1 /*
2 * Copyright (c) 1999, 2020, 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/osThread.hpp"
57 #include "runtime/perfMemory.hpp"
58 #include "runtime/sharedRuntime.hpp"
59 #include "runtime/statSampler.hpp"
60 #include "runtime/stubRoutines.hpp"
61 #include "runtime/thread.inline.hpp"
62 #include "runtime/threadCritical.hpp"
63 #include "runtime/threadSMR.hpp"
64 #include "runtime/timer.hpp"
65 #include "runtime/vm_version.hpp"
66 #include "semaphore_posix.hpp"
67 #include "services/attachListener.hpp"
68 #include "services/memTracker.hpp"
69 #include "services/runtimeService.hpp"
70 #include "utilities/align.hpp"
71 #include "utilities/decoder.hpp"
72 #include "utilities/defaultStream.hpp"
73 #include "utilities/events.hpp"
74 #include "utilities/elfFile.hpp"
75 #include "utilities/growableArray.hpp"
76 #include "utilities/macros.hpp"
77 #include "utilities/vmError.hpp"
78
79 // put OS-includes here
80 # include <sys/types.h>
81 # include <sys/mman.h>
82 # include <sys/stat.h>
83 # include <sys/select.h>
84 # include <pthread.h>
85 # include <signal.h>
86 # include <endian.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 pthread_t os::Linux::_main_thread;
151 int os::Linux::_page_size = -1;
152 bool os::Linux::_supports_fast_thread_cpu_time = false;
153 const char * os::Linux::_glibc_version = NULL;
154 const char * os::Linux::_libpthread_version = NULL;
155
156 static jlong initial_time_count=0;
157
158 static int clock_tics_per_sec = 100;
159
160 // If the VM might have been created on the primordial thread, we need to resolve the
161 // primordial thread stack bounds and check if the current thread might be the
162 // primordial thread in places. If we know that the primordial thread is never used,
163 // such as when the VM was created by one of the standard java launchers, we can
164 // avoid this
165 static bool suppress_primordial_thread_resolution = false;
166
167 // For diagnostics to print a message once. see run_periodic_checks
168 static sigset_t check_signal_done;
169 static bool check_signals = true;
170
171 // Signal number used to suspend/resume a thread
172
173 // do not use any signal number less than SIGSEGV, see 4355769
174 static int SR_signum = SIGUSR2;
175 sigset_t SR_sigset;
176
177 // utility functions
178
179 static int SR_initialize();
180
181 julong os::available_memory() {
182 return Linux::available_memory();
183 }
184
185 julong os::Linux::available_memory() {
186 // values in struct sysinfo are "unsigned long"
187 struct sysinfo si;
188 julong avail_mem;
189
190 if (OSContainer::is_containerized()) {
191 jlong mem_limit, mem_usage;
192 if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
193 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
194 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
195 }
196 if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
197 log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
198 }
199 if (mem_limit > 0 && mem_usage > 0 ) {
200 avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
201 log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
202 return avail_mem;
203 }
204 }
205
206 sysinfo(&si);
207 avail_mem = (julong)si.freeram * si.mem_unit;
208 log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
209 return avail_mem;
210 }
211
212 julong os::physical_memory() {
213 jlong phys_mem = 0;
214 if (OSContainer::is_containerized()) {
215 jlong mem_limit;
216 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
217 log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
218 return mem_limit;
219 }
220 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
221 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
222 }
223
224 phys_mem = Linux::physical_memory();
225 log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
226 return phys_mem;
227 }
228
229 static uint64_t initial_total_ticks = 0;
230 static uint64_t initial_steal_ticks = 0;
231 static bool has_initial_tick_info = false;
232
233 static void next_line(FILE *f) {
234 int c;
235 do {
236 c = fgetc(f);
237 } while (c != '\n' && c != EOF);
238 }
239
240 bool os::Linux::get_tick_information(CPUPerfTicks* pticks, int which_logical_cpu) {
241 FILE* fh;
242 uint64_t userTicks, niceTicks, systemTicks, idleTicks;
243 // since at least kernel 2.6 : iowait: time waiting for I/O to complete
244 // irq: time servicing interrupts; softirq: time servicing softirqs
245 uint64_t iowTicks = 0, irqTicks = 0, sirqTicks= 0;
246 // steal (since kernel 2.6.11): time spent in other OS when running in a virtualized environment
247 uint64_t stealTicks = 0;
248 // guest (since kernel 2.6.24): time spent running a virtual CPU for guest OS under the
249 // control of the Linux kernel
250 uint64_t guestNiceTicks = 0;
251 int logical_cpu = -1;
252 const int required_tickinfo_count = (which_logical_cpu == -1) ? 4 : 5;
253 int n;
254
255 memset(pticks, 0, sizeof(CPUPerfTicks));
256
257 if ((fh = fopen("/proc/stat", "r")) == NULL) {
258 return false;
259 }
260
261 if (which_logical_cpu == -1) {
262 n = fscanf(fh, "cpu " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
263 UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
264 UINT64_FORMAT " " UINT64_FORMAT " ",
265 &userTicks, &niceTicks, &systemTicks, &idleTicks,
266 &iowTicks, &irqTicks, &sirqTicks,
267 &stealTicks, &guestNiceTicks);
268 } else {
269 // Move to next line
270 next_line(fh);
271
272 // find the line for requested cpu faster to just iterate linefeeds?
273 for (int i = 0; i < which_logical_cpu; i++) {
274 next_line(fh);
275 }
276
277 n = fscanf(fh, "cpu%u " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
278 UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
279 UINT64_FORMAT " " UINT64_FORMAT " ",
280 &logical_cpu, &userTicks, &niceTicks,
281 &systemTicks, &idleTicks, &iowTicks, &irqTicks, &sirqTicks,
282 &stealTicks, &guestNiceTicks);
283 }
284
285 fclose(fh);
286 if (n < required_tickinfo_count || logical_cpu != which_logical_cpu) {
287 return false;
288 }
289 pticks->used = userTicks + niceTicks;
290 pticks->usedKernel = systemTicks + irqTicks + sirqTicks;
291 pticks->total = userTicks + niceTicks + systemTicks + idleTicks +
292 iowTicks + irqTicks + sirqTicks + stealTicks + guestNiceTicks;
293
294 if (n > required_tickinfo_count + 3) {
295 pticks->steal = stealTicks;
296 pticks->has_steal_ticks = true;
297 } else {
298 pticks->steal = 0;
299 pticks->has_steal_ticks = false;
300 }
301
302 return true;
303 }
304
305 // Return true if user is running as root.
306
307 bool os::have_special_privileges() {
308 static bool init = false;
309 static bool privileges = false;
310 if (!init) {
311 privileges = (getuid() != geteuid()) || (getgid() != getegid());
312 init = true;
313 }
314 return privileges;
315 }
316
317
318 #ifndef SYS_gettid
319 // i386: 224, ia64: 1105, amd64: 186, sparc 143
320 #ifdef __ia64__
321 #define SYS_gettid 1105
322 #else
323 #ifdef __i386__
324 #define SYS_gettid 224
325 #else
326 #ifdef __amd64__
327 #define SYS_gettid 186
328 #else
329 #ifdef __sparc__
330 #define SYS_gettid 143
331 #else
332 #error define gettid for the arch
333 #endif
334 #endif
335 #endif
336 #endif
337 #endif
338
339
340 // pid_t gettid()
341 //
342 // Returns the kernel thread id of the currently running thread. Kernel
343 // thread id is used to access /proc.
344 pid_t os::Linux::gettid() {
345 int rslt = syscall(SYS_gettid);
346 assert(rslt != -1, "must be."); // old linuxthreads implementation?
347 return (pid_t)rslt;
348 }
349
350 // Most versions of linux have a bug where the number of processors are
351 // determined by looking at the /proc file system. In a chroot environment,
352 // the system call returns 1.
353 static bool unsafe_chroot_detected = false;
354 static const char *unstable_chroot_error = "/proc file system not found.\n"
355 "Java may be unstable running multithreaded in a chroot "
356 "environment on Linux when /proc filesystem is not mounted.";
357
358 void os::Linux::initialize_system_info() {
359 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
360 if (processor_count() == 1) {
361 pid_t pid = os::Linux::gettid();
362 char fname[32];
363 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
364 FILE *fp = fopen(fname, "r");
365 if (fp == NULL) {
366 unsafe_chroot_detected = true;
367 } else {
368 fclose(fp);
369 }
370 }
371 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
372 assert(processor_count() > 0, "linux error");
373 }
374
375 void os::init_system_properties_values() {
376 // The next steps are taken in the product version:
377 //
378 // Obtain the JAVA_HOME value from the location of libjvm.so.
379 // This library should be located at:
380 // <JAVA_HOME>/lib/{client|server}/libjvm.so.
381 //
382 // If "/jre/lib/" appears at the right place in the path, then we
383 // assume libjvm.so is installed in a JDK and we use this path.
384 //
385 // Otherwise exit with message: "Could not create the Java virtual machine."
386 //
387 // The following extra steps are taken in the debugging version:
388 //
389 // If "/jre/lib/" does NOT appear at the right place in the path
390 // instead of exit check for $JAVA_HOME environment variable.
391 //
392 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
393 // then we append a fake suffix "hotspot/libjvm.so" to this path so
394 // it looks like libjvm.so is installed there
395 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
396 //
397 // Otherwise exit.
398 //
399 // Important note: if the location of libjvm.so changes this
400 // code needs to be changed accordingly.
401
402 // See ld(1):
403 // The linker uses the following search paths to locate required
404 // shared libraries:
405 // 1: ...
406 // ...
407 // 7: The default directories, normally /lib and /usr/lib.
408 #ifndef OVERRIDE_LIBPATH
409 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
410 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
411 #else
412 #define DEFAULT_LIBPATH "/lib:/usr/lib"
413 #endif
414 #else
415 #define DEFAULT_LIBPATH OVERRIDE_LIBPATH
416 #endif
417
418 // Base path of extensions installed on the system.
419 #define SYS_EXT_DIR "/usr/java/packages"
420 #define EXTENSIONS_DIR "/lib/ext"
421
422 // Buffer that fits several sprintfs.
423 // Note that the space for the colon and the trailing null are provided
424 // by the nulls included by the sizeof operator.
425 const size_t bufsize =
426 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
427 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
428 char *buf = NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
429
430 // sysclasspath, java_home, dll_dir
431 {
432 char *pslash;
433 os::jvm_path(buf, bufsize);
434
435 // Found the full path to libjvm.so.
436 // Now cut the path to <java_home>/jre if we can.
437 pslash = strrchr(buf, '/');
438 if (pslash != NULL) {
439 *pslash = '\0'; // Get rid of /libjvm.so.
440 }
441 pslash = strrchr(buf, '/');
442 if (pslash != NULL) {
443 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
444 }
445 Arguments::set_dll_dir(buf);
446
447 if (pslash != NULL) {
448 pslash = strrchr(buf, '/');
449 if (pslash != NULL) {
450 *pslash = '\0'; // Get rid of /lib.
451 }
452 }
453 Arguments::set_java_home(buf);
454 if (!set_boot_path('/', ':')) {
455 vm_exit_during_initialization("Failed setting boot class path.", NULL);
456 }
457 }
458
459 // Where to look for native libraries.
460 //
461 // Note: Due to a legacy implementation, most of the library path
462 // is set in the launcher. This was to accomodate linking restrictions
463 // on legacy Linux implementations (which are no longer supported).
464 // Eventually, all the library path setting will be done here.
465 //
466 // However, to prevent the proliferation of improperly built native
467 // libraries, the new path component /usr/java/packages is added here.
468 // Eventually, all the library path setting will be done here.
469 {
470 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
471 // should always exist (until the legacy problem cited above is
472 // addressed).
473 const char *v = ::getenv("LD_LIBRARY_PATH");
474 const char *v_colon = ":";
475 if (v == NULL) { v = ""; v_colon = ""; }
476 // That's +1 for the colon and +1 for the trailing '\0'.
477 char *ld_library_path = NEW_C_HEAP_ARRAY(char,
478 strlen(v) + 1 +
479 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
480 mtInternal);
481 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
482 Arguments::set_library_path(ld_library_path);
483 FREE_C_HEAP_ARRAY(char, ld_library_path);
484 }
485
486 // Extensions directories.
487 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
488 Arguments::set_ext_dirs(buf);
489
490 FREE_C_HEAP_ARRAY(char, buf);
491
492 #undef DEFAULT_LIBPATH
493 #undef SYS_EXT_DIR
494 #undef EXTENSIONS_DIR
495 }
496
497 ////////////////////////////////////////////////////////////////////////////////
498 // breakpoint support
499
500 void os::breakpoint() {
501 BREAKPOINT;
502 }
503
504 extern "C" void breakpoint() {
505 // use debugger to set breakpoint here
506 }
507
508 ////////////////////////////////////////////////////////////////////////////////
509 // signal support
510
511 debug_only(static bool signal_sets_initialized = false);
512 static sigset_t unblocked_sigs, vm_sigs;
513
514 void os::Linux::signal_sets_init() {
515 // Should also have an assertion stating we are still single-threaded.
516 assert(!signal_sets_initialized, "Already initialized");
517 // Fill in signals that are necessarily unblocked for all threads in
518 // the VM. Currently, we unblock the following signals:
519 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
520 // by -Xrs (=ReduceSignalUsage));
521 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
522 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
523 // the dispositions or masks wrt these signals.
