1 /*
2 * Copyright (c) 1999, 2019, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 // no precompiled headers
26 #include "jvm.h"
27 #include "classfile/classLoader.hpp"
28 #include "classfile/systemDictionary.hpp"
29 #include "classfile/vmSymbols.hpp"
30 #include "code/icBuffer.hpp"
31 #include "code/vtableStubs.hpp"
32 #include "compiler/compileBroker.hpp"
33 #include "compiler/disassembler.hpp"
34 #include "interpreter/interpreter.hpp"
35 #include "logging/log.hpp"
36 #include "logging/logStream.hpp"
37 #include "memory/allocation.inline.hpp"
38 #include "memory/filemap.hpp"
39 #include "oops/oop.inline.hpp"
40 #include "os_linux.inline.hpp"
41 #include "os_posix.inline.hpp"
42 #include "os_share_linux.hpp"
43 #include "osContainer_linux.hpp"
44 #include "prims/jniFastGetField.hpp"
45 #include "prims/jvm_misc.hpp"
46 #include "runtime/arguments.hpp"
47 #include "runtime/atomic.hpp"
48 #include "runtime/extendedPC.hpp"
49 #include "runtime/globals.hpp"
50 #include "runtime/interfaceSupport.inline.hpp"
51 #include "runtime/init.hpp"
52 #include "runtime/java.hpp"
53 #include "runtime/javaCalls.hpp"
54 #include "runtime/mutexLocker.hpp"
55 #include "runtime/objectMonitor.hpp"
56 #include "runtime/orderAccess.hpp"
57 #include "runtime/osThread.hpp"
58 #include "runtime/perfMemory.hpp"
59 #include "runtime/sharedRuntime.hpp"
60 #include "runtime/statSampler.hpp"
61 #include "runtime/stubRoutines.hpp"
62 #include "runtime/thread.inline.hpp"
63 #include "runtime/threadCritical.hpp"
64 #include "runtime/threadSMR.hpp"
65 #include "runtime/timer.hpp"
66 #include "runtime/vm_version.hpp"
67 #include "semaphore_posix.hpp"
68 #include "services/attachListener.hpp"
69 #include "services/memTracker.hpp"
70 #include "services/runtimeService.hpp"
71 #include "utilities/align.hpp"
72 #include "utilities/decoder.hpp"
73 #include "utilities/defaultStream.hpp"
74 #include "utilities/events.hpp"
75 #include "utilities/elfFile.hpp"
76 #include "utilities/growableArray.hpp"
77 #include "utilities/macros.hpp"
78 #include "utilities/vmError.hpp"
79
80 // put OS-includes here
81 # include <sys/types.h>
82 # include <sys/mman.h>
83 # include <sys/stat.h>
84 # include <sys/select.h>
85 # include <pthread.h>
86 # include <signal.h>
87 # include <errno.h>
88 # include <dlfcn.h>
89 # include <stdio.h>
90 # include <unistd.h>
91 # include <sys/resource.h>
92 # include <pthread.h>
93 # include <sys/stat.h>
94 # include <sys/time.h>
95 # include <sys/times.h>
96 # include <sys/utsname.h>
97 # include <sys/socket.h>
98 # include <sys/wait.h>
99 # include <pwd.h>
100 # include <poll.h>
101 # include <fcntl.h>
102 # include <string.h>
103 # include <syscall.h>
104 # include <sys/sysinfo.h>
105 # include <gnu/libc-version.h>
106 # include <sys/ipc.h>
107 # include <sys/shm.h>
108 # include <link.h>
109 # include <stdint.h>
110 # include <inttypes.h>
111 # include <sys/ioctl.h>
112
113 #ifndef _GNU_SOURCE
114 #define _GNU_SOURCE
115 #include <sched.h>
116 #undef _GNU_SOURCE
117 #else
118 #include <sched.h>
119 #endif
120
121 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
122 // getrusage() is prepared to handle the associated failure.
123 #ifndef RUSAGE_THREAD
124 #define RUSAGE_THREAD (1) /* only the calling thread */
125 #endif
126
127 #define MAX_PATH (2 * K)
128
129 #define MAX_SECS 100000000
130
131 // for timer info max values which include all bits
132 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
133
134 enum CoredumpFilterBit {
135 FILE_BACKED_PVT_BIT = 1 << 2,
136 FILE_BACKED_SHARED_BIT = 1 << 3,
137 LARGEPAGES_BIT = 1 << 6,
138 DAX_SHARED_BIT = 1 << 8
139 };
140
141 ////////////////////////////////////////////////////////////////////////////////
142 // global variables
143 julong os::Linux::_physical_memory = 0;
144
145 address os::Linux::_initial_thread_stack_bottom = NULL;
146 uintptr_t os::Linux::_initial_thread_stack_size = 0;
147
148 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
149 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
150 Mutex* os::Linux::_createThread_lock = NULL;
151 pthread_t os::Linux::_main_thread;
152 int os::Linux::_page_size = -1;
153 bool os::Linux::_supports_fast_thread_cpu_time = false;
154 uint32_t os::Linux::_os_version = 0;
155 const char * os::Linux::_glibc_version = NULL;
156 const char * os::Linux::_libpthread_version = NULL;
157
158 static jlong initial_time_count=0;
159
160 static int clock_tics_per_sec = 100;
161
162 // If the VM might have been created on the primordial thread, we need to resolve the
163 // primordial thread stack bounds and check if the current thread might be the
164 // primordial thread in places. If we know that the primordial thread is never used,
165 // such as when the VM was created by one of the standard java launchers, we can
166 // avoid this
167 static bool suppress_primordial_thread_resolution = false;
168
169 // For diagnostics to print a message once. see run_periodic_checks
170 static sigset_t check_signal_done;
171 static bool check_signals = true;
172
173 // Signal number used to suspend/resume a thread
174
175 // do not use any signal number less than SIGSEGV, see 4355769
176 static int SR_signum = SIGUSR2;
177 sigset_t SR_sigset;
178
179 // utility functions
180
181 static int SR_initialize();
182
183 julong os::available_memory() {
184 return Linux::available_memory();
185 }
186
187 julong os::Linux::available_memory() {
188 // values in struct sysinfo are "unsigned long"
189 struct sysinfo si;
190 julong avail_mem;
191
192 if (OSContainer::is_containerized()) {
193 jlong mem_limit, mem_usage;
194 if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
195 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
196 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
197 }
198 if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
199 log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
200 }
201 if (mem_limit > 0 && mem_usage > 0 ) {
202 avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
203 log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
204 return avail_mem;
205 }
206 }
207
208 sysinfo(&si);
209 avail_mem = (julong)si.freeram * si.mem_unit;
210 log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
211 return avail_mem;
212 }
213
214 julong os::physical_memory() {
215 jlong phys_mem = 0;
216 if (OSContainer::is_containerized()) {
217 jlong mem_limit;
218 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
219 log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
220 return mem_limit;
221 }
222 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
223 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
224 }
225
226 phys_mem = Linux::physical_memory();
227 log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
228 return phys_mem;
229 }
230
231 static uint64_t initial_total_ticks = 0;
232 static uint64_t initial_steal_ticks = 0;
233 static bool has_initial_tick_info = false;
234
235 static void next_line(FILE *f) {
236 int c;
237 do {
238 c = fgetc(f);
239 } while (c != '\n' && c != EOF);
240 }
241
242 bool os::Linux::get_tick_information(CPUPerfTicks* pticks, int which_logical_cpu) {
243 FILE* fh;
244 uint64_t userTicks, niceTicks, systemTicks, idleTicks;
245 // since at least kernel 2.6 : iowait: time waiting for I/O to complete
246 // irq: time servicing interrupts; softirq: time servicing softirqs
247 uint64_t iowTicks = 0, irqTicks = 0, sirqTicks= 0;
248 // steal (since kernel 2.6.11): time spent in other OS when running in a virtualized environment
249 uint64_t stealTicks = 0;
250 // guest (since kernel 2.6.24): time spent running a virtual CPU for guest OS under the
251 // control of the Linux kernel
252 uint64_t guestNiceTicks = 0;
253 int logical_cpu = -1;
254 const int required_tickinfo_count = (which_logical_cpu == -1) ? 4 : 5;
255 int n;
256
257 memset(pticks, 0, sizeof(CPUPerfTicks));
258
259 if ((fh = fopen("/proc/stat", "r")) == NULL) {
260 return false;
261 }
262
263 if (which_logical_cpu == -1) {
264 n = fscanf(fh, "cpu " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
265 UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
266 UINT64_FORMAT " " UINT64_FORMAT " ",
267 &userTicks, &niceTicks, &systemTicks, &idleTicks,
268 &iowTicks, &irqTicks, &sirqTicks,
269 &stealTicks, &guestNiceTicks);
270 } else {
271 // Move to next line
272 next_line(fh);
273
274 // find the line for requested cpu faster to just iterate linefeeds?
275 for (int i = 0; i < which_logical_cpu; i++) {
276 next_line(fh);
277 }
278
279 n = fscanf(fh, "cpu%u " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
280 UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
281 UINT64_FORMAT " " UINT64_FORMAT " ",
282 &logical_cpu, &userTicks, &niceTicks,
283 &systemTicks, &idleTicks, &iowTicks, &irqTicks, &sirqTicks,
284 &stealTicks, &guestNiceTicks);
285 }
286
287 fclose(fh);
288 if (n < required_tickinfo_count || logical_cpu != which_logical_cpu) {
289 return false;
290 }
291 pticks->used = userTicks + niceTicks;
292 pticks->usedKernel = systemTicks + irqTicks + sirqTicks;
293 pticks->total = userTicks + niceTicks + systemTicks + idleTicks +
294 iowTicks + irqTicks + sirqTicks + stealTicks + guestNiceTicks;
295
296 if (n > required_tickinfo_count + 3) {
297 pticks->steal = stealTicks;
298 pticks->has_steal_ticks = true;
299 } else {
300 pticks->steal = 0;
301 pticks->has_steal_ticks = false;
302 }
303
304 return true;
305 }
306
307 // Return true if user is running as root.
308
309 bool os::have_special_privileges() {
310 static bool init = false;
311 static bool privileges = false;
312 if (!init) {
313 privileges = (getuid() != geteuid()) || (getgid() != getegid());
314 init = true;
315 }
316 return privileges;
317 }
318
319
320 #ifndef SYS_gettid
321 // i386: 224, ia64: 1105, amd64: 186, sparc 143
322 #ifdef __ia64__
323 #define SYS_gettid 1105
324 #else
325 #ifdef __i386__
326 #define SYS_gettid 224
327 #else
328 #ifdef __amd64__
329 #define SYS_gettid 186
330 #else
331 #ifdef __sparc__
332 #define SYS_gettid 143
333 #else
334 #error define gettid for the arch
335 #endif
336 #endif
337 #endif
338 #endif
339 #endif
340
341
342 // pid_t gettid()
343 //
344 // Returns the kernel thread id of the currently running thread. Kernel
345 // thread id is used to access /proc.
346 pid_t os::Linux::gettid() {
347 int rslt = syscall(SYS_gettid);
348 assert(rslt != -1, "must be."); // old linuxthreads implementation?
349 return (pid_t)rslt;
350 }
351
352 // Most versions of linux have a bug where the number of processors are
353 // determined by looking at the /proc file system. In a chroot environment,
354 // the system call returns 1.
355 static bool unsafe_chroot_detected = false;
356 static const char *unstable_chroot_error = "/proc file system not found.\n"
357 "Java may be unstable running multithreaded in a chroot "
358 "environment on Linux when /proc filesystem is not mounted.";
359
360 void os::Linux::initialize_system_info() {
361 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
362 if (processor_count() == 1) {
363 pid_t pid = os::Linux::gettid();
364 char fname[32];
365 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
366 FILE *fp = fopen(fname, "r");
367 if (fp == NULL) {
368 unsafe_chroot_detected = true;
369 } else {
370 fclose(fp);
371 }
372 }
373 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
374 assert(processor_count() > 0, "linux error");
375 }
376
377 void os::init_system_properties_values() {
378 // The next steps are taken in the product version:
379 //
380 // Obtain the JAVA_HOME value from the location of libjvm.so.
381 // This library should be located at:
382 // <JAVA_HOME>/lib/{client|server}/libjvm.so.
383 //
384 // If "/jre/lib/" appears at the right place in the path, then we
385 // assume libjvm.so is installed in a JDK and we use this path.
386 //
387 // Otherwise exit with message: "Could not create the Java virtual machine."
388 //
389 // The following extra steps are taken in the debugging version:
390 //
391 // If "/jre/lib/" does NOT appear at the right place in the path
392 // instead of exit check for $JAVA_HOME environment variable.
393 //
394 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
395 // then we append a fake suffix "hotspot/libjvm.so" to this path so
396 // it looks like libjvm.so is installed there
397 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
398 //
399 // Otherwise exit.
400 //
401 // Important note: if the location of libjvm.so changes this
402 // code needs to be changed accordingly.
403
404 // See ld(1):
405 // The linker uses the following search paths to locate required
406 // shared libraries:
407 // 1: ...
408 // ...
409 // 7: The default directories, normally /lib and /usr/lib.
410 #ifndef OVERRIDE_LIBPATH
411 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
412 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
413 #else
414 #define DEFAULT_LIBPATH "/lib:/usr/lib"
415 #endif
416 #else
417 #define DEFAULT_LIBPATH OVERRIDE_LIBPATH
418 #endif
419
420 // Base path of extensions installed on the system.
421 #define SYS_EXT_DIR "/usr/java/packages"
422 #define EXTENSIONS_DIR "/lib/ext"
423
424 // Buffer that fits several sprintfs.
425 // Note that the space for the colon and the trailing null are provided
426 // by the nulls included by the sizeof operator.
427 const size_t bufsize =
428 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
429 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
430 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
431
432 // sysclasspath, java_home, dll_dir
433 {
434 char *pslash;
435 os::jvm_path(buf, bufsize);
436
437 // Found the full path to libjvm.so.
438 // Now cut the path to <java_home>/jre if we can.
439 pslash = strrchr(buf, '/');
440 if (pslash != NULL) {
441 *pslash = '\0'; // Get rid of /libjvm.so.
442 }
443 pslash = strrchr(buf, '/');
444 if (pslash != NULL) {
445 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
446 }
447 Arguments::set_dll_dir(buf);
448
449 if (pslash != NULL) {
450 pslash = strrchr(buf, '/');
451 if (pslash != NULL) {
452 *pslash = '\0'; // Get rid of /lib.
453 }
454 }
455 Arguments::set_java_home(buf);
456 if (!set_boot_path('/', ':')) {
457 vm_exit_during_initialization("Failed setting boot class path.", NULL);
458 }
459 }
460
461 // Where to look for native libraries.
462 //
463 // Note: Due to a legacy implementation, most of the library path
464 // is set in the launcher. This was to accomodate linking restrictions
465 // on legacy Linux implementations (which are no longer supported).
466 // Eventually, all the library path setting will be done here.
467 //
468 // However, to prevent the proliferation of improperly built native
469 // libraries, the new path component /usr/java/packages is added here.
470 // Eventually, all the library path setting will be done here.
471 {
472 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
473 // should always exist (until the legacy problem cited above is
474 // addressed).
475 const char *v = ::getenv("LD_LIBRARY_PATH");
476 const char *v_colon = ":";
477 if (v == NULL) { v = ""; v_colon = ""; }
478 // That's +1 for the colon and +1 for the trailing '\0'.
479 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
480 strlen(v) + 1 +
481 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
482 mtInternal);
483 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
484 Arguments::set_library_path(ld_library_path);
485 FREE_C_HEAP_ARRAY(char, ld_library_path);
486 }
487
488 // Extensions directories.
489 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
490 Arguments::set_ext_dirs(buf);
491
492 FREE_C_HEAP_ARRAY(char, buf);
493
494 #undef DEFAULT_LIBPATH
495 #undef SYS_EXT_DIR
496 #undef EXTENSIONS_DIR
497 }
498
499 ////////////////////////////////////////////////////////////////////////////////
500 // breakpoint support
501
502 void os::breakpoint() {
503 BREAKPOINT;
504 }
505
506 extern "C" void breakpoint() {
507 // use debugger to set breakpoint here
508 }
509
510 ////////////////////////////////////////////////////////////////////////////////
511 // signal support
512
513 debug_only(static bool signal_sets_initialized = false);
514 static sigset_t unblocked_sigs, vm_sigs;
515
516 void os::Linux::signal_sets_init() {
517 // Should also have an assertion stating we are still single-threaded.
518 assert(!signal_sets_initialized, "Already initialized");
519 // Fill in signals that are necessarily unblocked for all threads in
520 // the VM. Currently, we unblock the following signals:
521 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
522 // by -Xrs (=ReduceSignalUsage));
523 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
524 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
525 // the dispositions or masks wrt these signals.
526 // Programs embedding the VM that want to use the above signals for their
527 // own purposes must, at this time, use the "-Xrs" option to prevent
528 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
529 // (See bug 4345157, and other related bugs).
530 // In reality, though, unblocking these signals is really a nop, since
531 // these signals are not blocked by default.
532 sigemptyset(&unblocked_sigs);
533 sigaddset(&unblocked_sigs, SIGILL);
534 sigaddset(&unblocked_sigs, SIGSEGV);
535 sigaddset(&unblocked_sigs, SIGBUS);
536 sigaddset(&unblocked_sigs, SIGFPE);
537 #if defined(PPC64)
538 sigaddset(&unblocked_sigs, SIGTRAP);
539 #endif
540 sigaddset(&unblocked_sigs, SR_signum);
541
542 if (!ReduceSignalUsage) {
543 if (!os::Posix::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
544 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
545 }
546 if (!os::Posix::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
547 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
548 }
549 if (!os::Posix::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
550 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
551 }
552 }
553 // Fill in signals that are blocked by all but the VM thread.
554 sigemptyset(&vm_sigs);
555 if (!ReduceSignalUsage) {
556 sigaddset(&vm_sigs, BREAK_SIGNAL);
557 }
558 debug_only(signal_sets_initialized = true);
559
560 }
561
562 // These are signals that are unblocked while a thread is running Java.
563 // (For some reason, they get blocked by default.)
564 sigset_t* os::Linux::unblocked_signals() {
565 assert(signal_sets_initialized, "Not initialized");
566 return &unblocked_sigs;
567 }
568
569 // These are the signals that are blocked while a (non-VM) thread is
570 // running Java. Only the VM thread handles these signals.
571 sigset_t* os::Linux::vm_signals() {
572 assert(signal_sets_initialized, "Not initialized");
573 return &vm_sigs;
574 }
575
576 void os::Linux::hotspot_sigmask(Thread* thread) {
577
578 //Save caller's signal mask before setting VM signal mask
579 sigset_t caller_sigmask;
580 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
581
582 OSThread* osthread = thread->osthread();
583 osthread->set_caller_sigmask(caller_sigmask);
584
585 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
586
587 if (!ReduceSignalUsage) {
588 if (thread->is_VM_thread()) {
589 // Only the VM thread handles BREAK_SIGNAL ...
590 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
591 } else {
592 // ... all other threads block BREAK_SIGNAL
593 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
594 }
595 }
596 }
597
598 //////////////////////////////////////////////////////////////////////////////
599 // detecting pthread library
600
601 void os::Linux::libpthread_init() {
602 // Save glibc and pthread version strings.
603 #if !defined(_CS_GNU_LIBC_VERSION) || \
604 !defined(_CS_GNU_LIBPTHREAD_VERSION)
605 #error "glibc too old (< 2.3.2)"
606 #endif
607
608 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
609 assert(n > 0, "cannot retrieve glibc version");
610 char *str = (char *)malloc(n, mtInternal);
611 confstr(_CS_GNU_LIBC_VERSION, str, n);
612 os::Linux::set_glibc_version(str);
613
614 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
615 assert(n > 0, "cannot retrieve pthread version");
616 str = (char *)malloc(n, mtInternal);
617 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
618 os::Linux::set_libpthread_version(str);
619 }
620
621 /////////////////////////////////////////////////////////////////////////////
622 // thread stack expansion
623
624 // os::Linux::manually_expand_stack() takes care of expanding the thread
625 // stack. Note that this is normally not needed: pthread stacks allocate
626 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
627 // committed. Therefore it is not necessary to expand the stack manually.
628 //
629 // Manually expanding the stack was historically needed on LinuxThreads
630 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
631 // it is kept to deal with very rare corner cases:
632 //
633 // For one, user may run the VM on an own implementation of threads
634 // whose stacks are - like the old LinuxThreads - implemented using
635 // mmap(MAP_GROWSDOWN).
636 //
637 // Also, this coding may be needed if the VM is running on the primordial
638 // thread. Normally we avoid running on the primordial thread; however,
639 // user may still invoke the VM on the primordial thread.
640 //
641 // The following historical comment describes the details about running
642 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
643
644
645 // Force Linux kernel to expand current thread stack. If "bottom" is close
646 // to the stack guard, caller should block all signals.
647 //
648 // MAP_GROWSDOWN:
649 // A special mmap() flag that is used to implement thread stacks. It tells
650 // kernel that the memory region should extend downwards when needed. This
651 // allows early versions of LinuxThreads to only mmap the first few pages
652 // when creating a new thread. Linux kernel will automatically expand thread
653 // stack as needed (on page faults).
654 //
655 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
656 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
657 // region, it's hard to tell if the fault is due to a legitimate stack
658 // access or because of reading/writing non-exist memory (e.g. buffer
659 // overrun). As a rule, if the fault happens below current stack pointer,
660 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
661 // application (see Linux kernel fault.c).
662 //
663 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
664 // stack overflow detection.
665 //
666 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
667 // not use MAP_GROWSDOWN.
668 //
669 // To get around the problem and allow stack banging on Linux, we need to
670 // manually expand thread stack after receiving the SIGSEGV.
671 //
672 // There are two ways to expand thread stack to address "bottom", we used
673 // both of them in JVM before 1.5:
674 // 1. adjust stack pointer first so that it is below "bottom", and then
675 // touch "bottom"
676 // 2. mmap() the page in question
677 //
678 // Now alternate signal stack is gone, it's harder to use 2. For instance,
679 // if current sp is already near the lower end of page 101, and we need to
680 // call mmap() to map page 100, it is possible that part of the mmap() frame
681 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
682 // That will destroy the mmap() frame and cause VM to crash.
683 //
684 // The following code works by adjusting sp first, then accessing the "bottom"
685 // page to force a page fault. Linux kernel will then automatically expand the
686 // stack mapping.
687 //
688 // _expand_stack_to() assumes its frame size is less than page size, which
689 // should always be true if the function is not inlined.
690
691 static void NOINLINE _expand_stack_to(address bottom) {
692 address sp;
693 size_t size;
694 volatile char *p;
695
696 // Adjust bottom to point to the largest address within the same page, it
697 // gives us a one-page buffer if alloca() allocates slightly more memory.
698 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
699 bottom += os::Linux::page_size() - 1;
700
701 // sp might be slightly above current stack pointer; if that's the case, we
702 // will alloca() a little more space than necessary, which is OK. Don't use
703 // os::current_stack_pointer(), as its result can be slightly below current
704 // stack pointer, causing us to not alloca enough to reach "bottom".
705 sp = (address)&sp;
706
707 if (sp > bottom) {
708 size = sp - bottom;
709 p = (volatile char *)alloca(size);
710 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
711 p[0] = '\0';
712 }
713 }
714
715 void os::Linux::expand_stack_to(address bottom) {
716 _expand_stack_to(bottom);
717 }
718
719 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
720 assert(t!=NULL, "just checking");
721 assert(t->osthread()->expanding_stack(), "expand should be set");
722 assert(t->stack_base() != NULL, "stack_base was not initialized");
723
724 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
725 sigset_t mask_all, old_sigset;
726 sigfillset(&mask_all);
727 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
728 _expand_stack_to(addr);
729 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
730 return true;
731 }
732 return false;
733 }
734
735 //////////////////////////////////////////////////////////////////////////////
736 // create new thread
737
738 // Thread start routine for all newly created threads
739 static void *thread_native_entry(Thread *thread) {
740
741 thread->record_stack_base_and_size();
742
743 // Try to randomize the cache line index of hot stack frames.
744 // This helps when threads of the same stack traces evict each other's
745 // cache lines. The threads can be either from the same JVM instance, or
746 // from different JVM instances. The benefit is especially true for
747 // processors with hyperthreading technology.
