assert(status == 0, "gettime error");
1306 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1307 return result;
1308 } else {
1309 timeval time;
1310 int status = gettimeofday(&time, NULL);
1311 assert(status != -1, "linux error");
1312 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1313 return 1000 * usecs;
1314 }
1315 }
1316
1317 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1318 if (os::supports_monotonic_clock()) {
1319 info_ptr->max_value = ALL_64_BITS;
1320
1321 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1322 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1323 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1324 } else {
1325 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1326 info_ptr->max_value = ALL_64_BITS;
1327
1328 // gettimeofday is a real time clock so it skips
1329 info_ptr->may_skip_backward = true;
1330 info_ptr->may_skip_forward = true;
1331 }
1332
1333 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1334 }
1335
1336 // Return the real, user, and system times in seconds from an
1337 // arbitrary fixed point in the past.
1338 bool os::getTimesSecs(double* process_real_time,
1339 double* process_user_time,
1340 double* process_system_time) {
1341 struct tms ticks;
1342 clock_t real_ticks = times(&ticks);
1343
1344 if (real_ticks == (clock_t) (-1)) {
1345 return false;
1346 } else {
1347 double ticks_per_second = (double) clock_tics_per_sec;
1348 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1349 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1350 *process_real_time = ((double) real_ticks) / ticks_per_second;
1351
1352 return true;
1353 }
1354 }
1355
1356
1357 char * os::local_time_string(char *buf, size_t buflen) {
1358 struct tm t;
1359 time_t long_time;
1360 time(&long_time);
1361 localtime_r(&long_time, &t);
1362 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1363 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1364 t.tm_hour, t.tm_min, t.tm_sec);
1365 return buf;
1366 }
1367
1368 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1369 return localtime_r(clock, res);
1370 }
1371
1372 ////////////////////////////////////////////////////////////////////////////////
1373 // runtime exit support
1374
1375 // Note: os::shutdown() might be called very early during initialization, or
1376 // called from signal handler. Before adding something to os::shutdown(), make
1377 // sure it is async-safe and can handle partially initialized VM.
1378 void os::shutdown() {
1379
1380 // allow PerfMemory to attempt cleanup of any persistent resources
1381 perfMemory_exit();
1382
1383 // needs to remove object in file system
1384 AttachListener::abort();
1385
1386 // flush buffered output, finish log files
1387 ostream_abort();
1388
1389 // Check for abort hook
1390 abort_hook_t abort_hook = Arguments::abort_hook();
1391 if (abort_hook != NULL) {
1392 abort_hook();
1393 }
1394
1395 }
1396
1397 // Note: os::abort() might be called very early during initialization, or
1398 // called from signal handler. Before adding something to os::abort(), make
1399 // sure it is async-safe and can handle partially initialized VM.
1400 void os::abort(bool dump_core, void* siginfo, const void* context) {
1401 os::shutdown();
1402 if (dump_core) {
1403 #ifndef PRODUCT
1404 fdStream out(defaultStream::output_fd());
1405 out.print_raw("Current thread is ");
1406 char buf[16];
1407 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1408 out.print_raw_cr(buf);
1409 out.print_raw_cr("Dumping core ...");
1410 #endif
1411 ::abort(); // dump core
1412 }
1413
1414 ::exit(1);
1415 }
1416
1417 // Die immediately, no exit hook, no abort hook, no cleanup.
1418 void os::die() {
1419 ::abort();
1420 }
1421
1422
1423 // This method is a copy of JDK's sysGetLastErrorString
1424 // from src/solaris/hpi/src/system_md.c
1425
1426 size_t os::lasterror(char *buf, size_t len) {
1427 if (errno == 0) return 0;
1428
1429 const char *s = os::strerror(errno);
1430 size_t n = ::strlen(s);
1431 if (n >= len) {
1432 n = len - 1;
1433 }
1434 ::strncpy(buf, s, n);
1435 buf[n] = '\0';
1436 return n;
1437 }
1438
1439 // thread_id is kernel thread id (similar to Solaris LWP id)
1440 intx os::current_thread_id() { return os::Linux::gettid(); }
1441 int os::current_process_id() {
1442 return ::getpid();
1443 }
1444
1445 // DLL functions
1446
1447 const char* os::dll_file_extension() { return ".so"; }
1448
1449 // This must be hard coded because it's the system's temporary
1450 // directory not the java application's temp directory, ala java.io.tmpdir.
1451 const char* os::get_temp_directory() { return "/tmp"; }
1452
1453 static bool file_exists(const char* filename) {
1454 struct stat statbuf;
1455 if (filename == NULL || strlen(filename) == 0) {
1456 return false;
1457 }
1458 return os::stat(filename, &statbuf) == 0;
1459 }
1460
1461 // check if addr is inside libjvm.so
1462 bool os::address_is_in_vm(address addr) {
1463 static address libjvm_base_addr;
1464 Dl_info dlinfo;
1465
1466 if (libjvm_base_addr == NULL) {
1467 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1468 libjvm_base_addr = (address)dlinfo.dli_fbase;
1469 }
1470 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1471 }
1472
1473 if (dladdr((void *)addr, &dlinfo) != 0) {
1474 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1475 }
1476
1477 return false;
1478 }
1479
1480 bool os::dll_address_to_function_name(address addr, char *buf,
1481 int buflen, int *offset,
1482 bool demangle) {
1483 // buf is not optional, but offset is optional
1484 assert(buf != NULL, "sanity check");
1485
1486 Dl_info dlinfo;
1487
1488 if (dladdr((void*)addr, &dlinfo) != 0) {
1489 // see if we have a matching symbol
1490 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1491 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1492 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1493 }
1494 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1495 return true;
1496 }
1497 // no matching symbol so try for just file info
1498 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1499 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1500 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1501 return true;
1502 }
1503 }
1504 }
1505
1506 buf[0] = '\0';
1507 if (offset != NULL) *offset = -1;
1508 return false;
1509 }
1510
1511 struct _address_to_library_name {
1512 address addr; // input : memory address
1513 size_t buflen; // size of fname
1514 char* fname; // output: library name
1515 address base; // library base addr
1516 };
1517
1518 static int address_to_library_name_callback(struct dl_phdr_info *info,
1519 size_t size, void *data) {
1520 int i;
1521 bool found = false;
1522 address libbase = NULL;
1523 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1524
1525 // iterate through all loadable segments
1526 for (i = 0; i < info->dlpi_phnum; i++) {
1527 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1528 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1529 // base address of a library is the lowest address of its loaded
1530 // segments.
1531 if (libbase == NULL || libbase > segbase) {
1532 libbase = segbase;
1533 }
1534 // see if 'addr' is within current segment
1535 if (segbase <= d->addr &&
1536 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1537 found = true;
1538 }
1539 }
1540 }
1541
1542 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1543 // so dll_address_to_library_name() can fall through to use dladdr() which
1544 // can figure out executable name from argv[0].
1545 if (found && info->dlpi_name && info->dlpi_name[0]) {
1546 d->base = libbase;
1547 if (d->fname) {
1548 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1549 }
1550 return 1;
1551 }
1552 return 0;
1553 }
1554
1555 bool os::dll_address_to_library_name(address addr, char* buf,
1556 int buflen, int* offset) {
1557 // buf is not optional, but offset is optional
1558 assert(buf != NULL, "sanity check");
1559
1560 Dl_info dlinfo;
1561 struct _address_to_library_name data;
1562
1563 // There is a bug in old glibc dladdr() implementation that it could resolve
1564 // to wrong library name if the .so file has a base address != NULL. Here
1565 // we iterate through the program headers of all loaded libraries to find
1566 // out which library 'addr' really belongs to. This workaround can be
1567 // removed once the minimum requirement for glibc is moved to 2.3.x.
1568 data.addr = addr;
1569 data.fname = buf;
1570 data.buflen = buflen;
1571 data.base = NULL;
1572 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1573
1574 if (rslt) {
1575 // buf already contains library name
1576 if (offset) *offset = addr - data.base;
1577 return true;
1578 }
1579 if (dladdr((void*)addr, &dlinfo) != 0) {
1580 if (dlinfo.dli_fname != NULL) {
1581 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1582 }
1583 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1584 *offset = addr - (address)dlinfo.dli_fbase;
1585 }
1586 return true;
1587 }
1588
1589 buf[0] = '\0';
1590 if (offset) *offset = -1;
1591 return false;
1592 }
1593
1594 // Loads .dll/.so and
1595 // in case of error it checks if .dll/.so was built for the
1596 // same architecture as Hotspot is running on
1597
1598
1599 // Remember the stack's state. The Linux dynamic linker will change
1600 // the stack to 'executable' at most once, so we must safepoint only once.
1601 bool os::Linux::_stack_is_executable = false;
1602
1603 // VM operation that loads a library. This is necessary if stack protection
1604 // of the Java stacks can be lost during loading the library. If we
1605 // do not stop the Java threads, they can stack overflow before the stacks
1606 // are protected again.
1607 class VM_LinuxDllLoad: public VM_Operation {
1608 private:
1609 const char *_filename;
1610 char *_ebuf;
1611 int _ebuflen;
1612 void *_lib;
1613 public:
1614 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1615 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1616 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1617 void doit() {
1618 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1619 os::Linux::_stack_is_executable = true;
1620 }
1621 void* loaded_library() { return _lib; }
1622 };
1623
1624 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1625 void * result = NULL;
1626 bool load_attempted = false;
1627
1628 // Check whether the library to load might change execution rights
1629 // of the stack. If they are changed, the protection of the stack
1630 // guard pages will be lost. We need a safepoint to fix this.
1631 //
1632 // See Linux man page execstack(8) for more info.
1633 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1634 if (!ElfFile::specifies_noexecstack(filename)) {
1635 if (!is_init_completed()) {
1636 os::Linux::_stack_is_executable = true;
1637 // This is OK - No Java threads have been created yet, and hence no
1638 // stack guard pages to fix.
1639 //
1640 // This should happen only when you are building JDK7 using a very
1641 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1642 //
1643 // Dynamic loader will make all stacks executable after
1644 // this function returns, and will not do that again.
1645 assert(Threads::number_of_threads() == 0, "no Java threads should exist yet.");
1646 } else {
1647 warning("You have loaded library %s which might have disabled stack guard. "
1648 "The VM will try to fix the stack guard now.\n"
1649 "It's highly recommended that you fix the library with "
1650 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1651 filename);
1652
1653 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1654 JavaThread *jt = JavaThread::current();
1655 if (jt->thread_state() != _thread_in_native) {
1656 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1657 // that requires ExecStack. Cannot enter safe point. Let's give up.
1658 warning("Unable to fix stack guard. Giving up.");
1659 } else {
1660 if (!LoadExecStackDllInVMThread) {
1661 // This is for the case where the DLL has an static
1662 // constructor function that executes JNI code. We cannot
1663 // load such DLLs in the VMThread.
1664 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1665 }
1666
1667 ThreadInVMfromNative tiv(jt);
1668 debug_only(VMNativeEntryWrapper vew;)
1669
1670 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1671 VMThread::execute(&op);
1672 if (LoadExecStackDllInVMThread) {
1673 result = op.loaded_library();
1674 }
1675 load_attempted = true;
1676 }
1677 }
1678 }
1679 }
1680
1681 if (!load_attempted) {
1682 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1683 }
1684
1685 if (result != NULL) {
1686 // Successful loading
1687 return result;
1688 }
1689
1690 Elf32_Ehdr elf_head;
1691 int diag_msg_max_length=ebuflen-strlen(ebuf);
1692 char* diag_msg_buf=ebuf+strlen(ebuf);
1693
1694 if (diag_msg_max_length==0) {
1695 // No more space in ebuf for additional diagnostics message
1696 return NULL;
1697 }
1698
1699
1700 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1701
1702 if (file_descriptor < 0) {
1703 // Can't open library, report dlerror() message
1704 return NULL;
1705 }
1706
1707 bool failed_to_read_elf_head=
1708 (sizeof(elf_head)!=
1709 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1710
1711 ::close(file_descriptor);
1712 if (failed_to_read_elf_head) {
1713 // file i/o error - report dlerror() msg
1714 return NULL;
1715 }
1716
1717 typedef struct {
1718 Elf32_Half code; // Actual value as defined in elf.h
1719 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1720 unsigned char elf_class; // 32 or 64 bit
1721 unsigned char endianess; // MSB or LSB
1722 char* name; // String representation
1723 } arch_t;
1724
1725 #ifndef EM_486
1726 #define EM_486 6 /* Intel 80486 */
1727 #endif
1728 #ifndef EM_AARCH64
1729 #define EM_AARCH64 183 /* ARM AARCH64 */
1730 #endif
1731
1732 static const arch_t arch_array[]={
1733 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1734 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1735 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1736 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1737 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1738 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1739 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1740 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1741 #if defined(VM_LITTLE_ENDIAN)
1742 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1743 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1744 #else
1745 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1746 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1747 #endif
1748 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1749 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1750 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1751 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1752 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1753 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1754 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1755 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1756 };
1757
1758 #if (defined IA32)
1759 static Elf32_Half running_arch_code=EM_386;
1760 #elif (defined AMD64)
1761 static Elf32_Half running_arch_code=EM_X86_64;
1762 #elif (defined IA64)
1763 static Elf32_Half running_arch_code=EM_IA_64;
1764 #elif (defined __sparc) && (defined _LP64)
1765 static Elf32_Half running_arch_code=EM_SPARCV9;
1766 #elif (defined __sparc) && (!defined _LP64)
1767 static Elf32_Half running_arch_code=EM_SPARC;
1768 #elif (defined __powerpc64__)
1769 static Elf32_Half running_arch_code=EM_PPC64;
1770 #elif (defined __powerpc__)
1771 static Elf32_Half running_arch_code=EM_PPC;
1772 #elif (defined AARCH64)
1773 static Elf32_Half running_arch_code=EM_AARCH64;
1774 #elif (defined ARM)
1775 static Elf32_Half running_arch_code=EM_ARM;
1776 #elif (defined S390)
1777 static Elf32_Half running_arch_code=EM_S390;
1778 #elif (defined ALPHA)
1779 static Elf32_Half running_arch_code=EM_ALPHA;
1780 #elif (defined MIPSEL)
1781 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1782 #elif (defined PARISC)
1783 static Elf32_Half running_arch_code=EM_PARISC;
1784 #elif (defined MIPS)
1785 static Elf32_Half running_arch_code=EM_MIPS;
1786 #elif (defined M68K)
1787 static Elf32_Half running_arch_code=EM_68K;
1788 #elif (defined SH)
1789 static Elf32_Half running_arch_code=EM_SH;
1790 #else
1791 #error Method os::dll_load requires that one of following is defined:\
1792 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1793 #endif
1794
1795 // Identify compatability class for VM's architecture and library's architecture
1796 // Obtain string descriptions for architectures
1797
1798 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1799 int running_arch_index=-1;
1800
1801 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1802 if (running_arch_code == arch_array[i].code) {
1803 running_arch_index = i;
1804 }
1805 if (lib_arch.code == arch_array[i].code) {
1806 lib_arch.compat_class = arch_array[i].compat_class;
1807 lib_arch.name = arch_array[i].name;
1808 }
1809 }
1810
1811 assert(running_arch_index != -1,
1812 "Didn't find running architecture code (running_arch_code) in arch_array");
1813 if (running_arch_index == -1) {
1814 // Even though running architecture detection failed
1815 // we may still continue with reporting dlerror() message
1816 return NULL;
1817 }
1818
1819 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1820 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1821 return NULL;
1822 }
1823
1824 #ifndef S390
1825 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1826 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1827 return NULL;
1828 }
1829 #endif // !S390
1830
1831 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1832 if (lib_arch.name!=NULL) {
1833 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1834 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1835 lib_arch.name, arch_array[running_arch_index].name);
1836 } else {
1837 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1838 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1839 lib_arch.code,
1840 arch_array[running_arch_index].name);
1841 }
1842 }
1843
1844 return NULL;
1845 }
1846
1847 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1848 int ebuflen) {
1849 void * result = ::dlopen(filename, RTLD_LAZY);
1850 if (result == NULL) {
1851 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1852 ebuf[ebuflen-1] = '\0';
1853 }
1854 return result;
1855 }
1856
1857 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1858 int ebuflen) {
1859 void * result = NULL;
1860 if (LoadExecStackDllInVMThread) {
1861 result = dlopen_helper(filename, ebuf, ebuflen);
1862 }
1863
1864 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1865 // library that requires an executable stack, or which does not have this
1866 // stack attribute set, dlopen changes the stack attribute to executable. The
1867 // read protection of the guard pages gets lost.
1868 //
1869 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1870 // may have been queued at the same time.
1871
1872 if (!_stack_is_executable) {
1873 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
1874 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1875 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1876 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1877 warning("Attempt to reguard stack yellow zone failed.");
1878 }
1879 }
1880 }
1881 }
1882
1883 return result;
1884 }
1885
1886 void* os::dll_lookup(void* handle, const char* name) {
1887 void* res = dlsym(handle, name);
1888 return res;
1889 }
1890
1891 void* os::get_default_process_handle() {
1892 return (void*)::dlopen(NULL, RTLD_LAZY);
1893 }
1894
1895 static bool _print_ascii_file(const char* filename, outputStream* st) {
1896 int fd = ::open(filename, O_RDONLY);
1897 if (fd == -1) {
1898 return false;
1899 }
1900
1901 char buf[33];
1902 int bytes;
1903 buf[32] = '\0';
1904 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1905 st->print_raw(buf, bytes);
1906 }
1907
1908 ::close(fd);
1909
1910 return true;
1911 }
1912
1913 void os::print_dll_info(outputStream *st) {
1914 st->print_cr("Dynamic libraries:");
1915
1916 char fname[32];
1917 pid_t pid = os::Linux::gettid();
1918
1919 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1920
1921 if (!_print_ascii_file(fname, st)) {
1922 st->print("Can not get library information for pid = %d\n", pid);
1923 }
1924 }
1925
1926 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1927 FILE *procmapsFile = NULL;
1928
1929 // Open the procfs maps file for the current process
1930 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1931 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1932 char line[PATH_MAX + 100];
1933
1934 // Read line by line from 'file'
1935 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1936 u8 base, top, offset, inode;
1937 char permissions[5];
1938 char device[6];
1939 char name[PATH_MAX + 1];
1940
1941 // Parse fields from line
1942 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1943 &base, &top, permissions, &offset, device, &inode, name);
1944
1945 // Filter by device id '00:00' so that we only get file system mapped files.