524 // Programs embedding the VM that want to use the above signals for their
525 // own purposes must, at this time, use the "-Xrs" option to prevent
526 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
527 // (See bug 4345157, and other related bugs).
528 // In reality, though, unblocking these signals is really a nop, since
529 // these signals are not blocked by default.
530 sigemptyset(&unblocked_sigs);
531 sigaddset(&unblocked_sigs, SIGILL);
532 sigaddset(&unblocked_sigs, SIGSEGV);
533 sigaddset(&unblocked_sigs, SIGBUS);
534 sigaddset(&unblocked_sigs, SIGFPE);
535 #if defined(PPC64)
536 sigaddset(&unblocked_sigs, SIGTRAP);
537 #endif
538 sigaddset(&unblocked_sigs, SR_signum);
539
540 if (!ReduceSignalUsage) {
541 if (!os::Posix::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
542 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
543 }
544 if (!os::Posix::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
545 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
546 }
547 if (!os::Posix::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
548 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
549 }
550 }
551 // Fill in signals that are blocked by all but the VM thread.
552 sigemptyset(&vm_sigs);
553 if (!ReduceSignalUsage) {
554 sigaddset(&vm_sigs, BREAK_SIGNAL);
555 }
556 debug_only(signal_sets_initialized = true);
557
558 }
559
560 // These are signals that are unblocked while a thread is running Java.
561 // (For some reason, they get blocked by default.)
562 sigset_t* os::Linux::unblocked_signals() {
563 assert(signal_sets_initialized, "Not initialized");
564 return &unblocked_sigs;
565 }
566
567 // These are the signals that are blocked while a (non-VM) thread is
568 // running Java. Only the VM thread handles these signals.
569 sigset_t* os::Linux::vm_signals() {
570 assert(signal_sets_initialized, "Not initialized");
571 return &vm_sigs;
572 }
573
574 void os::Linux::hotspot_sigmask(Thread* thread) {
575
576 //Save caller's signal mask before setting VM signal mask
577 sigset_t caller_sigmask;
578 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
579
580 OSThread* osthread = thread->osthread();
581 osthread->set_caller_sigmask(caller_sigmask);
582
583 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
584
585 if (!ReduceSignalUsage) {
586 if (thread->is_VM_thread()) {
587 // Only the VM thread handles BREAK_SIGNAL ...
588 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
589 } else {
590 // ... all other threads block BREAK_SIGNAL
591 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
592 }
593 }
594 }
595
596 //////////////////////////////////////////////////////////////////////////////
597 // detecting pthread library
598
599 void os::Linux::libpthread_init() {
600 // Save glibc and pthread version strings.
601 #if !defined(_CS_GNU_LIBC_VERSION) || \
602 !defined(_CS_GNU_LIBPTHREAD_VERSION)
603 #error "glibc too old (< 2.3.2)"
604 #endif
605
606 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
607 assert(n > 0, "cannot retrieve glibc version");
608 char *str = (char *)malloc(n, mtInternal);
609 confstr(_CS_GNU_LIBC_VERSION, str, n);
610 os::Linux::set_glibc_version(str);
611
612 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
613 assert(n > 0, "cannot retrieve pthread version");
614 str = (char *)malloc(n, mtInternal);
615 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
616 os::Linux::set_libpthread_version(str);
617 }
618
619 /////////////////////////////////////////////////////////////////////////////
620 // thread stack expansion
621
622 // os::Linux::manually_expand_stack() takes care of expanding the thread
623 // stack. Note that this is normally not needed: pthread stacks allocate
624 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
625 // committed. Therefore it is not necessary to expand the stack manually.
626 //
627 // Manually expanding the stack was historically needed on LinuxThreads
628 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
629 // it is kept to deal with very rare corner cases:
630 //
631 // For one, user may run the VM on an own implementation of threads
632 // whose stacks are - like the old LinuxThreads - implemented using
633 // mmap(MAP_GROWSDOWN).
634 //
635 // Also, this coding may be needed if the VM is running on the primordial
636 // thread. Normally we avoid running on the primordial thread; however,
637 // user may still invoke the VM on the primordial thread.
638 //
639 // The following historical comment describes the details about running
640 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
641
642
643 // Force Linux kernel to expand current thread stack. If "bottom" is close
644 // to the stack guard, caller should block all signals.
645 //
646 // MAP_GROWSDOWN:
647 // A special mmap() flag that is used to implement thread stacks. It tells
648 // kernel that the memory region should extend downwards when needed. This
649 // allows early versions of LinuxThreads to only mmap the first few pages
650 // when creating a new thread. Linux kernel will automatically expand thread
651 // stack as needed (on page faults).
652 //
653 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
654 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
655 // region, it's hard to tell if the fault is due to a legitimate stack
656 // access or because of reading/writing non-exist memory (e.g. buffer
657 // overrun). As a rule, if the fault happens below current stack pointer,
658 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
659 // application (see Linux kernel fault.c).
660 //
661 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
662 // stack overflow detection.
663 //
664 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
665 // not use MAP_GROWSDOWN.
666 //
667 // To get around the problem and allow stack banging on Linux, we need to
668 // manually expand thread stack after receiving the SIGSEGV.
669 //
670 // There are two ways to expand thread stack to address "bottom", we used
671 // both of them in JVM before 1.5:
672 // 1. adjust stack pointer first so that it is below "bottom", and then
673 // touch "bottom"
674 // 2. mmap() the page in question
675 //
676 // Now alternate signal stack is gone, it's harder to use 2. For instance,
677 // if current sp is already near the lower end of page 101, and we need to
678 // call mmap() to map page 100, it is possible that part of the mmap() frame
679 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
680 // That will destroy the mmap() frame and cause VM to crash.
681 //
682 // The following code works by adjusting sp first, then accessing the "bottom"
683 // page to force a page fault. Linux kernel will then automatically expand the
684 // stack mapping.
685 //
686 // _expand_stack_to() assumes its frame size is less than page size, which
687 // should always be true if the function is not inlined.
688
689 static void NOINLINE _expand_stack_to(address bottom) {
690 address sp;
691 size_t size;
692 volatile char *p;
693
694 // Adjust bottom to point to the largest address within the same page, it
695 // gives us a one-page buffer if alloca() allocates slightly more memory.
696 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
697 bottom += os::Linux::page_size() - 1;
698
699 // sp might be slightly above current stack pointer; if that's the case, we
700 // will alloca() a little more space than necessary, which is OK. Don't use
701 // os::current_stack_pointer(), as its result can be slightly below current
702 // stack pointer, causing us to not alloca enough to reach "bottom".
703 sp = (address)&sp;
704
705 if (sp > bottom) {
706 size = sp - bottom;
707 p = (volatile char *)alloca(size);
708 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
709 p[0] = '\0';
710 }
711 }
712
713 void os::Linux::expand_stack_to(address bottom) {
714 _expand_stack_to(bottom);
715 }
716
717 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
718 assert(t!=NULL, "just checking");
719 assert(t->osthread()->expanding_stack(), "expand should be set");
720
721 if (t->is_in_usable_stack(addr)) {
722 sigset_t mask_all, old_sigset;
723 sigfillset(&mask_all);
724 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
725 _expand_stack_to(addr);
726 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
727 return true;
728 }
729 return false;
730 }
731
732 //////////////////////////////////////////////////////////////////////////////
733 // create new thread
734
735 // Thread start routine for all newly created threads
736 static void *thread_native_entry(Thread *thread) {
737
738 thread->record_stack_base_and_size();
739
740 // Try to randomize the cache line index of hot stack frames.
741 // This helps when threads of the same stack traces evict each other's
742 // cache lines. The threads can be either from the same JVM instance, or
743 // from different JVM instances. The benefit is especially true for
744 // processors with hyperthreading technology.
745 static int counter = 0;
746 int pid = os::current_process_id();
747 alloca(((pid ^ counter++) & 7) * 128);
748
749 thread->initialize_thread_current();
750
751 OSThread* osthread = thread->osthread();
752 Monitor* sync = osthread->startThread_lock();
753
754 osthread->set_thread_id(os::current_thread_id());
755
756 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
757 os::current_thread_id(), (uintx) pthread_self());
758
759 if (UseNUMA) {
760 int lgrp_id = os::numa_get_group_id();
761 if (lgrp_id != -1) {
762 thread->set_lgrp_id(lgrp_id);
763 }
764 }
765 // initialize signal mask for this thread
766 os::Linux::hotspot_sigmask(thread);
767
768 // initialize floating point control register
769 os::Linux::init_thread_fpu_state();
770
771 // handshaking with parent thread
772 {
773 MutexLocker ml(sync, Mutex::_no_safepoint_check_flag);
774
775 // notify parent thread
776 osthread->set_state(INITIALIZED);
777 sync->notify_all();
778
779 // wait until os::start_thread()
780 while (osthread->get_state() == INITIALIZED) {
781 sync->wait_without_safepoint_check();
782 }
783 }
784
785 assert(osthread->pthread_id() != 0, "pthread_id was not set as expected");
786
787 // call one more level start routine
788 thread->call_run();
789
790 // Note: at this point the thread object may already have deleted itself.
791 // Prevent dereferencing it from here on out.
792 thread = NULL;
793
794 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
795 os::current_thread_id(), (uintx) pthread_self());
796
797 return 0;
798 }
799
800 // On Linux, glibc places static TLS blocks (for __thread variables) on
801 // the thread stack. This decreases the stack size actually available
802 // to threads.
803 //
804 // For large static TLS sizes, this may cause threads to malfunction due
805 // to insufficient stack space. This is a well-known issue in glibc:
806 // http://sourceware.org/bugzilla/show_bug.cgi?id=11787.
807 //
808 // As a workaround, we call a private but assumed-stable glibc function,
809 // __pthread_get_minstack() to obtain the minstack size and derive the
810 // static TLS size from it. We then increase the user requested stack
811 // size by this TLS size.
812 //
813 // Due to compatibility concerns, this size adjustment is opt-in and
814 // controlled via AdjustStackSizeForTLS.
815 typedef size_t (*GetMinStack)(const pthread_attr_t *attr);
816
817 GetMinStack _get_minstack_func = NULL;
818
819 static void get_minstack_init() {
820 _get_minstack_func =
821 (GetMinStack)dlsym(RTLD_DEFAULT, "__pthread_get_minstack");
822 log_info(os, thread)("Lookup of __pthread_get_minstack %s",
823 _get_minstack_func == NULL ? "failed" : "succeeded");
824 }
825
826 // Returns the size of the static TLS area glibc puts on thread stacks.
827 // The value is cached on first use, which occurs when the first thread
828 // is created during VM initialization.
829 static size_t get_static_tls_area_size(const pthread_attr_t *attr) {
830 size_t tls_size = 0;
831 if (_get_minstack_func != NULL) {
832 // Obtain the pthread minstack size by calling __pthread_get_minstack.
833 size_t minstack_size = _get_minstack_func(attr);
834
835 // Remove non-TLS area size included in minstack size returned
836 // by __pthread_get_minstack() to get the static TLS size.
837 // In glibc before 2.27, minstack size includes guard_size.
838 // In glibc 2.27 and later, guard_size is automatically added
839 // to the stack size by pthread_create and is no longer included
840 // in minstack size. In both cases, the guard_size is taken into
841 // account, so there is no need to adjust the result for that.
842 //
843 // Although __pthread_get_minstack() is a private glibc function,
844 // it is expected to have a stable behavior across future glibc
845 // versions while glibc still allocates the static TLS blocks off
846 // the stack. Following is glibc 2.28 __pthread_get_minstack():
847 //
848 // size_t
849 // __pthread_get_minstack (const pthread_attr_t *attr)
850 // {
851 // return GLRO(dl_pagesize) + __static_tls_size + PTHREAD_STACK_MIN;
852 // }
853 //
854 //
855 // The following 'minstack_size > os::vm_page_size() + PTHREAD_STACK_MIN'
856 // if check is done for precaution.
857 if (minstack_size > (size_t)os::vm_page_size() + PTHREAD_STACK_MIN) {
858 tls_size = minstack_size - os::vm_page_size() - PTHREAD_STACK_MIN;
859 }
860 }
861
862 log_info(os, thread)("Stack size adjustment for TLS is " SIZE_FORMAT,
863 tls_size);
864 return tls_size;
865 }
866
867 bool os::create_thread(Thread* thread, ThreadType thr_type,
868 size_t req_stack_size) {
869 assert(thread->osthread() == NULL, "caller responsible");
870
871 // Allocate the OSThread object
872 OSThread* osthread = new OSThread(NULL, NULL);
873 if (osthread == NULL) {
874 return false;
875 }
876
877 // set the correct thread state
878 osthread->set_thread_type(thr_type);
879
880 // Initial state is ALLOCATED but not INITIALIZED
881 osthread->set_state(ALLOCATED);
882
883 thread->set_osthread(osthread);
884
885 // init thread attributes
886 pthread_attr_t attr;
887 pthread_attr_init(&attr);
888 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
889
890 // Calculate stack size if it's not specified by caller.
891 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
892 // In glibc versions prior to 2.7 the guard size mechanism
893 // is not implemented properly. The posix standard requires adding
894 // the size of the guard pages to the stack size, instead Linux
895 // takes the space out of 'stacksize'. Thus we adapt the requested
896 // stack_size by the size of the guard pages to mimick proper
897 // behaviour. However, be careful not to end up with a size
898 // of zero due to overflow. Don't add the guard page in that case.
899 size_t guard_size = os::Linux::default_guard_size(thr_type);
900 // Configure glibc guard page. Must happen before calling
901 // get_static_tls_area_size(), which uses the guard_size.