748 static int counter = 0;
749 int pid = os::current_process_id();
750 alloca(((pid ^ counter++) & 7) * 128);
751
752 thread->initialize_thread_current();
753
754 OSThread* osthread = thread->osthread();
755 Monitor* sync = osthread->startThread_lock();
756
757 osthread->set_thread_id(os::current_thread_id());
758
759 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
760 os::current_thread_id(), (uintx) pthread_self());
761
762 if (UseNUMA) {
763 int lgrp_id = os::numa_get_group_id();
764 if (lgrp_id != -1) {
765 thread->set_lgrp_id(lgrp_id);
766 }
767 }
768 // initialize signal mask for this thread
769 os::Linux::hotspot_sigmask(thread);
770
771 // initialize floating point control register
772 os::Linux::init_thread_fpu_state();
773
774 // handshaking with parent thread
775 {
776 MutexLocker ml(sync, Mutex::_no_safepoint_check_flag);
777
778 // notify parent thread
779 osthread->set_state(INITIALIZED);
780 sync->notify_all();
781
782 // wait until os::start_thread()
783 while (osthread->get_state() == INITIALIZED) {
784 sync->wait_without_safepoint_check();
785 }
786 }
787
788 assert(osthread->pthread_id() != 0, "pthread_id was not set as expected");
789
790 // call one more level start routine
791 thread->call_run();
792
793 // Note: at this point the thread object may already have deleted itself.
794 // Prevent dereferencing it from here on out.
795 thread = NULL;
796
797 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
798 os::current_thread_id(), (uintx) pthread_self());
799
800 return 0;
801 }
802
803 #ifdef __GLIBC__
804 // The code below is used to figure out the size of the TLS space used
805 // by the main executable (DSO's TLS space is allocated separately).
806 // TLS variables live in the stack space allocated for each thread, so
807 // the function is used to automatically adjust requested stack sizes.
808 // This code only works with glibc since it relies on an internal symbol
809 // in glibc. When doing a profile collection run with FDO, the
810 // compiler will define several TLS variables per object file (making
811 // this function call frequently necessary during FDO collection).
812 // p_dl_get_tls_static_info is initialized in InitGetTLSSize().
813 //
814 // The goal is to make sure that stacks have enough space for TLS.
815 // There is an upstream bug to make this happen in glibc:
816 // http://sourceware.org/bugzilla/show_bug.cgi?id=11787
817 // This code can presumably be retired when that fix is available
818 // pervasively.
819 extern "C" void _dl_get_tls_static_info(size_t*, size_t*) __attribute__((weak));
820
821 typedef void (*GetTLSType)(size_t*, size_t*);
822
823 GetTLSType p_dl_get_tls_static_info = NULL;
824
825 static void InitGetTLSSize() {
826 // This complicated sequence is used to support both static and
827 // dynamic linking when using the gold linker.
828 // _dl_get_tls_static_info is defined in libc.a and ld-linux.so.
829 // When linking statically, the symbol is available. When linking
830 // dynamically, the symbol is *not* available because gold does not
831 // consider ld-linux.so to be part of the link unless explicitly
832 // specified. The code below uses the static symbol if available.
833 // If not, it will use dlsym to search for the symbol at runtime.
834 p_dl_get_tls_static_info = &_dl_get_tls_static_info;
835 if (p_dl_get_tls_static_info == NULL) {
836 p_dl_get_tls_static_info =
837 (GetTLSType)dlsym(RTLD_NEXT, "_dl_get_tls_static_info");
838 }
839 }
840
841 static size_t GetTLSSize() {
842 size_t tls_size = 0;
843 size_t tls_align;
844
845 if (p_dl_get_tls_static_info) {
846 (*p_dl_get_tls_static_info)(&tls_size, &tls_align);
847 }
848 return tls_size;
849 }
850 #else
851 static void InitGetTLSSize() {
852 }
853 static size_t GetTLSSize() {
854 return 0;
855 }
856 #endif
857
858 bool os::create_thread(Thread* thread, ThreadType thr_type,
859 size_t req_stack_size) {
860 assert(thread->osthread() == NULL, "caller responsible");
861
862 // Allocate the OSThread object
863 OSThread* osthread = new OSThread(NULL, NULL);
864 if (osthread == NULL) {
865 return false;
866 }
867
868 // set the correct thread state
869 osthread->set_thread_type(thr_type);
870
871 // Initial state is ALLOCATED but not INITIALIZED
872 osthread->set_state(ALLOCATED);
873
874 thread->set_osthread(osthread);
875
876 // init thread attributes
877 pthread_attr_t attr;
878 pthread_attr_init(&attr);
879 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
880
881 // Calculate stack size if it's not specified by caller.
882 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
883 // In the Linux NPTL pthread implementation the guard size mechanism
884 // is not implemented properly. The posix standard requires adding
885 // the size of the guard pages to the stack size, instead Linux
886 // takes the space out of 'stacksize'. Thus we adapt the requested
887 // stack_size by the size of the guard pages to mimick proper
888 // behaviour. However, be careful not to end up with a size
889 // of zero due to overflow. Don't add the guard page in that case.
890 size_t guard_size = os::Linux::default_guard_size(thr_type);
891 if (stack_size <= SIZE_MAX - guard_size) {
892 stack_size += guard_size;
893 }
894 // Always, always, always give the user more for TLS, even if they
895 // haven't asked for it.
896 stack_size += GetTLSSize();
897 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
898
899 int status = pthread_attr_setstacksize(&attr, stack_size);
900 assert_status(status == 0, status, "pthread_attr_setstacksize");
901
902 // Configure glibc guard page.
903 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
904
905 ThreadState state;
906
907 {
908 pthread_t tid;
909 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
910
911 char buf[64];
912 if (ret == 0) {
913 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
914 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
915 } else {
916 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
917 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
918 // Log some OS information which might explain why creating the thread failed.
919 log_info(os, thread)("Number of threads approx. running in the VM: %d", Threads::number_of_threads());
920 LogStream st(Log(os, thread)::info());
921 os::Posix::print_rlimit_info(&st);
922 os::print_memory_info(&st);
923 os::Linux::print_proc_sys_info(&st);
924 os::Linux::print_container_info(&st);
925 }
926
927 pthread_attr_destroy(&attr);
928
929 if (ret != 0) {
930 // Need to clean up stuff we've allocated so far
931 thread->set_osthread(NULL);
932 delete osthread;
933 return false;
934 }
935
936 // Store pthread info into the OSThread
937 osthread->set_pthread_id(tid);
938
939 // Wait until child thread is either initialized or aborted
940 {
941 Monitor* sync_with_child = osthread->startThread_lock();
942 MutexLocker ml(sync_with_child, Mutex::_no_safepoint_check_flag);
943 while ((state = osthread->get_state()) == ALLOCATED) {
944 sync_with_child->wait_without_safepoint_check();
945 }
946 }
947 }
948
949 // Aborted due to thread limit being reached
950 if (state == ZOMBIE) {
951 thread->set_osthread(NULL);
952 delete osthread;
953 return false;
954 }
955
956 // The thread is returned suspended (in state INITIALIZED),
957 // and is started higher up in the call chain
958 assert(state == INITIALIZED, "race condition");
959 return true;
960 }
961
962 /////////////////////////////////////////////////////////////////////////////
963 // attach existing thread
964
965 // bootstrap the main thread
966 bool os::create_main_thread(JavaThread* thread) {
967 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
968 return create_attached_thread(thread);
969 }
970
971 bool os::create_attached_thread(JavaThread* thread) {
972 #ifdef ASSERT
973 thread->verify_not_published();
974 #endif
975
976 // Allocate the OSThread object
977 OSThread* osthread = new OSThread(NULL, NULL);
978
979 if (osthread == NULL) {
980 return false;
981 }
982
983 // Store pthread info into the OSThread
984 osthread->set_thread_id(os::Linux::gettid());
985 osthread->set_pthread_id(::pthread_self());
986
987 // initialize floating point control register
988 os::Linux::init_thread_fpu_state();
989
990 // Initial thread state is RUNNABLE
991 osthread->set_state(RUNNABLE);
992
993 thread->set_osthread(osthread);
994
995 if (UseNUMA) {
996 int lgrp_id = os::numa_get_group_id();
997 if (lgrp_id != -1) {
998 thread->set_lgrp_id(lgrp_id);
999 }
1000 }
1001
1002 if (os::is_primordial_thread()) {
1003 // If current thread is primordial thread, its stack is mapped on demand,
1004 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1005 // the entire stack region to avoid SEGV in stack banging.
1006 // It is also useful to get around the heap-stack-gap problem on SuSE
1007 // kernel (see 4821821 for details). We first expand stack to the top
1008 // of yellow zone, then enable stack yellow zone (order is significant,
1009 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1010 // is no gap between the last two virtual memory regions.
1011
1012 JavaThread *jt = (JavaThread *)thread;
1013 address addr = jt->stack_reserved_zone_base();
1014 assert(addr != NULL, "initialization problem?");
1015 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1016
1017 osthread->set_expanding_stack();
1018 os::Linux::manually_expand_stack(jt, addr);
1019 osthread->clear_expanding_stack();
1020 }
1021
1022 // initialize signal mask for this thread
1023 // and save the caller's signal mask
1024 os::Linux::hotspot_sigmask(thread);
1025
1026 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
1027 os::current_thread_id(), (uintx) pthread_self());
1028
1029 return true;
1030 }
1031
1032 void os::pd_start_thread(Thread* thread) {
1033 OSThread * osthread = thread->osthread();
1034 assert(osthread->get_state() != INITIALIZED, "just checking");
1035 Monitor* sync_with_child = osthread->startThread_lock();
1036 MutexLocker ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1037 sync_with_child->notify();
1038 }
1039
1040 // Free Linux resources related to the OSThread
1041 void os::free_thread(OSThread* osthread) {
1042 assert(osthread != NULL, "osthread not set");
1043
1044 // We are told to free resources of the argument thread,
1045 // but we can only really operate on the current thread.
1046 assert(Thread::current()->osthread() == osthread,
1047 "os::free_thread but not current thread");
1048
1049 #ifdef ASSERT
1050 sigset_t current;
1051 sigemptyset(¤t);
1052 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
1053 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
1054 #endif
1055
1056 // Restore caller's signal mask
1057 sigset_t sigmask = osthread->caller_sigmask();
1058 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1059
1060 delete osthread;
1061 }
1062
1063 //////////////////////////////////////////////////////////////////////////////
1064 // primordial thread
1065
1066 // Check if current thread is the primordial thread, similar to Solaris thr_main.
1067 bool os::is_primordial_thread(void) {
1068 if (suppress_primordial_thread_resolution) {
1069 return false;
1070 }
1071 char dummy;
1072 // If called before init complete, thread stack bottom will be null.
1073 // Can be called if fatal error occurs before initialization.
1074 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
1075 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
1076 os::Linux::initial_thread_stack_size() != 0,
1077 "os::init did not locate primordial thread's stack region");
1078 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
1079 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
1080 os::Linux::initial_thread_stack_size()) {
1081 return true;
1082 } else {
1083 return false;
1084 }
1085 }
1086
1087 // Find the virtual memory area that contains addr
1088 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1089 FILE *fp = fopen("/proc/self/maps", "r");
1090 if (fp) {
1091 address low, high;
1092 while (!feof(fp)) {
1093 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1094 if (low <= addr && addr < high) {
1095 if (vma_low) *vma_low = low;
1096 if (vma_high) *vma_high = high;
1097 fclose(fp);
1098 return true;
1099 }
1100 }
1101 for (;;) {
1102 int ch = fgetc(fp);
1103 if (ch == EOF || ch == (int)'\n') break;
1104 }
1105 }
1106 fclose(fp);
1107 }
1108 return false;
1109 }
1110
1111 // Locate primordial thread stack. This special handling of primordial thread stack
1112 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1113 // bogus value for the primordial process thread. While the launcher has created
1114 // the VM in a new thread since JDK 6, we still have to allow for the use of the
1115 // JNI invocation API from a primordial thread.
1116 void os::Linux::capture_initial_stack(size_t max_size) {
1117
1118 // max_size is either 0 (which means accept OS default for thread stacks) or
1119 // a user-specified value known to be at least the minimum needed. If we
1120 // are actually on the primordial thread we can make it appear that we have a
1121 // smaller max_size stack by inserting the guard pages at that location. But we
1122 // cannot do anything to emulate a larger stack than what has been provided by
1123 // the OS or threading library. In fact if we try to use a stack greater than
1124 // what is set by rlimit then we will crash the hosting process.
1125
1126 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
1127 // If this is "unlimited" then it will be a huge value.
1128 struct rlimit rlim;
1129 getrlimit(RLIMIT_STACK, &rlim);
1130 size_t stack_size = rlim.rlim_cur;
1131
1132 // 6308388: a bug in ld.so will relocate its own .data section to the
1133 // lower end of primordial stack; reduce ulimit -s value a little bit
1134 // so we won't install guard page on ld.so's data section.
1135 // But ensure we don't underflow the stack size - allow 1 page spare
1136 if (stack_size >= (size_t)(3 * page_size())) {
1137 stack_size -= 2 * page_size();
1138 }
1139
1140 // Try to figure out where the stack base (top) is. This is harder.
1141 //
1142 // When an application is started, glibc saves the initial stack pointer in
1143 // a global variable "__libc_stack_end", which is then used by system
1144 // libraries. __libc_stack_end should be pretty close to stack top. The
1145 // variable is available since the very early days. However, because it is
1146 // a private interface, it could disappear in the future.
1147 //
1148 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1149 // to __libc_stack_end, it is very close to stack top, but isn't the real
1150 // stack top. Note that /proc may not exist if VM is running as a chroot
1151 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1152 // /proc/<pid>/stat could change in the future (though unlikely).
1153 //
1154 // We try __libc_stack_end first. If that doesn't work, look for
1155 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1156 // as a hint, which should work well in most cases.
1157
1158 uintptr_t stack_start;
1159
1160 // try __libc_stack_end first
1161 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1162 if (p && *p) {
1163 stack_start = *p;
1164 } else {
1165 // see if we can get the start_stack field from /proc/self/stat
1166 FILE *fp;
1167 int pid;
1168 char state;
1169 int ppid;
1170 int pgrp;
1171 int session;
1172 int nr;
1173 int tpgrp;
1174 unsigned long flags;
1175 unsigned long minflt;
1176 unsigned long cminflt;
1177 unsigned long majflt;
1178 unsigned long cmajflt;
1179 unsigned long utime;
1180 unsigned long stime;
1181 long cutime;
1182 long cstime;
1183 long prio;
1184 long nice;
1185 long junk;
1186 long it_real;
1187 uintptr_t start;
1188 uintptr_t vsize;
1189 intptr_t rss;
1190 uintptr_t rsslim;
1191 uintptr_t scodes;
1192 uintptr_t ecode;
1193 int i;
1194
1195 // Figure what the primordial thread stack base is. Code is inspired
1196 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1197 // followed by command name surrounded by parentheses, state, etc.
1198 char stat[2048];
1199 int statlen;
1200
1201 fp = fopen("/proc/self/stat", "r");
1202 if (fp) {
1203 statlen = fread(stat, 1, 2047, fp);
1204 stat[statlen] = '\0';
1205 fclose(fp);
1206
1207 // Skip pid and the command string. Note that we could be dealing with
1208 // weird command names, e.g. user could decide to rename java launcher
1209 // to "java 1.4.2 :)", then the stat file would look like
1210 // 1234 (java 1.4.2 :)) R ... ...
1211 // We don't really need to know the command string, just find the last
1212 // occurrence of ")" and then start parsing from there. See bug 4726580.
1213 char * s = strrchr(stat, ')');
1214
1215 i = 0;
1216 if (s) {
1217 // Skip blank chars
1218 do { s++; } while (s && isspace(*s));
1219
1220 #define _UFM UINTX_FORMAT
1221 #define _DFM INTX_FORMAT
1222
1223 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1224 // 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
1225 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,
1226 &state, // 3 %c
1227 &ppid, // 4 %d
1228 &pgrp, // 5 %d
1229 &session, // 6 %d
1230 &nr, // 7 %d
1231 &tpgrp, // 8 %d
1232 &flags, // 9 %lu
1233 &minflt, // 10 %lu
1234 &cminflt, // 11 %lu
1235 &majflt, // 12 %lu
1236 &cmajflt, // 13 %lu
1237 &utime, // 14 %lu
1238 &stime, // 15 %lu
1239 &cutime, // 16 %ld
1240 &cstime, // 17 %ld
1241 &prio, // 18 %ld
1242 &nice, // 19 %ld
1243 &junk, // 20 %ld
1244 &it_real, // 21 %ld
1245 &start, // 22 UINTX_FORMAT
1246 &vsize, // 23 UINTX_FORMAT
1247 &rss, // 24 INTX_FORMAT
1248 &rsslim, // 25 UINTX_FORMAT
1249 &scodes, // 26 UINTX_FORMAT
1250 &ecode, // 27 UINTX_FORMAT
1251 &stack_start); // 28 UINTX_FORMAT
1252 }
1253
1254 #undef _UFM
1255 #undef _DFM
1256
1257 if (i != 28 - 2) {
1258 assert(false, "Bad conversion from /proc/self/stat");
1259 // product mode - assume we are the primordial thread, good luck in the
1260 // embedded case.
1261 warning("Can't detect primordial thread stack location - bad conversion");
1262 stack_start = (uintptr_t) &rlim;
1263 }
1264 } else {
1265 // For some reason we can't open /proc/self/stat (for example, running on
1266 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1267 // most cases, so don't abort:
1268 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1269 stack_start = (uintptr_t) &rlim;
1270 }
1271 }
1272
1273 // Now we have a pointer (stack_start) very close to the stack top, the
1274 // next thing to do is to figure out the exact location of stack top. We
1275 // can find out the virtual memory area that contains stack_start by
1276 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1277 // and its upper limit is the real stack top. (again, this would fail if
1278 // running inside chroot, because /proc may not exist.)
1279
1280 uintptr_t stack_top;
1281 address low, high;
1282 if (find_vma((address)stack_start, &low, &high)) {
1283 // success, "high" is the true stack top. (ignore "low", because initial
1284 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1285 stack_top = (uintptr_t)high;
1286 } else {
1287 // failed, likely because /proc/self/maps does not exist
1288 warning("Can't detect primordial thread stack location - find_vma failed");
1289 // best effort: stack_start is normally within a few pages below the real
1290 // stack top, use it as stack top, and reduce stack size so we won't put
1291 // guard page outside stack.
1292 stack_top = stack_start;
1293 stack_size -= 16 * page_size();
1294 }
1295
1296 // stack_top could be partially down the page so align it
1297 stack_top = align_up(stack_top, page_size());
1298
1299 // Allowed stack value is minimum of max_size and what we derived from rlimit
1300 if (max_size > 0) {
1301 _initial_thread_stack_size = MIN2(max_size, stack_size);
1302 } else {
1303 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1304 // clamp it at 8MB as we do on Solaris
1305 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1306 }
1307 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1308 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1309
1310 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1311
1312 if (log_is_enabled(Info, os, thread)) {
1313 // See if we seem to be on primordial process thread
1314 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1315 uintptr_t(&rlim) < stack_top;
1316
1317 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1318 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1319 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1320 stack_top, intptr_t(_initial_thread_stack_bottom));
1321 }
1322 }
1323
1324 ////////////////////////////////////////////////////////////////////////////////
1325 // time support
1326
1327 #ifndef SUPPORTS_CLOCK_MONOTONIC
1328 #error "Build platform doesn't support clock_gettime and related functionality"
1329 #endif
1330
1331 // Time since start-up in seconds to a fine granularity.
1332 // Used by VMSelfDestructTimer and the MemProfiler.
1333 double os::elapsedTime() {
1334
1335 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1336 }
1337
1338 jlong os::elapsed_counter() {
1339 return javaTimeNanos() - initial_time_count;
1340 }
1341
1342 jlong os::elapsed_frequency() {
1343 return NANOSECS_PER_SEC; // nanosecond resolution
1344 }
1345
1346 bool os::supports_vtime() { return true; }
1347 bool os::enable_vtime() { return false; }
1348 bool os::vtime_enabled() { return false; }
1349
1350 double os::elapsedVTime() {
1351 struct rusage usage;
1352 int retval = getrusage(RUSAGE_THREAD, &usage);
1353 if (retval == 0) {
1354 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);
1355 } else {
1356 // better than nothing, but not much
1357 return elapsedTime();
1358 }
1359 }
1360
1361 jlong os::javaTimeMillis() {
1362 timeval time;
1363 int status = gettimeofday(&time, NULL);
1364 assert(status != -1, "linux error");
1365 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1366 }
1367
1368 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1369 timeval time;
1370 int status = gettimeofday(&time, NULL);
1371 assert(status != -1, "linux error");
1372 seconds = jlong(time.tv_sec);
1373 nanos = jlong(time.tv_usec) * 1000;
1374 }
1375
1376 void os::Linux::fast_thread_clock_init() {
1377 if (!UseLinuxPosixThreadCPUClocks) {
1378 return;
1379 }
1380 clockid_t clockid;
1381 struct timespec tp;
1382 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1383 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1384
1385 // Switch to using fast clocks for thread cpu time if
1386 // the clock_getres() returns 0 error code.
1387 // Note, that some kernels may support the current thread
1388 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1389 // returned by the pthread_getcpuclockid().
1390 // If the fast Posix clocks are supported then the clock_getres()
1391 // must return at least tp.tv_sec == 0 which means a resolution
1392 // better than 1 sec. This is extra check for reliability.
1393
1394 if (pthread_getcpuclockid_func &&
1395 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1396 os::Posix::clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1397 _supports_fast_thread_cpu_time = true;
1398 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1399 }
1400 }
1401
1402 jlong os::javaTimeNanos() {
1403 if (os::supports_monotonic_clock()) {
1404 struct timespec tp;
1405 int status = os::Posix::clock_gettime(CLOCK_MONOTONIC, &tp);
1406 assert(status == 0, "gettime error");
1407 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1408 return result;
1409 } else {
1410 timeval time;
1411 int status = gettimeofday(&time, NULL);
1412 assert(status != -1, "linux error");
1413 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1414 return 1000 * usecs;
1415 }
1416 }
1417
1418 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1419 if (os::supports_monotonic_clock()) {
1420 info_ptr->max_value = ALL_64_BITS;
1421
1422 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1423 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1424 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1425 } else {
1426 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1427 info_ptr->max_value = ALL_64_BITS;
1428
1429 // gettimeofday is a real time clock so it skips
1430 info_ptr->may_skip_backward = true;
1431 info_ptr->may_skip_forward = true;
1432 }
1433
1434 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1435 }
1436
1437 // Return the real, user, and system times in seconds from an
1438 // arbitrary fixed point in the past.
1439 bool os::getTimesSecs(double* process_real_time,
1440 double* process_user_time,
1441 double* process_system_time) {
1442 struct tms ticks;
1443 clock_t real_ticks = times(&ticks);
1444
1445 if (real_ticks == (clock_t) (-1)) {
1446 return false;
1447 } else {
1448 double ticks_per_second = (double) clock_tics_per_sec;
1449 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1450 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1451 *process_real_time = ((double) real_ticks) / ticks_per_second;
1452
1453 return true;
1454 }
1455 }
1456
1457
1458 char * os::local_time_string(char *buf, size_t buflen) {
1459 struct tm t;
1460 time_t long_time;
1461 time(&long_time);
1462 localtime_r(&long_time, &t);
1463 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1464 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1465 t.tm_hour, t.tm_min, t.tm_sec);
1466 return buf;
1467 }
1468
1469 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1470 return localtime_r(clock, res);
1471 }
1472
1473 ////////////////////////////////////////////////////////////////////////////////
1474 // runtime exit support
1475
1476 // Note: os::shutdown() might be called very early during initialization, or
1477 // called from signal handler. Before adding something to os::shutdown(), make
1478 // sure it is async-safe and can handle partially initialized VM.
1479 void os::shutdown() {
1480
1481 // allow PerfMemory to attempt cleanup of any persistent resources
1482 perfMemory_exit();
1483
1484 // needs to remove object in file system
1485 AttachListener::abort();
1486
1487 // flush buffered output, finish log files
1488 ostream_abort();
1489
1490 // Check for abort hook
1491 abort_hook_t abort_hook = Arguments::abort_hook();
1492 if (abort_hook != NULL) {
1493 abort_hook();
1494 }
1495
1496 }
1497
1498 // Note: os::abort() might be called very early during initialization, or
1499 // called from signal handler. Before adding something to os::abort(), make
1500 // sure it is async-safe and can handle partially initialized VM.