1946 if (strcmp(device, "00:00") != 0) {
1947
1948 // Call callback with the fields of interest
1949 if(callback(name, (address)base, (address)top, param)) {
1950 // Oops abort, callback aborted
1951 fclose(procmapsFile);
1952 return 1;
1953 }
1954 }
1955 }
1956 fclose(procmapsFile);
1957 }
1958 return 0;
1959 }
1960
1961 void os::print_os_info_brief(outputStream* st) {
1962 os::Linux::print_distro_info(st);
1963
1964 os::Posix::print_uname_info(st);
1965
1966 os::Linux::print_libversion_info(st);
1967
1968 }
1969
1970 void os::print_os_info(outputStream* st) {
1971 st->print("OS:");
1972
1973 os::Linux::print_distro_info(st);
1974
1975 os::Posix::print_uname_info(st);
1976
1977 // Print warning if unsafe chroot environment detected
1978 if (unsafe_chroot_detected) {
1979 st->print("WARNING!! ");
1980 st->print_cr("%s", unstable_chroot_error);
1981 }
1982
1983 os::Linux::print_libversion_info(st);
1984
1985 os::Posix::print_rlimit_info(st);
1986
1987 os::Posix::print_load_average(st);
1988
1989 os::Linux::print_full_memory_info(st);
1990
1991 os::Linux::print_proc_sys_info(st);
1992
1993 os::Linux::print_container_info(st);
1994 }
1995
1996 // Try to identify popular distros.
1997 // Most Linux distributions have a /etc/XXX-release file, which contains
1998 // the OS version string. Newer Linux distributions have a /etc/lsb-release
1999 // file that also contains the OS version string. Some have more than one
2000 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2001 // /etc/redhat-release.), so the order is important.
2002 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2003 // their own specific XXX-release file as well as a redhat-release file.
2004 // Because of this the XXX-release file needs to be searched for before the
2005 // redhat-release file.
2006 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2007 // search for redhat-release / SuSE-release needs to be before lsb-release.
2008 // Since the lsb-release file is the new standard it needs to be searched
2009 // before the older style release files.
2010 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2011 // next to last resort. The os-release file is a new standard that contains
2012 // distribution information and the system-release file seems to be an old
2013 // standard that has been replaced by the lsb-release and os-release files.
2014 // Searching for the debian_version file is the last resort. It contains
2015 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2016 // "Debian " is printed before the contents of the debian_version file.
2017
2018 const char* distro_files[] = {
2019 "/etc/oracle-release",
2020 "/etc/mandriva-release",
2021 "/etc/mandrake-release",
2022 "/etc/sun-release",
2023 "/etc/redhat-release",
2024 "/etc/SuSE-release",
2025 "/etc/lsb-release",
2026 "/etc/turbolinux-release",
2027 "/etc/gentoo-release",
2028 "/etc/ltib-release",
2029 "/etc/angstrom-version",
2030 "/etc/system-release",
2031 "/etc/os-release",
2032 NULL };
2033
2034 void os::Linux::print_distro_info(outputStream* st) {
2035 for (int i = 0;; i++) {
2036 const char* file = distro_files[i];
2037 if (file == NULL) {
2038 break; // done
2039 }
2040 // If file prints, we found it.
2041 if (_print_ascii_file(file, st)) {
2042 return;
2043 }
2044 }
2045
2046 if (file_exists("/etc/debian_version")) {
2047 st->print("Debian ");
2048 _print_ascii_file("/etc/debian_version", st);
2049 } else {
2050 st->print("Linux");
2051 }
2052 st->cr();
2053 }
2054
2055 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2056 char buf[256];
2057 while (fgets(buf, sizeof(buf), fp)) {
2058 // Edit out extra stuff in expected format
2059 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2060 char* ptr = strstr(buf, "\""); // the name is in quotes
2061 if (ptr != NULL) {
2062 ptr++; // go beyond first quote
2063 char* nl = strchr(ptr, '\"');
2064 if (nl != NULL) *nl = '\0';
2065 strncpy(distro, ptr, length);
2066 } else {
2067 ptr = strstr(buf, "=");
2068 ptr++; // go beyond equals then
2069 char* nl = strchr(ptr, '\n');
2070 if (nl != NULL) *nl = '\0';
2071 strncpy(distro, ptr, length);
2072 }
2073 return;
2074 } else if (get_first_line) {
2075 char* nl = strchr(buf, '\n');
2076 if (nl != NULL) *nl = '\0';
2077 strncpy(distro, buf, length);
2078 return;
2079 }
2080 }
2081 // print last line and close
2082 char* nl = strchr(buf, '\n');
2083 if (nl != NULL) *nl = '\0';
2084 strncpy(distro, buf, length);
2085 }
2086
2087 static void parse_os_info(char* distro, size_t length, const char* file) {
2088 FILE* fp = fopen(file, "r");
2089 if (fp != NULL) {
2090 // if suse format, print out first line
2091 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2092 parse_os_info_helper(fp, distro, length, get_first_line);
2093 fclose(fp);
2094 }
2095 }
2096
2097 void os::get_summary_os_info(char* buf, size_t buflen) {
2098 for (int i = 0;; i++) {
2099 const char* file = distro_files[i];
2100 if (file == NULL) {
2101 break; // ran out of distro_files
2102 }
2103 if (file_exists(file)) {
2104 parse_os_info(buf, buflen, file);
2105 return;
2106 }
2107 }
2108 // special case for debian
2109 if (file_exists("/etc/debian_version")) {
2110 strncpy(buf, "Debian ", buflen);
2111 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2112 } else {
2113 strncpy(buf, "Linux", buflen);
2114 }
2115 }
2116
2117 void os::Linux::print_libversion_info(outputStream* st) {
2118 // libc, pthread
2119 st->print("libc:");
2120 st->print("%s ", os::Linux::glibc_version());
2121 st->print("%s ", os::Linux::libpthread_version());
2122 st->cr();
2123 }
2124
2125 void os::Linux::print_proc_sys_info(outputStream* st) {
2126 st->cr();
2127 st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):");
2128 _print_ascii_file("/proc/sys/kernel/threads-max", st);
2129 st->cr();
2130 st->cr();
2131
2132 st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):");
2133 _print_ascii_file("/proc/sys/vm/max_map_count", st);
2134 st->cr();
2135 st->cr();
2136
2137 st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):");
2138 _print_ascii_file("/proc/sys/kernel/pid_max", st);
2139 st->cr();
2140 st->cr();
2141 }
2142
2143 void os::Linux::print_full_memory_info(outputStream* st) {
2144 st->print("\n/proc/meminfo:\n");
2145 _print_ascii_file("/proc/meminfo", st);
2146 st->cr();
2147 }
2148
2149 void os::Linux::print_container_info(outputStream* st) {
2150 if (!OSContainer::is_containerized()) {
2151 return;
2152 }
2153
2154 st->print("container (cgroup) information:\n");
2155
2156 const char *p_ct = OSContainer::container_type();
2157 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed");
2158
2159 char *p = OSContainer::cpu_cpuset_cpus();
2160 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed");
2161 free(p);
2162
2163 p = OSContainer::cpu_cpuset_memory_nodes();
2164 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed");
2165 free(p);
2166
2167 int i = OSContainer::active_processor_count();
2168 if (i > 0) {
2169 st->print("active_processor_count: %d\n", i);
2170 } else {
2171 st->print("active_processor_count: failed\n");
2172 }
2173
2174 i = OSContainer::cpu_quota();
2175 st->print("cpu_quota: %d\n", i);
2176
2177 i = OSContainer::cpu_period();
2178 st->print("cpu_period: %d\n", i);
2179
2180 i = OSContainer::cpu_shares();
2181 st->print("cpu_shares: %d\n", i);
2182
2183 jlong j = OSContainer::memory_limit_in_bytes();
2184 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2185
2186 j = OSContainer::memory_and_swap_limit_in_bytes();
2187 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2188
2189 j = OSContainer::memory_soft_limit_in_bytes();
2190 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2191
2192 j = OSContainer::OSContainer::memory_usage_in_bytes();
2193 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2194
2195 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2196 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2197 st->cr();
2198 }
2199
2200 void os::print_memory_info(outputStream* st) {
2201
2202 st->print("Memory:");
2203 st->print(" %dk page", os::vm_page_size()>>10);
2204
2205 // values in struct sysinfo are "unsigned long"
2206 struct sysinfo si;
2207 sysinfo(&si);
2208
2209 st->print(", physical " UINT64_FORMAT "k",
2210 os::physical_memory() >> 10);
2211 st->print("(" UINT64_FORMAT "k free)",
2212 os::available_memory() >> 10);
2213 st->print(", swap " UINT64_FORMAT "k",
2214 ((jlong)si.totalswap * si.mem_unit) >> 10);
2215 st->print("(" UINT64_FORMAT "k free)",
2216 ((jlong)si.freeswap * si.mem_unit) >> 10);
2217 st->cr();
2218 }
2219
2220 // Print the first "model name" line and the first "flags" line
2221 // that we find and nothing more. We assume "model name" comes
2222 // before "flags" so if we find a second "model name", then the
2223 // "flags" field is considered missing.
2224 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2225 #if defined(IA32) || defined(AMD64)
2226 // Other platforms have less repetitive cpuinfo files
2227 FILE *fp = fopen("/proc/cpuinfo", "r");
2228 if (fp) {
2229 while (!feof(fp)) {
2230 if (fgets(buf, buflen, fp)) {
2231 // Assume model name comes before flags
2232 bool model_name_printed = false;
2233 if (strstr(buf, "model name") != NULL) {
2234 if (!model_name_printed) {
2235 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2236 st->print_raw(buf);
2237 model_name_printed = true;
2238 } else {
2239 // model name printed but not flags? Odd, just return
2240 fclose(fp);
2241 return true;
2242 }
2243 }
2244 // print the flags line too
2245 if (strstr(buf, "flags") != NULL) {
2246 st->print_raw(buf);
2247 fclose(fp);
2248 return true;
2249 }
2250 }
2251 }
2252 fclose(fp);
2253 }
2254 #endif // x86 platforms
2255 return false;
2256 }
2257
2258 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2259 // Only print the model name if the platform provides this as a summary
2260 if (!print_model_name_and_flags(st, buf, buflen)) {
2261 st->print("\n/proc/cpuinfo:\n");
2262 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2263 st->print_cr(" <Not Available>");
2264 }
2265 }
2266 }
2267
2268 #if defined(AMD64) || defined(IA32) || defined(X32)
2269 const char* search_string = "model name";
2270 #elif defined(M68K)
2271 const char* search_string = "CPU";
2272 #elif defined(PPC64)
2273 const char* search_string = "cpu";
2274 #elif defined(S390)
2275 const char* search_string = "processor";
2276 #elif defined(SPARC)
2277 const char* search_string = "cpu";
2278 #else
2279 const char* search_string = "Processor";
2280 #endif
2281
2282 // Parses the cpuinfo file for string representing the model name.
2283 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2284 FILE* fp = fopen("/proc/cpuinfo", "r");
2285 if (fp != NULL) {
2286 while (!feof(fp)) {
2287 char buf[256];
2288 if (fgets(buf, sizeof(buf), fp)) {
2289 char* start = strstr(buf, search_string);
2290 if (start != NULL) {
2291 char *ptr = start + strlen(search_string);
2292 char *end = buf + strlen(buf);
2293 while (ptr != end) {
2294 // skip whitespace and colon for the rest of the name.
2295 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2296 break;
2297 }
2298 ptr++;
2299 }
2300 if (ptr != end) {
2301 // reasonable string, get rid of newline and keep the rest
2302 char* nl = strchr(buf, '\n');
2303 if (nl != NULL) *nl = '\0';
2304 strncpy(cpuinfo, ptr, length);
2305 fclose(fp);
2306 return;
2307 }
2308 }
2309 }
2310 }
2311 fclose(fp);
2312 }
2313 // cpuinfo not found or parsing failed, just print generic string. The entire
2314 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2315 #if defined(AARCH64)
2316 strncpy(cpuinfo, "AArch64", length);
2317 #elif defined(AMD64)
2318 strncpy(cpuinfo, "x86_64", length);
2319 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2320 strncpy(cpuinfo, "ARM", length);
2321 #elif defined(IA32)
2322 strncpy(cpuinfo, "x86_32", length);
2323 #elif defined(IA64)
2324 strncpy(cpuinfo, "IA64", length);
2325 #elif defined(PPC)
2326 strncpy(cpuinfo, "PPC64", length);
2327 #elif defined(S390)
2328 strncpy(cpuinfo, "S390", length);
2329 #elif defined(SPARC)
2330 strncpy(cpuinfo, "sparcv9", length);
2331 #elif defined(ZERO_LIBARCH)
2332 strncpy(cpuinfo, ZERO_LIBARCH, length);
2333 #else
2334 strncpy(cpuinfo, "unknown", length);
2335 #endif
2336 }
2337
2338 static void print_signal_handler(outputStream* st, int sig,
2339 char* buf, size_t buflen);
2340
2341 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2342 st->print_cr("Signal Handlers:");
2343 print_signal_handler(st, SIGSEGV, buf, buflen);
2344 print_signal_handler(st, SIGBUS , buf, buflen);
2345 print_signal_handler(st, SIGFPE , buf, buflen);
2346 print_signal_handler(st, SIGPIPE, buf, buflen);
2347 print_signal_handler(st, SIGXFSZ, buf, buflen);
2348 print_signal_handler(st, SIGILL , buf, buflen);
2349 print_signal_handler(st, SR_signum, buf, buflen);
2350 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2351 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2352 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2353 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2354 #if defined(PPC64)
2355 print_signal_handler(st, SIGTRAP, buf, buflen);
2356 #endif
2357 }
2358
2359 static char saved_jvm_path[MAXPATHLEN] = {0};
2360
2361 // Find the full path to the current module, libjvm.so
2362 void os::jvm_path(char *buf, jint buflen) {
2363 // Error checking.
2364 if (buflen < MAXPATHLEN) {
2365 assert(false, "must use a large-enough buffer");
2366 buf[0] = '\0';
2367 return;
2368 }
2369 // Lazy resolve the path to current module.
2370 if (saved_jvm_path[0] != 0) {
2371 strcpy(buf, saved_jvm_path);
2372 return;
2373 }
2374
2375 char dli_fname[MAXPATHLEN];
2376 bool ret = dll_address_to_library_name(
2377 CAST_FROM_FN_PTR(address, os::jvm_path),
2378 dli_fname, sizeof(dli_fname), NULL);
2379 assert(ret, "cannot locate libjvm");
2380 char *rp = NULL;
2381 if (ret && dli_fname[0] != '\0') {
2382 rp = os::Posix::realpath(dli_fname, buf, buflen);
2383 }
2384 if (rp == NULL) {
2385 return;
2386 }
2387
2388 if (Arguments::sun_java_launcher_is_altjvm()) {
2389 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2390 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2391 // If "/jre/lib/" appears at the right place in the string, then
2392 // assume we are installed in a JDK and we're done. Otherwise, check
2393 // for a JAVA_HOME environment variable and fix up the path so it
2394 // looks like libjvm.so is installed there (append a fake suffix
2395 // hotspot/libjvm.so).
2396 const char *p = buf + strlen(buf) - 1;
2397 for (int count = 0; p > buf && count < 5; ++count) {
2398 for (--p; p > buf && *p != '/'; --p)
2399 /* empty */ ;
2400 }
2401
2402 if (strncmp(p, "/jre/lib/", 9) != 0) {
2403 // Look for JAVA_HOME in the environment.
2404 char* java_home_var = ::getenv("JAVA_HOME");
2405 if (java_home_var != NULL && java_home_var[0] != 0) {
2406 char* jrelib_p;
2407 int len;
2408
2409 // Check the current module name "libjvm.so".
2410 p = strrchr(buf, '/');
2411 if (p == NULL) {
2412 return;
2413 }
2414 assert(strstr(p, "/libjvm") == p, "invalid library name");
2415
2416 rp = os::Posix::realpath(java_home_var, buf, buflen);
2417 if (rp == NULL) {
2418 return;
2419 }
2420
2421 // determine if this is a legacy image or modules image
2422 // modules image doesn't have "jre" subdirectory
2423 len = strlen(buf);
2424 assert(len < buflen, "Ran out of buffer room");
2425 jrelib_p = buf + len;
2426 snprintf(jrelib_p, buflen-len, "/jre/lib");
2427 if (0 != access(buf, F_OK)) {
2428 snprintf(jrelib_p, buflen-len, "/lib");
2429 }
2430
2431 if (0 == access(buf, F_OK)) {
2432 // Use current module name "libjvm.so"
2433 len = strlen(buf);
2434 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2435 } else {
2436 // Go back to path of .so
2437 rp = os::Posix::realpath(dli_fname, buf, buflen);
2438 if (rp == NULL) {
2439 return;
2440 }
2441 }
2442 }
2443 }
2444 }
2445
2446 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2447 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2448 }
2449
2450 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2451 // no prefix required, not even "_"
2452 }
2453
2454 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2455 // no suffix required
2456 }
2457
2458 ////////////////////////////////////////////////////////////////////////////////
2459 // sun.misc.Signal support
2460
2461 static volatile jint sigint_count = 0;
2462
2463 static void UserHandler(int sig, void *siginfo, void *context) {
2464 // 4511530 - sem_post is serialized and handled by the manager thread. When
2465 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2466 // don't want to flood the manager thread with sem_post requests.
2467 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2468 return;
2469 }
2470
2471 // Ctrl-C is pressed during error reporting, likely because the error
2472 // handler fails to abort. Let VM die immediately.