902 pthread_attr_setguardsize(&attr, guard_size);
903
904 size_t stack_adjust_size = 0;
905 if (AdjustStackSizeForTLS) {
906 // Adjust the stack_size for on-stack TLS - see get_static_tls_area_size().
907 stack_adjust_size += get_static_tls_area_size(&attr);
908 } else {
909 stack_adjust_size += guard_size;
910 }
911
912 stack_adjust_size = align_up(stack_adjust_size, os::vm_page_size());
913 if (stack_size <= SIZE_MAX - stack_adjust_size) {
914 stack_size += stack_adjust_size;
915 }
916 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
917
918 int status = pthread_attr_setstacksize(&attr, stack_size);
919 assert_status(status == 0, status, "pthread_attr_setstacksize");
920
921 ThreadState state;
922
923 {
924 pthread_t tid;
925 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
926
927 char buf[64];
928 if (ret == 0) {
929 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
930 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
931 } else {
932 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
933 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
934 // Log some OS information which might explain why creating the thread failed.
935 log_info(os, thread)("Number of threads approx. running in the VM: %d", Threads::number_of_threads());
936 LogStream st(Log(os, thread)::info());
937 os::Posix::print_rlimit_info(&st);
938 os::print_memory_info(&st);
939 os::Linux::print_proc_sys_info(&st);
940 os::Linux::print_container_info(&st);
941 }
942
943 pthread_attr_destroy(&attr);
944
945 if (ret != 0) {
946 // Need to clean up stuff we've allocated so far
947 thread->set_osthread(NULL);
948 delete osthread;
949 return false;
950 }
951
952 // Store pthread info into the OSThread
953 osthread->set_pthread_id(tid);
954
955 // Wait until child thread is either initialized or aborted
956 {
957 Monitor* sync_with_child = osthread->startThread_lock();
958 MutexLocker ml(sync_with_child, Mutex::_no_safepoint_check_flag);
959 while ((state = osthread->get_state()) == ALLOCATED) {
960 sync_with_child->wait_without_safepoint_check();
961 }
962 }
963 }
964
965 // Aborted due to thread limit being reached
966 if (state == ZOMBIE) {
967 thread->set_osthread(NULL);
968 delete osthread;
969 return false;
970 }
971
972 // The thread is returned suspended (in state INITIALIZED),
973 // and is started higher up in the call chain
974 assert(state == INITIALIZED, "race condition");
975 return true;
976 }
977
978 /////////////////////////////////////////////////////////////////////////////
979 // attach existing thread
980
981 // bootstrap the main thread
982 bool os::create_main_thread(JavaThread* thread) {
983 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
984 return create_attached_thread(thread);
985 }
986
987 bool os::create_attached_thread(JavaThread* thread) {
988 #ifdef ASSERT
989 thread->verify_not_published();
990 #endif
991
992 // Allocate the OSThread object
993 OSThread* osthread = new OSThread(NULL, NULL);
994
995 if (osthread == NULL) {
996 return false;
997 }
998
999 // Store pthread info into the OSThread
1000 osthread->set_thread_id(os::Linux::gettid());
1001 osthread->set_pthread_id(::pthread_self());
1002
1003 // initialize floating point control register
1004 os::Linux::init_thread_fpu_state();
1005
1006 // Initial thread state is RUNNABLE
1007 osthread->set_state(RUNNABLE);
1008
1009 thread->set_osthread(osthread);
1010
1011 if (UseNUMA) {
1012 int lgrp_id = os::numa_get_group_id();
1013 if (lgrp_id != -1) {
1014 thread->set_lgrp_id(lgrp_id);
1015 }
1016 }
1017
1018 if (os::is_primordial_thread()) {
1019 // If current thread is primordial thread, its stack is mapped on demand,
1020 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1021 // the entire stack region to avoid SEGV in stack banging.
1022 // It is also useful to get around the heap-stack-gap problem on SuSE
1023 // kernel (see 4821821 for details). We first expand stack to the top
1024 // of yellow zone, then enable stack yellow zone (order is significant,
1025 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1026 // is no gap between the last two virtual memory regions.
1027
1028 JavaThread *jt = (JavaThread *)thread;
1029 address addr = jt->stack_reserved_zone_base();
1030 assert(addr != NULL, "initialization problem?");
1031 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1032
1033 osthread->set_expanding_stack();
1034 os::Linux::manually_expand_stack(jt, addr);
1035 osthread->clear_expanding_stack();
1036 }
1037
1038 // initialize signal mask for this thread
1039 // and save the caller's signal mask
1040 os::Linux::hotspot_sigmask(thread);
1041
1042 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
1043 os::current_thread_id(), (uintx) pthread_self());
1044
1045 return true;
1046 }
1047
1048 void os::pd_start_thread(Thread* thread) {
1049 OSThread * osthread = thread->osthread();
1050 assert(osthread->get_state() != INITIALIZED, "just checking");
1051 Monitor* sync_with_child = osthread->startThread_lock();
1052 MutexLocker ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1053 sync_with_child->notify();
1054 }
1055
1056 // Free Linux resources related to the OSThread
1057 void os::free_thread(OSThread* osthread) {
1058 assert(osthread != NULL, "osthread not set");
1059
1060 // We are told to free resources of the argument thread,
1061 // but we can only really operate on the current thread.
1062 assert(Thread::current()->osthread() == osthread,
1063 "os::free_thread but not current thread");
1064
1065 #ifdef ASSERT
1066 sigset_t current;
1067 sigemptyset(¤t);
1068 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
1069 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
1070 #endif
1071
1072 // Restore caller's signal mask
1073 sigset_t sigmask = osthread->caller_sigmask();
1074 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1075
1076 delete osthread;
1077 }
1078
1079 //////////////////////////////////////////////////////////////////////////////
1080 // primordial thread
1081
1082 // Check if current thread is the primordial thread, similar to Solaris thr_main.
1083 bool os::is_primordial_thread(void) {
1084 if (suppress_primordial_thread_resolution) {
1085 return false;
1086 }
1087 char dummy;
1088 // If called before init complete, thread stack bottom will be null.
1089 // Can be called if fatal error occurs before initialization.
1090 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
1091 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
1092 os::Linux::initial_thread_stack_size() != 0,
1093 "os::init did not locate primordial thread's stack region");
1094 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
1095 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
1096 os::Linux::initial_thread_stack_size()) {
1097 return true;
1098 } else {
1099 return false;
1100 }
1101 }
1102
1103 // Find the virtual memory area that contains addr
1104 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1105 FILE *fp = fopen("/proc/self/maps", "r");
1106 if (fp) {
1107 address low, high;
1108 while (!feof(fp)) {
1109 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1110 if (low <= addr && addr < high) {
1111 if (vma_low) *vma_low = low;
1112 if (vma_high) *vma_high = high;
1113 fclose(fp);
1114 return true;
1115 }
1116 }
1117 for (;;) {
1118 int ch = fgetc(fp);
1119 if (ch == EOF || ch == (int)'\n') break;
1120 }
1121 }
1122 fclose(fp);
1123 }
1124 return false;
1125 }
1126
1127 // Locate primordial thread stack. This special handling of primordial thread stack
1128 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1129 // bogus value for the primordial process thread. While the launcher has created
1130 // the VM in a new thread since JDK 6, we still have to allow for the use of the
1131 // JNI invocation API from a primordial thread.
1132 void os::Linux::capture_initial_stack(size_t max_size) {
1133
1134 // max_size is either 0 (which means accept OS default for thread stacks) or
1135 // a user-specified value known to be at least the minimum needed. If we
1136 // are actually on the primordial thread we can make it appear that we have a
1137 // smaller max_size stack by inserting the guard pages at that location. But we
1138 // cannot do anything to emulate a larger stack than what has been provided by
1139 // the OS or threading library. In fact if we try to use a stack greater than
1140 // what is set by rlimit then we will crash the hosting process.
1141
1142 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
1143 // If this is "unlimited" then it will be a huge value.
1144 struct rlimit rlim;
1145 getrlimit(RLIMIT_STACK, &rlim);
1146 size_t stack_size = rlim.rlim_cur;
1147
1148 // 6308388: a bug in ld.so will relocate its own .data section to the
1149 // lower end of primordial stack; reduce ulimit -s value a little bit
1150 // so we won't install guard page on ld.so's data section.
1151 // But ensure we don't underflow the stack size - allow 1 page spare
1152 if (stack_size >= (size_t)(3 * page_size())) {
1153 stack_size -= 2 * page_size();
1154 }
1155
1156 // Try to figure out where the stack base (top) is. This is harder.
1157 //
1158 // When an application is started, glibc saves the initial stack pointer in
1159 // a global variable "__libc_stack_end", which is then used by system
1160 // libraries. __libc_stack_end should be pretty close to stack top. The
1161 // variable is available since the very early days. However, because it is
1162 // a private interface, it could disappear in the future.
1163 //
1164 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1165 // to __libc_stack_end, it is very close to stack top, but isn't the real
1166 // stack top. Note that /proc may not exist if VM is running as a chroot
1167 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1168 // /proc/<pid>/stat could change in the future (though unlikely).
1169 //
1170 // We try __libc_stack_end first. If that doesn't work, look for
1171 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1172 // as a hint, which should work well in most cases.
1173
1174 uintptr_t stack_start;
1175
1176 // try __libc_stack_end first
1177 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1178 if (p && *p) {
1179 stack_start = *p;
1180 } else {
1181 // see if we can get the start_stack field from /proc/self/stat
1182 FILE *fp;
1183 int pid;
1184 char state;
1185 int ppid;
1186 int pgrp;
1187 int session;
1188 int nr;
1189 int tpgrp;
1190 unsigned long flags;
1191 unsigned long minflt;
1192 unsigned long cminflt;
1193 unsigned long majflt;
1194 unsigned long cmajflt;
1195 unsigned long utime;
1196 unsigned long stime;
1197 long cutime;
1198 long cstime;
1199 long prio;
1200 long nice;
1201 long junk;
1202 long it_real;
1203 uintptr_t start;
1204 uintptr_t vsize;
1205 intptr_t rss;
1206 uintptr_t rsslim;
1207 uintptr_t scodes;
1208 uintptr_t ecode;
1209 int i;
1210
1211 // Figure what the primordial thread stack base is. Code is inspired
1212 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1213 // followed by command name surrounded by parentheses, state, etc.
1214 char stat[2048];
1215 int statlen;
1216
1217 fp = fopen("/proc/self/stat", "r");
1218 if (fp) {
1219 statlen = fread(stat, 1, 2047, fp);
1220 stat[statlen] = '\0';
1221 fclose(fp);
1222
1223 // Skip pid and the command string. Note that we could be dealing with
1224 // weird command names, e.g. user could decide to rename java launcher
1225 // to "java 1.4.2 :)", then the stat file would look like
1226 // 1234 (java 1.4.2 :)) R ... ...
1227 // We don't really need to know the command string, just find the last
1228 // occurrence of ")" and then start parsing from there. See bug 4726580.
1229 char * s = strrchr(stat, ')');
1230
1231 i = 0;
1232 if (s) {
1233 // Skip blank chars
1234 do { s++; } while (s && isspace(*s));
1235
1236 #define _UFM UINTX_FORMAT
1237 #define _DFM INTX_FORMAT
1238
1239 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1240 // 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
1241 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,
1242 &state, // 3 %c
1243 &ppid, // 4 %d
1244 &pgrp, // 5 %d
1245 &session, // 6 %d
1246 &nr, // 7 %d
1247 &tpgrp, // 8 %d
1248 &flags, // 9 %lu
1249 &minflt, // 10 %lu
1250 &cminflt, // 11 %lu
1251 &majflt, // 12 %lu
1252 &cmajflt, // 13 %lu
1253 &utime, // 14 %lu
1254 &stime, // 15 %lu
1255 &cutime, // 16 %ld
1256 &cstime, // 17 %ld
1257 &prio, // 18 %ld
1258 &nice, // 19 %ld
1259 &junk, // 20 %ld
1260 &it_real, // 21 %ld
1261 &start, // 22 UINTX_FORMAT
1262 &vsize, // 23 UINTX_FORMAT
1263 &rss, // 24 INTX_FORMAT
1264 &rsslim, // 25 UINTX_FORMAT
1265 &scodes, // 26 UINTX_FORMAT
1266 &ecode, // 27 UINTX_FORMAT
1267 &stack_start); // 28 UINTX_FORMAT
1268 }
1269
1270 #undef _UFM
1271 #undef _DFM
1272
1273 if (i != 28 - 2) {
1274 assert(false, "Bad conversion from /proc/self/stat");
1275 // product mode - assume we are the primordial thread, good luck in the
1276 // embedded case.
1277 warning("Can't detect primordial thread stack location - bad conversion");
1278 stack_start = (uintptr_t) &rlim;
1279 }
1280 } else {
1281 // For some reason we can't open /proc/self/stat (for example, running on
1282 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1283 // most cases, so don't abort:
1284 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1285 stack_start = (uintptr_t) &rlim;
1286 }
1287 }
1288
1289 // Now we have a pointer (stack_start) very close to the stack top, the
1290 // next thing to do is to figure out the exact location of stack top. We
1291 // can find out the virtual memory area that contains stack_start by
1292 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1293 // and its upper limit is the real stack top. (again, this would fail if
1294 // running inside chroot, because /proc may not exist.)