1501 void os::abort(bool dump_core, void* siginfo, const void* context) {
1502 os::shutdown();
1503 if (dump_core) {
1504 if (DumpPrivateMappingsInCore) {
1505 ClassLoader::close_jrt_image();
1506 }
1507 #ifndef PRODUCT
1508 fdStream out(defaultStream::output_fd());
1509 out.print_raw("Current thread is ");
1510 char buf[16];
1511 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1512 out.print_raw_cr(buf);
1513 out.print_raw_cr("Dumping core ...");
1514 #endif
1515 ::abort(); // dump core
1516 }
1517
1518 ::exit(1);
1519 }
1520
1521 // Die immediately, no exit hook, no abort hook, no cleanup.
1522 void os::die() {
1523 ::abort();
1524 }
1525
1526 // thread_id is kernel thread id (similar to Solaris LWP id)
1527 intx os::current_thread_id() { return os::Linux::gettid(); }
1528 int os::current_process_id() {
1529 return ::getpid();
1530 }
1531
1532 // DLL functions
1533
1534 const char* os::dll_file_extension() { return ".so"; }
1535
1536 // This must be hard coded because it's the system's temporary
1537 // directory not the java application's temp directory, ala java.io.tmpdir.
1538 const char* os::get_temp_directory() { return "/tmp"; }
1539
1540 static bool file_exists(const char* filename) {
1541 struct stat statbuf;
1542 if (filename == NULL || strlen(filename) == 0) {
1543 return false;
1544 }
1545 return os::stat(filename, &statbuf) == 0;
1546 }
1547
1548 // check if addr is inside libjvm.so
1549 bool os::address_is_in_vm(address addr) {
1550 static address libjvm_base_addr;
1551 Dl_info dlinfo;
1552
1553 if (libjvm_base_addr == NULL) {
1554 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1555 libjvm_base_addr = (address)dlinfo.dli_fbase;
1556 }
1557 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1558 }
1559
1560 if (dladdr((void *)addr, &dlinfo) != 0) {
1561 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1562 }
1563
1564 return false;
1565 }
1566
1567 bool os::dll_address_to_function_name(address addr, char *buf,
1568 int buflen, int *offset,
1569 bool demangle) {
1570 // buf is not optional, but offset is optional
1571 assert(buf != NULL, "sanity check");
1572
1573 Dl_info dlinfo;
1574
1575 if (dladdr((void*)addr, &dlinfo) != 0) {
1576 // see if we have a matching symbol
1577 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1578 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1579 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1580 }
1581 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1582 return true;
1583 }
1584 // no matching symbol so try for just file info
1585 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1586 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1587 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1588 return true;
1589 }
1590 }
1591 }
1592
1593 buf[0] = '\0';
1594 if (offset != NULL) *offset = -1;
1595 return false;
1596 }
1597
1598 struct _address_to_library_name {
1599 address addr; // input : memory address
1600 size_t buflen; // size of fname
1601 char* fname; // output: library name
1602 address base; // library base addr
1603 };
1604
1605 static int address_to_library_name_callback(struct dl_phdr_info *info,
1606 size_t size, void *data) {
1607 int i;
1608 bool found = false;
1609 address libbase = NULL;
1610 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1611
1612 // iterate through all loadable segments
1613 for (i = 0; i < info->dlpi_phnum; i++) {
1614 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1615 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1616 // base address of a library is the lowest address of its loaded
1617 // segments.
1618 if (libbase == NULL || libbase > segbase) {
1619 libbase = segbase;
1620 }
1621 // see if 'addr' is within current segment
1622 if (segbase <= d->addr &&
1623 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1624 found = true;
1625 }
1626 }
1627 }
1628
1629 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1630 // so dll_address_to_library_name() can fall through to use dladdr() which
1631 // can figure out executable name from argv[0].
1632 if (found && info->dlpi_name && info->dlpi_name[0]) {
1633 d->base = libbase;
1634 if (d->fname) {
1635 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1636 }
1637 return 1;
1638 }
1639 return 0;
1640 }
1641
1642 bool os::dll_address_to_library_name(address addr, char* buf,
1643 int buflen, int* offset) {
1644 // buf is not optional, but offset is optional
1645 assert(buf != NULL, "sanity check");
1646
1647 Dl_info dlinfo;
1648 struct _address_to_library_name data;
1649
1650 // There is a bug in old glibc dladdr() implementation that it could resolve
1651 // to wrong library name if the .so file has a base address != NULL. Here
1652 // we iterate through the program headers of all loaded libraries to find
1653 // out which library 'addr' really belongs to. This workaround can be
1654 // removed once the minimum requirement for glibc is moved to 2.3.x.
1655 data.addr = addr;
1656 data.fname = buf;
1657 data.buflen = buflen;
1658 data.base = NULL;
1659 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1660
1661 if (rslt) {
1662 // buf already contains library name
1663 if (offset) *offset = addr - data.base;
1664 return true;
1665 }
1666 if (dladdr((void*)addr, &dlinfo) != 0) {
1667 if (dlinfo.dli_fname != NULL) {
1668 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1669 }
1670 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1671 *offset = addr - (address)dlinfo.dli_fbase;
1672 }
1673 return true;
1674 }
1675
1676 buf[0] = '\0';
1677 if (offset) *offset = -1;
1678 return false;
1679 }
1680
1681 // Loads .dll/.so and
1682 // in case of error it checks if .dll/.so was built for the
1683 // same architecture as Hotspot is running on
1684
1685
1686 // Remember the stack's state. The Linux dynamic linker will change
1687 // the stack to 'executable' at most once, so we must safepoint only once.
1688 bool os::Linux::_stack_is_executable = false;
1689
1690 // VM operation that loads a library. This is necessary if stack protection
1691 // of the Java stacks can be lost during loading the library. If we
1692 // do not stop the Java threads, they can stack overflow before the stacks
1693 // are protected again.
1694 class VM_LinuxDllLoad: public VM_Operation {
1695 private:
1696 const char *_filename;
1697 char *_ebuf;
1698 int _ebuflen;
1699 void *_lib;
1700 public:
1701 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1702 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1703 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1704 void doit() {
1705 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1706 os::Linux::_stack_is_executable = true;
1707 }
1708 void* loaded_library() { return _lib; }
1709 };
1710
1711 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1712 void * result = NULL;
1713 bool load_attempted = false;
1714
1715 // Check whether the library to load might change execution rights
1716 // of the stack. If they are changed, the protection of the stack
1717 // guard pages will be lost. We need a safepoint to fix this.
1718 //
1719 // See Linux man page execstack(8) for more info.
1720 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1721 if (!ElfFile::specifies_noexecstack(filename)) {
1722 if (!is_init_completed()) {
1723 os::Linux::_stack_is_executable = true;
1724 // This is OK - No Java threads have been created yet, and hence no
1725 // stack guard pages to fix.
1726 //
1727 // Dynamic loader will make all stacks executable after
1728 // this function returns, and will not do that again.
1729 assert(Threads::number_of_threads() == 0, "no Java threads should exist yet.");
1730 } else {
1731 warning("You have loaded library %s which might have disabled stack guard. "
1732 "The VM will try to fix the stack guard now.\n"
1733 "It's highly recommended that you fix the library with "
1734 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1735 filename);
1736
1737 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1738 JavaThread *jt = JavaThread::current();
1739 if (jt->thread_state() != _thread_in_native) {
1740 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1741 // that requires ExecStack. Cannot enter safe point. Let's give up.
1742 warning("Unable to fix stack guard. Giving up.");
1743 } else {
1744 if (!LoadExecStackDllInVMThread) {
1745 // This is for the case where the DLL has an static
1746 // constructor function that executes JNI code. We cannot
1747 // load such DLLs in the VMThread.
1748 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1749 }
1750
1751 ThreadInVMfromNative tiv(jt);
1752 debug_only(VMNativeEntryWrapper vew;)
1753
1754 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1755 VMThread::execute(&op);
1756 if (LoadExecStackDllInVMThread) {
1757 result = op.loaded_library();
1758 }
1759 load_attempted = true;
1760 }
1761 }
1762 }
1763 }
1764
1765 if (!load_attempted) {
1766 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1767 }
1768
1769 if (result != NULL) {
1770 // Successful loading
1771 return result;
1772 }
1773
1774 Elf32_Ehdr elf_head;
1775 int diag_msg_max_length=ebuflen-strlen(ebuf);
1776 char* diag_msg_buf=ebuf+strlen(ebuf);
1777
1778 if (diag_msg_max_length==0) {
1779 // No more space in ebuf for additional diagnostics message
1780 return NULL;
1781 }
1782
1783
1784 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1785
1786 if (file_descriptor < 0) {
1787 // Can't open library, report dlerror() message
1788 return NULL;
1789 }
1790
1791 bool failed_to_read_elf_head=
1792 (sizeof(elf_head)!=
1793 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1794
1795 ::close(file_descriptor);
1796 if (failed_to_read_elf_head) {
1797 // file i/o error - report dlerror() msg
1798 return NULL;
1799 }
1800
1801 typedef struct {
1802 Elf32_Half code; // Actual value as defined in elf.h
1803 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1804 unsigned char elf_class; // 32 or 64 bit
1805 unsigned char endianess; // MSB or LSB
1806 char* name; // String representation
1807 } arch_t;
1808
1809 #ifndef EM_486
1810 #define EM_486 6 /* Intel 80486 */
1811 #endif
1812 #ifndef EM_AARCH64
1813 #define EM_AARCH64 183 /* ARM AARCH64 */
1814 #endif
1815
1816 static const arch_t arch_array[]={
1817 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1818 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1819 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1820 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1821 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1822 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1823 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1824 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1825 #if defined(VM_LITTLE_ENDIAN)
1826 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1827 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1828 #else
1829 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1830 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1831 #endif
1832 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1833 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1834 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1835 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1836 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1837 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1838 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1839 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1840 };
1841
1842 #if (defined IA32)
1843 static Elf32_Half running_arch_code=EM_386;
1844 #elif (defined AMD64) || (defined X32)
1845 static Elf32_Half running_arch_code=EM_X86_64;
1846 #elif (defined IA64)
1847 static Elf32_Half running_arch_code=EM_IA_64;
1848 #elif (defined __sparc) && (defined _LP64)
1849 static Elf32_Half running_arch_code=EM_SPARCV9;
1850 #elif (defined __sparc) && (!defined _LP64)
1851 static Elf32_Half running_arch_code=EM_SPARC;
1852 #elif (defined __powerpc64__)
1853 static Elf32_Half running_arch_code=EM_PPC64;
1854 #elif (defined __powerpc__)
1855 static Elf32_Half running_arch_code=EM_PPC;
1856 #elif (defined AARCH64)
1857 static Elf32_Half running_arch_code=EM_AARCH64;
1858 #elif (defined ARM)
1859 static Elf32_Half running_arch_code=EM_ARM;
1860 #elif (defined S390)
1861 static Elf32_Half running_arch_code=EM_S390;
1862 #elif (defined ALPHA)
1863 static Elf32_Half running_arch_code=EM_ALPHA;
1864 #elif (defined MIPSEL)
1865 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1866 #elif (defined PARISC)
1867 static Elf32_Half running_arch_code=EM_PARISC;
1868 #elif (defined MIPS)
1869 static Elf32_Half running_arch_code=EM_MIPS;
1870 #elif (defined M68K)
1871 static Elf32_Half running_arch_code=EM_68K;
1872 #elif (defined SH)
1873 static Elf32_Half running_arch_code=EM_SH;
1874 #else
1875 #error Method os::dll_load requires that one of following is defined:\
1876 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1877 #endif
1878
1879 // Identify compatability class for VM's architecture and library's architecture
1880 // Obtain string descriptions for architectures
1881
1882 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1883 int running_arch_index=-1;
1884
1885 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1886 if (running_arch_code == arch_array[i].code) {
1887 running_arch_index = i;
1888 }
1889 if (lib_arch.code == arch_array[i].code) {
1890 lib_arch.compat_class = arch_array[i].compat_class;
1891 lib_arch.name = arch_array[i].name;
1892 }
1893 }
1894
1895 assert(running_arch_index != -1,
1896 "Didn't find running architecture code (running_arch_code) in arch_array");
1897 if (running_arch_index == -1) {
1898 // Even though running architecture detection failed
1899 // we may still continue with reporting dlerror() message
1900 return NULL;
1901 }
1902
1903 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1904 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1905 return NULL;
1906 }
1907
1908 #ifndef S390
1909 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1910 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1911 return NULL;
1912 }
1913 #endif // !S390
1914
1915 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1916 if (lib_arch.name!=NULL) {
1917 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1918 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1919 lib_arch.name, arch_array[running_arch_index].name);
1920 } else {
1921 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1922 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1923 lib_arch.code,
1924 arch_array[running_arch_index].name);
1925 }
1926 }
1927
1928 return NULL;
1929 }
1930
1931 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1932 int ebuflen) {
1933 void * result = ::dlopen(filename, RTLD_LAZY);
1934 if (result == NULL) {
1935 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1936 ebuf[ebuflen-1] = '\0';
1937 }
1938 return result;
1939 }
1940
1941 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1942 int ebuflen) {
1943 void * result = NULL;
1944 if (LoadExecStackDllInVMThread) {
1945 result = dlopen_helper(filename, ebuf, ebuflen);
1946 }
1947
1948 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1949 // library that requires an executable stack, or which does not have this
1950 // stack attribute set, dlopen changes the stack attribute to executable. The
1951 // read protection of the guard pages gets lost.
1952 //
1953 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1954 // may have been queued at the same time.
1955
1956 if (!_stack_is_executable) {
1957 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
1958 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1959 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1960 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1961 warning("Attempt to reguard stack yellow zone failed.");
1962 }
1963 }
1964 }
1965 }
1966
1967 return result;
1968 }
1969
1970 void* os::dll_lookup(void* handle, const char* name) {
1971 void* res = dlsym(handle, name);
1972 return res;
1973 }
1974
1975 void* os::get_default_process_handle() {
1976 return (void*)::dlopen(NULL, RTLD_LAZY);
1977 }
1978
1979 static bool _print_ascii_file(const char* filename, outputStream* st, const char* hdr = NULL) {
1980 int fd = ::open(filename, O_RDONLY);
1981 if (fd == -1) {
1982 return false;
1983 }
1984
1985 if (hdr != NULL) {
1986 st->print_cr("%s", hdr);
1987 }
1988
1989 char buf[33];
1990 int bytes;
1991 buf[32] = '\0';
1992 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1993 st->print_raw(buf, bytes);
1994 }
1995
1996 ::close(fd);
1997
1998 return true;
1999 }
2000
2001 void os::print_dll_info(outputStream *st) {
2002 st->print_cr("Dynamic libraries:");
2003
2004 char fname[32];
2005 pid_t pid = os::Linux::gettid();
2006
2007 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2008
2009 if (!_print_ascii_file(fname, st)) {
2010 st->print("Can not get library information for pid = %d\n", pid);
2011 }
2012 }
2013
2014 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2015 FILE *procmapsFile = NULL;
2016
2017 // Open the procfs maps file for the current process
2018 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2019 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2020 char line[PATH_MAX + 100];
2021
2022 // Read line by line from 'file'
2023 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2024 u8 base, top, offset, inode;
2025 char permissions[5];
2026 char device[6];
2027 char name[PATH_MAX + 1];
2028
2029 // Parse fields from line
2030 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %7s " INT64_FORMAT " %s",
2031 &base, &top, permissions, &offset, device, &inode, name);
2032
2033 // Filter by device id '00:00' so that we only get file system mapped files.
2034 if (strcmp(device, "00:00") != 0) {
2035
2036 // Call callback with the fields of interest
2037 if(callback(name, (address)base, (address)top, param)) {
2038 // Oops abort, callback aborted
2039 fclose(procmapsFile);
2040 return 1;
2041 }
2042 }
2043 }
2044 fclose(procmapsFile);
2045 }
2046 return 0;
2047 }
2048
2049 void os::print_os_info_brief(outputStream* st) {
2050 os::Linux::print_distro_info(st);
2051
2052 os::Posix::print_uname_info(st);
2053
2054 os::Linux::print_libversion_info(st);
2055
2056 }
2057
2058 void os::print_os_info(outputStream* st) {
2059 st->print("OS:");
2060
2061 os::Linux::print_distro_info(st);
2062
2063 os::Posix::print_uname_info(st);
2064
2065 // Print warning if unsafe chroot environment detected
2066 if (unsafe_chroot_detected) {
2067 st->print("WARNING!! ");
2068 st->print_cr("%s", unstable_chroot_error);
2069 }
2070
2071 os::Linux::print_libversion_info(st);
2072
2073 os::Posix::print_rlimit_info(st);
2074
2075 os::Posix::print_load_average(st);
2076
2077 os::Linux::print_full_memory_info(st);
2078
2079 os::Linux::print_proc_sys_info(st);
2080
2081 os::Linux::print_ld_preload_file(st);
2082
2083 os::Linux::print_container_info(st);
2084
2085 VM_Version::print_platform_virtualization_info(st);
2086
2087 os::Linux::print_steal_info(st);
2088 }
2089
2090 // Try to identify popular distros.
2091 // Most Linux distributions have a /etc/XXX-release file, which contains
2092 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2093 // file that also contains the OS version string. Some have more than one
2094 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2095 // /etc/redhat-release.), so the order is important.
2096 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2097 // their own specific XXX-release file as well as a redhat-release file.
2098 // Because of this the XXX-release file needs to be searched for before the
2099 // redhat-release file.
2100 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2101 // search for redhat-release / SuSE-release needs to be before lsb-release.
2102 // Since the lsb-release file is the new standard it needs to be searched
2103 // before the older style release files.
2104 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2105 // next to last resort. The os-release file is a new standard that contains
2106 // distribution information and the system-release file seems to be an old
2107 // standard that has been replaced by the lsb-release and os-release files.
2108 // Searching for the debian_version file is the last resort. It contains
2109 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2110 // "Debian " is printed before the contents of the debian_version file.
2111
2112 const char* distro_files[] = {
2113 "/etc/oracle-release",
2114 "/etc/mandriva-release",
2115 "/etc/mandrake-release",
2116 "/etc/sun-release",
2117 "/etc/redhat-release",
2118 "/etc/SuSE-release",
2119 "/etc/lsb-release",
2120 "/etc/turbolinux-release",
2121 "/etc/gentoo-release",
2122 "/etc/ltib-release",
2123 "/etc/angstrom-version",
2124 "/etc/system-release",
2125 "/etc/os-release",
2126 NULL };
2127
2128 void os::Linux::print_distro_info(outputStream* st) {
2129 for (int i = 0;; i++) {
2130 const char* file = distro_files[i];
2131 if (file == NULL) {
2132 break; // done
2133 }
2134 // If file prints, we found it.
2135 if (_print_ascii_file(file, st)) {
2136 return;
2137 }
2138 }
2139
2140 if (file_exists("/etc/debian_version")) {
2141 st->print("Debian ");
2142 _print_ascii_file("/etc/debian_version", st);
2143 } else {
2144 st->print("Linux");
2145 }
2146 st->cr();
2147 }
2148
2149 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2150 char buf[256];
2151 while (fgets(buf, sizeof(buf), fp)) {
2152 // Edit out extra stuff in expected format
2153 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2154 char* ptr = strstr(buf, "\""); // the name is in quotes
2155 if (ptr != NULL) {
2156 ptr++; // go beyond first quote
2157 char* nl = strchr(ptr, '\"');
2158 if (nl != NULL) *nl = '\0';
2159 strncpy(distro, ptr, length);
2160 } else {
2161 ptr = strstr(buf, "=");
2162 ptr++; // go beyond equals then
2163 char* nl = strchr(ptr, '\n');
2164 if (nl != NULL) *nl = '\0';
2165 strncpy(distro, ptr, length);
2166 }
2167 return;
2168 } else if (get_first_line) {
2169 char* nl = strchr(buf, '\n');
2170 if (nl != NULL) *nl = '\0';
2171 strncpy(distro, buf, length);
2172 return;
2173 }
2174 }
2175 // print last line and close
2176 char* nl = strchr(buf, '\n');
2177 if (nl != NULL) *nl = '\0';
2178 strncpy(distro, buf, length);
2179 }
2180
2181 static void parse_os_info(char* distro, size_t length, const char* file) {
2182 FILE* fp = fopen(file, "r");
2183 if (fp != NULL) {
2184 // if suse format, print out first line
2185 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2186 parse_os_info_helper(fp, distro, length, get_first_line);
2187 fclose(fp);
2188 }
2189 }
2190
2191 void os::get_summary_os_info(char* buf, size_t buflen) {
2192 for (int i = 0;; i++) {
2193 const char* file = distro_files[i];
2194 if (file == NULL) {
2195 break; // ran out of distro_files
2196 }
2197 if (file_exists(file)) {
2198 parse_os_info(buf, buflen, file);
2199 return;
2200 }
2201 }
2202 // special case for debian
2203 if (file_exists("/etc/debian_version")) {
2204 strncpy(buf, "Debian ", buflen);
2205 if (buflen > 7) {
2206 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2207 }
2208 } else {
2209 strncpy(buf, "Linux", buflen);
2210 }
2211 }
2212
2213 void os::Linux::print_libversion_info(outputStream* st) {
2214 // libc, pthread
2215 st->print("libc:");
2216 st->print("%s ", os::Linux::glibc_version());
2217 st->print("%s ", os::Linux::libpthread_version());
2218 st->cr();
2219 }
2220
2221 void os::Linux::print_proc_sys_info(outputStream* st) {
2222 st->cr();
2223 st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):");
2224 _print_ascii_file("/proc/sys/kernel/threads-max", st);
2225 st->cr();
2226 st->cr();
2227
2228 st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):");
2229 _print_ascii_file("/proc/sys/vm/max_map_count", st);
2230 st->cr();
2231 st->cr();
2232
2233 st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):");
2234 _print_ascii_file("/proc/sys/kernel/pid_max", st);
2235 st->cr();
2236 st->cr();
2237 }
2238
2239 void os::Linux::print_full_memory_info(outputStream* st) {
2240 st->print("\n/proc/meminfo:\n");
2241 _print_ascii_file("/proc/meminfo", st);
2242 st->cr();
2243 }
2244
2245 void os::Linux::print_ld_preload_file(outputStream* st) {
2246 _print_ascii_file("/etc/ld.so.preload", st, "\n/etc/ld.so.preload:");
2247 st->cr();
2248 }
2249
2250 void os::Linux::print_container_info(outputStream* st) {
2251 if (!OSContainer::is_containerized()) {
2252 return;
2253 }
2254
2255 st->print("container (cgroup) information:\n");
2256
2257 const char *p_ct = OSContainer::container_type();
2258 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "not supported");
2259
2260 char *p = OSContainer::cpu_cpuset_cpus();
2261 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "not supported");
2262 free(p);
2263
2264 p = OSContainer::cpu_cpuset_memory_nodes();
2265 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "not supported");
2266 free(p);
2267
2268 int i = OSContainer::active_processor_count();
2269 st->print("active_processor_count: ");
2270 if (i > 0) {
2271 st->print("%d\n", i);
2272 } else {
2273 st->print("not supported\n");
2274 }
2275
2276 i = OSContainer::cpu_quota();
2277 st->print("cpu_quota: ");
2278 if (i > 0) {
2279 st->print("%d\n", i);
2280 } else {
2281 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no quota");
2282 }
2283
2284 i = OSContainer::cpu_period();
2285 st->print("cpu_period: ");
2286 if (i > 0) {
2287 st->print("%d\n", i);
2288 } else {
2289 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no period");
2290 }
2291
2292 i = OSContainer::cpu_shares();
2293 st->print("cpu_shares: ");
2294 if (i > 0) {
2295 st->print("%d\n", i);
2296 } else {
2297 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no shares");
2298 }
2299
2300 jlong j = OSContainer::memory_limit_in_bytes();
2301 st->print("memory_limit_in_bytes: ");
2302 if (j > 0) {
2303 st->print(JLONG_FORMAT "\n", j);
2304 } else {
2305 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2306 }
2307
2308 j = OSContainer::memory_and_swap_limit_in_bytes();
2309 st->print("memory_and_swap_limit_in_bytes: ");
2310 if (j > 0) {
2311 st->print(JLONG_FORMAT "\n", j);
2312 } else {
2313 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2314 }
2315
2316 j = OSContainer::memory_soft_limit_in_bytes();
2317 st->print("memory_soft_limit_in_bytes: ");
2318 if (j > 0) {
2319 st->print(JLONG_FORMAT "\n", j);
2320 } else {
2321 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2322 }
2323
2324 j = OSContainer::OSContainer::memory_usage_in_bytes();
2325 st->print("memory_usage_in_bytes: ");
2326 if (j > 0) {
2327 st->print(JLONG_FORMAT "\n", j);
2328 } else {
2329 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2330 }
2331
2332 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2333 st->print("memory_max_usage_in_bytes: ");
2334 if (j > 0) {
2335 st->print(JLONG_FORMAT "\n", j);
2336 } else {
2337 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2338 }
2339 st->cr();
2340 }
2341
2342 void os::Linux::print_steal_info(outputStream* st) {
2343 if (has_initial_tick_info) {
2344 CPUPerfTicks pticks;
2345 bool res = os::Linux::get_tick_information(&pticks, -1);
2346
2347 if (res && pticks.has_steal_ticks) {
2348 uint64_t steal_ticks_difference = pticks.steal - initial_steal_ticks;
2349 uint64_t total_ticks_difference = pticks.total - initial_total_ticks;
2350 double steal_ticks_perc = 0.0;
2351 if (total_ticks_difference != 0) {
2352 steal_ticks_perc = (double) steal_ticks_difference / total_ticks_difference;
2353 }
2354 st->print_cr("Steal ticks since vm start: " UINT64_FORMAT, steal_ticks_difference);
2355 st->print_cr("Steal ticks percentage since vm start:%7.3f", steal_ticks_perc);
2356 }
2357 }
2358 }
2359
2360 void os::print_memory_info(outputStream* st) {
2361
2362 st->print("Memory:");
2363 st->print(" %dk page", os::vm_page_size()>>10);
2364
2365 // values in struct sysinfo are "unsigned long"
2366 struct sysinfo si;
2367 sysinfo(&si);
2368
2369 st->print(", physical " UINT64_FORMAT "k",
2370 os::physical_memory() >> 10);
2371 st->print("(" UINT64_FORMAT "k free)",
2372 os::available_memory() >> 10);
2373 st->print(", swap " UINT64_FORMAT "k",
2374 ((jlong)si.totalswap * si.mem_unit) >> 10);
2375 st->print("(" UINT64_FORMAT "k free)",
2376 ((jlong)si.freeswap * si.mem_unit) >> 10);
2377 st->cr();
2378 }
2379
2380 // Print the first "model name" line and the first "flags" line
2381 // that we find and nothing more. We assume "model name" comes
2382 // before "flags" so if we find a second "model name", then the
2383 // "flags" field is considered missing.