2473 if (sig == SIGINT && VMError::is_error_reported()) {
2474 os::die();
2475 }
2476
2477 os::signal_notify(sig);
2478 }
2479
2480 void* os::user_handler() {
2481 return CAST_FROM_FN_PTR(void*, UserHandler);
2482 }
2483
2484 static struct timespec create_semaphore_timespec(unsigned int sec, int nsec) {
2485 struct timespec ts;
2486 // Semaphore's are always associated with CLOCK_REALTIME
2487 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2488 // see os_posix.cpp for discussion on overflow checking
2489 if (sec >= MAX_SECS) {
2490 ts.tv_sec += MAX_SECS;
2491 ts.tv_nsec = 0;
2492 } else {
2493 ts.tv_sec += sec;
2494 ts.tv_nsec += nsec;
2495 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2496 ts.tv_nsec -= NANOSECS_PER_SEC;
2497 ++ts.tv_sec; // note: this must be <= max_secs
2498 }
2499 }
2500
2501 return ts;
2502 }
2503
2504 extern "C" {
2505 typedef void (*sa_handler_t)(int);
2506 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2507 }
2508
2509 void* os::signal(int signal_number, void* handler) {
2510 struct sigaction sigAct, oldSigAct;
2511
2512 sigfillset(&(sigAct.sa_mask));
2513 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2514 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2515
2516 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2517 // -1 means registration failed
2518 return (void *)-1;
2519 }
2520
2521 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2522 }
2523
2524 void os::signal_raise(int signal_number) {
2525 ::raise(signal_number);
2526 }
2527
2528 // The following code is moved from os.cpp for making this
2529 // code platform specific, which it is by its very nature.
2530
2531 // Will be modified when max signal is changed to be dynamic
2532 int os::sigexitnum_pd() {
2533 return NSIG;
2534 }
2535
2536 // a counter for each possible signal value
2537 static volatile jint pending_signals[NSIG+1] = { 0 };
2538
2539 // Linux(POSIX) specific hand shaking semaphore.
2540 static Semaphore* sig_sem = NULL;
2541 static PosixSemaphore sr_semaphore;
2542
2543 static void jdk_misc_signal_init() {
2544 // Initialize signal structures
2545 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2546
2547 // Initialize signal semaphore
2548 sig_sem = new Semaphore();
2549 }
2550
2551 void os::signal_notify(int sig) {
2552 if (sig_sem != NULL) {
2553 Atomic::inc(&pending_signals[sig]);
2554 sig_sem->signal();
2555 } else {
2556 // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2557 // initialization isn't called.
2558 assert(ReduceSignalUsage, "signal semaphore should be created");
2559 }
2560 }
2561
2562 static int check_pending_signals() {
2563 Atomic::store(0, &sigint_count);
2564 for (;;) {
2565 for (int i = 0; i < NSIG + 1; i++) {
2566 jint n = pending_signals[i];
2567 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2568 return i;
2569 }
2570 }
2571 JavaThread *thread = JavaThread::current();
2572 ThreadBlockInVM tbivm(thread);
2573
2574 bool threadIsSuspended;
2575 do {
2576 thread->set_suspend_equivalent();
2577 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2578 sig_sem->wait();
2579
2580 // were we externally suspended while we were waiting?
2581 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2582 if (threadIsSuspended) {
2583 // The semaphore has been incremented, but while we were waiting
2584 // another thread suspended us. We don't want to continue running
2585 // while suspended because that would surprise the thread that
2586 // suspended us.
2587 sig_sem->signal();
2588
2589 thread->java_suspend_self();
2590 }
2591 } while (threadIsSuspended);
2592 }
2593 }
2594
2595 int os::signal_wait() {
2596 return check_pending_signals();
2597 }
2598
2599 ////////////////////////////////////////////////////////////////////////////////
2600 // Virtual Memory
2601
2602 int os::vm_page_size() {
2603 // Seems redundant as all get out
2604 assert(os::Linux::page_size() != -1, "must call os::init");
2605 return os::Linux::page_size();
2606 }
2607
2608 // Solaris allocates memory by pages.
2609 int os::vm_allocation_granularity() {
2610 assert(os::Linux::page_size() != -1, "must call os::init");
2611 return os::Linux::page_size();
2612 }
2613
2614 // Rationale behind this function:
2615 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2616 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2617 // samples for JITted code. Here we create private executable mapping over the code cache
2618 // and then we can use standard (well, almost, as mapping can change) way to provide
2619 // info for the reporting script by storing timestamp and location of symbol
2620 void linux_wrap_code(char* base, size_t size) {
2621 static volatile jint cnt = 0;
2622
2623 if (!UseOprofile) {
2624 return;
2625 }
2626
2627 char buf[PATH_MAX+1];
2628 int num = Atomic::add(1, &cnt);
2629
2630 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2631 os::get_temp_directory(), os::current_process_id(), num);
2632 unlink(buf);
2633
2634 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2635
2636 if (fd != -1) {
2637 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2638 if (rv != (off_t)-1) {
2639 if (::write(fd, "", 1) == 1) {
2640 mmap(base, size,
2641 PROT_READ|PROT_WRITE|PROT_EXEC,
2642 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2643 }
2644 }
2645 ::close(fd);
2646 unlink(buf);
2647 }
2648 }
2649
2650 static bool recoverable_mmap_error(int err) {
2651 // See if the error is one we can let the caller handle. This
2652 // list of errno values comes from JBS-6843484. I can't find a
2653 // Linux man page that documents this specific set of errno
2654 // values so while this list currently matches Solaris, it may
2655 // change as we gain experience with this failure mode.
2656 switch (err) {
2657 case EBADF:
2658 case EINVAL:
2659 case ENOTSUP:
2660 // let the caller deal with these errors
2661 return true;
2662
2663 default:
2664 // Any remaining errors on this OS can cause our reserved mapping
2665 // to be lost. That can cause confusion where different data
2666 // structures think they have the same memory mapped. The worst
2667 // scenario is if both the VM and a library think they have the
2668 // same memory mapped.
2669 return false;
2670 }
2671 }
2672
2673 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2674 int err) {
2675 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2676 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2677 os::strerror(err), err);
2678 }
2679
2680 static void warn_fail_commit_memory(char* addr, size_t size,
2681 size_t alignment_hint, bool exec,
2682 int err) {
2683 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2684 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2685 alignment_hint, exec, os::strerror(err), err);
2686 }
2687
2688 // NOTE: Linux kernel does not really reserve the pages for us.
2689 // All it does is to check if there are enough free pages
2690 // left at the time of mmap(). This could be a potential
2691 // problem.
2692 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2693 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2694 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2695 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2696 if (res != (uintptr_t) MAP_FAILED) {
2697 if (UseNUMAInterleaving) {
2698 numa_make_global(addr, size);
2699 }
2700 return 0;
2701 }
2702
2703 int err = errno; // save errno from mmap() call above
2704
2705 if (!recoverable_mmap_error(err)) {
2706 warn_fail_commit_memory(addr, size, exec, err);
2707 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2708 }
2709
2710 return err;
2711 }
2712
2713 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2714 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2715 }
2716
2717 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2718 const char* mesg) {
2719 assert(mesg != NULL, "mesg must be specified");
2720 int err = os::Linux::commit_memory_impl(addr, size, exec);
2721 if (err != 0) {
2722 // the caller wants all commit errors to exit with the specified mesg:
2723 warn_fail_commit_memory(addr, size, exec, err);
2724 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2725 }
2726 }
2727
2728 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2729 #ifndef MAP_HUGETLB
2730 #define MAP_HUGETLB 0x40000
2731 #endif
2732
2733 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2734 #ifndef MADV_HUGEPAGE
2735 #define MADV_HUGEPAGE 14
2736 #endif
2737
2738 int os::Linux::commit_memory_impl(char* addr, size_t size,
2739 size_t alignment_hint, bool exec) {
2740 int err = os::Linux::commit_memory_impl(addr, size, exec);
2741 if (err == 0) {
2742 realign_memory(addr, size, alignment_hint);
2743 }
2744 return err;
2745 }
2746
2747 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2748 bool exec) {
2749 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2750 }
2751
2752 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2753 size_t alignment_hint, bool exec,
2754 const char* mesg) {
2755 assert(mesg != NULL, "mesg must be specified");
2756 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2757 if (err != 0) {
2758 // the caller wants all commit errors to exit with the specified mesg:
2759 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2760 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2761 }
2762 }
2763
2764 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2765 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2766 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2767 // be supported or the memory may already be backed by huge pages.
2768 ::madvise(addr, bytes, MADV_HUGEPAGE);
2769 }
2770 }
2771
2772 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2773 // This method works by doing an mmap over an existing mmaping and effectively discarding
2774 // the existing pages. However it won't work for SHM-based large pages that cannot be
2775 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2776 // small pages on top of the SHM segment. This method always works for small pages, so we
2777 // allow that in any case.
2778 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2779 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2780 }
2781 }
2782
2783 void os::numa_make_global(char *addr, size_t bytes) {
2784 Linux::numa_interleave_memory(addr, bytes);
2785 }
2786
2787 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2788 // bind policy to MPOL_PREFERRED for the current thread.
2789 #define USE_MPOL_PREFERRED 0
2790
2791 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2792 // To make NUMA and large pages more robust when both enabled, we need to ease
2793 // the requirements on where the memory should be allocated. MPOL_BIND is the
2794 // default policy and it will force memory to be allocated on the specified
2795 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2796 // the specified node, but will not force it. Using this policy will prevent
2797 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2798 // free large pages.
2799 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2800 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2801 }
2802
2803 bool os::numa_topology_changed() { return false; }
2804
2805 size_t os::numa_get_groups_num() {
2806 // Return just the number of nodes in which it's possible to allocate memory
2807 // (in numa terminology, configured nodes).
2808 return Linux::numa_num_configured_nodes();
2809 }
2810
2811 int os::numa_get_group_id() {
2812 int cpu_id = Linux::sched_getcpu();
2813 if (cpu_id != -1) {
2814 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2815 if (lgrp_id != -1) {
2816 return lgrp_id;
2817 }
2818 }
2819 return 0;
2820 }
2821
2822 int os::Linux::get_existing_num_nodes() {
2823 size_t node;
2824 size_t highest_node_number = Linux::numa_max_node();
2825 int num_nodes = 0;
2826
2827 // Get the total number of nodes in the system including nodes without memory.
2828 for (node = 0; node <= highest_node_number; node++) {
2829 if (isnode_in_existing_nodes(node)) {
2830 num_nodes++;
2831 }
2832 }
2833 return num_nodes;
2834 }
2835
2836 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2837 size_t highest_node_number = Linux::numa_max_node();
2838 size_t i = 0;
2839
2840 // Map all node ids in which is possible to allocate memory. Also nodes are
2841 // not always consecutively available, i.e. available from 0 to the highest
2842 // node number. If the nodes have been bound explicitly using numactl membind,
2843 // then allocate memory from those nodes only.
2844 for (size_t node = 0; node <= highest_node_number; node++) {
2845 if (Linux::isnode_in_bound_nodes(node)) {
2846 ids[i++] = node;
2847 }
2848 }
2849 return i;
2850 }
2851
2852 bool os::get_page_info(char *start, page_info* info) {
2853 return false;
2854 }
2855
2856 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2857 page_info* page_found) {
2858 return end;
2859 }
2860
2861
2862 int os::Linux::sched_getcpu_syscall(void) {
2863 unsigned int cpu = 0;
2864 int retval = -1;
2865
2866 #if defined(IA32)
2867 #ifndef SYS_getcpu
2868 #define SYS_getcpu 318
2869 #endif
2870 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2871 #elif defined(AMD64)
2872 // Unfortunately we have to bring all these macros here from vsyscall.h
2873 // to be able to compile on old linuxes.
2874 #define __NR_vgetcpu 2
2875 #define VSYSCALL_START (-10UL << 20)
2876 #define VSYSCALL_SIZE 1024
2877 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2878 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2879 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2880 retval = vgetcpu(&cpu, NULL, NULL);
2881 #endif
2882
2883 return (retval == -1) ? retval : cpu;
2884 }
2885
2886 void os::Linux::sched_getcpu_init() {
2887 // sched_getcpu() should be in libc.
2888 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2889 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2890
2891 // If it's not, try a direct syscall.
2892 if (sched_getcpu() == -1) {
2893 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2894 (void*)&sched_getcpu_syscall));
2895 }
2896
2897 if (sched_getcpu() == -1) {
2898 vm_exit_during_initialization("getcpu(2) system call not supported by kernel");
2899 }
2900 }
2901
2902 // Something to do with the numa-aware allocator needs these symbols
2903 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2904 extern "C" JNIEXPORT void numa_error(char *where) { }
2905
2906 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2907 // load symbol from base version instead.
2908 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2909 void *f = dlvsym(handle, name, "libnuma_1.1");
2910 if (f == NULL) {
2911 f = dlsym(handle, name);
2912 }
2913 return f;
2914 }
2915
2916 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2917 // Return NULL if the symbol is not defined in this particular version.
2918 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2919 return dlvsym(handle, name, "libnuma_1.2");
2920 }
2921
2922 bool os::Linux::libnuma_init() {
2923 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
2924 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2925 if (handle != NULL) {
2926 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2927 libnuma_dlsym(handle, "numa_node_to_cpus")));
2928 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2929 libnuma_dlsym(handle, "numa_max_node")));
2930 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2931 libnuma_dlsym(handle, "numa_num_configured_nodes")));
2932 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2933 libnuma_dlsym(handle, "numa_available")));
2934 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2935 libnuma_dlsym(handle, "numa_tonode_memory")));
2936 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2937 libnuma_dlsym(handle, "numa_interleave_memory")));
2938 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2939 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2940 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2941 libnuma_dlsym(handle, "numa_set_bind_policy")));
2942 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2943 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2944 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2945 libnuma_dlsym(handle, "numa_distance")));
2946 set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t,
2947 libnuma_v2_dlsym(handle, "numa_get_membind")));
2948
2949 if (numa_available() != -1) {
2950 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2951 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2952 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2953 // Create an index -> node mapping, since nodes are not always consecutive
2954 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2955 rebuild_nindex_to_node_map();
2956 // Create a cpu -> node mapping
2957 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2958 rebuild_cpu_to_node_map();
2959 return true;
2960 }
2961 }
2962 }
2963 return false;
2964 }
2965
2966 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2967 // Creating guard page is very expensive. Java thread has HotSpot
2968 // guard pages, only enable glibc guard page for non-Java threads.
2969 // (Remember: compiler thread is a Java thread, too!)
2970 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2971 }
2972
2973 void os::Linux::rebuild_nindex_to_node_map() {
2974 int highest_node_number = Linux::numa_max_node();
2975
2976 nindex_to_node()->clear();
2977 for (int node = 0; node <= highest_node_number; node++) {
2978 if (Linux::isnode_in_existing_nodes(node)) {
2979 nindex_to_node()->append(node);
2980 }
2981 }
2982 }
2983
2984 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2985 // The table is later used in get_node_by_cpu().
2986 void os::Linux::rebuild_cpu_to_node_map() {
2987 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2988 // in libnuma (possible values are starting from 16,
2989 // and continuing up with every other power of 2, but less
2990 // than the maximum number of CPUs supported by kernel), and
2991 // is a subject to change (in libnuma version 2 the requirements
2992 // are more reasonable) we'll just hardcode the number they use
2993 // in the library.
2994 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2995
2996 size_t cpu_num = processor_count();
2997 size_t cpu_map_size = NCPUS / BitsPerCLong;
2998 size_t cpu_map_valid_size =
2999 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
3000
3001 cpu_to_node()->clear();
3002 cpu_to_node()->at_grow(cpu_num - 1);
3003
3004 size_t node_num = get_existing_num_nodes();
3005
3006 int distance = 0;
3007 int closest_distance = INT_MAX;
3008 int closest_node = 0;
3009 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
3010 for (size_t i = 0; i < node_num; i++) {
3011 // Check if node is configured (not a memory-less node). If it is not, find
3012 // the closest configured node. Check also if node is bound, i.e. it's allowed
3013 // to allocate memory from the node. If it's not allowed, map cpus in that node
3014 // to the closest node from which memory allocation is allowed.
3015 if (!isnode_in_configured_nodes(nindex_to_node()->at(i)) ||
3016 !isnode_in_bound_nodes(nindex_to_node()->at(i))) {
3017 closest_distance = INT_MAX;
3018 // Check distance from all remaining nodes in the system. Ignore distance
3019 // from itself, from another non-configured node, and from another non-bound
3020 // node.
3021 for (size_t m = 0; m < node_num; m++) {
3022 if (m != i &&
3023 isnode_in_configured_nodes(nindex_to_node()->at(m)) &&
3024 isnode_in_bound_nodes(nindex_to_node()->at(m))) {
3025 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3026 // If a closest node is found, update. There is always at least one
3027 // configured and bound node in the system so there is always at least
3028 // one node close.
3029 if (distance != 0 && distance < closest_distance) {
3030 closest_distance = distance;
3031 closest_node = nindex_to_node()->at(m);
3032 }
3033 }
3034 }
3035 } else {
3036 // Current node is already a configured node.
3037 closest_node = nindex_to_node()->at(i);
3038 }
3039
3040 // Get cpus from the original node and map them to the closest node. If node
3041 // is a configured node (not a memory-less node), then original node and
3042 // closest node are the same.
3043 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3044 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3045 if (cpu_map[j] != 0) {
3046 for (size_t k = 0; k < BitsPerCLong; k++) {
3047 if (cpu_map[j] & (1UL << k)) {
3048 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3049 }
3050 }
3051 }
3052 }
3053 }
3054 }
3055 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3056 }
3057
3058 int os::Linux::get_node_by_cpu(int cpu_id) {
3059 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3060 return cpu_to_node()->at(cpu_id);
3061 }
3062 return -1;
3063 }
3064
3065 GrowableArray<int>* os::Linux::_cpu_to_node;
3066 GrowableArray<int>* os::Linux::_nindex_to_node;
3067 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3068 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3069 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3070 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3071 os::Linux::numa_available_func_t os::Linux::_numa_available;
3072 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3073 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3074 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3075 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3076 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3077 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3078 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind;
3079 unsigned long* os::Linux::_numa_all_nodes;
3080 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3081 struct bitmask* os::Linux::_numa_nodes_ptr;
3082
3083 bool os::pd_uncommit_memory(char* addr, size_t size) {
3084 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3085 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3086 return res != (uintptr_t) MAP_FAILED;
3087 }
3088
3089 static address get_stack_commited_bottom(address bottom, size_t size) {
3090 address nbot = bottom;
3091 address ntop = bottom + size;
3092
3093 size_t page_sz = os::vm_page_size();
3094 unsigned pages = size / page_sz;
3095
3096 unsigned char vec[1];
3097 unsigned imin = 1, imax = pages + 1, imid;
3098 int mincore_return_value = 0;
3099
3100 assert(imin <= imax, "Unexpected page size");
3101
3102 while (imin < imax) {
3103 imid = (imax + imin) / 2;
3104 nbot = ntop - (imid * page_sz);
3105
3106 // Use a trick with mincore to check whether the page is mapped or not.