1295
1296 uintptr_t stack_top;
1297 address low, high;
1298 if (find_vma((address)stack_start, &low, &high)) {
1299 // success, "high" is the true stack top. (ignore "low", because initial
1300 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1301 stack_top = (uintptr_t)high;
1302 } else {
1303 // failed, likely because /proc/self/maps does not exist
1304 warning("Can't detect primordial thread stack location - find_vma failed");
1305 // best effort: stack_start is normally within a few pages below the real
1306 // stack top, use it as stack top, and reduce stack size so we won't put
1307 // guard page outside stack.
1308 stack_top = stack_start;
1309 stack_size -= 16 * page_size();
1310 }
1311
1312 // stack_top could be partially down the page so align it
1313 stack_top = align_up(stack_top, page_size());
1314
1315 // Allowed stack value is minimum of max_size and what we derived from rlimit
1316 if (max_size > 0) {
1317 _initial_thread_stack_size = MIN2(max_size, stack_size);
1318 } else {
1319 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1320 // clamp it at 8MB as we do on Solaris
1321 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1322 }
1323 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1324 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1325
1326 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1327
1328 if (log_is_enabled(Info, os, thread)) {
1329 // See if we seem to be on primordial process thread
1330 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1331 uintptr_t(&rlim) < stack_top;
1332
1333 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1334 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1335 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1336 stack_top, intptr_t(_initial_thread_stack_bottom));
1337 }
1338 }
1339
1340 ////////////////////////////////////////////////////////////////////////////////
1341 // time support
1342
1343 #ifndef SUPPORTS_CLOCK_MONOTONIC
1344 #error "Build platform doesn't support clock_gettime and related functionality"
1345 #endif
1346
1347 // Time since start-up in seconds to a fine granularity.
1348 // Used by VMSelfDestructTimer and the MemProfiler.
1349 double os::elapsedTime() {
1350
1351 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1352 }
1353
1354 jlong os::elapsed_counter() {
1355 return javaTimeNanos() - initial_time_count;
1356 }
1357
1358 jlong os::elapsed_frequency() {
1359 return NANOSECS_PER_SEC; // nanosecond resolution
1360 }
1361
1362 bool os::supports_vtime() { return true; }
1363
1364 double os::elapsedVTime() {
1365 struct rusage usage;
1366 int retval = getrusage(RUSAGE_THREAD, &usage);
1367 if (retval == 0) {
1368 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);
1369 } else {
1370 // better than nothing, but not much
1371 return elapsedTime();
1372 }
1373 }
1374
1375 jlong os::javaTimeMillis() {
1376 timeval time;
1377 int status = gettimeofday(&time, NULL);
1378 assert(status != -1, "linux error");
1379 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1380 }
1381
1382 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1383 timeval time;
1384 int status = gettimeofday(&time, NULL);
1385 assert(status != -1, "linux error");
1386 seconds = jlong(time.tv_sec);
1387 nanos = jlong(time.tv_usec) * 1000;
1388 }
1389
1390 void os::Linux::fast_thread_clock_init() {
1391 if (!UseLinuxPosixThreadCPUClocks) {
1392 return;
1393 }
1394 clockid_t clockid;
1395 struct timespec tp;
1396 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1397 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1398
1399 // Switch to using fast clocks for thread cpu time if
1400 // the clock_getres() returns 0 error code.
1401 // Note, that some kernels may support the current thread
1402 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1403 // returned by the pthread_getcpuclockid().
1404 // If the fast Posix clocks are supported then the clock_getres()
1405 // must return at least tp.tv_sec == 0 which means a resolution
1406 // better than 1 sec. This is extra check for reliability.
1407
1408 if (pthread_getcpuclockid_func &&
1409 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1410 os::Posix::clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1411 _supports_fast_thread_cpu_time = true;
1412 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1413 }
1414 }
1415
1416 jlong os::javaTimeNanos() {
1417 if (os::supports_monotonic_clock()) {
1418 struct timespec tp;
1419 int status = os::Posix::clock_gettime(CLOCK_MONOTONIC, &tp);
1420 assert(status == 0, "gettime error");
1421 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1422 return result;
1423 } else {
1424 timeval time;
1425 int status = gettimeofday(&time, NULL);
1426 assert(status != -1, "linux error");
1427 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1428 return 1000 * usecs;
1429 }
1430 }
1431
1432 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1433 if (os::supports_monotonic_clock()) {
1434 info_ptr->max_value = ALL_64_BITS;
1435
1436 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1437 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1438 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1439 } else {
1440 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1441 info_ptr->max_value = ALL_64_BITS;
1442
1443 // gettimeofday is a real time clock so it skips
1444 info_ptr->may_skip_backward = true;
1445 info_ptr->may_skip_forward = true;
1446 }
1447
1448 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1449 }
1450
1451 // Return the real, user, and system times in seconds from an
1452 // arbitrary fixed point in the past.
1453 bool os::getTimesSecs(double* process_real_time,
1454 double* process_user_time,
1455 double* process_system_time) {
1456 struct tms ticks;
1457 clock_t real_ticks = times(&ticks);
1458
1459 if (real_ticks == (clock_t) (-1)) {
1460 return false;
1461 } else {
1462 double ticks_per_second = (double) clock_tics_per_sec;
1463 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1464 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1465 *process_real_time = ((double) real_ticks) / ticks_per_second;
1466
1467 return true;
1468 }
1469 }
1470
1471
1472 char * os::local_time_string(char *buf, size_t buflen) {
1473 struct tm t;
1474 time_t long_time;
1475 time(&long_time);
1476 localtime_r(&long_time, &t);
1477 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1478 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1479 t.tm_hour, t.tm_min, t.tm_sec);
1480 return buf;
1481 }
1482
1483 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1484 return localtime_r(clock, res);
1485 }
1486
1487 ////////////////////////////////////////////////////////////////////////////////
1488 // runtime exit support
1489
1490 // Note: os::shutdown() might be called very early during initialization, or
1491 // called from signal handler. Before adding something to os::shutdown(), make
1492 // sure it is async-safe and can handle partially initialized VM.
1493 void os::shutdown() {
1494
1495 // allow PerfMemory to attempt cleanup of any persistent resources
1496 perfMemory_exit();
1497
1498 // needs to remove object in file system
1499 AttachListener::abort();
1500
1501 // flush buffered output, finish log files
1502 ostream_abort();
1503
1504 // Check for abort hook
1505 abort_hook_t abort_hook = Arguments::abort_hook();
1506 if (abort_hook != NULL) {
1507 abort_hook();
1508 }
1509
1510 }
1511
1512 // Note: os::abort() might be called very early during initialization, or
1513 // called from signal handler. Before adding something to os::abort(), make
1514 // sure it is async-safe and can handle partially initialized VM.
1515 void os::abort(bool dump_core, void* siginfo, const void* context) {
1516 os::shutdown();
1517 if (dump_core) {
1518 if (DumpPrivateMappingsInCore) {
1519 ClassLoader::close_jrt_image();
1520 }
1521 #ifndef PRODUCT
1522 fdStream out(defaultStream::output_fd());
1523 out.print_raw("Current thread is ");
1524 char buf[16];
1525 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1526 out.print_raw_cr(buf);
1527 out.print_raw_cr("Dumping core ...");
1528 #endif
1529 ::abort(); // dump core
1530 }
1531
1532 ::exit(1);
1533 }
1534
1535 // Die immediately, no exit hook, no abort hook, no cleanup.
1536 // Dump a core file, if possible, for debugging.
1537 void os::die() {
1538 if (TestUnresponsiveErrorHandler && !CreateCoredumpOnCrash) {
1539 // For TimeoutInErrorHandlingTest.java, we just kill the VM
1540 // and don't take the time to generate a core file.
1541 os::signal_raise(SIGKILL);
1542 } else {
1543 ::abort();
1544 }
1545 }
1546
1547 // thread_id is kernel thread id (similar to Solaris LWP id)
1548 intx os::current_thread_id() { return os::Linux::gettid(); }
1549 int os::current_process_id() {
1550 return ::getpid();
1551 }
1552
1553 // DLL functions
1554
1555 const char* os::dll_file_extension() { return ".so"; }
1556
1557 // This must be hard coded because it's the system's temporary
1558 // directory not the java application's temp directory, ala java.io.tmpdir.
1559 const char* os::get_temp_directory() { return "/tmp"; }
1560
1561 static bool file_exists(const char* filename) {
1562 struct stat statbuf;
1563 if (filename == NULL || strlen(filename) == 0) {
1564 return false;
1565 }
1566 return os::stat(filename, &statbuf) == 0;
1567 }
1568
1569 // check if addr is inside libjvm.so
1570 bool os::address_is_in_vm(address addr) {
1571 static address libjvm_base_addr;
1572 Dl_info dlinfo;
1573
1574 if (libjvm_base_addr == NULL) {
1575 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1576 libjvm_base_addr = (address)dlinfo.dli_fbase;
1577 }
1578 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1579 }
1580
1581 if (dladdr((void *)addr, &dlinfo) != 0) {
1582 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1583 }
1584
1585 return false;
1586 }
1587
1588 bool os::dll_address_to_function_name(address addr, char *buf,
1589 int buflen, int *offset,
1590 bool demangle) {
1591 // buf is not optional, but offset is optional
1592 assert(buf != NULL, "sanity check");
1593
1594 Dl_info dlinfo;
1595
1596 if (dladdr((void*)addr, &dlinfo) != 0) {
1597 // see if we have a matching symbol
1598 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1599 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1600 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1601 }
1602 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1603 return true;
1604 }
1605 // no matching symbol so try for just file info
1606 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1607 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1608 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1609 return true;
1610 }
1611 }
1612 }
1613
1614 buf[0] = '\0';
1615 if (offset != NULL) *offset = -1;
1616 return false;
1617 }
1618
1619 struct _address_to_library_name {
1620 address addr; // input : memory address
1621 size_t buflen; // size of fname
1622 char* fname; // output: library name
1623 address base; // library base addr
1624 };
1625
1626 static int address_to_library_name_callback(struct dl_phdr_info *info,
1627 size_t size, void *data) {
1628 int i;
1629 bool found = false;
1630 address libbase = NULL;
1631 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1632
1633 // iterate through all loadable segments
1634 for (i = 0; i < info->dlpi_phnum; i++) {
1635 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1636 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1637 // base address of a library is the lowest address of its loaded
1638 // segments.
1639 if (libbase == NULL || libbase > segbase) {
1640 libbase = segbase;
1641 }
1642 // see if 'addr' is within current segment
1643 if (segbase <= d->addr &&
1644 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1645 found = true;
1646 }
1647 }
1648 }
1649
1650 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1651 // so dll_address_to_library_name() can fall through to use dladdr() which
1652 // can figure out executable name from argv[0].
1653 if (found && info->dlpi_name && info->dlpi_name[0]) {
1654 d->base = libbase;
1655 if (d->fname) {
1656 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1657 }
1658 return 1;
1659 }
1660 return 0;
1661 }
1662
1663 bool os::dll_address_to_library_name(address addr, char* buf,
1664 int buflen, int* offset) {
1665 // buf is not optional, but offset is optional
1666 assert(buf != NULL, "sanity check");
1667
1668 Dl_info dlinfo;
1669 struct _address_to_library_name data;
1670
1671 // There is a bug in old glibc dladdr() implementation that it could resolve
1672 // to wrong library name if the .so file has a base address != NULL. Here
1673 // we iterate through the program headers of all loaded libraries to find
1674 // out which library 'addr' really belongs to. This workaround can be
1675 // removed once the minimum requirement for glibc is moved to 2.3.x.
1676 data.addr = addr;
1677 data.fname = buf;
1678 data.buflen = buflen;
1679 data.base = NULL;
1680 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1681
1682 if (rslt) {
1683 // buf already contains library name
1684 if (offset) *offset = addr - data.base;
1685 return true;
1686 }
1687 if (dladdr((void*)addr, &dlinfo) != 0) {
1688 if (dlinfo.dli_fname != NULL) {
1689 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1690 }
1691 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1692 *offset = addr - (address)dlinfo.dli_fbase;
1693 }
1694 return true;
1695 }
1696
1697 buf[0] = '\0';
1698 if (offset) *offset = -1;
1699 return false;
1700 }
1701
1702 // Loads .dll/.so and
1703 // in case of error it checks if .dll/.so was built for the
1704 // same architecture as Hotspot is running on
1705
1706
1707 // Remember the stack's state. The Linux dynamic linker will change
1708 // the stack to 'executable' at most once, so we must safepoint only once.
1709 bool os::Linux::_stack_is_executable = false;
1710
1711 // VM operation that loads a library. This is necessary if stack protection
1712 // of the Java stacks can be lost during loading the library. If we
1713 // do not stop the Java threads, they can stack overflow before the stacks
1714 // are protected again.
1715 class VM_LinuxDllLoad: public VM_Operation {
1716 private:
1717 const char *_filename;
1718 char *_ebuf;
1719 int _ebuflen;
1720 void *_lib;
1721 public:
1722 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1723 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1724 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1725 void doit() {
1726 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1727 os::Linux::_stack_is_executable = true;
1728 }
1729 void* loaded_library() { return _lib; }
1730 };
1731
1732 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1733 void * result = NULL;
1734 bool load_attempted = false;
1735
1736 log_info(os)("attempting shared library load of %s", filename);
1737
1738 // Check whether the library to load might change execution rights
1739 // of the stack. If they are changed, the protection of the stack
1740 // guard pages will be lost. We need a safepoint to fix this.
1741 //
1742 // See Linux man page execstack(8) for more info.