2384 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2385 #if defined(IA32) || defined(AMD64)
2386 // Other platforms have less repetitive cpuinfo files
2387 FILE *fp = fopen("/proc/cpuinfo", "r");
2388 if (fp) {
2389 while (!feof(fp)) {
2390 if (fgets(buf, buflen, fp)) {
2391 // Assume model name comes before flags
2392 bool model_name_printed = false;
2393 if (strstr(buf, "model name") != NULL) {
2394 if (!model_name_printed) {
2395 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2396 st->print_raw(buf);
2397 model_name_printed = true;
2398 } else {
2399 // model name printed but not flags? Odd, just return
2400 fclose(fp);
2401 return true;
2402 }
2403 }
2404 // print the flags line too
2405 if (strstr(buf, "flags") != NULL) {
2406 st->print_raw(buf);
2407 fclose(fp);
2408 return true;
2409 }
2410 }
2411 }
2412 fclose(fp);
2413 }
2414 #endif // x86 platforms
2415 return false;
2416 }
2417
2418 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2419 // Only print the model name if the platform provides this as a summary
2420 if (!print_model_name_and_flags(st, buf, buflen)) {
2421 st->print("\n/proc/cpuinfo:\n");
2422 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2423 st->print_cr(" <Not Available>");
2424 }
2425 }
2426 }
2427
2428 #if defined(AMD64) || defined(IA32) || defined(X32)
2429 const char* search_string = "model name";
2430 #elif defined(M68K)
2431 const char* search_string = "CPU";
2432 #elif defined(PPC64)
2433 const char* search_string = "cpu";
2434 #elif defined(S390)
2435 const char* search_string = "machine =";
2436 #elif defined(SPARC)
2437 const char* search_string = "cpu";
2438 #else
2439 const char* search_string = "Processor";
2440 #endif
2441
2442 // Parses the cpuinfo file for string representing the model name.
2443 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2444 FILE* fp = fopen("/proc/cpuinfo", "r");
2445 if (fp != NULL) {
2446 while (!feof(fp)) {
2447 char buf[256];
2448 if (fgets(buf, sizeof(buf), fp)) {
2449 char* start = strstr(buf, search_string);
2450 if (start != NULL) {
2451 char *ptr = start + strlen(search_string);
2452 char *end = buf + strlen(buf);
2453 while (ptr != end) {
2454 // skip whitespace and colon for the rest of the name.
2455 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2456 break;
2457 }
2458 ptr++;
2459 }
2460 if (ptr != end) {
2461 // reasonable string, get rid of newline and keep the rest
2462 char* nl = strchr(buf, '\n');
2463 if (nl != NULL) *nl = '\0';
2464 strncpy(cpuinfo, ptr, length);
2465 fclose(fp);
2466 return;
2467 }
2468 }
2469 }
2470 }
2471 fclose(fp);
2472 }
2473 // cpuinfo not found or parsing failed, just print generic string. The entire
2474 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2475 #if defined(AARCH64)
2476 strncpy(cpuinfo, "AArch64", length);
2477 #elif defined(AMD64)
2478 strncpy(cpuinfo, "x86_64", length);
2479 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2480 strncpy(cpuinfo, "ARM", length);
2481 #elif defined(IA32)
2482 strncpy(cpuinfo, "x86_32", length);
2483 #elif defined(IA64)
2484 strncpy(cpuinfo, "IA64", length);
2485 #elif defined(PPC)
2486 strncpy(cpuinfo, "PPC64", length);
2487 #elif defined(S390)
2488 strncpy(cpuinfo, "S390", length);
2489 #elif defined(SPARC)
2490 strncpy(cpuinfo, "sparcv9", length);
2491 #elif defined(ZERO_LIBARCH)
2492 strncpy(cpuinfo, ZERO_LIBARCH, length);
2493 #else
2494 strncpy(cpuinfo, "unknown", length);
2495 #endif
2496 }
2497
2498 static void print_signal_handler(outputStream* st, int sig,
2499 char* buf, size_t buflen);
2500
2501 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2502 st->print_cr("Signal Handlers:");
2503 print_signal_handler(st, SIGSEGV, buf, buflen);
2504 print_signal_handler(st, SIGBUS , buf, buflen);
2505 print_signal_handler(st, SIGFPE , buf, buflen);
2506 print_signal_handler(st, SIGPIPE, buf, buflen);
2507 print_signal_handler(st, SIGXFSZ, buf, buflen);
2508 print_signal_handler(st, SIGILL , buf, buflen);
2509 print_signal_handler(st, SR_signum, buf, buflen);
2510 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2511 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2512 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2513 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2514 #if defined(PPC64)
2515 print_signal_handler(st, SIGTRAP, buf, buflen);
2516 #endif
2517 }
2518
2519 static char saved_jvm_path[MAXPATHLEN] = {0};
2520
2521 // Find the full path to the current module, libjvm.so
2522 void os::jvm_path(char *buf, jint buflen) {
2523 // Error checking.
2524 if (buflen < MAXPATHLEN) {
2525 assert(false, "must use a large-enough buffer");
2526 buf[0] = '\0';
2527 return;
2528 }
2529 // Lazy resolve the path to current module.
2530 if (saved_jvm_path[0] != 0) {
2531 strcpy(buf, saved_jvm_path);
2532 return;
2533 }
2534
2535 char dli_fname[MAXPATHLEN];
2536 bool ret = dll_address_to_library_name(
2537 CAST_FROM_FN_PTR(address, os::jvm_path),
2538 dli_fname, sizeof(dli_fname), NULL);
2539 assert(ret, "cannot locate libjvm");
2540 char *rp = NULL;
2541 if (ret && dli_fname[0] != '\0') {
2542 rp = os::Posix::realpath(dli_fname, buf, buflen);
2543 }
2544 if (rp == NULL) {
2545 return;
2546 }
2547
2548 if (Arguments::sun_java_launcher_is_altjvm()) {
2549 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2550 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2551 // If "/jre/lib/" appears at the right place in the string, then
2552 // assume we are installed in a JDK and we're done. Otherwise, check
2553 // for a JAVA_HOME environment variable and fix up the path so it
2554 // looks like libjvm.so is installed there (append a fake suffix
2555 // hotspot/libjvm.so).
2556 const char *p = buf + strlen(buf) - 1;
2557 for (int count = 0; p > buf && count < 5; ++count) {
2558 for (--p; p > buf && *p != '/'; --p)
2559 /* empty */ ;
2560 }
2561
2562 if (strncmp(p, "/jre/lib/", 9) != 0) {
2563 // Look for JAVA_HOME in the environment.
2564 char* java_home_var = ::getenv("JAVA_HOME");
2565 if (java_home_var != NULL && java_home_var[0] != 0) {
2566 char* jrelib_p;
2567 int len;
2568
2569 // Check the current module name "libjvm.so".
2570 p = strrchr(buf, '/');
2571 if (p == NULL) {
2572 return;
2573 }
2574 assert(strstr(p, "/libjvm") == p, "invalid library name");
2575
2576 rp = os::Posix::realpath(java_home_var, buf, buflen);
2577 if (rp == NULL) {
2578 return;
2579 }
2580
2581 // determine if this is a legacy image or modules image
2582 // modules image doesn't have "jre" subdirectory
2583 len = strlen(buf);
2584 assert(len < buflen, "Ran out of buffer room");
2585 jrelib_p = buf + len;
2586 snprintf(jrelib_p, buflen-len, "/jre/lib");
2587 if (0 != access(buf, F_OK)) {
2588 snprintf(jrelib_p, buflen-len, "/lib");
2589 }
2590
2591 if (0 == access(buf, F_OK)) {
2592 // Use current module name "libjvm.so"
2593 len = strlen(buf);
2594 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2595 } else {
2596 // Go back to path of .so
2597 rp = os::Posix::realpath(dli_fname, buf, buflen);
2598 if (rp == NULL) {
2599 return;
2600 }
2601 }
2602 }
2603 }
2604 }
2605
2606 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2607 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2608 }
2609
2610 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2611 // no prefix required, not even "_"
2612 }
2613
2614 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2615 // no suffix required
2616 }
2617
2618 ////////////////////////////////////////////////////////////////////////////////
2619 // sun.misc.Signal support
2620
2621 static volatile jint sigint_count = 0;
2622
2623 static void UserHandler(int sig, void *siginfo, void *context) {
2624 // 4511530 - sem_post is serialized and handled by the manager thread. When
2625 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2626 // don't want to flood the manager thread with sem_post requests.
2627 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2628 return;
2629 }
2630
2631 // Ctrl-C is pressed during error reporting, likely because the error
2632 // handler fails to abort. Let VM die immediately.
2633 if (sig == SIGINT && VMError::is_error_reported()) {
2634 os::die();
2635 }
2636
2637 os::signal_notify(sig);
2638 }
2639
2640 void* os::user_handler() {
2641 return CAST_FROM_FN_PTR(void*, UserHandler);
2642 }
2643
2644 extern "C" {
2645 typedef void (*sa_handler_t)(int);
2646 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2647 }
2648
2649 void* os::signal(int signal_number, void* handler) {
2650 struct sigaction sigAct, oldSigAct;
2651
2652 sigfillset(&(sigAct.sa_mask));
2653 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2654 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2655
2656 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2657 // -1 means registration failed
2658 return (void *)-1;
2659 }
2660
2661 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2662 }
2663
2664 void os::signal_raise(int signal_number) {
2665 ::raise(signal_number);
2666 }
2667
2668 // The following code is moved from os.cpp for making this
2669 // code platform specific, which it is by its very nature.
2670
2671 // Will be modified when max signal is changed to be dynamic
2672 int os::sigexitnum_pd() {
2673 return NSIG;
2674 }
2675
2676 // a counter for each possible signal value
2677 static volatile jint pending_signals[NSIG+1] = { 0 };
2678
2679 // Linux(POSIX) specific hand shaking semaphore.
2680 static Semaphore* sig_sem = NULL;
2681 static PosixSemaphore sr_semaphore;
2682
2683 static void jdk_misc_signal_init() {
2684 // Initialize signal structures
2685 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2686
2687 // Initialize signal semaphore
2688 sig_sem = new Semaphore();
2689 }
2690
2691 void os::signal_notify(int sig) {
2692 if (sig_sem != NULL) {
2693 Atomic::inc(&pending_signals[sig]);
2694 sig_sem->signal();
2695 } else {
2696 // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2697 // initialization isn't called.
2698 assert(ReduceSignalUsage, "signal semaphore should be created");
2699 }
2700 }
2701
2702 static int check_pending_signals() {
2703 Atomic::store(0, &sigint_count);
2704 for (;;) {
2705 for (int i = 0; i < NSIG + 1; i++) {
2706 jint n = pending_signals[i];
2707 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2708 return i;
2709 }
2710 }
2711 JavaThread *thread = JavaThread::current();
2712 ThreadBlockInVM tbivm(thread);
2713
2714 bool threadIsSuspended;
2715 do {
2716 thread->set_suspend_equivalent();
2717 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2718 sig_sem->wait();
2719
2720 // were we externally suspended while we were waiting?
2721 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2722 if (threadIsSuspended) {
2723 // The semaphore has been incremented, but while we were waiting
2724 // another thread suspended us. We don't want to continue running
2725 // while suspended because that would surprise the thread that
2726 // suspended us.
2727 sig_sem->signal();
2728
2729 thread->java_suspend_self();
2730 }
2731 } while (threadIsSuspended);
2732 }
2733 }
2734
2735 int os::signal_wait() {
2736 return check_pending_signals();
2737 }
2738
2739 ////////////////////////////////////////////////////////////////////////////////
2740 // Virtual Memory
2741
2742 int os::vm_page_size() {
2743 // Seems redundant as all get out
2744 assert(os::Linux::page_size() != -1, "must call os::init");
2745 return os::Linux::page_size();
2746 }
2747
2748 // Solaris allocates memory by pages.
2749 int os::vm_allocation_granularity() {
2750 assert(os::Linux::page_size() != -1, "must call os::init");
2751 return os::Linux::page_size();
2752 }
2753
2754 // Rationale behind this function:
2755 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2756 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2757 // samples for JITted code. Here we create private executable mapping over the code cache
2758 // and then we can use standard (well, almost, as mapping can change) way to provide
2759 // info for the reporting script by storing timestamp and location of symbol
2760 void linux_wrap_code(char* base, size_t size) {
2761 static volatile jint cnt = 0;
2762
2763 if (!UseOprofile) {
2764 return;
2765 }
2766
2767 char buf[PATH_MAX+1];
2768 int num = Atomic::add(1, &cnt);
2769
2770 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2771 os::get_temp_directory(), os::current_process_id(), num);
2772 unlink(buf);
2773
2774 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2775
2776 if (fd != -1) {
2777 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2778 if (rv != (off_t)-1) {
2779 if (::write(fd, "", 1) == 1) {
2780 mmap(base, size,
2781 PROT_READ|PROT_WRITE|PROT_EXEC,
2782 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2783 }
2784 }
2785 ::close(fd);
2786 unlink(buf);
2787 }
2788 }
2789
2790 static bool recoverable_mmap_error(int err) {
2791 // See if the error is one we can let the caller handle. This
2792 // list of errno values comes from JBS-6843484. I can't find a
2793 // Linux man page that documents this specific set of errno
2794 // values so while this list currently matches Solaris, it may
2795 // change as we gain experience with this failure mode.
2796 switch (err) {
2797 case EBADF:
2798 case EINVAL:
2799 case ENOTSUP:
2800 // let the caller deal with these errors
2801 return true;
2802
2803 default:
2804 // Any remaining errors on this OS can cause our reserved mapping
2805 // to be lost. That can cause confusion where different data
2806 // structures think they have the same memory mapped. The worst
2807 // scenario is if both the VM and a library think they have the
2808 // same memory mapped.
2809 return false;
2810 }
2811 }
2812
2813 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2814 int err) {
2815 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2816 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2817 os::strerror(err), err);
2818 }
2819
2820 static void warn_fail_commit_memory(char* addr, size_t size,
2821 size_t alignment_hint, bool exec,
2822 int err) {
2823 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2824 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2825 alignment_hint, exec, os::strerror(err), err);
2826 }
2827
2828 // NOTE: Linux kernel does not really reserve the pages for us.
2829 // All it does is to check if there are enough free pages
2830 // left at the time of mmap(). This could be a potential
2831 // problem.
2832 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2833 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2834 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2835 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2836 if (res != (uintptr_t) MAP_FAILED) {
2837 if (UseNUMAInterleaving) {
2838 numa_make_global(addr, size);
2839 }
2840 return 0;
2841 }
2842
2843 int err = errno; // save errno from mmap() call above
2844
2845 if (!recoverable_mmap_error(err)) {
2846 warn_fail_commit_memory(addr, size, exec, err);
2847 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2848 }
2849
2850 return err;
2851 }
2852
2853 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2854 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2855 }
2856
2857 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2858 const char* mesg) {
2859 assert(mesg != NULL, "mesg must be specified");
2860 int err = os::Linux::commit_memory_impl(addr, size, exec);
2861 if (err != 0) {
2862 // the caller wants all commit errors to exit with the specified mesg:
2863 warn_fail_commit_memory(addr, size, exec, err);
2864 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2865 }
2866 }
2867
2868 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2869 #ifndef MAP_HUGETLB
2870 #define MAP_HUGETLB 0x40000
2871 #endif
2872
2873 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2874 #ifndef MADV_HUGEPAGE
2875 #define MADV_HUGEPAGE 14
2876 #endif
2877
2878 int os::Linux::commit_memory_impl(char* addr, size_t size,
2879 size_t alignment_hint, bool exec) {
2880 int err = os::Linux::commit_memory_impl(addr, size, exec);
2881 if (err == 0) {
2882 realign_memory(addr, size, alignment_hint);
2883 }
2884 return err;
2885 }
2886
2887 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2888 bool exec) {
2889 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2890 }
2891
2892 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2893 size_t alignment_hint, bool exec,
2894 const char* mesg) {
2895 assert(mesg != NULL, "mesg must be specified");
2896 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2897 if (err != 0) {
2898 // the caller wants all commit errors to exit with the specified mesg:
2899 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2900 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2901 }
2902 }
2903
2904 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2905 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2906 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2907 // be supported or the memory may already be backed by huge pages.
2908 ::madvise(addr, bytes, MADV_HUGEPAGE);
2909 }
2910 }
2911
2912 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2913 // This method works by doing an mmap over an existing mmaping and effectively discarding
2914 // the existing pages. However it won't work for SHM-based large pages that cannot be
2915 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2916 // small pages on top of the SHM segment. This method always works for small pages, so we
2917 // allow that in any case.
2918 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2919 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2920 }
2921 }
2922
2923 void os::numa_make_global(char *addr, size_t bytes) {
2924 Linux::numa_interleave_memory(addr, bytes);
2925 }
2926
2927 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2928 // bind policy to MPOL_PREFERRED for the current thread.
2929 #define USE_MPOL_PREFERRED 0
2930
2931 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2932 // To make NUMA and large pages more robust when both enabled, we need to ease
2933 // the requirements on where the memory should be allocated. MPOL_BIND is the
2934 // default policy and it will force memory to be allocated on the specified
2935 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2936 // the specified node, but will not force it. Using this policy will prevent
2937 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2938 // free large pages.
2939 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2940 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2941 }
2942
2943 bool os::numa_topology_changed() { return false; }
2944
2945 size_t os::numa_get_groups_num() {
2946 // Return just the number of nodes in which it's possible to allocate memory
2947 // (in numa terminology, configured nodes).
2948 return Linux::numa_num_configured_nodes();
2949 }
2950
2951 int os::numa_get_group_id() {
2952 int cpu_id = Linux::sched_getcpu();
2953 if (cpu_id != -1) {
2954 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2955 if (lgrp_id != -1) {
2956 return lgrp_id;
2957 }
2958 }
2959 return 0;
2960 }
2961
2962 int os::Linux::get_existing_num_nodes() {
2963 int node;
2964 int highest_node_number = Linux::numa_max_node();
2965 int num_nodes = 0;
2966
2967 // Get the total number of nodes in the system including nodes without memory.
2968 for (node = 0; node <= highest_node_number; node++) {
2969 if (is_node_in_existing_nodes(node)) {
2970 num_nodes++;
2971 }
2972 }
2973 return num_nodes;
2974 }
2975
2976 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2977 int highest_node_number = Linux::numa_max_node();
2978 size_t i = 0;
2979
2980 // Map all node ids in which it is possible to allocate memory. Also nodes are
2981 // not always consecutively available, i.e. available from 0 to the highest
2982 // node number. If the nodes have been bound explicitly using numactl membind,
2983 // then allocate memory from those nodes only.
2984 for (int node = 0; node <= highest_node_number; node++) {
2985 if (Linux::is_node_in_bound_nodes((unsigned int)node)) {
2986 ids[i++] = node;
2987 }
2988 }
2989 return i;
2990 }
2991
2992 bool os::get_page_info(char *start, page_info* info) {
2993 return false;
2994 }
2995
2996 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2997 page_info* page_found) {
2998 return end;
2999 }
3000
3001
3002 int os::Linux::sched_getcpu_syscall(void) {
3003 unsigned int cpu = 0;
3004 int retval = -1;
3005
3006 #if defined(IA32)
3007 #ifndef SYS_getcpu
3008 #define SYS_getcpu 318
3009 #endif
3010 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
3011 #elif defined(AMD64)
3012 // Unfortunately we have to bring all these macros here from vsyscall.h
3013 // to be able to compile on old linuxes.
3014 #define __NR_vgetcpu 2
3015 #define VSYSCALL_START (-10UL << 20)
3016 #define VSYSCALL_SIZE 1024
3017 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
3018 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
3019 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
3020 retval = vgetcpu(&cpu, NULL, NULL);
3021 #endif
3022
3023 return (retval == -1) ? retval : cpu;
3024 }
3025
3026 void os::Linux::sched_getcpu_init() {
3027 // sched_getcpu() should be in libc.
3028 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
3029 dlsym(RTLD_DEFAULT, "sched_getcpu")));
3030
3031 // If it's not, try a direct syscall.
3032 if (sched_getcpu() == -1) {
3033 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
3034 (void*)&sched_getcpu_syscall));
3035 }
3036
3037 if (sched_getcpu() == -1) {
3038 vm_exit_during_initialization("getcpu(2) system call not supported by kernel");
3039 }
3040 }
3041
3042 // Something to do with the numa-aware allocator needs these symbols
3043 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
3044 extern "C" JNIEXPORT void numa_error(char *where) { }
3045
3046 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
3047 // load symbol from base version instead.
3048 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
3049 void *f = dlvsym(handle, name, "libnuma_1.1");
3050 if (f == NULL) {
3051 f = dlsym(handle, name);
3052 }
3053 return f;
3054 }
3055
3056 // Handle request to load libnuma symbol version 1.2 (API v2) only.
3057 // Return NULL if the symbol is not defined in this particular version.
3058 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
3059 return dlvsym(handle, name, "libnuma_1.2");
3060 }
3061
3062 bool os::Linux::libnuma_init() {
3063 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
3064 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
3065 if (handle != NULL) {
3066 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
3067 libnuma_dlsym(handle, "numa_node_to_cpus")));
3068 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
3069 libnuma_dlsym(handle, "numa_max_node")));
3070 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
3071 libnuma_dlsym(handle, "numa_num_configured_nodes")));
3072 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
3073 libnuma_dlsym(handle, "numa_available")));
3074 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
3075 libnuma_dlsym(handle, "numa_tonode_memory")));
3076 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
3077 libnuma_dlsym(handle, "numa_interleave_memory")));
3078 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
3079 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
3080 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
3081 libnuma_dlsym(handle, "numa_set_bind_policy")));
3082 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
3083 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
3084 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
3085 libnuma_dlsym(handle, "numa_distance")));
3086 set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t,
3087 libnuma_v2_dlsym(handle, "numa_get_membind")));
3088 set_numa_get_interleave_mask(CAST_TO_FN_PTR(numa_get_interleave_mask_func_t,
3089 libnuma_v2_dlsym(handle, "numa_get_interleave_mask")));
3090
3091 if (numa_available() != -1) {
3092 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
3093 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
3094 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
3095 set_numa_interleave_bitmask(_numa_get_interleave_mask());
3096 set_numa_membind_bitmask(_numa_get_membind());
3097 // Create an index -> node mapping, since nodes are not always consecutive
3098 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3099 rebuild_nindex_to_node_map();
3100 // Create a cpu -> node mapping
3101 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3102 rebuild_cpu_to_node_map();
3103 return true;
3104 }
3105 }
3106 }
3107 return false;
3108 }
3109
3110 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
3111 // Creating guard page is very expensive. Java thread has HotSpot
3112 // guard pages, only enable glibc guard page for non-Java threads.