3107 // mincore sets vec to 1 if page resides in memory and to 0 if page
3108 // is swapped output but if page we are asking for is unmapped
3109 // it returns -1,ENOMEM
3110 mincore_return_value = mincore(nbot, page_sz, vec);
3111
3112 if (mincore_return_value == -1) {
3113 // Page is not mapped go up
3114 // to find first mapped page
3115 if (errno != EAGAIN) {
3116 assert(errno == ENOMEM, "Unexpected mincore errno");
3117 imax = imid;
3118 }
3119 } else {
3120 // Page is mapped go down
3121 // to find first not mapped page
3122 imin = imid + 1;
3123 }
3124 }
3125
3126 nbot = nbot + page_sz;
3127
3128 // Adjust stack bottom one page up if last checked page is not mapped
3129 if (mincore_return_value == -1) {
3130 nbot = nbot + page_sz;
3131 }
3132
3133 return nbot;
3134 }
3135
3136 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3137 int mincore_return_value;
3138 const size_t stripe = 1024; // query this many pages each time
3139 unsigned char vec[stripe + 1];
3140 // set a guard
3141 vec[stripe] = 'X';
3142
3143 const size_t page_sz = os::vm_page_size();
3144 size_t pages = size / page_sz;
3145
3146 assert(is_aligned(start, page_sz), "Start address must be page aligned");
3147 assert(is_aligned(size, page_sz), "Size must be page aligned");
3148
3149 committed_start = NULL;
3150
3151 int loops = (pages + stripe - 1) / stripe;
3152 int committed_pages = 0;
3153 address loop_base = start;
3154 bool found_range = false;
3155
3156 for (int index = 0; index < loops && !found_range; index ++) {
3157 assert(pages > 0, "Nothing to do");
3158 int pages_to_query = (pages >= stripe) ? stripe : pages;
3159 pages -= pages_to_query;
3160
3161 // Get stable read
3162 while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3163
3164 // During shutdown, some memory goes away without properly notifying NMT,
3165 // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3166 // Bailout and return as not committed for now.
3167 if (mincore_return_value == -1 && errno == ENOMEM) {
3168 return false;
3169 }
3170
3171 assert(vec[stripe] == 'X', "overflow guard");
3172 assert(mincore_return_value == 0, "Range must be valid");
3173 // Process this stripe
3174 for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3175 if ((vec[vecIdx] & 0x01) == 0) { // not committed
3176 // End of current contiguous region
3177 if (committed_start != NULL) {
3178 found_range = true;
3179 break;
3180 }
3181 } else { // committed
3182 // Start of region
3183 if (committed_start == NULL) {
3184 committed_start = loop_base + page_sz * vecIdx;
3185 }
3186 committed_pages ++;
3187 }
3188 }
3189
3190 loop_base += pages_to_query * page_sz;
3191 }
3192
3193 if (committed_start != NULL) {
3194 assert(committed_pages > 0, "Must have committed region");
3195 assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3196 assert(committed_start >= start && committed_start < start + size, "Out of range");
3197 committed_size = page_sz * committed_pages;
3198 return true;
3199 } else {
3200 assert(committed_pages == 0, "Should not have committed region");
3201 return false;
3202 }
3203 }
3204
3205
3206 // Linux uses a growable mapping for the stack, and if the mapping for
3207 // the stack guard pages is not removed when we detach a thread the
3208 // stack cannot grow beyond the pages where the stack guard was
3209 // mapped. If at some point later in the process the stack expands to
3210 // that point, the Linux kernel cannot expand the stack any further
3211 // because the guard pages are in the way, and a segfault occurs.
3212 //
3213 // However, it's essential not to split the stack region by unmapping
3214 // a region (leaving a hole) that's already part of the stack mapping,
3215 // so if the stack mapping has already grown beyond the guard pages at
3216 // the time we create them, we have to truncate the stack mapping.
3217 // So, we need to know the extent of the stack mapping when
3218 // create_stack_guard_pages() is called.
3219
3220 // We only need this for stacks that are growable: at the time of
3221 // writing thread stacks don't use growable mappings (i.e. those
3222 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3223 // only applies to the main thread.
3224
3225 // If the (growable) stack mapping already extends beyond the point
3226 // where we're going to put our guard pages, truncate the mapping at
3227 // that point by munmap()ping it. This ensures that when we later
3228 // munmap() the guard pages we don't leave a hole in the stack
3229 // mapping. This only affects the main/primordial thread
3230
3231 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3232 if (os::is_primordial_thread()) {
3233 // As we manually grow stack up to bottom inside create_attached_thread(),
3234 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3235 // we don't need to do anything special.
3236 // Check it first, before calling heavy function.
3237 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3238 unsigned char vec[1];
3239
3240 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3241 // Fallback to slow path on all errors, including EAGAIN
3242 stack_extent = (uintptr_t) get_stack_commited_bottom(
3243 os::Linux::initial_thread_stack_bottom(),
3244 (size_t)addr - stack_extent);
3245 }
3246
3247 if (stack_extent < (uintptr_t)addr) {
3248 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3249 }
3250 }
3251
3252 return os::commit_memory(addr, size, !ExecMem);
3253 }
3254
3255 // If this is a growable mapping, remove the guard pages entirely by
3256 // munmap()ping them. If not, just call uncommit_memory(). This only
3257 // affects the main/primordial thread, but guard against future OS changes.
3258 // It's safe to always unmap guard pages for primordial thread because we
3259 // always place it right after end of the mapped region.
3260
3261 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3262 uintptr_t stack_extent, stack_base;
3263
3264 if (os::is_primordial_thread()) {
3265 return ::munmap(addr, size) == 0;
3266 }
3267
3268 return os::uncommit_memory(addr, size);
3269 }
3270
3271 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3272 // at 'requested_addr'. If there are existing memory mappings at the same
3273 // location, however, they will be overwritten. If 'fixed' is false,
3274 // 'requested_addr' is only treated as a hint, the return value may or
3275 // may not start from the requested address. Unlike Linux mmap(), this
3276 // function returns NULL to indicate failure.
3277 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3278 char * addr;
3279 int flags;
3280
3281 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3282 if (fixed) {
3283 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3284 flags |= MAP_FIXED;
3285 }
3286
3287 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3288 // touch an uncommitted page. Otherwise, the read/write might
3289 // succeed if we have enough swap space to back the physical page.
3290 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3291 flags, -1, 0);
3292
3293 return addr == MAP_FAILED ? NULL : addr;
3294 }
3295
3296 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3297 // (req_addr != NULL) or with a given alignment.
3298 // - bytes shall be a multiple of alignment.
3299 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3300 // - alignment sets the alignment at which memory shall be allocated.
3301 // It must be a multiple of allocation granularity.
3302 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3303 // req_addr or NULL.
3304 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3305
3306 size_t extra_size = bytes;
3307 if (req_addr == NULL && alignment > 0) {
3308 extra_size += alignment;
3309 }
3310
3311 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3312 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3313 -1, 0);
3314 if (start == MAP_FAILED) {
3315 start = NULL;
3316 } else {
3317 if (req_addr != NULL) {
3318 if (start != req_addr) {
3319 ::munmap(start, extra_size);
3320 start = NULL;
3321 }
3322 } else {
3323 char* const start_aligned = align_up(start, alignment);
3324 char* const end_aligned = start_aligned + bytes;
3325 char* const end = start + extra_size;
3326 if (start_aligned > start) {
3327 ::munmap(start, start_aligned - start);
3328 }
3329 if (end_aligned < end) {
3330 ::munmap(end_aligned, end - end_aligned);
3331 }
3332 start = start_aligned;
3333 }
3334 }
3335 return start;
3336 }
3337
3338 static int anon_munmap(char * addr, size_t size) {
3339 return ::munmap(addr, size) == 0;
3340 }
3341
3342 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3343 size_t alignment_hint) {
3344 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3345 }
3346
3347 bool os::pd_release_memory(char* addr, size_t size) {
3348 return anon_munmap(addr, size);
3349 }
3350
3351 static bool linux_mprotect(char* addr, size_t size, int prot) {
3352 // Linux wants the mprotect address argument to be page aligned.
3353 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3354
3355 // According to SUSv3, mprotect() should only be used with mappings
3356 // established by mmap(), and mmap() always maps whole pages. Unaligned
3357 // 'addr' likely indicates problem in the VM (e.g. trying to change
3358 // protection of malloc'ed or statically allocated memory). Check the
3359 // caller if you hit this assert.
3360 assert(addr == bottom, "sanity check");
3361
3362 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3363 return ::mprotect(bottom, size, prot) == 0;
3364 }
3365
3366 // Set protections specified
3367 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3368 bool is_committed) {
3369 unsigned int p = 0;
3370 switch (prot) {
3371 case MEM_PROT_NONE: p = PROT_NONE; break;
3372 case MEM_PROT_READ: p = PROT_READ; break;
3373 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3374 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3375 default:
3376 ShouldNotReachHere();
3377 }
3378 // is_committed is unused.
3379 return linux_mprotect(addr, bytes, p);
3380 }
3381
3382 bool os::guard_memory(char* addr, size_t size) {
3383 return linux_mprotect(addr, size, PROT_NONE);
3384 }
3385
3386 bool os::unguard_memory(char* addr, size_t size) {
3387 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3388 }
3389
3390 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3391 size_t page_size) {
3392 bool result = false;
3393 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3394 MAP_ANONYMOUS|MAP_PRIVATE,
3395 -1, 0);
3396 if (p != MAP_FAILED) {
3397 void *aligned_p = align_up(p, page_size);
3398
3399 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3400
3401 munmap(p, page_size * 2);
3402 }
3403
3404 if (warn && !result) {
3405 warning("TransparentHugePages is not supported by the operating system.");
3406 }
3407
3408 return result;
3409 }
3410
3411 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3412 bool result = false;
3413 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3414 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3415 -1, 0);
3416
3417 if (p != MAP_FAILED) {
3418 // We don't know if this really is a huge page or not.
3419 FILE *fp = fopen("/proc/self/maps", "r");
3420 if (fp) {
3421 while (!feof(fp)) {
3422 char chars[257];
3423 long x = 0;
3424 if (fgets(chars, sizeof(chars), fp)) {
3425 if (sscanf(chars, "%lx-%*x", &x) == 1
3426 && x == (long)p) {
3427 if (strstr (chars, "hugepage")) {
3428 result = true;
3429 break;
3430 }
3431 }
3432 }
3433 }
3434 fclose(fp);
3435 }
3436 munmap(p, page_size);
3437 }
3438
3439 if (warn && !result) {
3440 warning("HugeTLBFS is not supported by the operating system.");
3441 }
3442
3443 return result;
3444 }
3445
3446 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3447 //
3448 // From the coredump_filter documentation:
3449 //
3450 // - (bit 0) anonymous private memory
3451 // - (bit 1) anonymous shared memory
3452 // - (bit 2) file-backed private memory
3453 // - (bit 3) file-backed shared memory
3454 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3455 // effective only if the bit 2 is cleared)
3456 // - (bit 5) hugetlb private memory
3457 // - (bit 6) hugetlb shared memory
3458 // - (bit 7) dax private memory
3459 // - (bit 8) dax shared memory
3460 //
3461 static void set_coredump_filter(bool largepages, bool dax_shared) {
3462 FILE *f;
3463 long cdm;
3464 bool filter_changed = false;
3465
3466 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3467 return;
3468 }
3469
3470 if (fscanf(f, "%lx", &cdm) != 1) {
3471 fclose(f);
3472 return;
3473 }
3474
3475 rewind(f);
3476
3477 if (largepages && (cdm & LARGEPAGES_BIT) == 0) {
3478 cdm |= LARGEPAGES_BIT;
3479 filter_changed = true;
3480 }
3481 if (dax_shared && (cdm & DAX_SHARED_BIT) == 0) {
3482 cdm |= DAX_SHARED_BIT;
3483 filter_changed = true;
3484 }
3485 if (filter_changed) {
3486 fprintf(f, "%#lx", cdm);
3487 }
3488
3489 fclose(f);
3490 }
3491
3492 // Large page support
3493
3494 static size_t _large_page_size = 0;
3495
3496 size_t os::Linux::find_large_page_size() {
3497 size_t large_page_size = 0;
3498
3499 // large_page_size on Linux is used to round up heap size. x86 uses either
3500 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3501 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3502 // page as large as 256M.
3503 //
3504 // Here we try to figure out page size by parsing /proc/meminfo and looking
3505 // for a line with the following format:
3506 // Hugepagesize: 2048 kB
3507 //
3508 // If we can't determine the value (e.g. /proc is not mounted, or the text
3509 // format has been changed), we'll use the largest page size supported by
3510 // the processor.
3511
3512 #ifndef ZERO
3513 large_page_size =
3514 AARCH64_ONLY(2 * M)
3515 AMD64_ONLY(2 * M)
3516 ARM32_ONLY(2 * M)
3517 IA32_ONLY(4 * M)
3518 IA64_ONLY(256 * M)
3519 PPC_ONLY(4 * M)
3520 S390_ONLY(1 * M)
3521 SPARC_ONLY(4 * M);
3522 #endif // ZERO
3523
3524 FILE *fp = fopen("/proc/meminfo", "r");
3525 if (fp) {
3526 while (!feof(fp)) {
3527 int x = 0;
3528 char buf[16];
3529 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3530 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3531 large_page_size = x * K;
3532 break;
3533 }
3534 } else {
3535 // skip to next line
3536 for (;;) {
3537 int ch = fgetc(fp);
3538 if (ch == EOF || ch == (int)'\n') break;
3539 }
3540 }
3541 }
3542 fclose(fp);
3543 }
3544
3545 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3546 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3547 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3548 proper_unit_for_byte_size(large_page_size));
3549 }
3550
3551 return large_page_size;
3552 }
3553
3554 size_t os::Linux::setup_large_page_size() {
3555 _large_page_size = Linux::find_large_page_size();
3556 const size_t default_page_size = (size_t)Linux::page_size();
3557 if (_large_page_size > default_page_size) {
3558 _page_sizes[0] = _large_page_size;
3559 _page_sizes[1] = default_page_size;
3560 _page_sizes[2] = 0;
3561 }
3562
3563 return _large_page_size;
3564 }
3565
3566 bool os::Linux::setup_large_page_type(size_t page_size) {
3567 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3568 FLAG_IS_DEFAULT(UseSHM) &&
3569 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3570
3571 // The type of large pages has not been specified by the user.
3572
3573 // Try UseHugeTLBFS and then UseSHM.
3574 UseHugeTLBFS = UseSHM = true;
3575
3576 // Don't try UseTransparentHugePages since there are known
3577 // performance issues with it turned on. This might change in the future.
3578 UseTransparentHugePages = false;
3579 }
3580
3581 if (UseTransparentHugePages) {
3582 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3583 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3584 UseHugeTLBFS = false;
3585 UseSHM = false;
3586 return true;
3587 }
3588 UseTransparentHugePages = false;
3589 }
3590
3591 if (UseHugeTLBFS) {
3592 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3593 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3594 UseSHM = false;
3595 return true;
3596 }
3597 UseHugeTLBFS = false;
3598 }
3599
3600 return UseSHM;
3601 }
3602
3603 void os::large_page_init() {
3604 if (!UseLargePages &&
3605 !UseTransparentHugePages &&
3606 !UseHugeTLBFS &&
3607 !UseSHM) {
3608 // Not using large pages.
3609 return;
3610 }
3611
3612 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3613 // The user explicitly turned off large pages.
3614 // Ignore the rest of the large pages flags.
3615 UseTransparentHugePages = false;
3616 UseHugeTLBFS = false;
3617 UseSHM = false;
3618 return;
3619 }
3620
3621 size_t large_page_size = Linux::setup_large_page_size();
3622 UseLargePages = Linux::setup_large_page_type(large_page_size);
3623
3624 set_coredump_filter(true /*largepages*/, false /*dax_shared*/);
3625 }
3626
3627 #ifndef SHM_HUGETLB
3628 #define SHM_HUGETLB 04000
3629 #endif
3630
3631 #define shm_warning_format(format, ...) \
3632 do { \
3633 if (UseLargePages && \
3634 (!FLAG_IS_DEFAULT(UseLargePages) || \
3635 !FLAG_IS_DEFAULT(UseSHM) || \
3636 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3637 warning(format, __VA_ARGS__); \
3638 } \
3639 } while (0)
3640
3641 #define shm_warning(str) shm_warning_format("%s", str)
3642
3643 #define shm_warning_with_errno(str) \
3644 do { \
3645 int err = errno; \
3646 shm_warning_format(str " (error = %d)", err); \
3647 } while (0)
3648
3649 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3650 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3651
3652 if (!is_aligned(alignment, SHMLBA)) {
3653 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3654 return NULL;
3655 }
3656
3657 // To ensure that we get 'alignment' aligned memory from shmat,
3658 // we pre-reserve aligned virtual memory and then attach to that.
3659
3660 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3661 if (pre_reserved_addr == NULL) {
3662 // Couldn't pre-reserve aligned memory.
3663 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3664 return NULL;
3665 }
3666
3667 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3668 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3669
3670 if ((intptr_t)addr == -1) {
3671 int err = errno;
3672 shm_warning_with_errno("Failed to attach shared memory.");
3673
3674 assert(err != EACCES, "Unexpected error");
3675 assert(err != EIDRM, "Unexpected error");
3676 assert(err != EINVAL, "Unexpected error");
3677
3678 // Since we don't know if the kernel unmapped the pre-reserved memory area
3679 // we can't unmap it, since that would potentially unmap memory that was
3680 // mapped from other threads.