1743 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1744 if (!ElfFile::specifies_noexecstack(filename)) {
1745 if (!is_init_completed()) {
1746 os::Linux::_stack_is_executable = true;
1747 // This is OK - No Java threads have been created yet, and hence no
1748 // stack guard pages to fix.
1749 //
1750 // Dynamic loader will make all stacks executable after
1751 // this function returns, and will not do that again.
1752 assert(Threads::number_of_threads() == 0, "no Java threads should exist yet.");
1753 } else {
1754 warning("You have loaded library %s which might have disabled stack guard. "
1755 "The VM will try to fix the stack guard now.\n"
1756 "It's highly recommended that you fix the library with "
1757 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1758 filename);
1759
1760 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1761 JavaThread *jt = JavaThread::current();
1762 if (jt->thread_state() != _thread_in_native) {
1763 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1764 // that requires ExecStack. Cannot enter safe point. Let's give up.
1765 warning("Unable to fix stack guard. Giving up.");
1766 } else {
1767 if (!LoadExecStackDllInVMThread) {
1768 // This is for the case where the DLL has an static
1769 // constructor function that executes JNI code. We cannot
1770 // load such DLLs in the VMThread.
1771 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1772 }
1773
1774 ThreadInVMfromNative tiv(jt);
1775 debug_only(VMNativeEntryWrapper vew;)
1776
1777 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1778 VMThread::execute(&op);
1779 if (LoadExecStackDllInVMThread) {
1780 result = op.loaded_library();
1781 }
1782 load_attempted = true;
1783 }
1784 }
1785 }
1786 }
1787
1788 if (!load_attempted) {
1789 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1790 }
1791
1792 if (result != NULL) {
1793 // Successful loading
1794 return result;
1795 }
1796
1797 Elf32_Ehdr elf_head;
1798 int diag_msg_max_length=ebuflen-strlen(ebuf);
1799 char* diag_msg_buf=ebuf+strlen(ebuf);
1800
1801 if (diag_msg_max_length==0) {
1802 // No more space in ebuf for additional diagnostics message
1803 return NULL;
1804 }
1805
1806
1807 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1808
1809 if (file_descriptor < 0) {
1810 // Can't open library, report dlerror() message
1811 return NULL;
1812 }
1813
1814 bool failed_to_read_elf_head=
1815 (sizeof(elf_head)!=
1816 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1817
1818 ::close(file_descriptor);
1819 if (failed_to_read_elf_head) {
1820 // file i/o error - report dlerror() msg
1821 return NULL;
1822 }
1823
1824 if (elf_head.e_ident[EI_DATA] != LITTLE_ENDIAN_ONLY(ELFDATA2LSB) BIG_ENDIAN_ONLY(ELFDATA2MSB)) {
1825 // handle invalid/out of range endianness values
1826 if (elf_head.e_ident[EI_DATA] == 0 || elf_head.e_ident[EI_DATA] > 2) {
1827 return NULL;
1828 }
1829
1830 #if defined(VM_LITTLE_ENDIAN)
1831 // VM is LE, shared object BE
1832 elf_head.e_machine = be16toh(elf_head.e_machine);
1833 #else
1834 // VM is BE, shared object LE
1835 elf_head.e_machine = le16toh(elf_head.e_machine);
1836 #endif
1837 }
1838
1839 typedef struct {
1840 Elf32_Half code; // Actual value as defined in elf.h
1841 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1842 unsigned char elf_class; // 32 or 64 bit
1843 unsigned char endianness; // MSB or LSB
1844 char* name; // String representation
1845 } arch_t;
1846
1847 #ifndef EM_486
1848 #define EM_486 6 /* Intel 80486 */
1849 #endif
1850 #ifndef EM_AARCH64
1851 #define EM_AARCH64 183 /* ARM AARCH64 */
1852 #endif
1853
1854 static const arch_t arch_array[]={
1855 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1856 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1857 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1858 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1859 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1860 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1861 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1862 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1863 #if defined(VM_LITTLE_ENDIAN)
1864 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1865 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1866 #else
1867 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1868 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1869 #endif
1870 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1871 // we only support 64 bit z architecture
1872 {EM_S390, EM_S390, ELFCLASS64, ELFDATA2MSB, (char*)"IBM System/390"},
1873 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1874 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1875 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1876 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1877 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1878 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1879 };
1880
1881 #if (defined IA32)
1882 static Elf32_Half running_arch_code=EM_386;
1883 #elif (defined AMD64) || (defined X32)
1884 static Elf32_Half running_arch_code=EM_X86_64;
1885 #elif (defined IA64)
1886 static Elf32_Half running_arch_code=EM_IA_64;
1887 #elif (defined __sparc) && (defined _LP64)
1888 static Elf32_Half running_arch_code=EM_SPARCV9;
1889 #elif (defined __sparc) && (!defined _LP64)
1890 static Elf32_Half running_arch_code=EM_SPARC;
1891 #elif (defined __powerpc64__)
1892 static Elf32_Half running_arch_code=EM_PPC64;
1893 #elif (defined __powerpc__)
1894 static Elf32_Half running_arch_code=EM_PPC;
1895 #elif (defined AARCH64)
1896 static Elf32_Half running_arch_code=EM_AARCH64;
1897 #elif (defined ARM)
1898 static Elf32_Half running_arch_code=EM_ARM;
1899 #elif (defined S390)
1900 static Elf32_Half running_arch_code=EM_S390;
1901 #elif (defined ALPHA)
1902 static Elf32_Half running_arch_code=EM_ALPHA;
1903 #elif (defined MIPSEL)
1904 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1905 #elif (defined PARISC)
1906 static Elf32_Half running_arch_code=EM_PARISC;
1907 #elif (defined MIPS)
1908 static Elf32_Half running_arch_code=EM_MIPS;
1909 #elif (defined M68K)
1910 static Elf32_Half running_arch_code=EM_68K;
1911 #elif (defined SH)
1912 static Elf32_Half running_arch_code=EM_SH;
1913 #else
1914 #error Method os::dll_load requires that one of following is defined:\
1915 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1916 #endif
1917
1918 // Identify compatibility class for VM's architecture and library's architecture
1919 // Obtain string descriptions for architectures
1920
1921 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1922 int running_arch_index=-1;
1923
1924 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1925 if (running_arch_code == arch_array[i].code) {
1926 running_arch_index = i;
1927 }
1928 if (lib_arch.code == arch_array[i].code) {
1929 lib_arch.compat_class = arch_array[i].compat_class;
1930 lib_arch.name = arch_array[i].name;
1931 }
1932 }
1933
1934 assert(running_arch_index != -1,
1935 "Didn't find running architecture code (running_arch_code) in arch_array");
1936 if (running_arch_index == -1) {
1937 // Even though running architecture detection failed
1938 // we may still continue with reporting dlerror() message
1939 return NULL;
1940 }
1941
1942 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1943 if (lib_arch.name != NULL) {
1944 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1945 " (Possible cause: can't load %s .so on a %s platform)",
1946 lib_arch.name, arch_array[running_arch_index].name);
1947 } else {
1948 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1949 " (Possible cause: can't load this .so (machine code=0x%x) on a %s platform)",
1950 lib_arch.code, arch_array[running_arch_index].name);
1951 }
1952 return NULL;
1953 }
1954
1955 if (lib_arch.endianness != arch_array[running_arch_index].endianness) {
1956 ::snprintf(diag_msg_buf, diag_msg_max_length-1, " (Possible cause: endianness mismatch)");
1957 return NULL;
1958 }
1959
1960 // ELF file class/capacity : 0 - invalid, 1 - 32bit, 2 - 64bit
1961 if (lib_arch.elf_class > 2 || lib_arch.elf_class < 1) {
1962 ::snprintf(diag_msg_buf, diag_msg_max_length-1, " (Possible cause: invalid ELF file class)");
1963 return NULL;
1964 }
1965
1966 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1967 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1968 " (Possible cause: architecture word width mismatch, can't load %d-bit .so on a %d-bit platform)",
1969 (int) lib_arch.elf_class * 32, arch_array[running_arch_index].elf_class * 32);
1970 return NULL;
1971 }
1972
1973 return NULL;
1974 }
1975
1976 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1977 int ebuflen) {
1978 void * result = ::dlopen(filename, RTLD_LAZY);
1979 if (result == NULL) {
1980 const char* error_report = ::dlerror();
1981 if (error_report == NULL) {
1982 error_report = "dlerror returned no error description";
1983 }
1984 if (ebuf != NULL && ebuflen > 0) {
1985 ::strncpy(ebuf, error_report, ebuflen-1);
1986 ebuf[ebuflen-1]='\0';
1987 }
1988 Events::log(NULL, "Loading shared library %s failed, %s", filename, error_report);
1989 log_info(os)("shared library load of %s failed, %s", filename, error_report);
1990 } else {
1991 Events::log(NULL, "Loaded shared library %s", filename);
1992 log_info(os)("shared library load of %s was successful", filename);
1993 }
1994 return result;
1995 }
1996
1997 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1998 int ebuflen) {
1999 void * result = NULL;
2000 if (LoadExecStackDllInVMThread) {
2001 result = dlopen_helper(filename, ebuf, ebuflen);
2002 }
2003
2004 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2005 // library that requires an executable stack, or which does not have this
2006 // stack attribute set, dlopen changes the stack attribute to executable. The
2007 // read protection of the guard pages gets lost.
2008 //
2009 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2010 // may have been queued at the same time.
2011
2012 if (!_stack_is_executable) {
2013 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
2014 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2015 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
2016 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
2017 warning("Attempt to reguard stack yellow zone failed.");
2018 }
2019 }
2020 }
2021 }
2022
2023 return result;
2024 }
2025
2026 void* os::dll_lookup(void* handle, const char* name) {
2027 void* res = dlsym(handle, name);
2028 return res;
2029 }
2030
2031 void* os::get_default_process_handle() {
2032 return (void*)::dlopen(NULL, RTLD_LAZY);
2033 }
2034
2035 static bool _print_ascii_file(const char* filename, outputStream* st, const char* hdr = NULL) {
2036 int fd = ::open(filename, O_RDONLY);
2037 if (fd == -1) {
2038 return false;
2039 }
2040
2041 if (hdr != NULL) {
2042 st->print_cr("%s", hdr);
2043 }
2044
2045 char buf[33];
2046 int bytes;
2047 buf[32] = '\0';
2048 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
2049 st->print_raw(buf, bytes);
2050 }
2051
2052 ::close(fd);
2053
2054 return true;
2055 }
2056
2057 void os::print_dll_info(outputStream *st) {
2058 st->print_cr("Dynamic libraries:");
2059
2060 char fname[32];
2061 pid_t pid = os::Linux::gettid();
2062
2063 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2064
2065 if (!_print_ascii_file(fname, st)) {
2066 st->print("Can not get library information for pid = %d\n", pid);
2067 }
2068 }
2069
2070 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2071 FILE *procmapsFile = NULL;
2072
2073 // Open the procfs maps file for the current process
2074 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2075 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2076 char line[PATH_MAX + 100];
2077
2078 // Read line by line from 'file'
2079 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2080 u8 base, top, offset, inode;
2081 char permissions[5];
2082 char device[6];
2083 char name[sizeof(line)];
2084
2085 // Parse fields from line
2086 int matches = sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
2087 &base, &top, permissions, &offset, device, &inode, name);
2088 // the last entry 'name' is empty for some entries, so we might have 6 matches instead of 7 for some lines
2089 if (matches < 6) continue;
2090 if (matches == 6) name[0] = '\0';
2091
2092 // Filter by device id '00:00' so that we only get file system mapped files.
2093 if (strcmp(device, "00:00") != 0) {
2094
2095 // Call callback with the fields of interest
2096 if(callback(name, (address)base, (address)top, param)) {
2097 // Oops abort, callback aborted
2098 fclose(procmapsFile);
2099 return 1;
2100 }
2101 }
2102 }
2103 fclose(procmapsFile);
2104 }
2105 return 0;
2106 }
2107
2108 void os::print_os_info_brief(outputStream* st) {
2109 os::Linux::print_distro_info(st);
2110
2111 os::Posix::print_uname_info(st);
2112
2113 os::Linux::print_libversion_info(st);
2114
2115 }
2116
2117 void os::print_os_info(outputStream* st) {
2118 st->print("OS:");
2119
2120 os::Linux::print_distro_info(st);
2121
2122 os::Posix::print_uname_info(st);
2123
2124 os::Linux::print_uptime_info(st);
2125
2126 // Print warning if unsafe chroot environment detected
2127 if (unsafe_chroot_detected) {
2128 st->print("WARNING!! ");
2129 st->print_cr("%s", unstable_chroot_error);
2130 }
2131
2132 os::Linux::print_libversion_info(st);
2133
2134 os::Posix::print_rlimit_info(st);
2135
2136 os::Posix::print_load_average(st);
2137
2138 os::Linux::print_full_memory_info(st);
2139
2140 os::Linux::print_proc_sys_info(st);
2141
2142 os::Linux::print_ld_preload_file(st);
2143
2144 os::Linux::print_container_info(st);
2145
2146 VM_Version::print_platform_virtualization_info(st);
2147
2148 os::Linux::print_steal_info(st);
2149 }
2150
2151 // Try to identify popular distros.
2152 // Most Linux distributions have a /etc/XXX-release file, which contains
2153 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2154 // file that also contains the OS version string. Some have more than one
2155 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2156 // /etc/redhat-release.), so the order is important.