3113 // (Remember: compiler thread is a Java thread, too!)
3114 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
3115 }
3116
3117 void os::Linux::rebuild_nindex_to_node_map() {
3118 int highest_node_number = Linux::numa_max_node();
3119
3120 nindex_to_node()->clear();
3121 for (int node = 0; node <= highest_node_number; node++) {
3122 if (Linux::is_node_in_existing_nodes(node)) {
3123 nindex_to_node()->append(node);
3124 }
3125 }
3126 }
3127
3128 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
3129 // The table is later used in get_node_by_cpu().
3130 void os::Linux::rebuild_cpu_to_node_map() {
3131 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
3132 // in libnuma (possible values are starting from 16,
3133 // and continuing up with every other power of 2, but less
3134 // than the maximum number of CPUs supported by kernel), and
3135 // is a subject to change (in libnuma version 2 the requirements
3136 // are more reasonable) we'll just hardcode the number they use
3137 // in the library.
3138 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
3139
3140 size_t cpu_num = processor_count();
3141 size_t cpu_map_size = NCPUS / BitsPerCLong;
3142 size_t cpu_map_valid_size =
3143 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
3144
3145 cpu_to_node()->clear();
3146 cpu_to_node()->at_grow(cpu_num - 1);
3147
3148 size_t node_num = get_existing_num_nodes();
3149
3150 int distance = 0;
3151 int closest_distance = INT_MAX;
3152 int closest_node = 0;
3153 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
3154 for (size_t i = 0; i < node_num; i++) {
3155 // Check if node is configured (not a memory-less node). If it is not, find
3156 // the closest configured node. Check also if node is bound, i.e. it's allowed
3157 // to allocate memory from the node. If it's not allowed, map cpus in that node
3158 // to the closest node from which memory allocation is allowed.
3159 if (!is_node_in_configured_nodes(nindex_to_node()->at(i)) ||
3160 !is_node_in_bound_nodes(nindex_to_node()->at(i))) {
3161 closest_distance = INT_MAX;
3162 // Check distance from all remaining nodes in the system. Ignore distance
3163 // from itself, from another non-configured node, and from another non-bound
3164 // node.
3165 for (size_t m = 0; m < node_num; m++) {
3166 if (m != i &&
3167 is_node_in_configured_nodes(nindex_to_node()->at(m)) &&
3168 is_node_in_bound_nodes(nindex_to_node()->at(m))) {
3169 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3170 // If a closest node is found, update. There is always at least one
3171 // configured and bound node in the system so there is always at least
3172 // one node close.
3173 if (distance != 0 && distance < closest_distance) {
3174 closest_distance = distance;
3175 closest_node = nindex_to_node()->at(m);
3176 }
3177 }
3178 }
3179 } else {
3180 // Current node is already a configured node.
3181 closest_node = nindex_to_node()->at(i);
3182 }
3183
3184 // Get cpus from the original node and map them to the closest node. If node
3185 // is a configured node (not a memory-less node), then original node and
3186 // closest node are the same.
3187 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3188 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3189 if (cpu_map[j] != 0) {
3190 for (size_t k = 0; k < BitsPerCLong; k++) {
3191 if (cpu_map[j] & (1UL << k)) {
3192 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3193 }
3194 }
3195 }
3196 }
3197 }
3198 }
3199 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3200 }
3201
3202 int os::Linux::get_node_by_cpu(int cpu_id) {
3203 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3204 return cpu_to_node()->at(cpu_id);
3205 }
3206 return -1;
3207 }
3208
3209 GrowableArray<int>* os::Linux::_cpu_to_node;
3210 GrowableArray<int>* os::Linux::_nindex_to_node;
3211 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3212 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3213 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3214 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3215 os::Linux::numa_available_func_t os::Linux::_numa_available;
3216 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3217 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3218 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3219 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3220 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3221 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3222 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind;
3223 os::Linux::numa_get_interleave_mask_func_t os::Linux::_numa_get_interleave_mask;
3224 os::Linux::NumaAllocationPolicy os::Linux::_current_numa_policy;
3225 unsigned long* os::Linux::_numa_all_nodes;
3226 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3227 struct bitmask* os::Linux::_numa_nodes_ptr;
3228 struct bitmask* os::Linux::_numa_interleave_bitmask;
3229 struct bitmask* os::Linux::_numa_membind_bitmask;
3230
3231 bool os::pd_uncommit_memory(char* addr, size_t size) {
3232 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3233 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3234 return res != (uintptr_t) MAP_FAILED;
3235 }
3236
3237 static address get_stack_commited_bottom(address bottom, size_t size) {
3238 address nbot = bottom;
3239 address ntop = bottom + size;
3240
3241 size_t page_sz = os::vm_page_size();
3242 unsigned pages = size / page_sz;
3243
3244 unsigned char vec[1];
3245 unsigned imin = 1, imax = pages + 1, imid;
3246 int mincore_return_value = 0;
3247
3248 assert(imin <= imax, "Unexpected page size");
3249
3250 while (imin < imax) {
3251 imid = (imax + imin) / 2;
3252 nbot = ntop - (imid * page_sz);
3253
3254 // Use a trick with mincore to check whether the page is mapped or not.
3255 // mincore sets vec to 1 if page resides in memory and to 0 if page
3256 // is swapped output but if page we are asking for is unmapped
3257 // it returns -1,ENOMEM
3258 mincore_return_value = mincore(nbot, page_sz, vec);
3259
3260 if (mincore_return_value == -1) {
3261 // Page is not mapped go up
3262 // to find first mapped page
3263 if (errno != EAGAIN) {
3264 assert(errno == ENOMEM, "Unexpected mincore errno");
3265 imax = imid;
3266 }
3267 } else {
3268 // Page is mapped go down
3269 // to find first not mapped page
3270 imin = imid + 1;
3271 }
3272 }
3273
3274 nbot = nbot + page_sz;
3275
3276 // Adjust stack bottom one page up if last checked page is not mapped
3277 if (mincore_return_value == -1) {
3278 nbot = nbot + page_sz;
3279 }
3280
3281 return nbot;
3282 }
3283
3284 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3285 int mincore_return_value;
3286 const size_t stripe = 1024; // query this many pages each time
3287 unsigned char vec[stripe + 1];
3288 // set a guard
3289 vec[stripe] = 'X';
3290
3291 const size_t page_sz = os::vm_page_size();
3292 size_t pages = size / page_sz;
3293
3294 assert(is_aligned(start, page_sz), "Start address must be page aligned");
3295 assert(is_aligned(size, page_sz), "Size must be page aligned");
3296
3297 committed_start = NULL;
3298
3299 int loops = (pages + stripe - 1) / stripe;
3300 int committed_pages = 0;
3301 address loop_base = start;
3302 bool found_range = false;
3303
3304 for (int index = 0; index < loops && !found_range; index ++) {
3305 assert(pages > 0, "Nothing to do");
3306 int pages_to_query = (pages >= stripe) ? stripe : pages;
3307 pages -= pages_to_query;
3308
3309 // Get stable read
3310 while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3311
3312 // During shutdown, some memory goes away without properly notifying NMT,
3313 // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3314 // Bailout and return as not committed for now.
3315 if (mincore_return_value == -1 && errno == ENOMEM) {
3316 return false;
3317 }
3318
3319 assert(vec[stripe] == 'X', "overflow guard");
3320 assert(mincore_return_value == 0, "Range must be valid");
3321 // Process this stripe
3322 for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3323 if ((vec[vecIdx] & 0x01) == 0) { // not committed
3324 // End of current contiguous region
3325 if (committed_start != NULL) {
3326 found_range = true;
3327 break;
3328 }
3329 } else { // committed
3330 // Start of region
3331 if (committed_start == NULL) {
3332 committed_start = loop_base + page_sz * vecIdx;
3333 }
3334 committed_pages ++;
3335 }
3336 }
3337
3338 loop_base += pages_to_query * page_sz;
3339 }
3340
3341 if (committed_start != NULL) {
3342 assert(committed_pages > 0, "Must have committed region");
3343 assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3344 assert(committed_start >= start && committed_start < start + size, "Out of range");
3345 committed_size = page_sz * committed_pages;
3346 return true;
3347 } else {
3348 assert(committed_pages == 0, "Should not have committed region");
3349 return false;
3350 }
3351 }
3352
3353
3354 // Linux uses a growable mapping for the stack, and if the mapping for
3355 // the stack guard pages is not removed when we detach a thread the
3356 // stack cannot grow beyond the pages where the stack guard was
3357 // mapped. If at some point later in the process the stack expands to
3358 // that point, the Linux kernel cannot expand the stack any further
3359 // because the guard pages are in the way, and a segfault occurs.
3360 //
3361 // However, it's essential not to split the stack region by unmapping
3362 // a region (leaving a hole) that's already part of the stack mapping,
3363 // so if the stack mapping has already grown beyond the guard pages at
3364 // the time we create them, we have to truncate the stack mapping.
3365 // So, we need to know the extent of the stack mapping when
3366 // create_stack_guard_pages() is called.
3367
3368 // We only need this for stacks that are growable: at the time of
3369 // writing thread stacks don't use growable mappings (i.e. those
3370 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3371 // only applies to the main thread.
3372
3373 // If the (growable) stack mapping already extends beyond the point
3374 // where we're going to put our guard pages, truncate the mapping at
3375 // that point by munmap()ping it. This ensures that when we later
3376 // munmap() the guard pages we don't leave a hole in the stack
3377 // mapping. This only affects the main/primordial thread
3378
3379 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3380 if (os::is_primordial_thread()) {
3381 // As we manually grow stack up to bottom inside create_attached_thread(),
3382 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3383 // we don't need to do anything special.
3384 // Check it first, before calling heavy function.
3385 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3386 unsigned char vec[1];
3387
3388 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3389 // Fallback to slow path on all errors, including EAGAIN
3390 stack_extent = (uintptr_t) get_stack_commited_bottom(
3391 os::Linux::initial_thread_stack_bottom(),
3392 (size_t)addr - stack_extent);
3393 }
3394
3395 if (stack_extent < (uintptr_t)addr) {
3396 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3397 }
3398 }
3399
3400 return os::commit_memory(addr, size, !ExecMem);
3401 }
3402
3403 // If this is a growable mapping, remove the guard pages entirely by
3404 // munmap()ping them. If not, just call uncommit_memory(). This only
3405 // affects the main/primordial thread, but guard against future OS changes.
3406 // It's safe to always unmap guard pages for primordial thread because we
3407 // always place it right after end of the mapped region.
3408
3409 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3410 uintptr_t stack_extent, stack_base;
3411
3412 if (os::is_primordial_thread()) {
3413 return ::munmap(addr, size) == 0;
3414 }
3415
3416 return os::uncommit_memory(addr, size);
3417 }
3418
3419 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3420 // at 'requested_addr'. If there are existing memory mappings at the same
3421 // location, however, they will be overwritten. If 'fixed' is false,
3422 // 'requested_addr' is only treated as a hint, the return value may or
3423 // may not start from the requested address. Unlike Linux mmap(), this
3424 // function returns NULL to indicate failure.
3425 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3426 char * addr;
3427 int flags;
3428
3429 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3430 if (fixed) {
3431 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3432 flags |= MAP_FIXED;
3433 }
3434
3435 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3436 // touch an uncommitted page. Otherwise, the read/write might
3437 // succeed if we have enough swap space to back the physical page.
3438 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3439 flags, -1, 0);
3440
3441 return addr == MAP_FAILED ? NULL : addr;
3442 }
3443
3444 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3445 // (req_addr != NULL) or with a given alignment.
3446 // - bytes shall be a multiple of alignment.
3447 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3448 // - alignment sets the alignment at which memory shall be allocated.
3449 // It must be a multiple of allocation granularity.
3450 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3451 // req_addr or NULL.
3452 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3453
3454 size_t extra_size = bytes;
3455 if (req_addr == NULL && alignment > 0) {
3456 extra_size += alignment;
3457 }
3458
3459 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3460 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3461 -1, 0);
3462 if (start == MAP_FAILED) {
3463 start = NULL;
3464 } else {
3465 if (req_addr != NULL) {
3466 if (start != req_addr) {
3467 ::munmap(start, extra_size);
3468 start = NULL;
3469 }
3470 } else {
3471 char* const start_aligned = align_up(start, alignment);
3472 char* const end_aligned = start_aligned + bytes;
3473 char* const end = start + extra_size;
3474 if (start_aligned > start) {
3475 ::munmap(start, start_aligned - start);
3476 }
3477 if (end_aligned < end) {
3478 ::munmap(end_aligned, end - end_aligned);
3479 }
3480 start = start_aligned;
3481 }
3482 }
3483 return start;
3484 }
3485
3486 static int anon_munmap(char * addr, size_t size) {
3487 return ::munmap(addr, size) == 0;
3488 }
3489
3490 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3491 size_t alignment_hint) {
3492 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3493 }
3494
3495 bool os::pd_release_memory(char* addr, size_t size) {
3496 return anon_munmap(addr, size);
3497 }
3498
3499 static bool linux_mprotect(char* addr, size_t size, int prot) {
3500 // Linux wants the mprotect address argument to be page aligned.
3501 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3502
3503 // According to SUSv3, mprotect() should only be used with mappings
3504 // established by mmap(), and mmap() always maps whole pages. Unaligned
3505 // 'addr' likely indicates problem in the VM (e.g. trying to change
3506 // protection of malloc'ed or statically allocated memory). Check the
3507 // caller if you hit this assert.
3508 assert(addr == bottom, "sanity check");
3509
3510 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3511 return ::mprotect(bottom, size, prot) == 0;
3512 }
3513
3514 // Set protections specified
3515 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3516 bool is_committed) {
3517 unsigned int p = 0;
3518 switch (prot) {
3519 case MEM_PROT_NONE: p = PROT_NONE; break;
3520 case MEM_PROT_READ: p = PROT_READ; break;
3521 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3522 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3523 default:
3524 ShouldNotReachHere();
3525 }
3526 // is_committed is unused.
3527 return linux_mprotect(addr, bytes, p);
3528 }
3529
3530 bool os::guard_memory(char* addr, size_t size) {
3531 return linux_mprotect(addr, size, PROT_NONE);
3532 }
3533
3534 bool os::unguard_memory(char* addr, size_t size) {
3535 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3536 }
3537
3538 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3539 size_t page_size) {
3540 bool result = false;
3541 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3542 MAP_ANONYMOUS|MAP_PRIVATE,
3543 -1, 0);
3544 if (p != MAP_FAILED) {
3545 void *aligned_p = align_up(p, page_size);
3546
3547 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3548
3549 munmap(p, page_size * 2);
3550 }
3551
3552 if (warn && !result) {
3553 warning("TransparentHugePages is not supported by the operating system.");
3554 }
3555
3556 return result;
3557 }
3558
3559 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3560 bool result = false;
3561 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3562 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3563 -1, 0);
3564
3565 if (p != MAP_FAILED) {
3566 // We don't know if this really is a huge page or not.
3567 FILE *fp = fopen("/proc/self/maps", "r");
3568 if (fp) {
3569 while (!feof(fp)) {
3570 char chars[257];
3571 long x = 0;
3572 if (fgets(chars, sizeof(chars), fp)) {
3573 if (sscanf(chars, "%lx-%*x", &x) == 1
3574 && x == (long)p) {
3575 if (strstr (chars, "hugepage")) {
3576 result = true;
3577 break;
3578 }
3579 }
3580 }
3581 }
3582 fclose(fp);
3583 }
3584 munmap(p, page_size);
3585 }
3586
3587 if (warn && !result) {
3588 warning("HugeTLBFS is not supported by the operating system.");
3589 }
3590
3591 return result;
3592 }
3593
3594 // From the coredump_filter documentation:
3595 //
3596 // - (bit 0) anonymous private memory
3597 // - (bit 1) anonymous shared memory
3598 // - (bit 2) file-backed private memory
3599 // - (bit 3) file-backed shared memory
3600 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3601 // effective only if the bit 2 is cleared)
3602 // - (bit 5) hugetlb private memory
3603 // - (bit 6) hugetlb shared memory
3604 // - (bit 7) dax private memory
3605 // - (bit 8) dax shared memory
3606 //
3607 static void set_coredump_filter(CoredumpFilterBit bit) {
3608 FILE *f;
3609 long cdm;
3610
3611 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3612 return;
3613 }
3614
3615 if (fscanf(f, "%lx", &cdm) != 1) {
3616 fclose(f);
3617 return;
3618 }
3619
3620 long saved_cdm = cdm;
3621 rewind(f);
3622 cdm |= bit;
3623
3624 if (cdm != saved_cdm) {
3625 fprintf(f, "%#lx", cdm);
3626 }
3627
3628 fclose(f);
3629 }
3630
3631 // Large page support
3632
3633 static size_t _large_page_size = 0;
3634
3635 size_t os::Linux::find_large_page_size() {
3636 size_t large_page_size = 0;
3637
3638 // large_page_size on Linux is used to round up heap size. x86 uses either
3639 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3640 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3641 // page as large as 256M.
3642 //
3643 // Here we try to figure out page size by parsing /proc/meminfo and looking
3644 // for a line with the following format:
3645 // Hugepagesize: 2048 kB
3646 //
3647 // If we can't determine the value (e.g. /proc is not mounted, or the text
3648 // format has been changed), we'll use the largest page size supported by
3649 // the processor.
3650
3651 #ifndef ZERO
3652 large_page_size =
3653 AARCH64_ONLY(2 * M)
3654 AMD64_ONLY(2 * M)
3655 ARM32_ONLY(2 * M)
3656 IA32_ONLY(4 * M)
3657 IA64_ONLY(256 * M)
3658 PPC_ONLY(4 * M)
3659 S390_ONLY(1 * M)
3660 SPARC_ONLY(4 * M);
3661 #endif // ZERO
3662
3663 FILE *fp = fopen("/proc/meminfo", "r");
3664 if (fp) {
3665 while (!feof(fp)) {
3666 int x = 0;
3667 char buf[16];
3668 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3669 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3670 large_page_size = x * K;
3671 break;
3672 }
3673 } else {
3674 // skip to next line
3675 for (;;) {
3676 int ch = fgetc(fp);
3677 if (ch == EOF || ch == (int)'\n') break;
3678 }
3679 }
3680 }
3681 fclose(fp);
3682 }
3683
3684 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3685 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3686 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3687 proper_unit_for_byte_size(large_page_size));
3688 }
3689
3690 return large_page_size;
3691 }
3692
3693 size_t os::Linux::setup_large_page_size() {
3694 _large_page_size = Linux::find_large_page_size();
3695 const size_t default_page_size = (size_t)Linux::page_size();
3696 if (_large_page_size > default_page_size) {
3697 _page_sizes[0] = _large_page_size;
3698 _page_sizes[1] = default_page_size;
3699 _page_sizes[2] = 0;
3700 }
3701
3702 return _large_page_size;
3703 }
3704
3705 bool os::Linux::setup_large_page_type(size_t page_size) {
3706 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3707 FLAG_IS_DEFAULT(UseSHM) &&
3708 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3709
3710 // The type of large pages has not been specified by the user.
3711
3712 // Try UseHugeTLBFS and then UseSHM.
3713 UseHugeTLBFS = UseSHM = true;
3714
3715 // Don't try UseTransparentHugePages since there are known
3716 // performance issues with it turned on. This might change in the future.
3717 UseTransparentHugePages = false;
3718 }
3719
3720 if (UseTransparentHugePages) {
3721 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3722 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3723 UseHugeTLBFS = false;
3724 UseSHM = false;
3725 return true;
3726 }
3727 UseTransparentHugePages = false;
3728 }
3729
3730 if (UseHugeTLBFS) {
3731 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3732 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3733 UseSHM = false;
3734 return true;
3735 }
3736 UseHugeTLBFS = false;
3737 }
3738
3739 return UseSHM;
3740 }
3741
3742 void os::large_page_init() {
3743 if (!UseLargePages &&
3744 !UseTransparentHugePages &&
3745 !UseHugeTLBFS &&
3746 !UseSHM) {
3747 // Not using large pages.
3748 return;
3749 }
3750
3751 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3752 // The user explicitly turned off large pages.
3753 // Ignore the rest of the large pages flags.
3754 UseTransparentHugePages = false;
3755 UseHugeTLBFS = false;
3756 UseSHM = false;
3757 return;
3758 }
3759
3760 size_t large_page_size = Linux::setup_large_page_size();
3761 UseLargePages = Linux::setup_large_page_type(large_page_size);
3762
3763 set_coredump_filter(LARGEPAGES_BIT);
3764 }
3765
3766 #ifndef SHM_HUGETLB
3767 #define SHM_HUGETLB 04000
3768 #endif
3769
3770 #define shm_warning_format(format, ...) \
3771 do { \
3772 if (UseLargePages && \
3773 (!FLAG_IS_DEFAULT(UseLargePages) || \
3774 !FLAG_IS_DEFAULT(UseSHM) || \
3775 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3776 warning(format, __VA_ARGS__); \
3777 } \
3778 } while (0)
3779
3780 #define shm_warning(str) shm_warning_format("%s", str)
3781
3782 #define shm_warning_with_errno(str) \
3783 do { \
3784 int err = errno; \
3785 shm_warning_format(str " (error = %d)", err); \
3786 } while (0)
3787
3788 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3789 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3790
3791 if (!is_aligned(alignment, SHMLBA)) {
3792 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3793 return NULL;
3794 }
3795
3796 // To ensure that we get 'alignment' aligned memory from shmat,
3797 // we pre-reserve aligned virtual memory and then attach to that.
3798
3799 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3800 if (pre_reserved_addr == NULL) {
3801 // Couldn't pre-reserve aligned memory.
3802 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3803 return NULL;
3804 }
3805
3806 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3807 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3808
3809 if ((intptr_t)addr == -1) {
3810 int err = errno;
3811 shm_warning_with_errno("Failed to attach shared memory.");
3812
3813 assert(err != EACCES, "Unexpected error");
3814 assert(err != EIDRM, "Unexpected error");
3815 assert(err != EINVAL, "Unexpected error");
3816
3817 // Since we don't know if the kernel unmapped the pre-reserved memory area
3818 // we can't unmap it, since that would potentially unmap memory that was
3819 // mapped from other threads.
3820 return NULL;
3821 }
3822
3823 return addr;
3824 }
3825
3826 static char* shmat_at_address(int shmid, char* req_addr) {
3827 if (!is_aligned(req_addr, SHMLBA)) {
3828 assert(false, "Requested address needs to be SHMLBA aligned");
3829 return NULL;
3830 }
3831
3832 char* addr = (char*)shmat(shmid, req_addr, 0);
3833
3834 if ((intptr_t)addr == -1) {
3835 shm_warning_with_errno("Failed to attach shared memory.");
3836 return NULL;
3837 }
3838
3839 return addr;
3840 }
3841
3842 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3843 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3844 if (req_addr != NULL) {
3845 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3846 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3847 return shmat_at_address(shmid, req_addr);
3848 }
3849
3850 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3851 // return large page size aligned memory addresses when req_addr == NULL.
3852 // However, if the alignment is larger than the large page size, we have
3853 // to manually ensure that the memory returned is 'alignment' aligned.
3854 if (alignment > os::large_page_size()) {
3855 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3856 return shmat_with_alignment(shmid, bytes, alignment);
3857 } else {
3858 return shmat_at_address(shmid, NULL);
3859 }
3860 }
3861
3862 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3863 char* req_addr, bool exec) {
3864 // "exec" is passed in but not used. Creating the shared image for
3865 // the code cache doesn't have an SHM_X executable permission to check.
3866 assert(UseLargePages && UseSHM, "only for SHM large pages");
3867 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3868 assert(is_aligned(req_addr, alignment), "Unaligned address");
3869
3870 if (!is_aligned(bytes, os::large_page_size())) {
3871 return NULL; // Fallback to small pages.
3872 }
3873
3874 // Create a large shared memory region to attach to based on size.
3875 // Currently, size is the total size of the heap.
3876 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3877 if (shmid == -1) {
3878 // Possible reasons for shmget failure:
3879 // 1. shmmax is too small for Java heap.
3880 // > check shmmax value: cat /proc/sys/kernel/shmmax
3881 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3882 // 2. not enough large page memory.