3681 return NULL;
3682 }
3683
3684 return addr;
3685 }
3686
3687 static char* shmat_at_address(int shmid, char* req_addr) {
3688 if (!is_aligned(req_addr, SHMLBA)) {
3689 assert(false, "Requested address needs to be SHMLBA aligned");
3690 return NULL;
3691 }
3692
3693 char* addr = (char*)shmat(shmid, req_addr, 0);
3694
3695 if ((intptr_t)addr == -1) {
3696 shm_warning_with_errno("Failed to attach shared memory.");
3697 return NULL;
3698 }
3699
3700 return addr;
3701 }
3702
3703 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3704 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3705 if (req_addr != NULL) {
3706 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3707 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3708 return shmat_at_address(shmid, req_addr);
3709 }
3710
3711 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3712 // return large page size aligned memory addresses when req_addr == NULL.
3713 // However, if the alignment is larger than the large page size, we have
3714 // to manually ensure that the memory returned is 'alignment' aligned.
3715 if (alignment > os::large_page_size()) {
3716 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3717 return shmat_with_alignment(shmid, bytes, alignment);
3718 } else {
3719 return shmat_at_address(shmid, NULL);
3720 }
3721 }
3722
3723 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3724 char* req_addr, bool exec) {
3725 // "exec" is passed in but not used. Creating the shared image for
3726 // the code cache doesn't have an SHM_X executable permission to check.
3727 assert(UseLargePages && UseSHM, "only for SHM large pages");
3728 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3729 assert(is_aligned(req_addr, alignment), "Unaligned address");
3730
3731 if (!is_aligned(bytes, os::large_page_size())) {
3732 return NULL; // Fallback to small pages.
3733 }
3734
3735 // Create a large shared memory region to attach to based on size.
3736 // Currently, size is the total size of the heap.
3737 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3738 if (shmid == -1) {
3739 // Possible reasons for shmget failure:
3740 // 1. shmmax is too small for Java heap.
3741 // > check shmmax value: cat /proc/sys/kernel/shmmax
3742 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3743 // 2. not enough large page memory.
3744 // > check available large pages: cat /proc/meminfo
3745 // > increase amount of large pages:
3746 // echo new_value > /proc/sys/vm/nr_hugepages
3747 // Note 1: different Linux may use different name for this property,
3748 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3749 // Note 2: it's possible there's enough physical memory available but
3750 // they are so fragmented after a long run that they can't
3751 // coalesce into large pages. Try to reserve large pages when
3752 // the system is still "fresh".
3753 shm_warning_with_errno("Failed to reserve shared memory.");
3754 return NULL;
3755 }
3756
3757 // Attach to the region.
3758 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3759
3760 // Remove shmid. If shmat() is successful, the actual shared memory segment
3761 // will be deleted when it's detached by shmdt() or when the process
3762 // terminates. If shmat() is not successful this will remove the shared
3763 // segment immediately.
3764 shmctl(shmid, IPC_RMID, NULL);
3765
3766 return addr;
3767 }
3768
3769 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3770 int error) {
3771 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3772
3773 bool warn_on_failure = UseLargePages &&
3774 (!FLAG_IS_DEFAULT(UseLargePages) ||
3775 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3776 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3777
3778 if (warn_on_failure) {
3779 char msg[128];
3780 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3781 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3782 warning("%s", msg);
3783 }
3784 }
3785
3786 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3787 char* req_addr,
3788 bool exec) {
3789 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3790 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3791 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3792
3793 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3794 char* addr = (char*)::mmap(req_addr, bytes, prot,
3795 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3796 -1, 0);
3797
3798 if (addr == MAP_FAILED) {
3799 warn_on_large_pages_failure(req_addr, bytes, errno);
3800 return NULL;
3801 }
3802
3803 assert(is_aligned(addr, os::large_page_size()), "Must be");
3804
3805 return addr;
3806 }
3807
3808 // Reserve memory using mmap(MAP_HUGETLB).
3809 // - bytes shall be a multiple of alignment.
3810 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3811 // - alignment sets the alignment at which memory shall be allocated.
3812 // It must be a multiple of allocation granularity.
3813 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3814 // req_addr or NULL.
3815 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3816 size_t alignment,
3817 char* req_addr,
3818 bool exec) {
3819 size_t large_page_size = os::large_page_size();
3820 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3821
3822 assert(is_aligned(req_addr, alignment), "Must be");
3823 assert(is_aligned(bytes, alignment), "Must be");
3824
3825 // First reserve - but not commit - the address range in small pages.
3826 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3827
3828 if (start == NULL) {
3829 return NULL;
3830 }
3831
3832 assert(is_aligned(start, alignment), "Must be");
3833
3834 char* end = start + bytes;
3835
3836 // Find the regions of the allocated chunk that can be promoted to large pages.
3837 char* lp_start = align_up(start, large_page_size);
3838 char* lp_end = align_down(end, large_page_size);
3839
3840 size_t lp_bytes = lp_end - lp_start;
3841
3842 assert(is_aligned(lp_bytes, large_page_size), "Must be");
3843
3844 if (lp_bytes == 0) {
3845 // The mapped region doesn't even span the start and the end of a large page.
3846 // Fall back to allocate a non-special area.
3847 ::munmap(start, end - start);
3848 return NULL;
3849 }
3850
3851 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3852
3853 void* result;
3854
3855 // Commit small-paged leading area.
3856 if (start != lp_start) {
3857 result = ::mmap(start, lp_start - start, prot,
3858 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3859 -1, 0);
3860 if (result == MAP_FAILED) {
3861 ::munmap(lp_start, end - lp_start);
3862 return NULL;
3863 }
3864 }
3865
3866 // Commit large-paged area.
3867 result = ::mmap(lp_start, lp_bytes, prot,
3868 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3869 -1, 0);
3870 if (result == MAP_FAILED) {
3871 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3872 // If the mmap above fails, the large pages region will be unmapped and we
3873 // have regions before and after with small pages. Release these regions.
3874 //
3875 // | mapped | unmapped | mapped |
3876 // ^ ^ ^ ^
3877 // start lp_start lp_end end
3878 //
3879 ::munmap(start, lp_start - start);
3880 ::munmap(lp_end, end - lp_end);
3881 return NULL;
3882 }
3883
3884 // Commit small-paged trailing area.
3885 if (lp_end != end) {
3886 result = ::mmap(lp_end, end - lp_end, prot,
3887 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3888 -1, 0);
3889 if (result == MAP_FAILED) {
3890 ::munmap(start, lp_end - start);
3891 return NULL;
3892 }
3893 }
3894
3895 return start;
3896 }
3897
3898 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3899 size_t alignment,
3900 char* req_addr,
3901 bool exec) {
3902 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3903 assert(is_aligned(req_addr, alignment), "Must be");
3904 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3905 assert(is_power_of_2(os::large_page_size()), "Must be");
3906 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3907
3908 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3909 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3910 } else {
3911 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3912 }
3913 }
3914
3915 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3916 char* req_addr, bool exec) {
3917 assert(UseLargePages, "only for large pages");
3918
3919 char* addr;
3920 if (UseSHM) {
3921 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3922 } else {
3923 assert(UseHugeTLBFS, "must be");
3924 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3925 }
3926
3927 if (addr != NULL) {
3928 if (UseNUMAInterleaving) {
3929 numa_make_global(addr, bytes);
3930 }
3931
3932 // The memory is committed
3933 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3934 }
3935
3936 return addr;
3937 }
3938
3939 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3940 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3941 return shmdt(base) == 0;
3942 }
3943
3944 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3945 return pd_release_memory(base, bytes);
3946 }
3947
3948 bool os::release_memory_special(char* base, size_t bytes) {
3949 bool res;
3950 if (MemTracker::tracking_level() > NMT_minimal) {
3951 Tracker tkr(Tracker::release);
3952 res = os::Linux::release_memory_special_impl(base, bytes);
3953 if (res) {
3954 tkr.record((address)base, bytes);
3955 }
3956
3957 } else {
3958 res = os::Linux::release_memory_special_impl(base, bytes);
3959 }
3960 return res;
3961 }
3962
3963 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3964 assert(UseLargePages, "only for large pages");
3965 bool res;
3966
3967 if (UseSHM) {
3968 res = os::Linux::release_memory_special_shm(base, bytes);
3969 } else {
3970 assert(UseHugeTLBFS, "must be");
3971 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3972 }
3973 return res;
3974 }
3975
3976 size_t os::large_page_size() {
3977 return _large_page_size;
3978 }
3979
3980 // With SysV SHM the entire memory region must be allocated as shared
3981 // memory.
3982 // HugeTLBFS allows application to commit large page memory on demand.
3983 // However, when committing memory with HugeTLBFS fails, the region
3984 // that was supposed to be committed will lose the old reservation
3985 // and allow other threads to steal that memory region. Because of this
3986 // behavior we can't commit HugeTLBFS memory.
3987 bool os::can_commit_large_page_memory() {
3988 return UseTransparentHugePages;
3989 }
3990
3991 bool os::can_execute_large_page_memory() {
3992 return UseTransparentHugePages || UseHugeTLBFS;
3993 }
3994
3995 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
3996 assert(file_desc >= 0, "file_desc is not valid");
3997 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
3998 if (result != NULL) {
3999 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
4000 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
4001 }
4002 }
4003 return result;
4004 }
4005
4006 // Reserve memory at an arbitrary address, only if that area is
4007 // available (and not reserved for something else).
4008
4009 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
4010 const int max_tries = 10;
4011 char* base[max_tries];
4012 size_t size[max_tries];
4013 const size_t gap = 0x000000;
4014
4015 // Assert only that the size is a multiple of the page size, since
4016 // that's all that mmap requires, and since that's all we really know
4017 // about at this low abstraction level. If we need higher alignment,
4018 // we can either pass an alignment to this method or verify alignment
4019 // in one of the methods further up the call chain. See bug 5044738.
4020 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
4021
4022 // Repeatedly allocate blocks until the block is allocated at the
4023 // right spot.
4024
4025 // Linux mmap allows caller to pass an address as hint; give it a try first,
4026 // if kernel honors the hint then we can return immediately.
4027 char * addr = anon_mmap(requested_addr, bytes, false);
4028 if (addr == requested_addr) {
4029 return requested_addr;
4030 }
4031
4032 if (addr != NULL) {
4033 // mmap() is successful but it fails to reserve at the requested address
4034 anon_munmap(addr, bytes);
4035 }
4036
4037 int i;
4038 for (i = 0; i < max_tries; ++i) {
4039 base[i] = reserve_memory(bytes);
4040
4041 if (base[i] != NULL) {
4042 // Is this the block we wanted?
4043 if (base[i] == requested_addr) {
4044 size[i] = bytes;
4045 break;
4046 }
4047
4048 // Does this overlap the block we wanted? Give back the overlapped
4049 // parts and try again.
4050
4051 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
4052 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
4053 unmap_memory(base[i], top_overlap);
4054 base[i] += top_overlap;
4055 size[i] = bytes - top_overlap;
4056 } else {
4057 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
4058 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
4059 unmap_memory(requested_addr, bottom_overlap);
4060 size[i] = bytes - bottom_overlap;
4061 } else {
4062 size[i] = bytes;
4063 }
4064 }
4065 }
4066 }
4067
4068 // Give back the unused reserved pieces.
4069
4070 for (int j = 0; j < i; ++j) {
4071 if (base[j] != NULL) {
4072 unmap_memory(base[j], size[j]);
4073 }
4074 }
4075
4076 if (i < max_tries) {
4077 return requested_addr;
4078 } else {
4079 return NULL;
4080 }
4081 }
4082
4083 size_t os::read(int fd, void *buf, unsigned int nBytes) {
4084 return ::read(fd, buf, nBytes);
4085 }
4086
4087 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
4088 return ::pread(fd, buf, nBytes, offset);
4089 }
4090
4091 // Short sleep, direct OS call.
4092 //
4093 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
4094 // sched_yield(2) will actually give up the CPU:
4095 //
4096 // * Alone on this pariticular CPU, keeps running.
4097 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
4098 // (pre 2.6.39).
4099 //
4100 // So calling this with 0 is an alternative.
4101 //
4102 void os::naked_short_sleep(jlong ms) {
4103 struct timespec req;
4104
4105 assert(ms < 1000, "Un-interruptable sleep, short time use only");
4106 req.tv_sec = 0;
4107 if (ms > 0) {
4108 req.tv_nsec = (ms % 1000) * 1000000;
4109 } else {
4110 req.tv_nsec = 1;
4111 }
4112
4113 nanosleep(&req, NULL);
4114
4115 return;
4116 }
4117
4118 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4119 void os::infinite_sleep() {
4120 while (true) { // sleep forever ...
4121 ::sleep(100); // ... 100 seconds at a time
4122 }
4123 }
4124
4125 // Used to convert frequent JVM_Yield() to nops
4126 bool os::dont_yield() {
4127 return DontYieldALot;
4128 }
4129
4130 void os::naked_yield() {
4131 sched_yield();
4132 }
4133
4134 ////////////////////////////////////////////////////////////////////////////////
4135 // thread priority support
4136
4137 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4138 // only supports dynamic priority, static priority must be zero. For real-time
4139 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4140 // However, for large multi-threaded applications, SCHED_RR is not only slower
4141 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4142 // of 5 runs - Sep 2005).
4143 //
4144 // The following code actually changes the niceness of kernel-thread/LWP. It
4145 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4146 // not the entire user process, and user level threads are 1:1 mapped to kernel
4147 // threads. It has always been the case, but could change in the future. For
4148 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4149 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4150
4151 int os::java_to_os_priority[CriticalPriority + 1] = {
4152 19, // 0 Entry should never be used
4153
4154 4, // 1 MinPriority
4155 3, // 2
4156 2, // 3
4157
4158 1, // 4
4159 0, // 5 NormPriority
4160 -1, // 6
4161
4162 -2, // 7
4163 -3, // 8
4164 -4, // 9 NearMaxPriority
4165
4166 -5, // 10 MaxPriority
4167
4168 -5 // 11 CriticalPriority
4169 };
4170
4171 static int prio_init() {
4172 if (ThreadPriorityPolicy == 1) {
4173 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4174 // if effective uid is not root. Perhaps, a more elegant way of doing
4175 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4176 if (geteuid() != 0) {
4177 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4178 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4179 }
4180 ThreadPriorityPolicy = 0;
4181 }
4182 }
4183 if (UseCriticalJavaThreadPriority) {
4184 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4185 }
4186 return 0;
4187 }
4188
4189 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4190 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4191
4192 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4193 return (ret == 0) ? OS_OK : OS_ERR;
4194 }
4195
4196 OSReturn os::get_native_priority(const Thread* const thread,
4197 int *priority_ptr) {
4198 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4199 *priority_ptr = java_to_os_priority[NormPriority];
4200 return OS_OK;
4201 }
4202
4203 errno = 0;
4204 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4205 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4206 }
4207
4208 // Hint to the underlying OS that a task switch would not be good.
4209 // Void return because it's a hint and can fail.
4210 void os::hint_no_preempt() {}
4211
4212 ////////////////////////////////////////////////////////////////////////////////
4213 // suspend/resume support
4214
4215 // The low-level signal-based suspend/resume support is a remnant from the
4216 // old VM-suspension that used to be for java-suspension, safepoints etc,
4217 // within hotspot. Currently used by JFR's OSThreadSampler
4218 //
4219 // The remaining code is greatly simplified from the more general suspension
4220 // code that used to be used.
4221 //
4222 // The protocol is quite simple:
4223 // - suspend:
4224 // - sends a signal to the target thread
4225 // - polls the suspend state of the osthread using a yield loop
4226 // - target thread signal handler (SR_handler) sets suspend state
4227 // and blocks in sigsuspend until continued
4228 // - resume:
4229 // - sets target osthread state to continue
4230 // - sends signal to end the sigsuspend loop in the SR_handler
4231 //
4232 // Note that the SR_lock plays no role in this suspend/resume protocol,
4233 // but is checked for NULL in SR_handler as a thread termination indicator.
4234 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4235 //
4236 // Note that resume_clear_context() and suspend_save_context() are needed
4237 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4238 // which in part is used by:
4239 // - Forte Analyzer: AsyncGetCallTrace()
4240 // - StackBanging: get_frame_at_stack_banging_point()
4241
4242 static void resume_clear_context(OSThread *osthread) {
4243 osthread->set_ucontext(NULL);
4244 osthread->set_siginfo(NULL);
4245 }
4246
4247 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4248 ucontext_t* context) {
4249 osthread->set_ucontext(context);
4250 osthread->set_siginfo(siginfo);
4251 }
4252
4253 // Handler function invoked when a thread's execution is suspended or
4254 // resumed. We have to be careful that only async-safe functions are
4255 // called here (Note: most pthread functions are not async safe and
4256 // should be avoided.)
4257 //
4258 // Note: sigwait() is a more natural fit than sigsuspend() from an
4259 // interface point of view, but sigwait() prevents the signal hander
4260 // from being run. libpthread would get very confused by not having
4261 // its signal handlers run and prevents sigwait()'s use with the
4262 // mutex granting granting signal.
4263 //
4264 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4265 //
4266 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4267 // Save and restore errno to avoid confusing native code with EINTR
4268 // after sigsuspend.
4269 int old_errno = errno;
4270
4271 Thread* thread = Thread::current_or_null_safe();
4272 assert(thread != NULL, "Missing current thread in SR_handler");
4273
4274 // On some systems we have seen signal delivery get "stuck" until the signal
4275 // mask is changed as part of thread termination. Check that the current thread
4276 // has not already terminated (via SR_lock()) - else the following assertion
4277 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4278 // destructor has completed.