2157 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2158 // their own specific XXX-release file as well as a redhat-release file.
2159 // Because of this the XXX-release file needs to be searched for before the
2160 // redhat-release file.
2161 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2162 // search for redhat-release / SuSE-release needs to be before lsb-release.
2163 // Since the lsb-release file is the new standard it needs to be searched
2164 // before the older style release files.
2165 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2166 // next to last resort. The os-release file is a new standard that contains
2167 // distribution information and the system-release file seems to be an old
2168 // standard that has been replaced by the lsb-release and os-release files.
2169 // Searching for the debian_version file is the last resort. It contains
2170 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2171 // "Debian " is printed before the contents of the debian_version file.
2172
2173 const char* distro_files[] = {
2174 "/etc/oracle-release",
2175 "/etc/mandriva-release",
2176 "/etc/mandrake-release",
2177 "/etc/sun-release",
2178 "/etc/redhat-release",
2179 "/etc/SuSE-release",
2180 "/etc/lsb-release",
2181 "/etc/turbolinux-release",
2182 "/etc/gentoo-release",
2183 "/etc/ltib-release",
2184 "/etc/angstrom-version",
2185 "/etc/system-release",
2186 "/etc/os-release",
2187 NULL };
2188
2189 void os::Linux::print_distro_info(outputStream* st) {
2190 for (int i = 0;; i++) {
2191 const char* file = distro_files[i];
2192 if (file == NULL) {
2193 break; // done
2194 }
2195 // If file prints, we found it.
2196 if (_print_ascii_file(file, st)) {
2197 return;
2198 }
2199 }
2200
2201 if (file_exists("/etc/debian_version")) {
2202 st->print("Debian ");
2203 _print_ascii_file("/etc/debian_version", st);
2204 } else {
2205 st->print("Linux");
2206 }
2207 st->cr();
2208 }
2209
2210 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2211 char buf[256];
2212 while (fgets(buf, sizeof(buf), fp)) {
2213 // Edit out extra stuff in expected format
2214 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2215 char* ptr = strstr(buf, "\""); // the name is in quotes
2216 if (ptr != NULL) {
2217 ptr++; // go beyond first quote
2218 char* nl = strchr(ptr, '\"');
2219 if (nl != NULL) *nl = '\0';
2220 strncpy(distro, ptr, length);
2221 } else {
2222 ptr = strstr(buf, "=");
2223 ptr++; // go beyond equals then
2224 char* nl = strchr(ptr, '\n');
2225 if (nl != NULL) *nl = '\0';
2226 strncpy(distro, ptr, length);
2227 }
2228 return;
2229 } else if (get_first_line) {
2230 char* nl = strchr(buf, '\n');
2231 if (nl != NULL) *nl = '\0';
2232 strncpy(distro, buf, length);
2233 return;
2234 }
2235 }
2236 // print last line and close
2237 char* nl = strchr(buf, '\n');
2238 if (nl != NULL) *nl = '\0';
2239 strncpy(distro, buf, length);
2240 }
2241
2242 static void parse_os_info(char* distro, size_t length, const char* file) {
2243 FILE* fp = fopen(file, "r");
2244 if (fp != NULL) {
2245 // if suse format, print out first line
2246 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2247 parse_os_info_helper(fp, distro, length, get_first_line);
2248 fclose(fp);
2249 }
2250 }
2251
2252 void os::get_summary_os_info(char* buf, size_t buflen) {
2253 for (int i = 0;; i++) {
2254 const char* file = distro_files[i];
2255 if (file == NULL) {
2256 break; // ran out of distro_files
2257 }
2258 if (file_exists(file)) {
2259 parse_os_info(buf, buflen, file);
2260 return;
2261 }
2262 }
2263 // special case for debian
2264 if (file_exists("/etc/debian_version")) {
2265 strncpy(buf, "Debian ", buflen);
2266 if (buflen > 7) {
2267 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2268 }
2269 } else {
2270 strncpy(buf, "Linux", buflen);
2271 }
2272 }
2273
2274 void os::Linux::print_libversion_info(outputStream* st) {
2275 // libc, pthread
2276 st->print("libc:");
2277 st->print("%s ", os::Linux::glibc_version());
2278 st->print("%s ", os::Linux::libpthread_version());
2279 st->cr();
2280 }
2281
2282 void os::Linux::print_proc_sys_info(outputStream* st) {
2283 st->cr();
2284 st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):");
2285 _print_ascii_file("/proc/sys/kernel/threads-max", st);
2286 st->cr();
2287 st->cr();
2288
2289 st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):");
2290 _print_ascii_file("/proc/sys/vm/max_map_count", st);
2291 st->cr();
2292 st->cr();
2293
2294 st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):");
2295 _print_ascii_file("/proc/sys/kernel/pid_max", st);
2296 st->cr();
2297 st->cr();
2298 }
2299
2300 void os::Linux::print_full_memory_info(outputStream* st) {
2301 st->print("\n/proc/meminfo:\n");
2302 _print_ascii_file("/proc/meminfo", st);
2303 st->cr();
2304 }
2305
2306 void os::Linux::print_ld_preload_file(outputStream* st) {
2307 _print_ascii_file("/etc/ld.so.preload", st, "\n/etc/ld.so.preload:");
2308 st->cr();
2309 }
2310
2311 void os::Linux::print_uptime_info(outputStream* st) {
2312 struct sysinfo sinfo;
2313 int ret = sysinfo(&sinfo);
2314 if (ret == 0) {
2315 os::print_dhm(st, "OS uptime:", (long) sinfo.uptime);
2316 }
2317 }
2318
2319
2320 void os::Linux::print_container_info(outputStream* st) {
2321 if (!OSContainer::is_containerized()) {
2322 return;
2323 }
2324
2325 st->print("container (cgroup) information:\n");
2326
2327 const char *p_ct = OSContainer::container_type();
2328 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "not supported");
2329
2330 char *p = OSContainer::cpu_cpuset_cpus();
2331 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "not supported");
2332 free(p);
2333
2334 p = OSContainer::cpu_cpuset_memory_nodes();
2335 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "not supported");
2336 free(p);
2337
2338 int i = OSContainer::active_processor_count();
2339 st->print("active_processor_count: ");
2340 if (i > 0) {
2341 st->print("%d\n", i);
2342 } else {
2343 st->print("not supported\n");
2344 }
2345
2346 i = OSContainer::cpu_quota();
2347 st->print("cpu_quota: ");
2348 if (i > 0) {
2349 st->print("%d\n", i);
2350 } else {
2351 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no quota");
2352 }
2353
2354 i = OSContainer::cpu_period();
2355 st->print("cpu_period: ");
2356 if (i > 0) {
2357 st->print("%d\n", i);
2358 } else {
2359 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no period");
2360 }
2361
2362 i = OSContainer::cpu_shares();
2363 st->print("cpu_shares: ");
2364 if (i > 0) {
2365 st->print("%d\n", i);
2366 } else {
2367 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no shares");
2368 }
2369
2370 jlong j = OSContainer::memory_limit_in_bytes();
2371 st->print("memory_limit_in_bytes: ");
2372 if (j > 0) {
2373 st->print(JLONG_FORMAT "\n", j);
2374 } else {
2375 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2376 }
2377
2378 j = OSContainer::memory_and_swap_limit_in_bytes();
2379 st->print("memory_and_swap_limit_in_bytes: ");
2380 if (j > 0) {
2381 st->print(JLONG_FORMAT "\n", j);
2382 } else {
2383 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2384 }
2385
2386 j = OSContainer::memory_soft_limit_in_bytes();
2387 st->print("memory_soft_limit_in_bytes: ");
2388 if (j > 0) {
2389 st->print(JLONG_FORMAT "\n", j);
2390 } else {
2391 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2392 }
2393
2394 j = OSContainer::OSContainer::memory_usage_in_bytes();
2395 st->print("memory_usage_in_bytes: ");
2396 if (j > 0) {
2397 st->print(JLONG_FORMAT "\n", j);
2398 } else {
2399 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2400 }
2401
2402 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2403 st->print("memory_max_usage_in_bytes: ");
2404 if (j > 0) {
2405 st->print(JLONG_FORMAT "\n", j);
2406 } else {
2407 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2408 }
2409 st->cr();
2410 }
2411
2412 void os::Linux::print_steal_info(outputStream* st) {
2413 if (has_initial_tick_info) {
2414 CPUPerfTicks pticks;
2415 bool res = os::Linux::get_tick_information(&pticks, -1);
2416
2417 if (res && pticks.has_steal_ticks) {
2418 uint64_t steal_ticks_difference = pticks.steal - initial_steal_ticks;
2419 uint64_t total_ticks_difference = pticks.total - initial_total_ticks;
2420 double steal_ticks_perc = 0.0;
2421 if (total_ticks_difference != 0) {
2422 steal_ticks_perc = (double) steal_ticks_difference / total_ticks_difference;
2423 }
2424 st->print_cr("Steal ticks since vm start: " UINT64_FORMAT, steal_ticks_difference);
2425 st->print_cr("Steal ticks percentage since vm start:%7.3f", steal_ticks_perc);
2426 }
2427 }
2428 }
2429
2430 void os::print_memory_info(outputStream* st) {
2431
2432 st->print("Memory:");
2433 st->print(" %dk page", os::vm_page_size()>>10);
2434
2435 // values in struct sysinfo are "unsigned long"
2436 struct sysinfo si;
2437 sysinfo(&si);
2438
2439 st->print(", physical " UINT64_FORMAT "k",
2440 os::physical_memory() >> 10);
2441 st->print("(" UINT64_FORMAT "k free)",
2442 os::available_memory() >> 10);
2443 st->print(", swap " UINT64_FORMAT "k",
2444 ((jlong)si.totalswap * si.mem_unit) >> 10);
2445 st->print("(" UINT64_FORMAT "k free)",
2446 ((jlong)si.freeswap * si.mem_unit) >> 10);
2447 st->cr();
2448 }
2449
2450 // Print the first "model name" line and the first "flags" line
2451 // that we find and nothing more. We assume "model name" comes
2452 // before "flags" so if we find a second "model name", then the
2453 // "flags" field is considered missing.
2454 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2455 #if defined(IA32) || defined(AMD64)
2456 // Other platforms have less repetitive cpuinfo files
2457 FILE *fp = fopen("/proc/cpuinfo", "r");
2458 if (fp) {
2459 while (!feof(fp)) {
2460 if (fgets(buf, buflen, fp)) {
2461 // Assume model name comes before flags
2462 bool model_name_printed = false;
2463 if (strstr(buf, "model name") != NULL) {
2464 if (!model_name_printed) {
2465 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2466 st->print_raw(buf);
2467 model_name_printed = true;
2468 } else {
2469 // model name printed but not flags? Odd, just return
2470 fclose(fp);
2471 return true;
2472 }
2473 }
2474 // print the flags line too
2475 if (strstr(buf, "flags") != NULL) {
2476 st->print_raw(buf);
2477 fclose(fp);
2478 return true;
2479 }
2480 }
2481 }
2482 fclose(fp);
2483 }
2484 #endif // x86 platforms
2485 return false;
2486 }
2487
2488 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2489 // Only print the model name if the platform provides this as a summary
2490 if (!print_model_name_and_flags(st, buf, buflen)) {
2491 st->print("\n/proc/cpuinfo:\n");
2492 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2493 st->print_cr(" <Not Available>");
2494 }
2495 }
2496 }
2497
2498 #if defined(AMD64) || defined(IA32) || defined(X32)
2499 const char* search_string = "model name";
2500 #elif defined(M68K)
2501 const char* search_string = "CPU";
2502 #elif defined(PPC64)
2503 const char* search_string = "cpu";
2504 #elif defined(S390)
2505 const char* search_string = "machine =";
2506 #elif defined(SPARC)
2507 const char* search_string = "cpu";
2508 #else
2509 const char* search_string = "Processor";
2510 #endif
2511
2512 // Parses the cpuinfo file for string representing the model name.
2513 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2514 FILE* fp = fopen("/proc/cpuinfo", "r");
2515 if (fp != NULL) {
2516 while (!feof(fp)) {
2517 char buf[256];
2518 if (fgets(buf, sizeof(buf), fp)) {
2519 char* start = strstr(buf, search_string);
2520 if (start != NULL) {
2521 char *ptr = start + strlen(search_string);
2522 char *end = buf + strlen(buf);
2523 while (ptr != end) {
2524 // skip whitespace and colon for the rest of the name.