3883 // > check available large pages: cat /proc/meminfo
3884 // > increase amount of large pages:
3885 // echo new_value > /proc/sys/vm/nr_hugepages
3886 // Note 1: different Linux may use different name for this property,
3887 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3888 // Note 2: it's possible there's enough physical memory available but
3889 // they are so fragmented after a long run that they can't
3890 // coalesce into large pages. Try to reserve large pages when
3891 // the system is still "fresh".
3892 shm_warning_with_errno("Failed to reserve shared memory.");
3893 return NULL;
3894 }
3895
3896 // Attach to the region.
3897 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3898
3899 // Remove shmid. If shmat() is successful, the actual shared memory segment
3900 // will be deleted when it's detached by shmdt() or when the process
3901 // terminates. If shmat() is not successful this will remove the shared
3902 // segment immediately.
3903 shmctl(shmid, IPC_RMID, NULL);
3904
3905 return addr;
3906 }
3907
3908 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3909 int error) {
3910 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3911
3912 bool warn_on_failure = UseLargePages &&
3913 (!FLAG_IS_DEFAULT(UseLargePages) ||
3914 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3915 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3916
3917 if (warn_on_failure) {
3918 char msg[128];
3919 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3920 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3921 warning("%s", msg);
3922 }
3923 }
3924
3925 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3926 char* req_addr,
3927 bool exec) {
3928 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3929 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3930 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3931
3932 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3933 char* addr = (char*)::mmap(req_addr, bytes, prot,
3934 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3935 -1, 0);
3936
3937 if (addr == MAP_FAILED) {
3938 warn_on_large_pages_failure(req_addr, bytes, errno);
3939 return NULL;
3940 }
3941
3942 assert(is_aligned(addr, os::large_page_size()), "Must be");
3943
3944 return addr;
3945 }
3946
3947 // Reserve memory using mmap(MAP_HUGETLB).
3948 // - bytes shall be a multiple of alignment.
3949 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3950 // - alignment sets the alignment at which memory shall be allocated.
3951 // It must be a multiple of allocation granularity.
3952 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3953 // req_addr or NULL.
3954 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3955 size_t alignment,
3956 char* req_addr,
3957 bool exec) {
3958 size_t large_page_size = os::large_page_size();
3959 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3960
3961 assert(is_aligned(req_addr, alignment), "Must be");
3962 assert(is_aligned(bytes, alignment), "Must be");
3963
3964 // First reserve - but not commit - the address range in small pages.
3965 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3966
3967 if (start == NULL) {
3968 return NULL;
3969 }
3970
3971 assert(is_aligned(start, alignment), "Must be");
3972
3973 char* end = start + bytes;
3974
3975 // Find the regions of the allocated chunk that can be promoted to large pages.
3976 char* lp_start = align_up(start, large_page_size);
3977 char* lp_end = align_down(end, large_page_size);
3978
3979 size_t lp_bytes = lp_end - lp_start;
3980
3981 assert(is_aligned(lp_bytes, large_page_size), "Must be");
3982
3983 if (lp_bytes == 0) {
3984 // The mapped region doesn't even span the start and the end of a large page.
3985 // Fall back to allocate a non-special area.
3986 ::munmap(start, end - start);
3987 return NULL;
3988 }
3989
3990 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3991
3992 void* result;
3993
3994 // Commit small-paged leading area.
3995 if (start != lp_start) {
3996 result = ::mmap(start, lp_start - start, prot,
3997 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3998 -1, 0);
3999 if (result == MAP_FAILED) {
4000 ::munmap(lp_start, end - lp_start);
4001 return NULL;
4002 }
4003 }
4004
4005 // Commit large-paged area.
4006 result = ::mmap(lp_start, lp_bytes, prot,
4007 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
4008 -1, 0);
4009 if (result == MAP_FAILED) {
4010 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
4011 // If the mmap above fails, the large pages region will be unmapped and we
4012 // have regions before and after with small pages. Release these regions.
4013 //
4014 // | mapped | unmapped | mapped |
4015 // ^ ^ ^ ^
4016 // start lp_start lp_end end
4017 //
4018 ::munmap(start, lp_start - start);
4019 ::munmap(lp_end, end - lp_end);
4020 return NULL;
4021 }
4022
4023 // Commit small-paged trailing area.
4024 if (lp_end != end) {
4025 result = ::mmap(lp_end, end - lp_end, prot,
4026 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
4027 -1, 0);
4028 if (result == MAP_FAILED) {
4029 ::munmap(start, lp_end - start);
4030 return NULL;
4031 }
4032 }
4033
4034 return start;
4035 }
4036
4037 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
4038 size_t alignment,
4039 char* req_addr,
4040 bool exec) {
4041 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
4042 assert(is_aligned(req_addr, alignment), "Must be");
4043 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
4044 assert(is_power_of_2(os::large_page_size()), "Must be");
4045 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
4046
4047 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
4048 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
4049 } else {
4050 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
4051 }
4052 }
4053
4054 char* os::reserve_memory_special(size_t bytes, size_t alignment,
4055 char* req_addr, bool exec) {
4056 assert(UseLargePages, "only for large pages");
4057
4058 char* addr;
4059 if (UseSHM) {
4060 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
4061 } else {
4062 assert(UseHugeTLBFS, "must be");
4063 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
4064 }
4065
4066 if (addr != NULL) {
4067 if (UseNUMAInterleaving) {
4068 numa_make_global(addr, bytes);
4069 }
4070
4071 // The memory is committed
4072 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
4073 }
4074
4075 return addr;
4076 }
4077
4078 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
4079 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
4080 return shmdt(base) == 0;
4081 }
4082
4083 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
4084 return pd_release_memory(base, bytes);
4085 }
4086
4087 bool os::release_memory_special(char* base, size_t bytes) {
4088 bool res;
4089 if (MemTracker::tracking_level() > NMT_minimal) {
4090 Tracker tkr(Tracker::release);
4091 res = os::Linux::release_memory_special_impl(base, bytes);
4092 if (res) {
4093 tkr.record((address)base, bytes);
4094 }
4095
4096 } else {
4097 res = os::Linux::release_memory_special_impl(base, bytes);
4098 }
4099 return res;
4100 }
4101
4102 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
4103 assert(UseLargePages, "only for large pages");
4104 bool res;
4105
4106 if (UseSHM) {
4107 res = os::Linux::release_memory_special_shm(base, bytes);
4108 } else {
4109 assert(UseHugeTLBFS, "must be");
4110 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
4111 }
4112 return res;
4113 }
4114
4115 size_t os::large_page_size() {
4116 return _large_page_size;
4117 }
4118
4119 // With SysV SHM the entire memory region must be allocated as shared
4120 // memory.
4121 // HugeTLBFS allows application to commit large page memory on demand.
4122 // However, when committing memory with HugeTLBFS fails, the region
4123 // that was supposed to be committed will lose the old reservation
4124 // and allow other threads to steal that memory region. Because of this
4125 // behavior we can't commit HugeTLBFS memory.
4126 bool os::can_commit_large_page_memory() {
4127 return UseTransparentHugePages;
4128 }
4129
4130 bool os::can_execute_large_page_memory() {
4131 return UseTransparentHugePages || UseHugeTLBFS;
4132 }
4133
4134 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
4135 assert(file_desc >= 0, "file_desc is not valid");
4136 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
4137 if (result != NULL) {
4138 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
4139 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
4140 }
4141 }
4142 return result;
4143 }
4144
4145 // Reserve memory at an arbitrary address, only if that area is
4146 // available (and not reserved for something else).
4147
4148 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
4149 const int max_tries = 10;
4150 char* base[max_tries];
4151 size_t size[max_tries];
4152 const size_t gap = 0x000000;
4153
4154 // Assert only that the size is a multiple of the page size, since
4155 // that's all that mmap requires, and since that's all we really know
4156 // about at this low abstraction level. If we need higher alignment,
4157 // we can either pass an alignment to this method or verify alignment
4158 // in one of the methods further up the call chain. See bug 5044738.
4159 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
4160
4161 // Repeatedly allocate blocks until the block is allocated at the
4162 // right spot.
4163
4164 // Linux mmap allows caller to pass an address as hint; give it a try first,
4165 // if kernel honors the hint then we can return immediately.
4166 char * addr = anon_mmap(requested_addr, bytes, false);
4167 if (addr == requested_addr) {
4168 return requested_addr;
4169 }
4170
4171 if (addr != NULL) {
4172 // mmap() is successful but it fails to reserve at the requested address
4173 anon_munmap(addr, bytes);
4174 }
4175
4176 int i;
4177 for (i = 0; i < max_tries; ++i) {
4178 base[i] = reserve_memory(bytes);
4179
4180 if (base[i] != NULL) {
4181 // Is this the block we wanted?
4182 if (base[i] == requested_addr) {
4183 size[i] = bytes;
4184 break;
4185 }
4186
4187 // Does this overlap the block we wanted? Give back the overlapped
4188 // parts and try again.
4189
4190 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
4191 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
4192 unmap_memory(base[i], top_overlap);
4193 base[i] += top_overlap;
4194 size[i] = bytes - top_overlap;
4195 } else {
4196 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
4197 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
4198 unmap_memory(requested_addr, bottom_overlap);
4199 size[i] = bytes - bottom_overlap;
4200 } else {
4201 size[i] = bytes;
4202 }
4203 }
4204 }
4205 }
4206
4207 // Give back the unused reserved pieces.
4208
4209 for (int j = 0; j < i; ++j) {
4210 if (base[j] != NULL) {
4211 unmap_memory(base[j], size[j]);
4212 }
4213 }
4214
4215 if (i < max_tries) {
4216 return requested_addr;
4217 } else {
4218 return NULL;
4219 }
4220 }
4221
4222 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4223 void os::infinite_sleep() {
4224 while (true) { // sleep forever ...
4225 ::sleep(100); // ... 100 seconds at a time
4226 }
4227 }
4228
4229 // Used to convert frequent JVM_Yield() to nops
4230 bool os::dont_yield() {
4231 return DontYieldALot;
4232 }
4233
4234 // Linux CFS scheduler (since 2.6.23) does not guarantee sched_yield(2) will
4235 // actually give up the CPU. Since skip buddy (v2.6.28):
4236 //
4237 // * Sets the yielding task as skip buddy for current CPU's run queue.
4238 // * Picks next from run queue, if empty, picks a skip buddy (can be the yielding task).
4239 // * Clears skip buddies for this run queue (yielding task no longer a skip buddy).
4240 //
4241 // An alternative is calling os::naked_short_nanosleep with a small number to avoid
4242 // getting re-scheduled immediately.
4243 //
4244 void os::naked_yield() {
4245 sched_yield();
4246 }
4247
4248 ////////////////////////////////////////////////////////////////////////////////
4249 // thread priority support
4250
4251 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4252 // only supports dynamic priority, static priority must be zero. For real-time
4253 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4254 // However, for large multi-threaded applications, SCHED_RR is not only slower
4255 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4256 // of 5 runs - Sep 2005).
4257 //
4258 // The following code actually changes the niceness of kernel-thread/LWP. It
4259 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4260 // not the entire user process, and user level threads are 1:1 mapped to kernel
4261 // threads. It has always been the case, but could change in the future. For
4262 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4263 // It is only used when ThreadPriorityPolicy=1 and may require system level permission
4264 // (e.g., root privilege or CAP_SYS_NICE capability).
4265
4266 int os::java_to_os_priority[CriticalPriority + 1] = {
4267 19, // 0 Entry should never be used
4268
4269 4, // 1 MinPriority
4270 3, // 2
4271 2, // 3
4272
4273 1, // 4
4274 0, // 5 NormPriority
4275 -1, // 6
4276
4277 -2, // 7
4278 -3, // 8
4279 -4, // 9 NearMaxPriority
4280
4281 -5, // 10 MaxPriority
4282
4283 -5 // 11 CriticalPriority
4284 };
4285
4286 static int prio_init() {
4287 if (ThreadPriorityPolicy == 1) {
4288 if (geteuid() != 0) {
4289 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4290 warning("-XX:ThreadPriorityPolicy=1 may require system level permission, " \
4291 "e.g., being the root user. If the necessary permission is not " \
4292 "possessed, changes to priority will be silently ignored.");
4293 }
4294 }
4295 }
4296 if (UseCriticalJavaThreadPriority) {
4297 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4298 }
4299 return 0;
4300 }
4301
4302 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4303 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4304
4305 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4306 return (ret == 0) ? OS_OK : OS_ERR;
4307 }
4308
4309 OSReturn os::get_native_priority(const Thread* const thread,
4310 int *priority_ptr) {
4311 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4312 *priority_ptr = java_to_os_priority[NormPriority];
4313 return OS_OK;
4314 }
4315
4316 errno = 0;
4317 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4318 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4319 }
4320
4321 ////////////////////////////////////////////////////////////////////////////////
4322 // suspend/resume support
4323
4324 // The low-level signal-based suspend/resume support is a remnant from the
4325 // old VM-suspension that used to be for java-suspension, safepoints etc,
4326 // within hotspot. Currently used by JFR's OSThreadSampler
4327 //
4328 // The remaining code is greatly simplified from the more general suspension
4329 // code that used to be used.
4330 //
4331 // The protocol is quite simple:
4332 // - suspend:
4333 // - sends a signal to the target thread
4334 // - polls the suspend state of the osthread using a yield loop
4335 // - target thread signal handler (SR_handler) sets suspend state
4336 // and blocks in sigsuspend until continued
4337 // - resume:
4338 // - sets target osthread state to continue
4339 // - sends signal to end the sigsuspend loop in the SR_handler
4340 //
4341 // Note that the SR_lock plays no role in this suspend/resume protocol,
4342 // but is checked for NULL in SR_handler as a thread termination indicator.
4343 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4344 //
4345 // Note that resume_clear_context() and suspend_save_context() are needed
4346 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4347 // which in part is used by:
4348 // - Forte Analyzer: AsyncGetCallTrace()
4349 // - StackBanging: get_frame_at_stack_banging_point()
4350
4351 static void resume_clear_context(OSThread *osthread) {
4352 osthread->set_ucontext(NULL);
4353 osthread->set_siginfo(NULL);
4354 }
4355
4356 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4357 ucontext_t* context) {
4358 osthread->set_ucontext(context);
4359 osthread->set_siginfo(siginfo);
4360 }
4361
4362 // Handler function invoked when a thread's execution is suspended or
4363 // resumed. We have to be careful that only async-safe functions are
4364 // called here (Note: most pthread functions are not async safe and
4365 // should be avoided.)
4366 //
4367 // Note: sigwait() is a more natural fit than sigsuspend() from an
4368 // interface point of view, but sigwait() prevents the signal hander
4369 // from being run. libpthread would get very confused by not having
4370 // its signal handlers run and prevents sigwait()'s use with the
4371 // mutex granting granting signal.
4372 //
4373 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4374 //
4375 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4376 // Save and restore errno to avoid confusing native code with EINTR
4377 // after sigsuspend.
4378 int old_errno = errno;
4379
4380 Thread* thread = Thread::current_or_null_safe();
4381 assert(thread != NULL, "Missing current thread in SR_handler");
4382
4383 // On some systems we have seen signal delivery get "stuck" until the signal
4384 // mask is changed as part of thread termination. Check that the current thread
4385 // has not already terminated (via SR_lock()) - else the following assertion
4386 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4387 // destructor has completed.
4388
4389 if (thread->SR_lock() == NULL) {
4390 return;
4391 }
4392
4393 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4394
4395 OSThread* osthread = thread->osthread();
4396
4397 os::SuspendResume::State current = osthread->sr.state();
4398 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4399 suspend_save_context(osthread, siginfo, context);
4400
4401 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4402 os::SuspendResume::State state = osthread->sr.suspended();
4403 if (state == os::SuspendResume::SR_SUSPENDED) {
4404 sigset_t suspend_set; // signals for sigsuspend()
4405 sigemptyset(&suspend_set);
4406 // get current set of blocked signals and unblock resume signal
4407 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4408 sigdelset(&suspend_set, SR_signum);
4409
4410 sr_semaphore.signal();
4411 // wait here until we are resumed
4412 while (1) {
4413 sigsuspend(&suspend_set);
4414
4415 os::SuspendResume::State result = osthread->sr.running();
4416 if (result == os::SuspendResume::SR_RUNNING) {
4417 sr_semaphore.signal();
4418 break;
4419 }
4420 }
4421
4422 } else if (state == os::SuspendResume::SR_RUNNING) {
4423 // request was cancelled, continue
4424 } else {
4425 ShouldNotReachHere();
4426 }
4427
4428 resume_clear_context(osthread);
4429 } else if (current == os::SuspendResume::SR_RUNNING) {
4430 // request was cancelled, continue
4431 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4432 // ignore
4433 } else {
4434 // ignore
4435 }
4436
4437 errno = old_errno;
4438 }
4439
4440 static int SR_initialize() {
4441 struct sigaction act;
4442 char *s;
4443
4444 // Get signal number to use for suspend/resume
4445 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4446 int sig = ::strtol(s, 0, 10);
4447 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4448 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4449 SR_signum = sig;
4450 } else {
4451 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4452 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4453 }
4454 }
4455
4456 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4457 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4458
4459 sigemptyset(&SR_sigset);
4460 sigaddset(&SR_sigset, SR_signum);
4461
4462 // Set up signal handler for suspend/resume
4463 act.sa_flags = SA_RESTART|SA_SIGINFO;
4464 act.sa_handler = (void (*)(int)) SR_handler;
4465
4466 // SR_signum is blocked by default.
4467 // 4528190 - We also need to block pthread restart signal (32 on all
4468 // supported Linux platforms). Note that LinuxThreads need to block
4469 // this signal for all threads to work properly. So we don't have
4470 // to use hard-coded signal number when setting up the mask.
4471 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4472
4473 if (sigaction(SR_signum, &act, 0) == -1) {
4474 return -1;
4475 }
4476
4477 // Save signal flag
4478 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4479 return 0;
4480 }
4481
4482 static int sr_notify(OSThread* osthread) {
4483 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4484 assert_status(status == 0, status, "pthread_kill");
4485 return status;
4486 }
4487
4488 // "Randomly" selected value for how long we want to spin
4489 // before bailing out on suspending a thread, also how often
4490 // we send a signal to a thread we want to resume
4491 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4492 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4493
4494 // returns true on success and false on error - really an error is fatal
4495 // but this seems the normal response to library errors
4496 static bool do_suspend(OSThread* osthread) {
4497 assert(osthread->sr.is_running(), "thread should be running");
4498 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4499
4500 // mark as suspended and send signal
4501 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4502 // failed to switch, state wasn't running?
4503 ShouldNotReachHere();
4504 return false;
4505 }
4506
4507 if (sr_notify(osthread) != 0) {
4508 ShouldNotReachHere();
4509 }
4510
4511 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4512 while (true) {
4513 if (sr_semaphore.timedwait(2)) {
4514 break;
4515 } else {
4516 // timeout
4517 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4518 if (cancelled == os::SuspendResume::SR_RUNNING) {
4519 return false;
4520 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4521 // make sure that we consume the signal on the semaphore as well
4522 sr_semaphore.wait();
4523 break;
4524 } else {
4525 ShouldNotReachHere();
4526 return false;
4527 }
4528 }
4529 }
4530
4531 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4532 return true;
4533 }
4534
4535 static void do_resume(OSThread* osthread) {
4536 assert(osthread->sr.is_suspended(), "thread should be suspended");
4537 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4538
4539 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4540 // failed to switch to WAKEUP_REQUEST
4541 ShouldNotReachHere();
4542 return;
4543 }
4544
4545 while (true) {
4546 if (sr_notify(osthread) == 0) {
4547 if (sr_semaphore.timedwait(2)) {
4548 if (osthread->sr.is_running()) {
4549 return;
4550 }
4551 }
4552 } else {
4553 ShouldNotReachHere();
4554 }
4555 }
4556
4557 guarantee(osthread->sr.is_running(), "Must be running!");
4558 }
4559
4560 ///////////////////////////////////////////////////////////////////////////////////
4561 // signal handling (except suspend/resume)
4562
4563 // This routine may be used by user applications as a "hook" to catch signals.
4564 // The user-defined signal handler must pass unrecognized signals to this
4565 // routine, and if it returns true (non-zero), then the signal handler must
4566 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4567 // routine will never retun false (zero), but instead will execute a VM panic
4568 // routine kill the process.
4569 //
4570 // If this routine returns false, it is OK to call it again. This allows
4571 // the user-defined signal handler to perform checks either before or after
4572 // the VM performs its own checks. Naturally, the user code would be making
4573 // a serious error if it tried to handle an exception (such as a null check
4574 // or breakpoint) that the VM was generating for its own correct operation.
4575 //
4576 // This routine may recognize any of the following kinds of signals:
4577 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4578 // It should be consulted by handlers for any of those signals.
4579 //
4580 // The caller of this routine must pass in the three arguments supplied
4581 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4582 // field of the structure passed to sigaction(). This routine assumes that
4583 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4584 //
4585 // Note that the VM will print warnings if it detects conflicting signal
4586 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4587 //
4588 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4589 siginfo_t* siginfo,
4590 void* ucontext,
4591 int abort_if_unrecognized);
4592
4593 static void signalHandler(int sig, siginfo_t* info, void* uc) {
4594 assert(info != NULL && uc != NULL, "it must be old kernel");
4595 int orig_errno = errno; // Preserve errno value over signal handler.
4596 JVM_handle_linux_signal(sig, info, uc, true);
4597 errno = orig_errno;
4598 }
4599
4600
4601 // This boolean allows users to forward their own non-matching signals
4602 // to JVM_handle_linux_signal, harmlessly.
4603 bool os::Linux::signal_handlers_are_installed = false;
4604
4605 // For signal-chaining
4606 bool os::Linux::libjsig_is_loaded = false;
4607 typedef struct sigaction *(*get_signal_t)(int);
4608 get_signal_t os::Linux::get_signal_action = NULL;
4609
4610 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4611 struct sigaction *actp = NULL;
4612
4613 if (libjsig_is_loaded) {
4614 // Retrieve the old signal handler from libjsig
4615 actp = (*get_signal_action)(sig);
4616 }
4617 if (actp == NULL) {
4618 // Retrieve the preinstalled signal handler from jvm
4619 actp = os::Posix::get_preinstalled_handler(sig);
4620 }
4621
4622 return actp;
4623 }
4624
4625 static bool call_chained_handler(struct sigaction *actp, int sig,
4626 siginfo_t *siginfo, void *context) {
4627 // Call the old signal handler
4628 if (actp->sa_handler == SIG_DFL) {
4629 // It's more reasonable to let jvm treat it as an unexpected exception
4630 // instead of taking the default action.
4631 return false;
4632 } else if (actp->sa_handler != SIG_IGN) {
4633 if ((actp->sa_flags & SA_NODEFER) == 0) {
4634 // automaticlly block the signal
4635 sigaddset(&(actp->sa_mask), sig);
4636 }
4637
4638 sa_handler_t hand = NULL;
4639 sa_sigaction_t sa = NULL;
4640 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4641 // retrieve the chained handler
4642 if (siginfo_flag_set) {
4643 sa = actp->sa_sigaction;
4644 } else {
4645 hand = actp->sa_handler;
4646 }
4647
4648 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4649 actp->sa_handler = SIG_DFL;
4650 }
4651
4652 // try to honor the signal mask
4653 sigset_t oset;
4654 sigemptyset(&oset);
4655 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4656
4657 // call into the chained handler
4658 if (siginfo_flag_set) {
4659 (*sa)(sig, siginfo, context);
4660 } else {
4661 (*hand)(sig);
4662 }
4663
4664 // restore the signal mask
4665 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4666 }
4667 // Tell jvm's signal handler the signal is taken care of.
4668 return true;
4669 }
4670
4671 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4672 bool chained = false;
4673 // signal-chaining
4674 if (UseSignalChaining) {
4675 struct sigaction *actp = get_chained_signal_action(sig);
4676 if (actp != NULL) {
4677 chained = call_chained_handler(actp, sig, siginfo, context);
4678 }
4679 }
4680 return chained;
4681 }
4682
4683 // for diagnostic
4684 int sigflags[NSIG];
4685
4686 int os::Linux::get_our_sigflags(int sig) {
4687 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4688 return sigflags[sig];
4689 }
4690
4691 void os::Linux::set_our_sigflags(int sig, int flags) {
4692 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4693 if (sig > 0 && sig < NSIG) {
4694 sigflags[sig] = flags;
4695 }
4696 }
4697
4698 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4699 // Check for overwrite.
4700 struct sigaction oldAct;
4701 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4702
4703 void* oldhand = oldAct.sa_sigaction
4704 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4705 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4706 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4707 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4708 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4709 if (AllowUserSignalHandlers || !set_installed) {
4710 // Do not overwrite; user takes responsibility to forward to us.