4279
4280 if (thread->SR_lock() == NULL) {
4281 return;
4282 }
4283
4284 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4285
4286 OSThread* osthread = thread->osthread();
4287
4288 os::SuspendResume::State current = osthread->sr.state();
4289 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4290 suspend_save_context(osthread, siginfo, context);
4291
4292 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4293 os::SuspendResume::State state = osthread->sr.suspended();
4294 if (state == os::SuspendResume::SR_SUSPENDED) {
4295 sigset_t suspend_set; // signals for sigsuspend()
4296 sigemptyset(&suspend_set);
4297 // get current set of blocked signals and unblock resume signal
4298 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4299 sigdelset(&suspend_set, SR_signum);
4300
4301 sr_semaphore.signal();
4302 // wait here until we are resumed
4303 while (1) {
4304 sigsuspend(&suspend_set);
4305
4306 os::SuspendResume::State result = osthread->sr.running();
4307 if (result == os::SuspendResume::SR_RUNNING) {
4308 sr_semaphore.signal();
4309 break;
4310 }
4311 }
4312
4313 } else if (state == os::SuspendResume::SR_RUNNING) {
4314 // request was cancelled, continue
4315 } else {
4316 ShouldNotReachHere();
4317 }
4318
4319 resume_clear_context(osthread);
4320 } else if (current == os::SuspendResume::SR_RUNNING) {
4321 // request was cancelled, continue
4322 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4323 // ignore
4324 } else {
4325 // ignore
4326 }
4327
4328 errno = old_errno;
4329 }
4330
4331 static int SR_initialize() {
4332 struct sigaction act;
4333 char *s;
4334
4335 // Get signal number to use for suspend/resume
4336 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4337 int sig = ::strtol(s, 0, 10);
4338 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4339 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4340 SR_signum = sig;
4341 } else {
4342 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4343 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4344 }
4345 }
4346
4347 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4348 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4349
4350 sigemptyset(&SR_sigset);
4351 sigaddset(&SR_sigset, SR_signum);
4352
4353 // Set up signal handler for suspend/resume
4354 act.sa_flags = SA_RESTART|SA_SIGINFO;
4355 act.sa_handler = (void (*)(int)) SR_handler;
4356
4357 // SR_signum is blocked by default.
4358 // 4528190 - We also need to block pthread restart signal (32 on all
4359 // supported Linux platforms). Note that LinuxThreads need to block
4360 // this signal for all threads to work properly. So we don't have
4361 // to use hard-coded signal number when setting up the mask.
4362 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4363
4364 if (sigaction(SR_signum, &act, 0) == -1) {
4365 return -1;
4366 }
4367
4368 // Save signal flag
4369 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4370 return 0;
4371 }
4372
4373 static int sr_notify(OSThread* osthread) {
4374 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4375 assert_status(status == 0, status, "pthread_kill");
4376 return status;
4377 }
4378
4379 // "Randomly" selected value for how long we want to spin
4380 // before bailing out on suspending a thread, also how often
4381 // we send a signal to a thread we want to resume
4382 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4383 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4384
4385 // returns true on success and false on error - really an error is fatal
4386 // but this seems the normal response to library errors
4387 static bool do_suspend(OSThread* osthread) {
4388 assert(osthread->sr.is_running(), "thread should be running");
4389 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4390
4391 // mark as suspended and send signal
4392 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4393 // failed to switch, state wasn't running?
4394 ShouldNotReachHere();
4395 return false;
4396 }
4397
4398 if (sr_notify(osthread) != 0) {
4399 ShouldNotReachHere();
4400 }
4401
4402 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4403 while (true) {
4404 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4405 break;
4406 } else {
4407 // timeout
4408 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4409 if (cancelled == os::SuspendResume::SR_RUNNING) {
4410 return false;
4411 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4412 // make sure that we consume the signal on the semaphore as well
4413 sr_semaphore.wait();
4414 break;
4415 } else {
4416 ShouldNotReachHere();
4417 return false;
4418 }
4419 }
4420 }
4421
4422 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4423 return true;
4424 }
4425
4426 static void do_resume(OSThread* osthread) {
4427 assert(osthread->sr.is_suspended(), "thread should be suspended");
4428 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4429
4430 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4431 // failed to switch to WAKEUP_REQUEST
4432 ShouldNotReachHere();
4433 return;
4434 }
4435
4436 while (true) {
4437 if (sr_notify(osthread) == 0) {
4438 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4439 if (osthread->sr.is_running()) {
4440 return;
4441 }
4442 }
4443 } else {
4444 ShouldNotReachHere();
4445 }
4446 }
4447
4448 guarantee(osthread->sr.is_running(), "Must be running!");
4449 }
4450
4451 ///////////////////////////////////////////////////////////////////////////////////
4452 // signal handling (except suspend/resume)
4453
4454 // This routine may be used by user applications as a "hook" to catch signals.
4455 // The user-defined signal handler must pass unrecognized signals to this
4456 // routine, and if it returns true (non-zero), then the signal handler must
4457 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4458 // routine will never retun false (zero), but instead will execute a VM panic
4459 // routine kill the process.
4460 //
4461 // If this routine returns false, it is OK to call it again. This allows
4462 // the user-defined signal handler to perform checks either before or after
4463 // the VM performs its own checks. Naturally, the user code would be making
4464 // a serious error if it tried to handle an exception (such as a null check
4465 // or breakpoint) that the VM was generating for its own correct operation.
4466 //
4467 // This routine may recognize any of the following kinds of signals:
4468 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4469 // It should be consulted by handlers for any of those signals.
4470 //
4471 // The caller of this routine must pass in the three arguments supplied
4472 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4473 // field of the structure passed to sigaction(). This routine assumes that
4474 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4475 //
4476 // Note that the VM will print warnings if it detects conflicting signal
4477 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4478 //
4479 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4480 siginfo_t* siginfo,
4481 void* ucontext,
4482 int abort_if_unrecognized);
4483
4484 static void signalHandler(int sig, siginfo_t* info, void* uc) {
4485 assert(info != NULL && uc != NULL, "it must be old kernel");
4486 int orig_errno = errno; // Preserve errno value over signal handler.
4487 JVM_handle_linux_signal(sig, info, uc, true);
4488 errno = orig_errno;
4489 }
4490
4491
4492 // This boolean allows users to forward their own non-matching signals
4493 // to JVM_handle_linux_signal, harmlessly.
4494 bool os::Linux::signal_handlers_are_installed = false;
4495
4496 // For signal-chaining
4497 struct sigaction sigact[NSIG];
4498 uint64_t sigs = 0;
4499 #if (64 < NSIG-1)
4500 #error "Not all signals can be encoded in sigs. Adapt its type!"
4501 #endif
4502 bool os::Linux::libjsig_is_loaded = false;
4503 typedef struct sigaction *(*get_signal_t)(int);
4504 get_signal_t os::Linux::get_signal_action = NULL;
4505
4506 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4507 struct sigaction *actp = NULL;
4508
4509 if (libjsig_is_loaded) {
4510 // Retrieve the old signal handler from libjsig
4511 actp = (*get_signal_action)(sig);
4512 }
4513 if (actp == NULL) {
4514 // Retrieve the preinstalled signal handler from jvm
4515 actp = get_preinstalled_handler(sig);
4516 }
4517
4518 return actp;
4519 }
4520
4521 static bool call_chained_handler(struct sigaction *actp, int sig,
4522 siginfo_t *siginfo, void *context) {
4523 // Call the old signal handler
4524 if (actp->sa_handler == SIG_DFL) {
4525 // It's more reasonable to let jvm treat it as an unexpected exception
4526 // instead of taking the default action.
4527 return false;
4528 } else if (actp->sa_handler != SIG_IGN) {
4529 if ((actp->sa_flags & SA_NODEFER) == 0) {
4530 // automaticlly block the signal
4531 sigaddset(&(actp->sa_mask), sig);
4532 }
4533
4534 sa_handler_t hand = NULL;
4535 sa_sigaction_t sa = NULL;
4536 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4537 // retrieve the chained handler
4538 if (siginfo_flag_set) {
4539 sa = actp->sa_sigaction;
4540 } else {
4541 hand = actp->sa_handler;
4542 }
4543
4544 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4545 actp->sa_handler = SIG_DFL;
4546 }
4547
4548 // try to honor the signal mask
4549 sigset_t oset;
4550 sigemptyset(&oset);
4551 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4552
4553 // call into the chained handler
4554 if (siginfo_flag_set) {
4555 (*sa)(sig, siginfo, context);
4556 } else {
4557 (*hand)(sig);
4558 }
4559
4560 // restore the signal mask
4561 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4562 }
4563 // Tell jvm's signal handler the signal is taken care of.
4564 return true;
4565 }
4566
4567 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4568 bool chained = false;
4569 // signal-chaining
4570 if (UseSignalChaining) {
4571 struct sigaction *actp = get_chained_signal_action(sig);
4572 if (actp != NULL) {
4573 chained = call_chained_handler(actp, sig, siginfo, context);
4574 }
4575 }
4576 return chained;
4577 }
4578
4579 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4580 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4581 return &sigact[sig];
4582 }
4583 return NULL;
4584 }
4585
4586 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4587 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4588 sigact[sig] = oldAct;
4589 sigs |= (uint64_t)1 << (sig-1);
4590 }
4591
4592 // for diagnostic
4593 int sigflags[NSIG];
4594
4595 int os::Linux::get_our_sigflags(int sig) {
4596 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4597 return sigflags[sig];
4598 }
4599
4600 void os::Linux::set_our_sigflags(int sig, int flags) {
4601 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4602 if (sig > 0 && sig < NSIG) {
4603 sigflags[sig] = flags;
4604 }
4605 }
4606
4607 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4608 // Check for overwrite.
4609 struct sigaction oldAct;
4610 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4611
4612 void* oldhand = oldAct.sa_sigaction
4613 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4614 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4615 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4616 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4617 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4618 if (AllowUserSignalHandlers || !set_installed) {
4619 // Do not overwrite; user takes responsibility to forward to us.
4620 return;
4621 } else if (UseSignalChaining) {
4622 // save the old handler in jvm
4623 save_preinstalled_handler(sig, oldAct);
4624 // libjsig also interposes the sigaction() call below and saves the
4625 // old sigaction on it own.
4626 } else {
4627 fatal("Encountered unexpected pre-existing sigaction handler "
4628 "%#lx for signal %d.", (long)oldhand, sig);
4629 }
4630 }
4631
4632 struct sigaction sigAct;
4633 sigfillset(&(sigAct.sa_mask));
4634 sigAct.sa_handler = SIG_DFL;
4635 if (!set_installed) {
4636 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4637 } else {
4638 sigAct.sa_sigaction = signalHandler;
4639 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4640 }
4641 // Save flags, which are set by ours
4642 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4643 sigflags[sig] = sigAct.sa_flags;
4644
4645 int ret = sigaction(sig, &sigAct, &oldAct);
4646 assert(ret == 0, "check");
4647
4648 void* oldhand2 = oldAct.sa_sigaction
4649 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4650 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4651 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4652 }
4653
4654 // install signal handlers for signals that HotSpot needs to
4655 // handle in order to support Java-level exception handling.
4656
4657 void os::Linux::install_signal_handlers() {
4658 if (!signal_handlers_are_installed) {
4659 signal_handlers_are_installed = true;
4660
4661 // signal-chaining
4662 typedef void (*signal_setting_t)();
4663 signal_setting_t begin_signal_setting = NULL;
4664 signal_setting_t end_signal_setting = NULL;
4665 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4666 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4667 if (begin_signal_setting != NULL) {
4668 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4669 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4670 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4671 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4672 libjsig_is_loaded = true;
4673 assert(UseSignalChaining, "should enable signal-chaining");
4674 }
4675 if (libjsig_is_loaded) {
4676 // Tell libjsig jvm is setting signal handlers
4677 (*begin_signal_setting)();
4678 }
4679
4680 set_signal_handler(SIGSEGV, true);
4681 set_signal_handler(SIGPIPE, true);
4682 set_signal_handler(SIGBUS, true);
4683 set_signal_handler(SIGILL, true);
4684 set_signal_handler(SIGFPE, true);
4685 #if defined(PPC64)
4686 set_signal_handler(SIGTRAP, true);
4687 #endif
4688 set_signal_handler(SIGXFSZ, true);
4689
4690 if (libjsig_is_loaded) {
4691 // Tell libjsig jvm finishes setting signal handlers
4692 (*end_signal_setting)();
4693 }
4694
4695 // We don't activate signal checker if libjsig is in place, we trust ourselves
4696 // and if UserSignalHandler is installed all bets are off.
4697 // Log that signal checking is off only if -verbose:jni is specified.
4698 if (CheckJNICalls) {
4699 if (libjsig_is_loaded) {
4700 if (PrintJNIResolving) {
4701 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4702 }
4703 check_signals = false;
4704 }
4705 if (AllowUserSignalHandlers) {
4706 if (PrintJNIResolving) {
4707 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4708 }
4709 check_signals = false;
4710 }
4711 }
4712 }
4713 }
4714
4715 // This is the fastest way to get thread cpu time on Linux.
4716 // Returns cpu time (user+sys) for any thread, not only for current.
4717 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4718 // It might work on 2.6.10+ with a special kernel/glibc patch.
4719 // For reference, please, see IEEE Std 1003.1-2004:
4720 // http://www.unix.org/single_unix_specification
4721
4722 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4723 struct timespec tp;
4724 int rc = os::Linux::clock_gettime(clockid, &tp);
4725 assert(rc == 0, "clock_gettime is expected to return 0 code");
4726
4727 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4728 }
4729
4730 void os::Linux::initialize_os_info() {
4731 assert(_os_version == 0, "OS info already initialized");
4732
4733 struct utsname _uname;
4734
4735 uint32_t major;
4736 uint32_t minor;
4737 uint32_t fix;
4738
4739 int rc;
4740
4741 // Kernel version is unknown if
4742 // verification below fails.
4743 _os_version = 0x01000000;
4744
4745 rc = uname(&_uname);
4746 if (rc != -1) {
4747
4748 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4749 if (rc == 3) {
4750
4751 if (major < 256 && minor < 256 && fix < 256) {
4752 // Kernel version format is as expected,
4753 // set it overriding unknown state.
4754 _os_version = (major << 16) |
4755 (minor << 8 ) |
4756 (fix << 0 ) ;
4757 }
4758 }
4759 }
4760 }
4761
4762 uint32_t os::Linux::os_version() {
4763 assert(_os_version != 0, "not initialized");
4764 return _os_version & 0x00FFFFFF;
4765 }
4766
4767 bool os::Linux::os_version_is_known() {
4768 assert(_os_version != 0, "not initialized");
4769 return _os_version & 0x01000000 ? false : true;
4770 }
4771
4772 /////
4773 // glibc on Linux platform uses non-documented flag
4774 // to indicate, that some special sort of signal
4775 // trampoline is used.
4776 // We will never set this flag, and we should
4777 // ignore this flag in our diagnostic
4778 #ifdef SIGNIFICANT_SIGNAL_MASK
4779 #undef SIGNIFICANT_SIGNAL_MASK
4780 #endif
4781 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4782
4783 static const char* get_signal_handler_name(address handler,
4784 char* buf, int buflen) {
4785 int offset = 0;
4786 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4787 if (found) {
4788 // skip directory names
4789 const char *p1, *p2;
4790 p1 = buf;
4791 size_t len = strlen(os::file_separator());
4792 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4793 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4794 } else {
4795 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4796 }
4797 return buf;
4798 }
4799
4800 static void print_signal_handler(outputStream* st, int sig,
4801 char* buf, size_t buflen) {
4802 struct sigaction sa;
4803
4804 sigaction(sig, NULL, &sa);
4805
4806 // See comment for SIGNIFICANT_SIGNAL_MASK define
4807 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4808
4809 st->print("%s: ", os::exception_name(sig, buf, buflen));
4810
4811 address handler = (sa.sa_flags & SA_SIGINFO)
4812 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4813 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4814
4815 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4816 st->print("SIG_DFL");
4817 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4818 st->print("SIG_IGN");
4819 } else {
4820 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4821 }
4822
4823 st->print(", sa_mask[0]=");
4824 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4825
4826 address rh = VMError::get_resetted_sighandler(sig);
4827 // May be, handler was resetted by VMError?
4828 if (rh != NULL) {
4829 handler = rh;
4830 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4831 }
4832
4833 st->print(", sa_flags=");
4834 os::Posix::print_sa_flags(st, sa.sa_flags);
4835
4836 // Check: is it our handler?
4837 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4838 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4839 // It is our signal handler
4840 // check for flags, reset system-used one!
4841 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4842 st->print(
4843 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4844 os::Linux::get_our_sigflags(sig));
4845 }
4846 }
4847 st->cr();
4848 }
4849
4850
4851 #define DO_SIGNAL_CHECK(sig) \
4852 do { \
4853 if (!sigismember(&check_signal_done, sig)) { \
4854 os::Linux::check_signal_handler(sig); \
4855 } \
4856 } while (0)
4857
4858 // This method is a periodic task to check for misbehaving JNI applications
4859 // under CheckJNI, we can add any periodic checks here
4860
4861 void os::run_periodic_checks() {
4862 if (check_signals == false) return;
4863
4864 // SEGV and BUS if overridden could potentially prevent
4865 // generation of hs*.log in the event of a crash, debugging
4866 // such a case can be very challenging, so we absolutely
4867 // check the following for a good measure:
4868 DO_SIGNAL_CHECK(SIGSEGV);
4869 DO_SIGNAL_CHECK(SIGILL);
4870 DO_SIGNAL_CHECK(SIGFPE);
4871 DO_SIGNAL_CHECK(SIGBUS);
4872 DO_SIGNAL_CHECK(SIGPIPE);
4873 DO_SIGNAL_CHECK(SIGXFSZ);
4874 #if defined(PPC64)
4875 DO_SIGNAL_CHECK(SIGTRAP);
4876 #endif
4877
4878 // ReduceSignalUsage allows the user to override these handlers
4879 // see comments at the very top and jvm_md.h
4880 if (!ReduceSignalUsage) {
4881 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4882 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4883 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4884 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4885 }
4886
4887 DO_SIGNAL_CHECK(SR_signum);
4888 }
4889
4890 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4891
4892 static os_sigaction_t os_sigaction = NULL;
4893
4894 void os::Linux::check_signal_handler(int sig) {
4895 char buf[O_BUFLEN];
4896 address jvmHandler = NULL;
4897
4898
4899 struct sigaction act;
4900 if (os_sigaction == NULL) {
4901 // only trust the default sigaction, in case it has been interposed
4902 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4903 if (os_sigaction == NULL) return;
4904 }
4905
4906 os_sigaction(sig, (struct sigaction*)NULL, &act);
4907
4908
4909 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4910
4911 address thisHandler = (act.sa_flags & SA_SIGINFO)
4912 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4913 : CAST_FROM_FN_PTR(address, act.sa_handler);
4914
4915
4916 switch (sig) {
4917 case SIGSEGV:
4918 case SIGBUS:
4919 case SIGFPE:
4920 case SIGPIPE:
4921 case SIGILL:
4922 case SIGXFSZ:
4923 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4924 break;
4925
4926 case SHUTDOWN1_SIGNAL:
4927 case SHUTDOWN2_SIGNAL:
4928 case SHUTDOWN3_SIGNAL:
4929 case BREAK_SIGNAL:
4930 jvmHandler = (address)user_handler();
4931 break;
4932
4933 default:
4934 if (sig == SR_signum) {
4935 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4936 } else {
4937 return;
4938 }
4939 break;
4940 }
4941
4942 if (thisHandler != jvmHandler) {
4943 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4944 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4945 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4946 // No need to check this sig any longer
4947 sigaddset(&check_signal_done, sig);
4948 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4949 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4950 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4951 exception_name(sig, buf, O_BUFLEN));
4952 }
4953 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4954 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4955 tty->print("expected:");
4956 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4957 tty->cr();
4958 tty->print(" found:");
4959 os::Posix::print_sa_flags(tty, act.sa_flags);
4960 tty->cr();
4961 // No need to check this sig any longer
4962 sigaddset(&check_signal_done, sig);
4963 }
4964
4965 // Dump all the signal
4966 if (sigismember(&check_signal_done, sig)) {
4967 print_signal_handlers(tty, buf, O_BUFLEN);
4968 }
4969 }
4970
4971 extern void report_error(char* file_name, int line_no, char* title,
4972 char* format, ...);
4973
4974 // this is called _before_ most of the global arguments have been parsed
4975 void os::init(void) {
4976 char dummy; // used to get a guess on initial stack address
4977
4978 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4979
4980 init_random(1234567);
4981
4982 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4983 if (Linux::page_size() == -1) {
4984 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4985 os::strerror(errno));
4986 }
4987 init_page_sizes((size_t) Linux::page_size());
4988
4989 Linux::initialize_system_info();
4990
4991 Linux::initialize_os_info();
4992
4993 // _main_thread points to the thread that created/loaded the JVM.