2525 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2526 break;
2527 }
2528 ptr++;
2529 }
2530 if (ptr != end) {
2531 // reasonable string, get rid of newline and keep the rest
2532 char* nl = strchr(buf, '\n');
2533 if (nl != NULL) *nl = '\0';
2534 strncpy(cpuinfo, ptr, length);
2535 fclose(fp);
2536 return;
2537 }
2538 }
2539 }
2540 }
2541 fclose(fp);
2542 }
2543 // cpuinfo not found or parsing failed, just print generic string. The entire
2544 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2545 #if defined(AARCH64)
2546 strncpy(cpuinfo, "AArch64", length);
2547 #elif defined(AMD64)
2548 strncpy(cpuinfo, "x86_64", length);
2549 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2550 strncpy(cpuinfo, "ARM", length);
2551 #elif defined(IA32)
2552 strncpy(cpuinfo, "x86_32", length);
2553 #elif defined(IA64)
2554 strncpy(cpuinfo, "IA64", length);
2555 #elif defined(PPC)
2556 strncpy(cpuinfo, "PPC64", length);
2557 #elif defined(S390)
2558 strncpy(cpuinfo, "S390", length);
2559 #elif defined(SPARC)
2560 strncpy(cpuinfo, "sparcv9", length);
2561 #elif defined(ZERO_LIBARCH)
2562 strncpy(cpuinfo, ZERO_LIBARCH, length);
2563 #else
2564 strncpy(cpuinfo, "unknown", length);
2565 #endif
2566 }
2567
2568 static void print_signal_handler(outputStream* st, int sig,
2569 char* buf, size_t buflen);
2570
2571 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2572 st->print_cr("Signal Handlers:");
2573 print_signal_handler(st, SIGSEGV, buf, buflen);
2574 print_signal_handler(st, SIGBUS , buf, buflen);
2575 print_signal_handler(st, SIGFPE , buf, buflen);
2576 print_signal_handler(st, SIGPIPE, buf, buflen);
2577 print_signal_handler(st, SIGXFSZ, buf, buflen);
2578 print_signal_handler(st, SIGILL , buf, buflen);
2579 print_signal_handler(st, SR_signum, buf, buflen);
2580 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2581 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2582 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2583 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2584 #if defined(PPC64)
2585 print_signal_handler(st, SIGTRAP, buf, buflen);
2586 #endif
2587 }
2588
2589 static char saved_jvm_path[MAXPATHLEN] = {0};
2590
2591 // Find the full path to the current module, libjvm.so
2592 void os::jvm_path(char *buf, jint buflen) {
2593 // Error checking.
2594 if (buflen < MAXPATHLEN) {
2595 assert(false, "must use a large-enough buffer");
2596 buf[0] = '\0';
2597 return;
2598 }
2599 // Lazy resolve the path to current module.
2600 if (saved_jvm_path[0] != 0) {
2601 strcpy(buf, saved_jvm_path);
2602 return;
2603 }
2604
2605 char dli_fname[MAXPATHLEN];
2606 bool ret = dll_address_to_library_name(
2607 CAST_FROM_FN_PTR(address, os::jvm_path),
2608 dli_fname, sizeof(dli_fname), NULL);
2609 assert(ret, "cannot locate libjvm");
2610 char *rp = NULL;
2611 if (ret && dli_fname[0] != '\0') {
2612 rp = os::Posix::realpath(dli_fname, buf, buflen);
2613 }
2614 if (rp == NULL) {
2615 return;
2616 }
2617
2618 if (Arguments::sun_java_launcher_is_altjvm()) {
2619 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2620 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2621 // If "/jre/lib/" appears at the right place in the string, then
2622 // assume we are installed in a JDK and we're done. Otherwise, check
2623 // for a JAVA_HOME environment variable and fix up the path so it
2624 // looks like libjvm.so is installed there (append a fake suffix
2625 // hotspot/libjvm.so).
2626 const char *p = buf + strlen(buf) - 1;
2627 for (int count = 0; p > buf && count < 5; ++count) {
2628 for (--p; p > buf && *p != '/'; --p)
2629 /* empty */ ;
2630 }
2631
2632 if (strncmp(p, "/jre/lib/", 9) != 0) {
2633 // Look for JAVA_HOME in the environment.
2634 char* java_home_var = ::getenv("JAVA_HOME");
2635 if (java_home_var != NULL && java_home_var[0] != 0) {
2636 char* jrelib_p;
2637 int len;
2638
2639 // Check the current module name "libjvm.so".
2640 p = strrchr(buf, '/');
2641 if (p == NULL) {
2642 return;
2643 }
2644 assert(strstr(p, "/libjvm") == p, "invalid library name");
2645
2646 rp = os::Posix::realpath(java_home_var, buf, buflen);
2647 if (rp == NULL) {
2648 return;
2649 }
2650
2651 // determine if this is a legacy image or modules image
2652 // modules image doesn't have "jre" subdirectory
2653 len = strlen(buf);
2654 assert(len < buflen, "Ran out of buffer room");
2655 jrelib_p = buf + len;
2656 snprintf(jrelib_p, buflen-len, "/jre/lib");
2657 if (0 != access(buf, F_OK)) {
2658 snprintf(jrelib_p, buflen-len, "/lib");
2659 }
2660
2661 if (0 == access(buf, F_OK)) {
2662 // Use current module name "libjvm.so"
2663 len = strlen(buf);
2664 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2665 } else {
2666 // Go back to path of .so
2667 rp = os::Posix::realpath(dli_fname, buf, buflen);
2668 if (rp == NULL) {
2669 return;
2670 }
2671 }
2672 }
2673 }
2674 }
2675
2676 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2677 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2678 }
2679
2680 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2681 // no prefix required, not even "_"
2682 }
2683
2684 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2685 // no suffix required
2686 }
2687
2688 ////////////////////////////////////////////////////////////////////////////////
2689 // sun.misc.Signal support
2690
2691 static void UserHandler(int sig, void *siginfo, void *context) {
2692 // Ctrl-C is pressed during error reporting, likely because the error
2693 // handler fails to abort. Let VM die immediately.
2694 if (sig == SIGINT && VMError::is_error_reported()) {
2695 os::die();
2696 }
2697
2698 os::signal_notify(sig);
2699 }
2700
2701 void* os::user_handler() {
2702 return CAST_FROM_FN_PTR(void*, UserHandler);
2703 }
2704
2705 extern "C" {
2706 typedef void (*sa_handler_t)(int);
2707 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2708 }
2709
2710 void* os::signal(int signal_number, void* handler) {
2711 struct sigaction sigAct, oldSigAct;
2712
2713 sigfillset(&(sigAct.sa_mask));
2714 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2715 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2716
2717 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2718 // -1 means registration failed
2719 return (void *)-1;
2720 }
2721
2722 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2723 }
2724
2725 void os::signal_raise(int signal_number) {
2726 ::raise(signal_number);
2727 }
2728
2729 // The following code is moved from os.cpp for making this
2730 // code platform specific, which it is by its very nature.
2731
2732 // Will be modified when max signal is changed to be dynamic
2733 int os::sigexitnum_pd() {
2734 return NSIG;
2735 }
2736
2737 // a counter for each possible signal value
2738 static volatile jint pending_signals[NSIG+1] = { 0 };
2739
2740 // Linux(POSIX) specific hand shaking semaphore.
2741 static Semaphore* sig_sem = NULL;
2742 static PosixSemaphore sr_semaphore;
2743
2744 static void jdk_misc_signal_init() {
2745 // Initialize signal structures
2746 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2747
2748 // Initialize signal semaphore
2749 sig_sem = new Semaphore();
2750 }
2751
2752 void os::signal_notify(int sig) {
2753 if (sig_sem != NULL) {
2754 Atomic::inc(&pending_signals[sig]);
2755 sig_sem->signal();
2756 } else {
2757 // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2758 // initialization isn't called.
2759 assert(ReduceSignalUsage, "signal semaphore should be created");
2760 }
2761 }
2762
2763 static int check_pending_signals() {
2764 for (;;) {
2765 for (int i = 0; i < NSIG + 1; i++) {
2766 jint n = pending_signals[i];
2767 if (n > 0 && n == Atomic::cmpxchg(&pending_signals[i], n, n - 1)) {
2768 return i;
2769 }
2770 }
2771 JavaThread *thread = JavaThread::current();
2772 ThreadBlockInVM tbivm(thread);
2773
2774 bool threadIsSuspended;
2775 do {
2776 thread->set_suspend_equivalent();
2777 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2778 sig_sem->wait();
2779
2780 // were we externally suspended while we were waiting?
2781 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2782 if (threadIsSuspended) {
2783 // The semaphore has been incremented, but while we were waiting
2784 // another thread suspended us. We don't want to continue running
2785 // while suspended because that would surprise the thread that
2786 // suspended us.
2787 sig_sem->signal();
2788
2789 thread->java_suspend_self();
2790 }
2791 } while (threadIsSuspended);
2792 }
2793 }
2794
2795 int os::signal_wait() {
2796 return check_pending_signals();
2797 }
2798
2799 ////////////////////////////////////////////////////////////////////////////////
2800 // Virtual Memory
2801
2802 int os::vm_page_size() {
2803 // Seems redundant as all get out
2804 assert(os::Linux::page_size() != -1, "must call os::init");
2805 return os::Linux::page_size();
2806 }
2807
2808 // Solaris allocates memory by pages.
2809 int os::vm_allocation_granularity() {
2810 assert(os::Linux::page_size() != -1, "must call os::init");
2811 return os::Linux::page_size();
2812 }
2813
2814 // Rationale behind this function:
2815 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2816 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2817 // samples for JITted code. Here we create private executable mapping over the code cache
2818 // and then we can use standard (well, almost, as mapping can change) way to provide
2819 // info for the reporting script by storing timestamp and location of symbol
2820 void linux_wrap_code(char* base, size_t size) {
2821 static volatile jint cnt = 0;
2822
2823 if (!UseOprofile) {
2824 return;
2825 }
2826
2827 char buf[PATH_MAX+1];
2828 int num = Atomic::add(&cnt, 1);
2829
2830 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2831 os::get_temp_directory(), os::current_process_id(), num);
2832 unlink(buf);
2833
2834 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2835
2836 if (fd != -1) {
2837 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2838 if (rv != (off_t)-1) {
2839 if (::write(fd, "", 1) == 1) {
2840 mmap(base, size,
2841 PROT_READ|PROT_WRITE|PROT_EXEC,
2842 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2843 }
2844 }
2845 ::close(fd);
2846 unlink(buf);
2847 }
2848 }
2849
2850 static bool recoverable_mmap_error(int err) {
2851 // See if the error is one we can let the caller handle. This
2852 // list of errno values comes from JBS-6843484. I can't find a
2853 // Linux man page that documents this specific set of errno
2854 // values so while this list currently matches Solaris, it may
2855 // change as we gain experience with this failure mode.
2856 switch (err) {
2857 case EBADF:
2858 case EINVAL:
2859 case ENOTSUP:
2860 // let the caller deal with these errors
2861 return true;
2862
2863 default:
2864 // Any remaining errors on this OS can cause our reserved mapping
2865 // to be lost. That can cause confusion where different data
2866 // structures think they have the same memory mapped. The worst
2867 // scenario is if both the VM and a library think they have the
2868 // same memory mapped.
2869 return false;
2870 }
2871 }
2872
2873 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2874 int err) {
2875 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2876 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2877 os::strerror(err), err);
2878 }
2879
2880 static void warn_fail_commit_memory(char* addr, size_t size,
2881 size_t alignment_hint, bool exec,
2882 int err) {
2883 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2884 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2885 alignment_hint, exec, os::strerror(err), err);
2886 }
2887
2888 // NOTE: Linux kernel does not really reserve the pages for us.
2889 // All it does is to check if there are enough free pages
2890 // left at the time of mmap(). This could be a potential
2891 // problem.
2892 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2893 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2894 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2895 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2896 if (res != (uintptr_t) MAP_FAILED) {
2897 if (UseNUMAInterleaving) {
2898 numa_make_global(addr, size);
2899 }
2900 return 0;
2901 }
2902
2903 int err = errno; // save errno from mmap() call above
2904
2905 if (!recoverable_mmap_error(err)) {
2906 warn_fail_commit_memory(addr, size, exec, err);
2907 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2908 }
2909
2910 return err;
2911 }
2912
2913 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2914 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2915 }
2916
2917 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2918 const char* mesg) {
2919 assert(mesg != NULL, "mesg must be specified");
2920 int err = os::Linux::commit_memory_impl(addr, size, exec);
2921 if (err != 0) {
2922 // the caller wants all commit errors to exit with the specified mesg:
2923 warn_fail_commit_memory(addr, size, exec, err);
2924 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2925 }
2926 }
2927
2928 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2929 #ifndef MAP_HUGETLB
2930 #define MAP_HUGETLB 0x40000
2931 #endif
2932
2933 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2934 #ifndef MADV_HUGEPAGE
2935 #define MADV_HUGEPAGE 14
2936 #endif
2937
2938 int os::Linux::commit_memory_impl(char* addr, size_t size,
2939 size_t alignment_hint, bool exec) {
2940 int err = os::Linux::commit_memory_impl(addr, size, exec);
2941 if (err == 0) {
2942 realign_memory(addr, size, alignment_hint);
2943 }
2944 return err;
2945 }
2946
2947 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2948 bool exec) {
2949 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2950 }
2951
2952 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2953 size_t alignment_hint, bool exec,
2954 const char* mesg) {
2955 assert(mesg != NULL, "mesg must be specified");
2956 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2957 if (err != 0) {
2958 // the caller wants all commit errors to exit with the specified mesg:
2959 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2960 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2961 }
2962 }
2963
2964 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2965 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2966 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2967 // be supported or the memory may already be backed by huge pages.
2968 ::madvise(addr, bytes, MADV_HUGEPAGE);
2969 }
2970 }
2971
2972 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2973 // This method works by doing an mmap over an existing mmaping and effectively discarding
2974 // the existing pages. However it won't work for SHM-based large pages that cannot be
2975 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2976 // small pages on top of the SHM segment. This method always works for small pages, so we
2977 // allow that in any case.
2978 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2979 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2980 }
2981 }
2982
2983 void os::numa_make_global(char *addr, size_t bytes) {
2984 Linux::numa_interleave_memory(addr, bytes);
2985 }
2986
2987 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2988 // bind policy to MPOL_PREFERRED for the current thread.