4711 return;
4712 } else if (UseSignalChaining) {
4713 // save the old handler in jvm
4714 os::Posix::save_preinstalled_handler(sig, oldAct);
4715 // libjsig also interposes the sigaction() call below and saves the
4716 // old sigaction on it own.
4717 } else {
4718 fatal("Encountered unexpected pre-existing sigaction handler "
4719 "%#lx for signal %d.", (long)oldhand, sig);
4720 }
4721 }
4722
4723 struct sigaction sigAct;
4724 sigfillset(&(sigAct.sa_mask));
4725 sigAct.sa_handler = SIG_DFL;
4726 if (!set_installed) {
4727 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4728 } else {
4729 sigAct.sa_sigaction = signalHandler;
4730 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4731 }
4732 // Save flags, which are set by ours
4733 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4734 sigflags[sig] = sigAct.sa_flags;
4735
4736 int ret = sigaction(sig, &sigAct, &oldAct);
4737 assert(ret == 0, "check");
4738
4739 void* oldhand2 = oldAct.sa_sigaction
4740 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4741 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4742 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4743 }
4744
4745 // install signal handlers for signals that HotSpot needs to
4746 // handle in order to support Java-level exception handling.
4747
4748 void os::Linux::install_signal_handlers() {
4749 if (!signal_handlers_are_installed) {
4750 signal_handlers_are_installed = true;
4751
4752 // signal-chaining
4753 typedef void (*signal_setting_t)();
4754 signal_setting_t begin_signal_setting = NULL;
4755 signal_setting_t end_signal_setting = NULL;
4756 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4757 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4758 if (begin_signal_setting != NULL) {
4759 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4760 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4761 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4762 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4763 libjsig_is_loaded = true;
4764 assert(UseSignalChaining, "should enable signal-chaining");
4765 }
4766 if (libjsig_is_loaded) {
4767 // Tell libjsig jvm is setting signal handlers
4768 (*begin_signal_setting)();
4769 }
4770
4771 set_signal_handler(SIGSEGV, true);
4772 set_signal_handler(SIGPIPE, true);
4773 set_signal_handler(SIGBUS, true);
4774 set_signal_handler(SIGILL, true);
4775 set_signal_handler(SIGFPE, true);
4776 #if defined(PPC64)
4777 set_signal_handler(SIGTRAP, true);
4778 #endif
4779 set_signal_handler(SIGXFSZ, true);
4780
4781 if (libjsig_is_loaded) {
4782 // Tell libjsig jvm finishes setting signal handlers
4783 (*end_signal_setting)();
4784 }
4785
4786 // We don't activate signal checker if libjsig is in place, we trust ourselves
4787 // and if UserSignalHandler is installed all bets are off.
4788 // Log that signal checking is off only if -verbose:jni is specified.
4789 if (CheckJNICalls) {
4790 if (libjsig_is_loaded) {
4791 if (PrintJNIResolving) {
4792 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4793 }
4794 check_signals = false;
4795 }
4796 if (AllowUserSignalHandlers) {
4797 if (PrintJNIResolving) {
4798 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4799 }
4800 check_signals = false;
4801 }
4802 }
4803 }
4804 }
4805
4806 // This is the fastest way to get thread cpu time on Linux.
4807 // Returns cpu time (user+sys) for any thread, not only for current.
4808 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4809 // It might work on 2.6.10+ with a special kernel/glibc patch.
4810 // For reference, please, see IEEE Std 1003.1-2004:
4811 // http://www.unix.org/single_unix_specification
4812
4813 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4814 struct timespec tp;
4815 int rc = os::Posix::clock_gettime(clockid, &tp);
4816 assert(rc == 0, "clock_gettime is expected to return 0 code");
4817
4818 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4819 }
4820
4821 void os::Linux::initialize_os_info() {
4822 assert(_os_version == 0, "OS info already initialized");
4823
4824 struct utsname _uname;
4825
4826 uint32_t major;
4827 uint32_t minor;
4828 uint32_t fix;
4829
4830 int rc;
4831
4832 // Kernel version is unknown if
4833 // verification below fails.
4834 _os_version = 0x01000000;
4835
4836 rc = uname(&_uname);
4837 if (rc != -1) {
4838
4839 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4840 if (rc == 3) {
4841
4842 if (major < 256 && minor < 256 && fix < 256) {
4843 // Kernel version format is as expected,
4844 // set it overriding unknown state.
4845 _os_version = (major << 16) |
4846 (minor << 8 ) |
4847 (fix << 0 ) ;
4848 }
4849 }
4850 }
4851 }
4852
4853 uint32_t os::Linux::os_version() {
4854 assert(_os_version != 0, "not initialized");
4855 return _os_version & 0x00FFFFFF;
4856 }
4857
4858 bool os::Linux::os_version_is_known() {
4859 assert(_os_version != 0, "not initialized");
4860 return _os_version & 0x01000000 ? false : true;
4861 }
4862
4863 /////
4864 // glibc on Linux platform uses non-documented flag
4865 // to indicate, that some special sort of signal
4866 // trampoline is used.
4867 // We will never set this flag, and we should
4868 // ignore this flag in our diagnostic
4869 #ifdef SIGNIFICANT_SIGNAL_MASK
4870 #undef SIGNIFICANT_SIGNAL_MASK
4871 #endif
4872 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4873
4874 static const char* get_signal_handler_name(address handler,
4875 char* buf, int buflen) {
4876 int offset = 0;
4877 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4878 if (found) {
4879 // skip directory names
4880 const char *p1, *p2;
4881 p1 = buf;
4882 size_t len = strlen(os::file_separator());
4883 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4884 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4885 } else {
4886 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4887 }
4888 return buf;
4889 }
4890
4891 static void print_signal_handler(outputStream* st, int sig,
4892 char* buf, size_t buflen) {
4893 struct sigaction sa;
4894
4895 sigaction(sig, NULL, &sa);
4896
4897 // See comment for SIGNIFICANT_SIGNAL_MASK define
4898 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4899
4900 st->print("%s: ", os::exception_name(sig, buf, buflen));
4901
4902 address handler = (sa.sa_flags & SA_SIGINFO)
4903 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4904 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4905
4906 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4907 st->print("SIG_DFL");
4908 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4909 st->print("SIG_IGN");
4910 } else {
4911 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4912 }
4913
4914 st->print(", sa_mask[0]=");
4915 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4916
4917 address rh = VMError::get_resetted_sighandler(sig);
4918 // May be, handler was resetted by VMError?
4919 if (rh != NULL) {
4920 handler = rh;
4921 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4922 }
4923
4924 st->print(", sa_flags=");
4925 os::Posix::print_sa_flags(st, sa.sa_flags);
4926
4927 // Check: is it our handler?
4928 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4929 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4930 // It is our signal handler
4931 // check for flags, reset system-used one!
4932 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4933 st->print(
4934 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4935 os::Linux::get_our_sigflags(sig));
4936 }
4937 }
4938 st->cr();
4939 }
4940
4941
4942 #define DO_SIGNAL_CHECK(sig) \
4943 do { \
4944 if (!sigismember(&check_signal_done, sig)) { \
4945 os::Linux::check_signal_handler(sig); \
4946 } \
4947 } while (0)
4948
4949 // This method is a periodic task to check for misbehaving JNI applications
4950 // under CheckJNI, we can add any periodic checks here
4951
4952 void os::run_periodic_checks() {
4953 if (check_signals == false) return;
4954
4955 // SEGV and BUS if overridden could potentially prevent
4956 // generation of hs*.log in the event of a crash, debugging
4957 // such a case can be very challenging, so we absolutely
4958 // check the following for a good measure:
4959 DO_SIGNAL_CHECK(SIGSEGV);
4960 DO_SIGNAL_CHECK(SIGILL);
4961 DO_SIGNAL_CHECK(SIGFPE);
4962 DO_SIGNAL_CHECK(SIGBUS);
4963 DO_SIGNAL_CHECK(SIGPIPE);
4964 DO_SIGNAL_CHECK(SIGXFSZ);
4965 #if defined(PPC64)
4966 DO_SIGNAL_CHECK(SIGTRAP);
4967 #endif
4968
4969 // ReduceSignalUsage allows the user to override these handlers
4970 // see comments at the very top and jvm_md.h
4971 if (!ReduceSignalUsage) {
4972 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4973 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4974 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4975 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4976 }
4977
4978 DO_SIGNAL_CHECK(SR_signum);
4979 }
4980
4981 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4982
4983 static os_sigaction_t os_sigaction = NULL;
4984
4985 void os::Linux::check_signal_handler(int sig) {
4986 char buf[O_BUFLEN];
4987 address jvmHandler = NULL;
4988
4989
4990 struct sigaction act;
4991 if (os_sigaction == NULL) {
4992 // only trust the default sigaction, in case it has been interposed
4993 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4994 if (os_sigaction == NULL) return;
4995 }
4996
4997 os_sigaction(sig, (struct sigaction*)NULL, &act);
4998
4999
5000 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
5001
5002 address thisHandler = (act.sa_flags & SA_SIGINFO)
5003 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
5004 : CAST_FROM_FN_PTR(address, act.sa_handler);
5005
5006
5007 switch (sig) {
5008 case SIGSEGV:
5009 case SIGBUS:
5010 case SIGFPE:
5011 case SIGPIPE:
5012 case SIGILL:
5013 case SIGXFSZ:
5014 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
5015 break;
5016
5017 case SHUTDOWN1_SIGNAL:
5018 case SHUTDOWN2_SIGNAL:
5019 case SHUTDOWN3_SIGNAL:
5020 case BREAK_SIGNAL:
5021 jvmHandler = (address)user_handler();
5022 break;
5023
5024 default:
5025 if (sig == SR_signum) {
5026 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
5027 } else {
5028 return;
5029 }
5030 break;
5031 }
5032
5033 if (thisHandler != jvmHandler) {
5034 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
5035 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
5036 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
5037 // No need to check this sig any longer
5038 sigaddset(&check_signal_done, sig);
5039 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
5040 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
5041 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
5042 exception_name(sig, buf, O_BUFLEN));
5043 }
5044 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
5045 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
5046 tty->print("expected:");
5047 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
5048 tty->cr();
5049 tty->print(" found:");
5050 os::Posix::print_sa_flags(tty, act.sa_flags);
5051 tty->cr();
5052 // No need to check this sig any longer
5053 sigaddset(&check_signal_done, sig);
5054 }
5055
5056 // Dump all the signal
5057 if (sigismember(&check_signal_done, sig)) {
5058 print_signal_handlers(tty, buf, O_BUFLEN);
5059 }
5060 }
5061
5062 extern void report_error(char* file_name, int line_no, char* title,
5063 char* format, ...);
5064
5065 // this is called _before_ most of the global arguments have been parsed
5066 void os::init(void) {
5067 char dummy; // used to get a guess on initial stack address
5068
5069 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
5070
5071 init_random(1234567);
5072
5073 Linux::set_page_size(sysconf(_SC_PAGESIZE));
5074 if (Linux::page_size() == -1) {
5075 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
5076 os::strerror(errno));
5077 }
5078 init_page_sizes((size_t) Linux::page_size());
5079
5080 Linux::initialize_system_info();
5081
5082 Linux::initialize_os_info();
5083
5084 os::Linux::CPUPerfTicks pticks;
5085 bool res = os::Linux::get_tick_information(&pticks, -1);
5086
5087 if (res && pticks.has_steal_ticks) {
5088 has_initial_tick_info = true;
5089 initial_total_ticks = pticks.total;
5090 initial_steal_ticks = pticks.steal;
5091 }
5092
5093 // _main_thread points to the thread that created/loaded the JVM.
5094 Linux::_main_thread = pthread_self();
5095
5096 // retrieve entry point for pthread_setname_np
5097 Linux::_pthread_setname_np =
5098 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
5099
5100 os::Posix::init();
5101
5102 initial_time_count = javaTimeNanos();
5103
5104 // Always warn if no monotonic clock available
5105 if (!os::Posix::supports_monotonic_clock()) {
5106 warning("No monotonic clock was available - timed services may " \
5107 "be adversely affected if the time-of-day clock changes");
5108 }
5109 }
5110
5111 // To install functions for atexit system call
5112 extern "C" {
5113 static void perfMemory_exit_helper() {
5114 perfMemory_exit();
5115 }
5116 }
5117
5118 void os::pd_init_container_support() {
5119 OSContainer::init();
5120 }
5121
5122 void os::Linux::numa_init() {
5123
5124 // Java can be invoked as
5125 // 1. Without numactl and heap will be allocated/configured on all nodes as
5126 // per the system policy.
5127 // 2. With numactl --interleave:
5128 // Use numa_get_interleave_mask(v2) API to get nodes bitmask. The same
5129 // API for membind case bitmask is reset.
5130 // Interleave is only hint and Kernel can fallback to other nodes if
5131 // no memory is available on the target nodes.
5132 // 3. With numactl --membind:
5133 // Use numa_get_membind(v2) API to get nodes bitmask. The same API for
5134 // interleave case returns bitmask of all nodes.
5135 // numa_all_nodes_ptr holds bitmask of all nodes.
5136 // numa_get_interleave_mask(v2) and numa_get_membind(v2) APIs returns correct
5137 // bitmask when externally configured to run on all or fewer nodes.
5138
5139 if (!Linux::libnuma_init()) {
5140 UseNUMA = false;
5141 } else {
5142 if ((Linux::numa_max_node() < 1) || Linux::is_bound_to_single_node()) {
5143 // If there's only one node (they start from 0) or if the process
5144 // is bound explicitly to a single node using membind, disable NUMA.
5145 UseNUMA = false;
5146 } else {
5147
5148 LogTarget(Info,os) log;
5149 LogStream ls(log);
5150
5151 Linux::set_configured_numa_policy(Linux::identify_numa_policy());
5152
5153 struct bitmask* bmp = Linux::_numa_membind_bitmask;
5154 const char* numa_mode = "membind";
5155
5156 if (Linux::is_running_in_interleave_mode()) {
5157 bmp = Linux::_numa_interleave_bitmask;
5158 numa_mode = "interleave";
5159 }
5160
5161 ls.print("UseNUMA is enabled and invoked in '%s' mode."
5162 " Heap will be configured using NUMA memory nodes:", numa_mode);
5163
5164 for (int node = 0; node <= Linux::numa_max_node(); node++) {
5165 if (Linux::_numa_bitmask_isbitset(bmp, node)) {
5166 ls.print(" %d", node);
5167 }
5168 }
5169 }
5170 }
5171
5172 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5173 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5174 // we can make the adaptive lgrp chunk resizing work. If the user specified both
5175 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5176 // and disable adaptive resizing.
5177 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5178 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5179 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5180 UseAdaptiveSizePolicy = false;
5181 UseAdaptiveNUMAChunkSizing = false;
5182 }
5183 }
5184
5185 if (!UseNUMA && ForceNUMA) {
5186 UseNUMA = true;
5187 }
5188 }
5189
5190 // this is called _after_ the global arguments have been parsed
5191 jint os::init_2(void) {
5192
5193 // This could be set after os::Posix::init() but all platforms
5194 // have to set it the same so we have to mirror Solaris.
5195 DEBUG_ONLY(os::set_mutex_init_done();)
5196
5197 os::Posix::init_2();
5198
5199 Linux::fast_thread_clock_init();
5200
5201 // initialize suspend/resume support - must do this before signal_sets_init()
5202 if (SR_initialize() != 0) {
5203 perror("SR_initialize failed");
5204 return JNI_ERR;
5205 }
5206
5207 Linux::signal_sets_init();
5208 Linux::install_signal_handlers();
5209 // Initialize data for jdk.internal.misc.Signal
5210 if (!ReduceSignalUsage) {
5211 jdk_misc_signal_init();
5212 }
5213
5214 InitGetTLSSize();
5215
5216 // Check and sets minimum stack sizes against command line options
5217 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
5218 return JNI_ERR;
5219 }
5220
5221 #if defined(IA32)
5222 // Need to ensure we've determined the process's initial stack to
5223 // perform the workaround
5224 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5225 workaround_expand_exec_shield_cs_limit();
5226 #else
5227 suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
5228 if (!suppress_primordial_thread_resolution) {
5229 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5230 }
5231 #endif
5232
5233 Linux::libpthread_init();
5234 Linux::sched_getcpu_init();
5235 log_info(os)("HotSpot is running with %s, %s",
5236 Linux::glibc_version(), Linux::libpthread_version());
5237
5238 if (UseNUMA) {
5239 Linux::numa_init();
5240 }
5241
5242 if (MaxFDLimit) {
5243 // set the number of file descriptors to max. print out error
5244 // if getrlimit/setrlimit fails but continue regardless.
5245 struct rlimit nbr_files;
5246 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5247 if (status != 0) {
5248 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5249 } else {
5250 nbr_files.rlim_cur = nbr_files.rlim_max;
5251 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5252 if (status != 0) {
5253 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5254 }
5255 }
5256 }
5257
5258 // Initialize lock used to serialize thread creation (see os::create_thread)
5259 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5260
5261 // at-exit methods are called in the reverse order of their registration.
5262 // atexit functions are called on return from main or as a result of a
5263 // call to exit(3C). There can be only 32 of these functions registered
5264 // and atexit() does not set errno.
5265
5266 if (PerfAllowAtExitRegistration) {
5267 // only register atexit functions if PerfAllowAtExitRegistration is set.
5268 // atexit functions can be delayed until process exit time, which
5269 // can be problematic for embedded VM situations. Embedded VMs should
5270 // call DestroyJavaVM() to assure that VM resources are released.
5271
5272 // note: perfMemory_exit_helper atexit function may be removed in
5273 // the future if the appropriate cleanup code can be added to the
5274 // VM_Exit VMOperation's doit method.
5275 if (atexit(perfMemory_exit_helper) != 0) {
5276 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5277 }
5278 }
5279
5280 // initialize thread priority policy
5281 prio_init();
5282
5283 if (!FLAG_IS_DEFAULT(AllocateHeapAt) || !FLAG_IS_DEFAULT(AllocateOldGenAt)) {
5284 set_coredump_filter(DAX_SHARED_BIT);
5285 }
5286
5287 if (DumpPrivateMappingsInCore) {
5288 set_coredump_filter(FILE_BACKED_PVT_BIT);
5289 }
5290
5291 if (DumpSharedMappingsInCore) {
5292 set_coredump_filter(FILE_BACKED_SHARED_BIT);
5293 }
5294
5295 return JNI_OK;
5296 }
5297
5298 // Mark the polling page as unreadable
5299 void os::make_polling_page_unreadable(void) {
5300 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5301 fatal("Could not disable polling page");
5302 }
5303 }
5304
5305 // Mark the polling page as readable
5306 void os::make_polling_page_readable(void) {
5307 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5308 fatal("Could not enable polling page");
5309 }
5310 }
5311
5312 // older glibc versions don't have this macro (which expands to
5313 // an optimized bit-counting function) so we have to roll our own
5314 #ifndef CPU_COUNT
5315
5316 static int _cpu_count(const cpu_set_t* cpus) {
5317 int count = 0;
5318 // only look up to the number of configured processors
5319 for (int i = 0; i < os::processor_count(); i++) {
5320 if (CPU_ISSET(i, cpus)) {
5321 count++;
5322 }
5323 }
5324 return count;
5325 }
5326
5327 #define CPU_COUNT(cpus) _cpu_count(cpus)
5328
5329 #endif // CPU_COUNT
5330
5331 // Get the current number of available processors for this process.
5332 // This value can change at any time during a process's lifetime.
5333 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5334 // If it appears there may be more than 1024 processors then we do a
5335 // dynamic check - see 6515172 for details.
5336 // If anything goes wrong we fallback to returning the number of online
5337 // processors - which can be greater than the number available to the process.
5338 int os::Linux::active_processor_count() {
5339 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5340 cpu_set_t* cpus_p = &cpus;
5341 int cpus_size = sizeof(cpu_set_t);
5342
5343 int configured_cpus = os::processor_count(); // upper bound on available cpus
5344 int cpu_count = 0;
5345
5346 // old build platforms may not support dynamic cpu sets
5347 #ifdef CPU_ALLOC
5348
5349 // To enable easy testing of the dynamic path on different platforms we
5350 // introduce a diagnostic flag: UseCpuAllocPath
5351 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5352 // kernel may use a mask bigger than cpu_set_t
5353 log_trace(os)("active_processor_count: using dynamic path %s"
5354 "- configured processors: %d",
5355 UseCpuAllocPath ? "(forced) " : "",
5356 configured_cpus);
5357 cpus_p = CPU_ALLOC(configured_cpus);
5358 if (cpus_p != NULL) {
5359 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5360 // zero it just to be safe
5361 CPU_ZERO_S(cpus_size, cpus_p);
5362 }
5363 else {
5364 // failed to allocate so fallback to online cpus
5365 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5366 log_trace(os)("active_processor_count: "
5367 "CPU_ALLOC failed (%s) - using "
5368 "online processor count: %d",
5369 os::strerror(errno), online_cpus);
5370 return online_cpus;
5371 }
5372 }
5373 else {
5374 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5375 configured_cpus);
5376 }
5377 #else // CPU_ALLOC
5378 // these stubs won't be executed
5379 #define CPU_COUNT_S(size, cpus) -1
5380 #define CPU_FREE(cpus)
5381
5382 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5383 configured_cpus);
5384 #endif // CPU_ALLOC
5385
5386 // pid 0 means the current thread - which we have to assume represents the process
5387 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5388 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5389 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5390 }
5391 else {
5392 cpu_count = CPU_COUNT(cpus_p);
5393 }
5394 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5395 }
5396 else {
5397 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5398 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5399 "which may exceed available processors", os::strerror(errno), cpu_count);
5400 }
5401
5402 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5403 CPU_FREE(cpus_p);
5404 }
5405
5406 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5407 return cpu_count;
5408 }
5409
5410 // Determine the active processor count from one of
5411 // three different sources:
5412 //
5413 // 1. User option -XX:ActiveProcessorCount
5414 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5415 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5416 //
5417 // Option 1, if specified, will always override.
5418 // If the cgroup subsystem is active and configured, we
5419 // will return the min of the cgroup and option 2 results.
5420 // This is required since tools, such as numactl, that
5421 // alter cpu affinity do not update cgroup subsystem
5422 // cpuset configuration files.
5423 int os::active_processor_count() {
5424 // User has overridden the number of active processors
5425 if (ActiveProcessorCount > 0) {
5426 log_trace(os)("active_processor_count: "
5427 "active processor count set by user : %d",
5428 ActiveProcessorCount);
5429 return ActiveProcessorCount;
5430 }
5431
5432 int active_cpus;
5433 if (OSContainer::is_containerized()) {
5434 active_cpus = OSContainer::active_processor_count();
5435 log_trace(os)("active_processor_count: determined by OSContainer: %d",
5436 active_cpus);
5437 } else {
5438 active_cpus = os::Linux::active_processor_count();
5439 }
5440
5441 return active_cpus;
5442 }
5443
5444 uint os::processor_id() {
5445 const int id = Linux::sched_getcpu();
5446 assert(id >= 0 && id < _processor_count, "Invalid processor id");
5447 return (uint)id;
5448 }
5449
5450 void os::set_native_thread_name(const char *name) {
5451 if (Linux::_pthread_setname_np) {
5452 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5453 snprintf(buf, sizeof(buf), "%s", name);
5454 buf[sizeof(buf) - 1] = '\0';
5455 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5456 // ERANGE should not happen; all other errors should just be ignored.
5457 assert(rc != ERANGE, "pthread_setname_np failed");
5458 }
5459 }
5460
5461 bool os::distribute_processes(uint length, uint* distribution) {
5462 // Not yet implemented.
5463 return false;
5464 }
5465
5466 bool os::bind_to_processor(uint processor_id) {
5467 // Not yet implemented.