4994 Linux::_main_thread = pthread_self();
4995
4996 Linux::clock_init();
4997 initial_time_count = javaTimeNanos();
4998
4999 // retrieve entry point for pthread_setname_np
5000 Linux::_pthread_setname_np =
5001 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
5002
5003 os::Posix::init();
5004 }
5005
5006 // To install functions for atexit system call
5007 extern "C" {
5008 static void perfMemory_exit_helper() {
5009 perfMemory_exit();
5010 }
5011 }
5012
5013 void os::pd_init_container_support() {
5014 OSContainer::init();
5015 }
5016
5017 // this is called _after_ the global arguments have been parsed
5018 jint os::init_2(void) {
5019
5020 os::Posix::init_2();
5021
5022 Linux::fast_thread_clock_init();
5023
5024 // initialize suspend/resume support - must do this before signal_sets_init()
5025 if (SR_initialize() != 0) {
5026 perror("SR_initialize failed");
5027 return JNI_ERR;
5028 }
5029
5030 Linux::signal_sets_init();
5031 Linux::install_signal_handlers();
5032 // Initialize data for jdk.internal.misc.Signal
5033 if (!ReduceSignalUsage) {
5034 jdk_misc_signal_init();
5035 }
5036
5037 // Check and sets minimum stack sizes against command line options
5038 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
5039 return JNI_ERR;
5040 }
5041
5042 suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
5043 if (!suppress_primordial_thread_resolution) {
5044 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5045 }
5046
5047 #if defined(IA32)
5048 workaround_expand_exec_shield_cs_limit();
5049 #endif
5050
5051 Linux::libpthread_init();
5052 Linux::sched_getcpu_init();
5053 log_info(os)("HotSpot is running with %s, %s",
5054 Linux::glibc_version(), Linux::libpthread_version());
5055
5056 if (UseNUMA) {
5057 if (!Linux::libnuma_init()) {
5058 UseNUMA = false;
5059 } else {
5060 if ((Linux::numa_max_node() < 1) || Linux::isbound_to_single_node()) {
5061 // If there's only one node (they start from 0) or if the process
5062 // is bound explicitly to a single node using membind, disable NUMA.
5063 UseNUMA = false;
5064 }
5065 }
5066
5067 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5068 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5069 // we can make the adaptive lgrp chunk resizing work. If the user specified both
5070 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5071 // and disable adaptive resizing.
5072 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5073 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5074 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5075 UseAdaptiveSizePolicy = false;
5076 UseAdaptiveNUMAChunkSizing = false;
5077 }
5078 }
5079
5080 if (!UseNUMA && ForceNUMA) {
5081 UseNUMA = true;
5082 }
5083 }
5084
5085 if (MaxFDLimit) {
5086 // set the number of file descriptors to max. print out error
5087 // if getrlimit/setrlimit fails but continue regardless.
5088 struct rlimit nbr_files;
5089 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5090 if (status != 0) {
5091 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5092 } else {
5093 nbr_files.rlim_cur = nbr_files.rlim_max;
5094 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5095 if (status != 0) {
5096 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5097 }
5098 }
5099 }
5100
5101 // Initialize lock used to serialize thread creation (see os::create_thread)
5102 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5103
5104 // at-exit methods are called in the reverse order of their registration.
5105 // atexit functions are called on return from main or as a result of a
5106 // call to exit(3C). There can be only 32 of these functions registered
5107 // and atexit() does not set errno.
5108
5109 if (PerfAllowAtExitRegistration) {
5110 // only register atexit functions if PerfAllowAtExitRegistration is set.
5111 // atexit functions can be delayed until process exit time, which
5112 // can be problematic for embedded VM situations. Embedded VMs should
5113 // call DestroyJavaVM() to assure that VM resources are released.
5114
5115 // note: perfMemory_exit_helper atexit function may be removed in
5116 // the future if the appropriate cleanup code can be added to the
5117 // VM_Exit VMOperation's doit method.
5118 if (atexit(perfMemory_exit_helper) != 0) {
5119 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5120 }
5121 }
5122
5123 // initialize thread priority policy
5124 prio_init();
5125
5126 if (!FLAG_IS_DEFAULT(AllocateHeapAt)) {
5127 set_coredump_filter(false /*largepages*/, true /*dax_shared*/);
5128 }
5129 return JNI_OK;
5130 }
5131
5132 // Mark the polling page as unreadable
5133 void os::make_polling_page_unreadable(void) {
5134 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5135 fatal("Could not disable polling page");
5136 }
5137 }
5138
5139 // Mark the polling page as readable
5140 void os::make_polling_page_readable(void) {
5141 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5142 fatal("Could not enable polling page");
5143 }
5144 }
5145
5146 // older glibc versions don't have this macro (which expands to
5147 // an optimized bit-counting function) so we have to roll our own
5148 #ifndef CPU_COUNT
5149
5150 static int _cpu_count(const cpu_set_t* cpus) {
5151 int count = 0;
5152 // only look up to the number of configured processors
5153 for (int i = 0; i < os::processor_count(); i++) {
5154 if (CPU_ISSET(i, cpus)) {
5155 count++;
5156 }
5157 }
5158 return count;
5159 }
5160
5161 #define CPU_COUNT(cpus) _cpu_count(cpus)
5162
5163 #endif // CPU_COUNT
5164
5165 // Get the current number of available processors for this process.
5166 // This value can change at any time during a process's lifetime.
5167 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5168 // If it appears there may be more than 1024 processors then we do a
5169 // dynamic check - see 6515172 for details.
5170 // If anything goes wrong we fallback to returning the number of online
5171 // processors - which can be greater than the number available to the process.
5172 int os::Linux::active_processor_count() {
5173 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5174 cpu_set_t* cpus_p = &cpus;
5175 int cpus_size = sizeof(cpu_set_t);
5176
5177 int configured_cpus = os::processor_count(); // upper bound on available cpus
5178 int cpu_count = 0;
5179
5180 // old build platforms may not support dynamic cpu sets
5181 #ifdef CPU_ALLOC
5182
5183 // To enable easy testing of the dynamic path on different platforms we
5184 // introduce a diagnostic flag: UseCpuAllocPath
5185 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5186 // kernel may use a mask bigger than cpu_set_t
5187 log_trace(os)("active_processor_count: using dynamic path %s"
5188 "- configured processors: %d",
5189 UseCpuAllocPath ? "(forced) " : "",
5190 configured_cpus);
5191 cpus_p = CPU_ALLOC(configured_cpus);
5192 if (cpus_p != NULL) {
5193 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5194 // zero it just to be safe
5195 CPU_ZERO_S(cpus_size, cpus_p);
5196 }
5197 else {
5198 // failed to allocate so fallback to online cpus
5199 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5200 log_trace(os)("active_processor_count: "
5201 "CPU_ALLOC failed (%s) - using "
5202 "online processor count: %d",
5203 os::strerror(errno), online_cpus);
5204 return online_cpus;
5205 }
5206 }
5207 else {
5208 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5209 configured_cpus);
5210 }
5211 #else // CPU_ALLOC
5212 // these stubs won't be executed
5213 #define CPU_COUNT_S(size, cpus) -1
5214 #define CPU_FREE(cpus)
5215
5216 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5217 configured_cpus);
5218 #endif // CPU_ALLOC
5219
5220 // pid 0 means the current thread - which we have to assume represents the process
5221 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5222 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5223 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5224 }
5225 else {
5226 cpu_count = CPU_COUNT(cpus_p);
5227 }
5228 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5229 }
5230 else {
5231 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5232 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5233 "which may exceed available processors", os::strerror(errno), cpu_count);
5234 }
5235
5236 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5237 CPU_FREE(cpus_p);
5238 }
5239
5240 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5241 return cpu_count;
5242 }
5243
5244 // Determine the active processor count from one of
5245 // three different sources:
5246 //
5247 // 1. User option -XX:ActiveProcessorCount
5248 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5249 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5250 //
5251 // Option 1, if specified, will always override.
5252 // If the cgroup subsystem is active and configured, we
5253 // will return the min of the cgroup and option 2 results.
5254 // This is required since tools, such as numactl, that
5255 // alter cpu affinity do not update cgroup subsystem
5256 // cpuset configuration files.
5257 int os::active_processor_count() {
5258 // User has overridden the number of active processors
5259 if (ActiveProcessorCount > 0) {
5260 log_trace(os)("active_processor_count: "
5261 "active processor count set by user : %d",
5262 ActiveProcessorCount);
5263 return ActiveProcessorCount;
5264 }
5265
5266 int active_cpus;
5267 if (OSContainer::is_containerized()) {
5268 active_cpus = OSContainer::active_processor_count();
5269 log_trace(os)("active_processor_count: determined by OSContainer: %d",
5270 active_cpus);
5271 } else {
5272 active_cpus = os::Linux::active_processor_count();
5273 }
5274
5275 return active_cpus;
5276 }
5277
5278 uint os::processor_id() {
5279 const int id = Linux::sched_getcpu();
5280 assert(id >= 0 && id < _processor_count, "Invalid processor id");
5281 return (uint)id;
5282 }
5283
5284 void os::set_native_thread_name(const char *name) {
5285 if (Linux::_pthread_setname_np) {
5286 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5287 snprintf(buf, sizeof(buf), "%s", name);
5288 buf[sizeof(buf) - 1] = '\0';
5289 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5290 // ERANGE should not happen; all other errors should just be ignored.
5291 assert(rc != ERANGE, "pthread_setname_np failed");
5292 }
5293 }
5294
5295 bool os::distribute_processes(uint length, uint* distribution) {
5296 // Not yet implemented.
5297 return false;
5298 }
5299
5300 bool os::bind_to_processor(uint processor_id) {
5301 // Not yet implemented.
5302 return false;
5303 }
5304
5305 ///
5306
5307 void os::SuspendedThreadTask::internal_do_task() {
5308 if (do_suspend(_thread->osthread())) {
5309 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5310 do_task(context);
5311 do_resume(_thread->osthread());
5312 }
5313 }
5314
5315 ////////////////////////////////////////////////////////////////////////////////
5316 // debug support
5317
5318 bool os::find(address addr, outputStream* st) {
5319 Dl_info dlinfo;
5320 memset(&dlinfo, 0, sizeof(dlinfo));
5321 if (dladdr(addr, &dlinfo) != 0) {
5322 st->print(PTR_FORMAT ": ", p2i(addr));
5323 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5324 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5325 p2i(addr) - p2i(dlinfo.dli_saddr));
5326 } else if (dlinfo.dli_fbase != NULL) {
5327 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5328 } else {
5329 st->print("<absolute address>");
5330 }
5331 if (dlinfo.dli_fname != NULL) {
5332 st->print(" in %s", dlinfo.dli_fname);
5333 }
5334 if (dlinfo.dli_fbase != NULL) {
5335 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5336 }
5337 st->cr();
5338
5339 if (Verbose) {
5340 // decode some bytes around the PC
5341 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5342 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5343 address lowest = (address) dlinfo.dli_sname;
5344 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5345 if (begin < lowest) begin = lowest;
5346 Dl_info dlinfo2;
5347 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5348 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5349 end = (address) dlinfo2.dli_saddr;
5350 }
5351 Disassembler::decode(begin, end, st);
5352 }
5353 return true;
5354 }
5355 return false;
5356 }
5357
5358 ////////////////////////////////////////////////////////////////////////////////
5359 // misc
5360
5361 // This does not do anything on Linux. This is basically a hook for being
5362 // able to use structured exception handling (thread-local exception filters)
5363 // on, e.g., Win32.
5364 void
5365 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5366 JavaCallArguments* args, Thread* thread) {
5367 f(value, method, args, thread);
5368 }
5369
5370 void os::print_statistics() {
5371 }
5372
5373 bool os::message_box(const char* title, const char* message) {
5374 int i;
5375 fdStream err(defaultStream::error_fd());
5376 for (i = 0; i < 78; i++) err.print_raw("=");
5377 err.cr();
5378 err.print_raw_cr(title);
5379 for (i = 0; i < 78; i++) err.print_raw("-");
5380 err.cr();
5381 err.print_raw_cr(message);
5382 for (i = 0; i < 78; i++) err.print_raw("=");
5383 err.cr();
5384
5385 char buf[16];
5386 // Prevent process from exiting upon "read error" without consuming all CPU
5387 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5388
5389 return buf[0] == 'y' || buf[0] == 'Y';
5390 }
5391
5392 // Is a (classpath) directory empty?
5393 bool os::dir_is_empty(const char* path) {
5394 DIR *dir = NULL;
5395 struct dirent *ptr;
5396
5397 dir = opendir(path);
5398 if (dir == NULL) return true;
5399
5400 // Scan the directory
5401 bool result = true;
5402 char buf[sizeof(struct dirent) + MAX_PATH];
5403 while (result && (ptr = ::readdir(dir)) != NULL) {
5404 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5405 result = false;
5406 }
5407 }
5408 closedir(dir);
5409 return result;
5410 }
5411
5412 // This code originates from JDK's sysOpen and open64_w
5413 // from src/solaris/hpi/src/system_md.c
5414
5415 int os::open(const char *path, int oflag, int mode) {
5416 if (strlen(path) > MAX_PATH - 1) {
5417 errno = ENAMETOOLONG;
5418 return -1;
5419 }
5420
5421 // All file descriptors that are opened in the Java process and not
5422 // specifically destined for a subprocess should have the close-on-exec
5423 // flag set. If we don't set it, then careless 3rd party native code
5424 // might fork and exec without closing all appropriate file descriptors
5425 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5426 // turn might:
5427 //
5428 // - cause end-of-file to fail to be detected on some file
5429 // descriptors, resulting in mysterious hangs, or
5430 //
5431 // - might cause an fopen in the subprocess to fail on a system
5432 // suffering from bug 1085341.
5433 //
5434 // (Yes, the default setting of the close-on-exec flag is a Unix
5435 // design flaw)
5436 //
5437 // See:
5438 // 1085341: 32-bit stdio routines should support file descriptors >255
5439 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5440 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5441 //
5442 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5443 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5444 // because it saves a system call and removes a small window where the flag
5445 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5446 // and we fall back to using FD_CLOEXEC (see below).
5447 #ifdef O_CLOEXEC
5448 oflag |= O_CLOEXEC;
5449 #endif
5450
5451 int fd = ::open64(path, oflag, mode);
5452 if (fd == -1) return -1;
5453
5454 //If the open succeeded, the file might still be a directory
5455 {
5456 struct stat64 buf64;
5457 int ret = ::fstat64(fd, &buf64);
5458 int st_mode = buf64.st_mode;
5459
5460 if (ret != -1) {
5461 if ((st_mode & S_IFMT) == S_IFDIR) {
5462 errno = EISDIR;
5463 ::close(fd);
5464 return -1;
5465 }
5466 } else {
5467 ::close(fd);
5468 return -1;
5469 }
5470 }
5471
5472 #ifdef FD_CLOEXEC
5473 // Validate that the use of the O_CLOEXEC flag on open above worked.
5474 // With recent kernels, we will perform this check exactly once.
5475 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5476 if (!O_CLOEXEC_is_known_to_work) {
5477 int flags = ::fcntl(fd, F_GETFD);
5478 if (flags != -1) {
5479 if ((flags & FD_CLOEXEC) != 0)
5480 O_CLOEXEC_is_known_to_work = 1;
5481 else
5482 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5483 }
5484 }
5485 #endif
5486
5487 return fd;
5488 }
5489
5490
5491 // create binary file, rewriting existing file if required
5492 int os::create_binary_file(const char* path, bool rewrite_existing) {
5493 int oflags = O_WRONLY | O_CREAT;
5494 if (!rewrite_existing) {
5495 oflags |= O_EXCL;
5496 }
5497 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5498 }
5499
5500 // return current position of file pointer
5501 jlong os::current_file_offset(int fd) {
5502 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5503 }
5504
5505 // move file pointer to the specified offset
5506 jlong os::seek_to_file_offset(int fd, jlong offset) {
5507 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5508 }
5509
5510 // This code originates from JDK's sysAvailable
5511 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5512
5513 int os::available(int fd, jlong *bytes) {
5514 jlong cur, end;
5515 int mode;
5516 struct stat64 buf64;
5517
5518 if (::fstat64(fd, &buf64) >= 0) {
5519 mode = buf64.st_mode;
5520 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5521 int n;
5522 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5523 *bytes = n;
5524 return 1;
5525 }
5526 }
5527 }
5528 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5529 return 0;
5530 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5531 return 0;
5532 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5533 return 0;
5534 }
5535 *bytes = end - cur;
5536 return 1;
5537 }
5538
5539 // Map a block of memory.