5468 return false;
5469 }
5470
5471 ///
5472
5473 void os::SuspendedThreadTask::internal_do_task() {
5474 if (do_suspend(_thread->osthread())) {
5475 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5476 do_task(context);
5477 do_resume(_thread->osthread());
5478 }
5479 }
5480
5481 ////////////////////////////////////////////////////////////////////////////////
5482 // debug support
5483
5484 bool os::find(address addr, outputStream* st) {
5485 Dl_info dlinfo;
5486 memset(&dlinfo, 0, sizeof(dlinfo));
5487 if (dladdr(addr, &dlinfo) != 0) {
5488 st->print(PTR_FORMAT ": ", p2i(addr));
5489 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5490 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5491 p2i(addr) - p2i(dlinfo.dli_saddr));
5492 } else if (dlinfo.dli_fbase != NULL) {
5493 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5494 } else {
5495 st->print("<absolute address>");
5496 }
5497 if (dlinfo.dli_fname != NULL) {
5498 st->print(" in %s", dlinfo.dli_fname);
5499 }
5500 if (dlinfo.dli_fbase != NULL) {
5501 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5502 }
5503 st->cr();
5504
5505 if (Verbose) {
5506 // decode some bytes around the PC
5507 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5508 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5509 address lowest = (address) dlinfo.dli_sname;
5510 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5511 if (begin < lowest) begin = lowest;
5512 Dl_info dlinfo2;
5513 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5514 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5515 end = (address) dlinfo2.dli_saddr;
5516 }
5517 Disassembler::decode(begin, end, st);
5518 }
5519 return true;
5520 }
5521 return false;
5522 }
5523
5524 ////////////////////////////////////////////////////////////////////////////////
5525 // misc
5526
5527 // This does not do anything on Linux. This is basically a hook for being
5528 // able to use structured exception handling (thread-local exception filters)
5529 // on, e.g., Win32.
5530 void
5531 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5532 JavaCallArguments* args, Thread* thread) {
5533 f(value, method, args, thread);
5534 }
5535
5536 void os::print_statistics() {
5537 }
5538
5539 bool os::message_box(const char* title, const char* message) {
5540 int i;
5541 fdStream err(defaultStream::error_fd());
5542 for (i = 0; i < 78; i++) err.print_raw("=");
5543 err.cr();
5544 err.print_raw_cr(title);
5545 for (i = 0; i < 78; i++) err.print_raw("-");
5546 err.cr();
5547 err.print_raw_cr(message);
5548 for (i = 0; i < 78; i++) err.print_raw("=");
5549 err.cr();
5550
5551 char buf[16];
5552 // Prevent process from exiting upon "read error" without consuming all CPU
5553 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5554
5555 return buf[0] == 'y' || buf[0] == 'Y';
5556 }
5557
5558 // Is a (classpath) directory empty?
5559 bool os::dir_is_empty(const char* path) {
5560 DIR *dir = NULL;
5561 struct dirent *ptr;
5562
5563 dir = opendir(path);
5564 if (dir == NULL) return true;
5565
5566 // Scan the directory
5567 bool result = true;
5568 while (result && (ptr = readdir(dir)) != NULL) {
5569 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5570 result = false;
5571 }
5572 }
5573 closedir(dir);
5574 return result;
5575 }
5576
5577 // This code originates from JDK's sysOpen and open64_w
5578 // from src/solaris/hpi/src/system_md.c
5579
5580 int os::open(const char *path, int oflag, int mode) {
5581 if (strlen(path) > MAX_PATH - 1) {
5582 errno = ENAMETOOLONG;
5583 return -1;
5584 }
5585
5586 // All file descriptors that are opened in the Java process and not
5587 // specifically destined for a subprocess should have the close-on-exec
5588 // flag set. If we don't set it, then careless 3rd party native code
5589 // might fork and exec without closing all appropriate file descriptors
5590 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5591 // turn might:
5592 //
5593 // - cause end-of-file to fail to be detected on some file
5594 // descriptors, resulting in mysterious hangs, or
5595 //
5596 // - might cause an fopen in the subprocess to fail on a system
5597 // suffering from bug 1085341.
5598 //
5599 // (Yes, the default setting of the close-on-exec flag is a Unix
5600 // design flaw)
5601 //
5602 // See:
5603 // 1085341: 32-bit stdio routines should support file descriptors >255
5604 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5605 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5606 //
5607 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5608 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5609 // because it saves a system call and removes a small window where the flag
5610 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5611 // and we fall back to using FD_CLOEXEC (see below).
5612 #ifdef O_CLOEXEC
5613 oflag |= O_CLOEXEC;
5614 #endif
5615
5616 int fd = ::open64(path, oflag, mode);
5617 if (fd == -1) return -1;
5618
5619 //If the open succeeded, the file might still be a directory
5620 {
5621 struct stat64 buf64;
5622 int ret = ::fstat64(fd, &buf64);
5623 int st_mode = buf64.st_mode;
5624
5625 if (ret != -1) {
5626 if ((st_mode & S_IFMT) == S_IFDIR) {
5627 errno = EISDIR;
5628 ::close(fd);
5629 return -1;
5630 }
5631 } else {
5632 ::close(fd);
5633 return -1;
5634 }
5635 }
5636
5637 #ifdef FD_CLOEXEC
5638 // Validate that the use of the O_CLOEXEC flag on open above worked.
5639 // With recent kernels, we will perform this check exactly once.
5640 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5641 if (!O_CLOEXEC_is_known_to_work) {
5642 int flags = ::fcntl(fd, F_GETFD);
5643 if (flags != -1) {
5644 if ((flags & FD_CLOEXEC) != 0)
5645 O_CLOEXEC_is_known_to_work = 1;
5646 else
5647 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5648 }
5649 }
5650 #endif
5651
5652 return fd;
5653 }
5654
5655
5656 // create binary file, rewriting existing file if required
5657 int os::create_binary_file(const char* path, bool rewrite_existing) {
5658 int oflags = O_WRONLY | O_CREAT;
5659 if (!rewrite_existing) {
5660 oflags |= O_EXCL;
5661 }
5662 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5663 }
5664
5665 // return current position of file pointer
5666 jlong os::current_file_offset(int fd) {
5667 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5668 }
5669
5670 // move file pointer to the specified offset
5671 jlong os::seek_to_file_offset(int fd, jlong offset) {
5672 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5673 }
5674
5675 // This code originates from JDK's sysAvailable
5676 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5677
5678 int os::available(int fd, jlong *bytes) {
5679 jlong cur, end;
5680 int mode;
5681 struct stat64 buf64;
5682
5683 if (::fstat64(fd, &buf64) >= 0) {
5684 mode = buf64.st_mode;
5685 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5686 int n;
5687 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5688 *bytes = n;
5689 return 1;
5690 }
5691 }
5692 }
5693 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5694 return 0;
5695 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5696 return 0;
5697 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5698 return 0;
5699 }
5700 *bytes = end - cur;
5701 return 1;
5702 }
5703
5704 // Map a block of memory.
5705 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5706 char *addr, size_t bytes, bool read_only,
5707 bool allow_exec) {
5708 int prot;
5709 int flags = MAP_PRIVATE;
5710
5711 if (read_only) {
5712 prot = PROT_READ;
5713 } else {
5714 prot = PROT_READ | PROT_WRITE;
5715 }
5716
5717 if (allow_exec) {
5718 prot |= PROT_EXEC;
5719 }
5720
5721 if (addr != NULL) {
5722 flags |= MAP_FIXED;
5723 }
5724
5725 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5726 fd, file_offset);
5727 if (mapped_address == MAP_FAILED) {
5728 return NULL;
5729 }
5730 return mapped_address;
5731 }
5732
5733
5734 // Remap a block of memory.
5735 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5736 char *addr, size_t bytes, bool read_only,
5737 bool allow_exec) {
5738 // same as map_memory() on this OS
5739 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5740 allow_exec);
5741 }
5742
5743
5744 // Unmap a block of memory.
5745 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5746 return munmap(addr, bytes) == 0;
5747 }
5748
5749 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5750
5751 static jlong fast_cpu_time(Thread *thread) {
5752 clockid_t clockid;
5753 int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(),
5754 &clockid);
5755 if (rc == 0) {
5756 return os::Linux::fast_thread_cpu_time(clockid);
5757 } else {
5758 // It's possible to encounter a terminated native thread that failed
5759 // to detach itself from the VM - which should result in ESRCH.
5760 assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed");
5761 return -1;
5762 }
5763 }
5764
5765 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5766 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5767 // of a thread.
5768 //
5769 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5770 // the fast estimate available on the platform.
5771
5772 jlong os::current_thread_cpu_time() {
5773 if (os::Linux::supports_fast_thread_cpu_time()) {
5774 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5775 } else {
5776 // return user + sys since the cost is the same
5777 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5778 }
5779 }
5780
5781 jlong os::thread_cpu_time(Thread* thread) {
5782 // consistent with what current_thread_cpu_time() returns
5783 if (os::Linux::supports_fast_thread_cpu_time()) {
5784 return fast_cpu_time(thread);
5785 } else {
5786 return slow_thread_cpu_time(thread, true /* user + sys */);
5787 }
5788 }
5789
5790 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5791 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5792 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5793 } else {
5794 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5795 }
5796 }
5797
5798 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5799 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5800 return fast_cpu_time(thread);
5801 } else {
5802 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5803 }
5804 }
5805
5806 // -1 on error.
5807 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5808 pid_t tid = thread->osthread()->thread_id();
5809 char *s;
5810 char stat[2048];
5811 int statlen;
5812 char proc_name[64];
5813 int count;
5814 long sys_time, user_time;
5815 char cdummy;
5816 int idummy;
5817 long ldummy;
5818 FILE *fp;
5819
5820 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5821 fp = fopen(proc_name, "r");
5822 if (fp == NULL) return -1;
5823 statlen = fread(stat, 1, 2047, fp);
5824 stat[statlen] = '\0';
5825 fclose(fp);
5826
5827 // Skip pid and the command string. Note that we could be dealing with
5828 // weird command names, e.g. user could decide to rename java launcher
5829 // to "java 1.4.2 :)", then the stat file would look like
5830 // 1234 (java 1.4.2 :)) R ... ...
5831 // We don't really need to know the command string, just find the last
5832 // occurrence of ")" and then start parsing from there. See bug 4726580.
5833 s = strrchr(stat, ')');
5834 if (s == NULL) return -1;
5835
5836 // Skip blank chars
5837 do { s++; } while (s && isspace(*s));
5838
5839 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5840 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5841 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5842 &user_time, &sys_time);
5843 if (count != 13) return -1;
5844 if (user_sys_cpu_time) {
5845 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5846 } else {
5847 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5848 }
5849 }
5850
5851 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5852 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5853 info_ptr->may_skip_backward = false; // elapsed time not wall time
5854 info_ptr->may_skip_forward = false; // elapsed time not wall time
5855 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5856 }
5857
5858 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5859 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5860 info_ptr->may_skip_backward = false; // elapsed time not wall time
5861 info_ptr->may_skip_forward = false; // elapsed time not wall time
5862 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5863 }
5864
5865 bool os::is_thread_cpu_time_supported() {
5866 return true;
5867 }
5868
5869 // System loadavg support. Returns -1 if load average cannot be obtained.
5870 // Linux doesn't yet have a (official) notion of processor sets,
5871 // so just return the system wide load average.
5872 int os::loadavg(double loadavg[], int nelem) {
5873 return ::getloadavg(loadavg, nelem);
5874 }
5875
5876 void os::pause() {
5877 char filename[MAX_PATH];
5878 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5879 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5880 } else {
5881 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5882 }
5883
5884 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5885 if (fd != -1) {
5886 struct stat buf;
5887 ::close(fd);
5888 while (::stat(filename, &buf) == 0) {
5889 (void)::poll(NULL, 0, 100);
5890 }
5891 } else {
5892 jio_fprintf(stderr,
5893 "Could not open pause file '%s', continuing immediately.\n", filename);
5894 }
5895 }
5896
5897 extern char** environ;
5898
5899 // Run the specified command in a separate process. Return its exit value,
5900 // or -1 on failure (e.g. can't fork a new process).
5901 // Unlike system(), this function can be called from signal handler. It
5902 // doesn't block SIGINT et al.
5903 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) {
5904 const char * argv[4] = {"sh", "-c", cmd, NULL};
5905
5906 pid_t pid ;
5907
5908 if (use_vfork_if_available) {
5909 pid = vfork();
5910 } else {
5911 pid = fork();
5912 }
5913
5914 if (pid < 0) {
5915 // fork failed
5916 return -1;
5917
5918 } else if (pid == 0) {
5919 // child process
5920
5921 execve("/bin/sh", (char* const*)argv, environ);
5922
5923 // execve failed
5924 _exit(-1);
5925
5926 } else {
5927 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5928 // care about the actual exit code, for now.
5929
5930 int status;
5931
5932 // Wait for the child process to exit. This returns immediately if
5933 // the child has already exited. */
5934 while (waitpid(pid, &status, 0) < 0) {
5935 switch (errno) {
5936 case ECHILD: return 0;
5937 case EINTR: break;
5938 default: return -1;
5939 }
5940 }
5941
5942 if (WIFEXITED(status)) {
5943 // The child exited normally; get its exit code.
5944 return WEXITSTATUS(status);
5945 } else if (WIFSIGNALED(status)) {
5946 // The child exited because of a signal
5947 // The best value to return is 0x80 + signal number,
5948 // because that is what all Unix shells do, and because
5949 // it allows callers to distinguish between process exit and
5950 // process death by signal.
5951 return 0x80 + WTERMSIG(status);
5952 } else {
5953 // Unknown exit code; pass it through
5954 return status;
5955 }
5956 }
5957 }
5958
5959 // Get the default path to the core file
5960 // Returns the length of the string
5961 int os::get_core_path(char* buffer, size_t bufferSize) {
5962 /*
5963 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5964 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5965 */
5966 const int core_pattern_len = 129;
5967 char core_pattern[core_pattern_len] = {0};
5968
5969 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5970 if (core_pattern_file == -1) {
5971 return -1;
5972 }
5973
5974 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5975 ::close(core_pattern_file);
5976 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5977 return -1;
5978 }
5979 if (core_pattern[ret-1] == '\n') {
5980 core_pattern[ret-1] = '\0';
5981 } else {
5982 core_pattern[ret] = '\0';
5983 }
5984
5985 // Replace the %p in the core pattern with the process id. NOTE: we do this
5986 // only if the pattern doesn't start with "|", and we support only one %p in
5987 // the pattern.
5988 char *pid_pos = strstr(core_pattern, "%p");
5989 const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : ""; // skip over the "%p"
5990 int written;
5991
5992 if (core_pattern[0] == '/') {
5993 if (pid_pos != NULL) {
5994 *pid_pos = '\0';
5995 written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern,
5996 current_process_id(), tail);
5997 } else {
5998 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5999 }
6000 } else {
6001 char cwd[PATH_MAX];
6002
6003 const char* p = get_current_directory(cwd, PATH_MAX);
6004 if (p == NULL) {
6005 return -1;
6006 }
6007
6008 if (core_pattern[0] == '|') {
6009 written = jio_snprintf(buffer, bufferSize,
6010 "\"%s\" (or dumping to %s/core.%d)",
6011 &core_pattern[1], p, current_process_id());
6012 } else if (pid_pos != NULL) {
6013 *pid_pos = '\0';
6014 written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern,
6015 current_process_id(), tail);
6016 } else {
6017 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6018 }
6019 }
6020
6021 if (written < 0) {
6022 return -1;
6023 }
6024
6025 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6026 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6027
6028 if (core_uses_pid_file != -1) {
6029 char core_uses_pid = 0;
6030 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6031 ::close(core_uses_pid_file);
6032
6033 if (core_uses_pid == '1') {
6034 jio_snprintf(buffer + written, bufferSize - written,
6035 ".%d", current_process_id());
6036 }
6037 }
6038 }
6039
6040 return strlen(buffer);
6041 }
6042
6043 bool os::start_debugging(char *buf, int buflen) {
6044 int len = (int)strlen(buf);
6045 char *p = &buf[len];
6046
6047 jio_snprintf(p, buflen-len,
6048 "\n\n"
6049 "Do you want to debug the problem?\n\n"
6050 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
6051 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
6052 "Otherwise, press RETURN to abort...",
6053 os::current_process_id(), os::current_process_id(),
6054 os::current_thread_id(), os::current_thread_id());
6055
6056 bool yes = os::message_box("Unexpected Error", buf);
6057
6058 if (yes) {
6059 // yes, user asked VM to launch debugger
6060 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
6061 os::current_process_id(), os::current_process_id());
6062
6063 os::fork_and_exec(buf);
6064 yes = false;
6065 }
6066 return yes;
6067 }
6068
6069
6070 // Java/Compiler thread:
6071 //
6072 // Low memory addresses
6073 // P0 +------------------------+
6074 // | |\ Java thread created by VM does not have glibc
6075 // | glibc guard page | - guard page, attached Java thread usually has
6076 // | |/ 1 glibc guard page.
6077 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6078 // | |\
6079 // | HotSpot Guard Pages | - red, yellow and reserved pages
6080 // | |/
6081 // +------------------------+ JavaThread::stack_reserved_zone_base()
6082 // | |\
6083 // | Normal Stack | -
6084 // | |/
6085 // P2 +------------------------+ Thread::stack_base()
6086 //
6087 // Non-Java thread:
6088 //
6089 // Low memory addresses
6090 // P0 +------------------------+
6091 // | |\
6092 // | glibc guard page | - usually 1 page
6093 // | |/
6094 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6095 // | |\
6096 // | Normal Stack | -
6097 // | |/
6098 // P2 +------------------------+ Thread::stack_base()
6099 //
6100 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
6101 // returned from pthread_attr_getstack().
6102 // ** Due to NPTL implementation error, linux takes the glibc guard page out
6103 // of the stack size given in pthread_attr. We work around this for
6104 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
6105 //
6106 #ifndef ZERO
6107 static void current_stack_region(address * bottom, size_t * size) {
6108 if (os::is_primordial_thread()) {
6109 // primordial thread needs special handling because pthread_getattr_np()
6110 // may return bogus value.
6111 *bottom = os::Linux::initial_thread_stack_bottom();
6112 *size = os::Linux::initial_thread_stack_size();
6113 } else {
6114 pthread_attr_t attr;
6115
6116 int rslt = pthread_getattr_np(pthread_self(), &attr);
6117
6118 // JVM needs to know exact stack location, abort if it fails
6119 if (rslt != 0) {
6120 if (rslt == ENOMEM) {
6121 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
6122 } else {
6123 fatal("pthread_getattr_np failed with error = %d", rslt);
6124 }
6125 }
6126
6127 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
6128 fatal("Cannot locate current stack attributes!");
6129 }
6130
6131 // Work around NPTL stack guard error.
6132 size_t guard_size = 0;
6133 rslt = pthread_attr_getguardsize(&attr, &guard_size);
6134 if (rslt != 0) {
6135 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
6136 }
6137 *bottom += guard_size;
6138 *size -= guard_size;
6139
6140 pthread_attr_destroy(&attr);
6141
6142 }
6143 assert(os::current_stack_pointer() >= *bottom &&
6144 os::current_stack_pointer() < *bottom + *size, "just checking");
6145 }
6146
6147 address os::current_stack_base() {
6148 address bottom;
6149 size_t size;
6150 current_stack_region(&bottom, &size);
6151 return (bottom + size);
6152 }
6153
6154 size_t os::current_stack_size() {
6155 // This stack size includes the usable stack and HotSpot guard pages
6156 // (for the threads that have Hotspot guard pages).
6157 address bottom;
6158 size_t size;
6159 current_stack_region(&bottom, &size);
6160 return size;
6161 }
6162 #endif
6163
6164 static inline struct timespec get_mtime(const char* filename) {
6165 struct stat st;
6166 int ret = os::stat(filename, &st);
6167 assert(ret == 0, "failed to stat() file '%s': %s", filename, os::strerror(errno));
6168 return st.st_mtim;
6169 }
6170
6171 int os::compare_file_modified_times(const char* file1, const char* file2) {
6172 struct timespec filetime1 = get_mtime(file1);
6173 struct timespec filetime2 = get_mtime(file2);
6174 int diff = filetime1.tv_sec - filetime2.tv_sec;
6175 if (diff == 0) {
6176 return filetime1.tv_nsec - filetime2.tv_nsec;
6177 }
6178 return diff;
6179 }
6180
6181 /////////////// Unit tests ///////////////
6182
6183 #ifndef PRODUCT
6184
6185 class TestReserveMemorySpecial : AllStatic {
6186 public:
6187 static void small_page_write(void* addr, size_t size) {
6188 size_t page_size = os::vm_page_size();
6189
6190 char* end = (char*)addr + size;
6191 for (char* p = (char*)addr; p < end; p += page_size) {
6192 *p = 1;
6193 }
6194 }
6195
6196 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6197 if (!UseHugeTLBFS) {
6198 return;
6199 }
6200
6201 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6202
6203 if (addr != NULL) {
6204 small_page_write(addr, size);
6205
6206 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6207 }
6208 }
6209
6210 static void test_reserve_memory_special_huge_tlbfs_only() {
6211 if (!UseHugeTLBFS) {
6212 return;
6213 }
6214
6215 size_t lp = os::large_page_size();
6216
6217 for (size_t size = lp; size <= lp * 10; size += lp) {
6218 test_reserve_memory_special_huge_tlbfs_only(size);
6219 }
6220 }
6221
6222 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6223 size_t lp = os::large_page_size();
6224 size_t ag = os::vm_allocation_granularity();
6225
6226 // sizes to test
6227 const size_t sizes[] = {
6228 lp, lp + ag, lp + lp / 2, lp * 2,
6229 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6230 lp * 10, lp * 10 + lp / 2
6231 };
6232 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6233
6234 // For each size/alignment combination, we test three scenarios:
6235 // 1) with req_addr == NULL
6236 // 2) with a non-null req_addr at which we expect to successfully allocate
6237 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6238 // expect the allocation to either fail or to ignore req_addr
6239
6240 // Pre-allocate two areas; they shall be as large as the largest allocation
6241 // and aligned to the largest alignment we will be testing.
6242 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6243 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6244 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6245 -1, 0);
6246 assert(mapping1 != MAP_FAILED, "should work");
6247
6248 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6249 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6250 -1, 0);
6251 assert(mapping2 != MAP_FAILED, "should work");
6252
6253 // Unmap the first mapping, but leave the second mapping intact: the first
6254 // mapping will serve as a value for a "good" req_addr (case 2). The second
6255 // mapping, still intact, as "bad" req_addr (case 3).
6256 ::munmap(mapping1, mapping_size);
6257
6258 // Case 1
6259 for (int i = 0; i < num_sizes; i++) {
6260 const size_t size = sizes[i];
6261 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6262 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6263 if (p != NULL) {
6264 assert(is_aligned(p, alignment), "must be");
6265 small_page_write(p, size);
6266 os::Linux::release_memory_special_huge_tlbfs(p, size);
6267 }
6268 }
6269 }
6270
6271 // Case 2
6272 for (int i = 0; i < num_sizes; i++) {
6273 const size_t size = sizes[i];
6274 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6275 char* const req_addr = align_up(mapping1, alignment);
6276 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6277 if (p != NULL) {
6278 assert(p == req_addr, "must be");
6279 small_page_write(p, size);
6280 os::Linux::release_memory_special_huge_tlbfs(p, size);
6281 }
6282 }
6283 }
6284
6285 // Case 3
6286 for (int i = 0; i < num_sizes; i++) {
6287 const size_t size = sizes[i];
6288 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6289 char* const req_addr = align_up(mapping2, alignment);
6290 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6291 // as the area around req_addr contains already existing mappings, the API should always
6292 // return NULL (as per contract, it cannot return another address)
6293 assert(p == NULL, "must be");
6294 }
6295 }
6296
6297 ::munmap(mapping2, mapping_size);
6298
6299 }
6300
6301 static void test_reserve_memory_special_huge_tlbfs() {
6302 if (!UseHugeTLBFS) {
6303 return;
6304 }
6305
6306 test_reserve_memory_special_huge_tlbfs_only();
6307 test_reserve_memory_special_huge_tlbfs_mixed();
6308 }
6309
6310 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6311 if (!UseSHM) {
6312 return;
6313 }
6314
6315 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6316
6317 if (addr != NULL) {
6318 assert(is_aligned(addr, alignment), "Check");
6319 assert(is_aligned(addr, os::large_page_size()), "Check");
6320
6321 small_page_write(addr, size);
6322
6323 os::Linux::release_memory_special_shm(addr, size);
6324 }
6325 }
6326
6327 static void test_reserve_memory_special_shm() {
6328 size_t lp = os::large_page_size();
6329 size_t ag = os::vm_allocation_granularity();
6330
6331 for (size_t size = ag; size < lp * 3; size += ag) {
6332 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6333 test_reserve_memory_special_shm(size, alignment);
6334 }
6335 }
6336 }
6337
6338 static void test() {
6339 test_reserve_memory_special_huge_tlbfs();
6340 test_reserve_memory_special_shm();
6341 }
6342 };
6343
6344 void TestReserveMemorySpecial_test() {
6345 TestReserveMemorySpecial::test();
6346 }
6347
6348 #endif