5540 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5541 char *addr, size_t bytes, bool read_only,
5542 bool allow_exec) {
5543 int prot;
5544 int flags = MAP_PRIVATE;
5545
5546 if (read_only) {
5547 prot = PROT_READ;
5548 } else {
5549 prot = PROT_READ | PROT_WRITE;
5550 }
5551
5552 if (allow_exec) {
5553 prot |= PROT_EXEC;
5554 }
5555
5556 if (addr != NULL) {
5557 flags |= MAP_FIXED;
5558 }
5559
5560 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5561 fd, file_offset);
5562 if (mapped_address == MAP_FAILED) {
5563 return NULL;
5564 }
5565 return mapped_address;
5566 }
5567
5568
5569 // Remap a block of memory.
5570 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5571 char *addr, size_t bytes, bool read_only,
5572 bool allow_exec) {
5573 // same as map_memory() on this OS
5574 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5575 allow_exec);
5576 }
5577
5578
5579 // Unmap a block of memory.
5580 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5581 return munmap(addr, bytes) == 0;
5582 }
5583
5584 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5585
5586 static clockid_t thread_cpu_clockid(Thread* thread) {
5587 pthread_t tid = thread->osthread()->pthread_id();
5588 clockid_t clockid;
5589
5590 // Get thread clockid
5591 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5592 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5593 return clockid;
5594 }
5595
5596 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5597 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5598 // of a thread.
5599 //
5600 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5601 // the fast estimate available on the platform.
5602
5603 jlong os::current_thread_cpu_time() {
5604 if (os::Linux::supports_fast_thread_cpu_time()) {
5605 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5606 } else {
5607 // return user + sys since the cost is the same
5608 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5609 }
5610 }
5611
5612 jlong os::thread_cpu_time(Thread* thread) {
5613 // consistent with what current_thread_cpu_time() returns
5614 if (os::Linux::supports_fast_thread_cpu_time()) {
5615 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5616 } else {
5617 return slow_thread_cpu_time(thread, true /* user + sys */);
5618 }
5619 }
5620
5621 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5622 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5623 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5624 } else {
5625 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5626 }
5627 }
5628
5629 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5630 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5631 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5632 } else {
5633 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5634 }
5635 }
5636
5637 // -1 on error.
5638 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5639 pid_t tid = thread->osthread()->thread_id();
5640 char *s;
5641 char stat[2048];
5642 int statlen;
5643 char proc_name[64];
5644 int count;
5645 long sys_time, user_time;
5646 char cdummy;
5647 int idummy;
5648 long ldummy;
5649 FILE *fp;
5650
5651 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5652 fp = fopen(proc_name, "r");
5653 if (fp == NULL) return -1;
5654 statlen = fread(stat, 1, 2047, fp);
5655 stat[statlen] = '\0';
5656 fclose(fp);
5657
5658 // Skip pid and the command string. Note that we could be dealing with
5659 // weird command names, e.g. user could decide to rename java launcher
5660 // to "java 1.4.2 :)", then the stat file would look like
5661 // 1234 (java 1.4.2 :)) R ... ...
5662 // We don't really need to know the command string, just find the last
5663 // occurrence of ")" and then start parsing from there. See bug 4726580.
5664 s = strrchr(stat, ')');
5665 if (s == NULL) return -1;
5666
5667 // Skip blank chars
5668 do { s++; } while (s && isspace(*s));
5669
5670 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5671 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5672 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5673 &user_time, &sys_time);
5674 if (count != 13) return -1;
5675 if (user_sys_cpu_time) {
5676 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5677 } else {
5678 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5679 }
5680 }
5681
5682 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5683 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5684 info_ptr->may_skip_backward = false; // elapsed time not wall time
5685 info_ptr->may_skip_forward = false; // elapsed time not wall time
5686 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5687 }
5688
5689 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5690 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5691 info_ptr->may_skip_backward = false; // elapsed time not wall time
5692 info_ptr->may_skip_forward = false; // elapsed time not wall time
5693 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5694 }
5695
5696 bool os::is_thread_cpu_time_supported() {
5697 return true;
5698 }
5699
5700 // System loadavg support. Returns -1 if load average cannot be obtained.
5701 // Linux doesn't yet have a (official) notion of processor sets,
5702 // so just return the system wide load average.
5703 int os::loadavg(double loadavg[], int nelem) {
5704 return ::getloadavg(loadavg, nelem);
5705 }
5706
5707 void os::pause() {
5708 char filename[MAX_PATH];
5709 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5710 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5711 } else {
5712 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5713 }
5714
5715 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5716 if (fd != -1) {
5717 struct stat buf;
5718 ::close(fd);
5719 while (::stat(filename, &buf) == 0) {
5720 (void)::poll(NULL, 0, 100);
5721 }
5722 } else {
5723 jio_fprintf(stderr,
5724 "Could not open pause file '%s', continuing immediately.\n", filename);
5725 }
5726 }
5727
5728 extern char** environ;
5729
5730 // Run the specified command in a separate process. Return its exit value,
5731 // or -1 on failure (e.g. can't fork a new process).
5732 // Unlike system(), this function can be called from signal handler. It
5733 // doesn't block SIGINT et al.
5734 int os::fork_and_exec(char* cmd) {
5735 const char * argv[4] = {"sh", "-c", cmd, NULL};
5736
5737 pid_t pid = fork();
5738
5739 if (pid < 0) {
5740 // fork failed
5741 return -1;
5742
5743 } else if (pid == 0) {
5744 // child process
5745
5746 execve("/bin/sh", (char* const*)argv, environ);
5747
5748 // execve failed
5749 _exit(-1);
5750
5751 } else {
5752 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5753 // care about the actual exit code, for now.
5754
5755 int status;
5756
5757 // Wait for the child process to exit. This returns immediately if
5758 // the child has already exited. */
5759 while (waitpid(pid, &status, 0) < 0) {
5760 switch (errno) {
5761 case ECHILD: return 0;
5762 case EINTR: break;
5763 default: return -1;
5764 }
5765 }
5766
5767 if (WIFEXITED(status)) {
5768 // The child exited normally; get its exit code.
5769 return WEXITSTATUS(status);
5770 } else if (WIFSIGNALED(status)) {
5771 // The child exited because of a signal
5772 // The best value to return is 0x80 + signal number,
5773 // because that is what all Unix shells do, and because
5774 // it allows callers to distinguish between process exit and
5775 // process death by signal.
5776 return 0x80 + WTERMSIG(status);
5777 } else {
5778 // Unknown exit code; pass it through
5779 return status;
5780 }
5781 }
5782 }
5783
5784 // Get the default path to the core file
5785 // Returns the length of the string
5786 int os::get_core_path(char* buffer, size_t bufferSize) {
5787 /*
5788 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5789 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5790 */
5791 const int core_pattern_len = 129;
5792 char core_pattern[core_pattern_len] = {0};
5793
5794 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5795 if (core_pattern_file == -1) {
5796 return -1;
5797 }
5798
5799 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5800 ::close(core_pattern_file);
5801 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5802 return -1;
5803 }
5804 if (core_pattern[ret-1] == '\n') {
5805 core_pattern[ret-1] = '\0';
5806 } else {
5807 core_pattern[ret] = '\0';
5808 }
5809
5810 char *pid_pos = strstr(core_pattern, "%p");
5811 int written;
5812
5813 if (core_pattern[0] == '/') {
5814 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5815 } else {
5816 char cwd[PATH_MAX];
5817
5818 const char* p = get_current_directory(cwd, PATH_MAX);
5819 if (p == NULL) {
5820 return -1;
5821 }
5822
5823 if (core_pattern[0] == '|') {
5824 written = jio_snprintf(buffer, bufferSize,
5825 "\"%s\" (or dumping to %s/core.%d)",
5826 &core_pattern[1], p, current_process_id());
5827 } else {
5828 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5829 }
5830 }
5831
5832 if (written < 0) {
5833 return -1;
5834 }
5835
5836 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5837 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5838
5839 if (core_uses_pid_file != -1) {
5840 char core_uses_pid = 0;
5841 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5842 ::close(core_uses_pid_file);
5843
5844 if (core_uses_pid == '1') {
5845 jio_snprintf(buffer + written, bufferSize - written,
5846 ".%d", current_process_id());
5847 }
5848 }
5849 }
5850
5851 return strlen(buffer);
5852 }
5853
5854 bool os::start_debugging(char *buf, int buflen) {
5855 int len = (int)strlen(buf);
5856 char *p = &buf[len];
5857
5858 jio_snprintf(p, buflen-len,
5859 "\n\n"
5860 "Do you want to debug the problem?\n\n"
5861 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5862 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5863 "Otherwise, press RETURN to abort...",
5864 os::current_process_id(), os::current_process_id(),
5865 os::current_thread_id(), os::current_thread_id());
5866
5867 bool yes = os::message_box("Unexpected Error", buf);
5868
5869 if (yes) {
5870 // yes, user asked VM to launch debugger
5871 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5872 os::current_process_id(), os::current_process_id());
5873
5874 os::fork_and_exec(buf);
5875 yes = false;
5876 }
5877 return yes;
5878 }
5879
5880
5881 // Java/Compiler thread:
5882 //
5883 // Low memory addresses
5884 // P0 +------------------------+
5885 // | |\ Java thread created by VM does not have glibc
5886 // | glibc guard page | - guard page, attached Java thread usually has
5887 // | |/ 1 glibc guard page.
5888 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5889 // | |\
5890 // | HotSpot Guard Pages | - red, yellow and reserved pages
5891 // | |/
5892 // +------------------------+ JavaThread::stack_reserved_zone_base()
5893 // | |\
5894 // | Normal Stack | -
5895 // | |/
5896 // P2 +------------------------+ Thread::stack_base()
5897 //
5898 // Non-Java thread:
5899 //
5900 // Low memory addresses
5901 // P0 +------------------------+
5902 // | |\
5903 // | glibc guard page | - usually 1 page
5904 // | |/
5905 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5906 // | |\
5907 // | Normal Stack | -
5908 // | |/
5909 // P2 +------------------------+ Thread::stack_base()
5910 //
5911 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
5912 // returned from pthread_attr_getstack().
5913 // ** Due to NPTL implementation error, linux takes the glibc guard page out
5914 // of the stack size given in pthread_attr. We work around this for
5915 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
5916 //
5917 #ifndef ZERO
5918 static void current_stack_region(address * bottom, size_t * size) {
5919 if (os::is_primordial_thread()) {
5920 // primordial thread needs special handling because pthread_getattr_np()
5921 // may return bogus value.
5922 *bottom = os::Linux::initial_thread_stack_bottom();
5923 *size = os::Linux::initial_thread_stack_size();
5924 } else {
5925 pthread_attr_t attr;
5926
5927 int rslt = pthread_getattr_np(pthread_self(), &attr);
5928
5929 // JVM needs to know exact stack location, abort if it fails
5930 if (rslt != 0) {
5931 if (rslt == ENOMEM) {
5932 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
5933 } else {
5934 fatal("pthread_getattr_np failed with error = %d", rslt);
5935 }
5936 }
5937
5938 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
5939 fatal("Cannot locate current stack attributes!");
5940 }
5941
5942 // Work around NPTL stack guard error.
5943 size_t guard_size = 0;
5944 rslt = pthread_attr_getguardsize(&attr, &guard_size);
5945 if (rslt != 0) {
5946 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
5947 }
5948 *bottom += guard_size;
5949 *size -= guard_size;
5950
5951 pthread_attr_destroy(&attr);
5952
5953 }
5954 assert(os::current_stack_pointer() >= *bottom &&
5955 os::current_stack_pointer() < *bottom + *size, "just checking");
5956 }
5957
5958 address os::current_stack_base() {
5959 address bottom;
5960 size_t size;
5961 current_stack_region(&bottom, &size);
5962 return (bottom + size);
5963 }
5964
5965 size_t os::current_stack_size() {
5966 // This stack size includes the usable stack and HotSpot guard pages
5967 // (for the threads that have Hotspot guard pages).
5968 address bottom;
5969 size_t size;
5970 current_stack_region(&bottom, &size);
5971 return size;
5972 }
5973 #endif
5974
5975 static inline struct timespec get_mtime(const char* filename) {
5976 struct stat st;
5977 int ret = os::stat(filename, &st);
5978 assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
5979 return st.st_mtim;
5980 }
5981
5982 int os::compare_file_modified_times(const char* file1, const char* file2) {
5983 struct timespec filetime1 = get_mtime(file1);
5984 struct timespec filetime2 = get_mtime(file2);
5985 int diff = filetime1.tv_sec - filetime2.tv_sec;
5986 if (diff == 0) {
5987 return filetime1.tv_nsec - filetime2.tv_nsec;
5988 }
5989 return diff;
5990 }
5991
5992 /////////////// Unit tests ///////////////
5993
5994 #ifndef PRODUCT
5995
5996 #define test_log(...) \
5997 do { \
5998 if (VerboseInternalVMTests) { \
5999 tty->print_cr(__VA_ARGS__); \
6000 tty->flush(); \
6001 } \
6002 } while (false)
6003
6004 class TestReserveMemorySpecial : AllStatic {
6005 public:
6006 static void small_page_write(void* addr, size_t size) {
6007 size_t page_size = os::vm_page_size();
6008
6009 char* end = (char*)addr + size;
6010 for (char* p = (char*)addr; p < end; p += page_size) {
6011 *p = 1;
6012 }
6013 }
6014
6015 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6016 if (!UseHugeTLBFS) {
6017 return;
6018 }
6019
6020 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6021
6022 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6023
6024 if (addr != NULL) {
6025 small_page_write(addr, size);
6026
6027 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6028 }
6029 }
6030
6031 static void test_reserve_memory_special_huge_tlbfs_only() {
6032 if (!UseHugeTLBFS) {
6033 return;
6034 }
6035
6036 size_t lp = os::large_page_size();
6037
6038 for (size_t size = lp; size <= lp * 10; size += lp) {
6039 test_reserve_memory_special_huge_tlbfs_only(size);
6040 }
6041 }
6042
6043 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6044 size_t lp = os::large_page_size();
6045 size_t ag = os::vm_allocation_granularity();
6046
6047 // sizes to test
6048 const size_t sizes[] = {
6049 lp, lp + ag, lp + lp / 2, lp * 2,
6050 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6051 lp * 10, lp * 10 + lp / 2
6052 };
6053 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6054
6055 // For each size/alignment combination, we test three scenarios:
6056 // 1) with req_addr == NULL
6057 // 2) with a non-null req_addr at which we expect to successfully allocate
6058 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6059 // expect the allocation to either fail or to ignore req_addr
6060
6061 // Pre-allocate two areas; they shall be as large as the largest allocation
6062 // and aligned to the largest alignment we will be testing.
6063 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6064 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6065 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6066 -1, 0);
6067 assert(mapping1 != MAP_FAILED, "should work");
6068
6069 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6070 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6071 -1, 0);
6072 assert(mapping2 != MAP_FAILED, "should work");
6073
6074 // Unmap the first mapping, but leave the second mapping intact: the first
6075 // mapping will serve as a value for a "good" req_addr (case 2). The second
6076 // mapping, still intact, as "bad" req_addr (case 3).
6077 ::munmap(mapping1, mapping_size);
6078
6079 // Case 1
6080 test_log("%s, req_addr NULL:", __FUNCTION__);
6081 test_log("size align result");
6082
6083 for (int i = 0; i < num_sizes; i++) {
6084 const size_t size = sizes[i];
6085 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6086 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6087 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6088 size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6089 if (p != NULL) {
6090 assert(is_aligned(p, alignment), "must be");
6091 small_page_write(p, size);
6092 os::Linux::release_memory_special_huge_tlbfs(p, size);
6093 }
6094 }
6095 }
6096
6097 // Case 2
6098 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6099 test_log("size align req_addr result");
6100
6101 for (int i = 0; i < num_sizes; i++) {
6102 const size_t size = sizes[i];
6103 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6104 char* const req_addr = align_up(mapping1, alignment);
6105 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6106 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6107 size, alignment, p2i(req_addr), p2i(p),
6108 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6109 if (p != NULL) {
6110 assert(p == req_addr, "must be");
6111 small_page_write(p, size);
6112 os::Linux::release_memory_special_huge_tlbfs(p, size);
6113 }
6114 }
6115 }
6116
6117 // Case 3
6118 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6119 test_log("size align req_addr result");
6120
6121 for (int i = 0; i < num_sizes; i++) {
6122 const size_t size = sizes[i];
6123 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6124 char* const req_addr = align_up(mapping2, alignment);
6125 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6126 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6127 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6128 // as the area around req_addr contains already existing mappings, the API should always
6129 // return NULL (as per contract, it cannot return another address)
6130 assert(p == NULL, "must be");
6131 }
6132 }
6133
6134 ::munmap(mapping2, mapping_size);
6135
6136 }
6137
6138 static void test_reserve_memory_special_huge_tlbfs() {
6139 if (!UseHugeTLBFS) {
6140 return;
6141 }
6142
6143 test_reserve_memory_special_huge_tlbfs_only();
6144 test_reserve_memory_special_huge_tlbfs_mixed();
6145 }
6146
6147 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6148 if (!UseSHM) {
6149 return;
6150 }
6151
6152 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6153
6154 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6155
6156 if (addr != NULL) {
6157 assert(is_aligned(addr, alignment), "Check");
6158 assert(is_aligned(addr, os::large_page_size()), "Check");
6159
6160 small_page_write(addr, size);
6161
6162 os::Linux::release_memory_special_shm(addr, size);
6163 }
6164 }
6165
6166 static void test_reserve_memory_special_shm() {
6167 size_t lp = os::large_page_size();
6168 size_t ag = os::vm_allocation_granularity();
6169
6170 for (size_t size = ag; size < lp * 3; size += ag) {
6171 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6172 test_reserve_memory_special_shm(size, alignment);
6173 }
6174 }
6175 }
6176
6177 static void test() {
6178 test_reserve_memory_special_huge_tlbfs();
6179 test_reserve_memory_special_shm();
6180 }
6181 };
6182
6183 void TestReserveMemorySpecial_test() {
6184 TestReserveMemorySpecial::test();
6185 }
6186
6187 #endif