2989 #define USE_MPOL_PREFERRED 0
2990
2991 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2992 // To make NUMA and large pages more robust when both enabled, we need to ease
2993 // the requirements on where the memory should be allocated. MPOL_BIND is the
2994 // default policy and it will force memory to be allocated on the specified
2995 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2996 // the specified node, but will not force it. Using this policy will prevent
2997 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2998 // free large pages.
2999 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
3000 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
3001 }
3002
3003 bool os::numa_topology_changed() { return false; }
3004
3005 size_t os::numa_get_groups_num() {
3006 // Return just the number of nodes in which it's possible to allocate memory
3007 // (in numa terminology, configured nodes).
3008 return Linux::numa_num_configured_nodes();
3009 }
3010
3011 int os::numa_get_group_id() {
3012 int cpu_id = Linux::sched_getcpu();
3013 if (cpu_id != -1) {
3014 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
3015 if (lgrp_id != -1) {
3016 return lgrp_id;
3017 }
3018 }
3019 return 0;
3020 }
3021
3022 int os::numa_get_group_id_for_address(const void* address) {
3023 void** pages = const_cast<void**>(&address);
3024 int id = -1;
3025
3026 if (os::Linux::numa_move_pages(0, 1, pages, NULL, &id, 0) == -1) {
3027 return -1;
3028 }
3029 if (id < 0) {
3030 return -1;
3031 }
3032 return id;
3033 }
3034
3035 int os::Linux::get_existing_num_nodes() {
3036 int node;
3037 int highest_node_number = Linux::numa_max_node();
3038 int num_nodes = 0;
3039
3040 // Get the total number of nodes in the system including nodes without memory.
3041 for (node = 0; node <= highest_node_number; node++) {
3042 if (is_node_in_existing_nodes(node)) {
3043 num_nodes++;
3044 }
3045 }
3046 return num_nodes;
3047 }
3048
3049 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
3050 int highest_node_number = Linux::numa_max_node();
3051 size_t i = 0;
3052
3053 // Map all node ids in which it is possible to allocate memory. Also nodes are
3054 // not always consecutively available, i.e. available from 0 to the highest
3055 // node number. If the nodes have been bound explicitly using numactl membind,
3056 // then allocate memory from those nodes only.
3057 for (int node = 0; node <= highest_node_number; node++) {
3058 if (Linux::is_node_in_bound_nodes((unsigned int)node)) {
3059 ids[i++] = node;
3060 }
3061 }
3062 return i;
3063 }
3064
3065 bool os::get_page_info(char *start, page_info* info) {
3066 return false;
3067 }
3068
3069 char *os::scan_pages(char *start, char* end, page_info* page_expected,
3070 page_info* page_found) {
3071 return end;
3072 }
3073
3074
3075 int os::Linux::sched_getcpu_syscall(void) {
3076 unsigned int cpu = 0;
3077 int retval = -1;
3078
3079 #if defined(IA32)
3080 #ifndef SYS_getcpu
3081 #define SYS_getcpu 318
3082 #endif
3083 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
3084 #elif defined(AMD64)
3085 // Unfortunately we have to bring all these macros here from vsyscall.h
3086 // to be able to compile on old linuxes.
3087 #define __NR_vgetcpu 2
3088 #define VSYSCALL_START (-10UL << 20)
3089 #define VSYSCALL_SIZE 1024
3090 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
3091 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
3092 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
3093 retval = vgetcpu(&cpu, NULL, NULL);
3094 #endif
3095
3096 return (retval == -1) ? retval : cpu;
3097 }
3098
3099 void os::Linux::sched_getcpu_init() {
3100 // sched_getcpu() should be in libc.
3101 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
3102 dlsym(RTLD_DEFAULT, "sched_getcpu")));
3103
3104 // If it's not, try a direct syscall.
3105 if (sched_getcpu() == -1) {
3106 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
3107 (void*)&sched_getcpu_syscall));
3108 }
3109
3110 if (sched_getcpu() == -1) {
3111 vm_exit_during_initialization("getcpu(2) system call not supported by kernel");
3112 }
3113 }
3114
3115 // Something to do with the numa-aware allocator needs these symbols
3116 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
3117 extern "C" JNIEXPORT void numa_error(char *where) { }
3118
3119 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
3120 // load symbol from base version instead.
3121 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
3122 void *f = dlvsym(handle, name, "libnuma_1.1");
3123 if (f == NULL) {
3124 f = dlsym(handle, name);
3125 }
3126 return f;
3127 }
3128
3129 // Handle request to load libnuma symbol version 1.2 (API v2) only.
3130 // Return NULL if the symbol is not defined in this particular version.
3131 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
3132 return dlvsym(handle, name, "libnuma_1.2");
3133 }
3134
3135 bool os::Linux::libnuma_init() {
3136 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
3137 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
3138 if (handle != NULL) {
3139 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
3140 libnuma_dlsym(handle, "numa_node_to_cpus")));
3141 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
3142 libnuma_dlsym(handle, "numa_max_node")));
3143 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
3144 libnuma_dlsym(handle, "numa_num_configured_nodes")));
3145 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
3146 libnuma_dlsym(handle, "numa_available")));
3147 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
3148 libnuma_dlsym(handle, "numa_tonode_memory")));
3149 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
3150 libnuma_dlsym(handle, "numa_interleave_memory")));
3151 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
3152 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
3153 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
3154 libnuma_dlsym(handle, "numa_set_bind_policy")));
3155 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
3156 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
3157 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
3158 libnuma_dlsym(handle, "numa_distance")));
3159 set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t,
3160 libnuma_v2_dlsym(handle, "numa_get_membind")));
3161 set_numa_get_interleave_mask(CAST_TO_FN_PTR(numa_get_interleave_mask_func_t,
3162 libnuma_v2_dlsym(handle, "numa_get_interleave_mask")));
3163 set_numa_move_pages(CAST_TO_FN_PTR(numa_move_pages_func_t,
3164 libnuma_dlsym(handle, "numa_move_pages")));
3165 set_numa_set_preferred(CAST_TO_FN_PTR(numa_set_preferred_func_t,
3166 libnuma_dlsym(handle, "numa_set_preferred")));
3167
3168 if (numa_available() != -1) {
3169 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
3170 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
3171 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
3172 set_numa_interleave_bitmask(_numa_get_interleave_mask());
3173 set_numa_membind_bitmask(_numa_get_membind());
3174 // Create an index -> node mapping, since nodes are not always consecutive
3175 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3176 rebuild_nindex_to_node_map();
3177 // Create a cpu -> node mapping
3178 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3179 rebuild_cpu_to_node_map();
3180 return true;
3181 }
3182 }
3183 }
3184 return false;
3185 }
3186
3187 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
3188 // Creating guard page is very expensive. Java thread has HotSpot
3189 // guard pages, only enable glibc guard page for non-Java threads.
3190 // (Remember: compiler thread is a Java thread, too!)
3191 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
3192 }
3193
3194 void os::Linux::rebuild_nindex_to_node_map() {
3195 int highest_node_number = Linux::numa_max_node();
3196
3197 nindex_to_node()->clear();
3198 for (int node = 0; node <= highest_node_number; node++) {
3199 if (Linux::is_node_in_existing_nodes(node)) {
3200 nindex_to_node()->append(node);
3201 }
3202 }
3203 }
3204
3205 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
3206 // The table is later used in get_node_by_cpu().
3207 void os::Linux::rebuild_cpu_to_node_map() {
3208 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
3209 // in libnuma (possible values are starting from 16,
3210 // and continuing up with every other power of 2, but less
3211 // than the maximum number of CPUs supported by kernel), and
3212 // is a subject to change (in libnuma version 2 the requirements
3213 // are more reasonable) we'll just hardcode the number they use
3214 // in the library.
3215 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
3216
3217 size_t cpu_num = processor_count();
3218 size_t cpu_map_size = NCPUS / BitsPerCLong;
3219 size_t cpu_map_valid_size =
3220 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
3221
3222 cpu_to_node()->clear();
3223 cpu_to_node()->at_grow(cpu_num - 1);
3224
3225 size_t node_num = get_existing_num_nodes();
3226
3227 int distance = 0;
3228 int closest_distance = INT_MAX;
3229 int closest_node = 0;
3230 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
3231 for (size_t i = 0; i < node_num; i++) {
3232 // Check if node is configured (not a memory-less node). If it is not, find
3233 // the closest configured node. Check also if node is bound, i.e. it's allowed
3234 // to allocate memory from the node. If it's not allowed, map cpus in that node
3235 // to the closest node from which memory allocation is allowed.
3236 if (!is_node_in_configured_nodes(nindex_to_node()->at(i)) ||
3237 !is_node_in_bound_nodes(nindex_to_node()->at(i))) {
3238 closest_distance = INT_MAX;
3239 // Check distance from all remaining nodes in the system. Ignore distance
3240 // from itself, from another non-configured node, and from another non-bound
3241 // node.
3242 for (size_t m = 0; m < node_num; m++) {
3243 if (m != i &&
3244 is_node_in_configured_nodes(nindex_to_node()->at(m)) &&
3245 is_node_in_bound_nodes(nindex_to_node()->at(m))) {
3246 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3247 // If a closest node is found, update. There is always at least one
3248 // configured and bound node in the system so there is always at least
3249 // one node close.
3250 if (distance != 0 && distance < closest_distance) {
3251 closest_distance = distance;
3252 closest_node = nindex_to_node()->at(m);
3253 }
3254 }
3255 }
3256 } else {
3257 // Current node is already a configured node.
3258 closest_node = nindex_to_node()->at(i);
3259 }
3260
3261 // Get cpus from the original node and map them to the closest node. If node
3262 // is a configured node (not a memory-less node), then original node and
3263 // closest node are the same.
3264 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3265 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3266 if (cpu_map[j] != 0) {
3267 for (size_t k = 0; k < BitsPerCLong; k++) {
3268 if (cpu_map[j] & (1UL << k)) {
3269 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3270 }
3271 }
3272 }
3273 }
3274 }
3275 }
3276 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3277 }
3278
3279 int os::Linux::get_node_by_cpu(int cpu_id) {
3280 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3281 return cpu_to_node()->at(cpu_id);
3282 }
3283 return -1;
3284 }
3285
3286 GrowableArray<int>* os::Linux::_cpu_to_node;
3287 GrowableArray<int>* os::Linux::_nindex_to_node;
3288 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3289 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3290 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3291 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3292 os::Linux::numa_available_func_t os::Linux::_numa_available;
3293 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3294 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3295 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3296 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3297 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3298 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3299 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind;
3300 os::Linux::numa_get_interleave_mask_func_t os::Linux::_numa_get_interleave_mask;
3301 os::Linux::numa_move_pages_func_t os::Linux::_numa_move_pages;
3302 os::Linux::numa_set_preferred_func_t os::Linux::_numa_set_preferred;
3303 os::Linux::NumaAllocationPolicy os::Linux::_current_numa_policy;
3304 unsigned long* os::Linux::_numa_all_nodes;
3305 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3306 struct bitmask* os::Linux::_numa_nodes_ptr;
3307 struct bitmask* os::Linux::_numa_interleave_bitmask;
3308 struct bitmask* os::Linux::_numa_membind_bitmask;
3309
3310 bool os::pd_uncommit_memory(char* addr, size_t size) {
3311 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3312 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3313 return res != (uintptr_t) MAP_FAILED;
3314 }
3315
3316 static address get_stack_commited_bottom(address bottom, size_t size) {
3317 address nbot = bottom;
3318 address ntop = bottom + size;
3319
3320 size_t page_sz = os::vm_page_size();
3321 unsigned pages = size / page_sz;
3322
3323 unsigned char vec[1];
3324 unsigned imin = 1, imax = pages + 1, imid;
3325 int mincore_return_value = 0;
3326
3327 assert(imin <= imax, "Unexpected page size");
3328
3329 while (imin < imax) {
3330 imid = (imax + imin) / 2;
3331 nbot = ntop - (imid * page_sz);
3332
3333 // Use a trick with mincore to check whether the page is mapped or not.
3334 // mincore sets vec to 1 if page resides in memory and to 0 if page
3335 // is swapped output but if page we are asking for is unmapped
3336 // it returns -1,ENOMEM
3337 mincore_return_value = mincore(nbot, page_sz, vec);
3338
3339 if (mincore_return_value == -1) {
3340 // Page is not mapped go up
3341 // to find first mapped page
3342 if (errno != EAGAIN) {
3343 assert(errno == ENOMEM, "Unexpected mincore errno");
3344 imax = imid;
3345 }
3346 } else {
3347 // Page is mapped go down
3348 // to find first not mapped page
3349 imin = imid + 1;
3350 }
3351 }
3352
3353 nbot = nbot + page_sz;
3354
3355 // Adjust stack bottom one page up if last checked page is not mapped
3356 if (mincore_return_value == -1) {
3357 nbot = nbot + page_sz;
3358 }
3359
3360 return nbot;
3361 }
3362
3363 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3364 int mincore_return_value;
3365 const size_t stripe = 1024; // query this many pages each time
3366 unsigned char vec[stripe + 1];
3367 // set a guard
3368 vec[stripe] = 'X';
3369
3370 const size_t page_sz = os::vm_page_size();
3371 size_t pages = size / page_sz;
3372
3373 assert(is_aligned(start, page_sz), "Start address must be page aligned");
3374 assert(is_aligned(size, page_sz), "Size must be page aligned");
3375
3376 committed_start = NULL;
3377
3378 int loops = (pages + stripe - 1) / stripe;
3379 int committed_pages = 0;
3380 address loop_base = start;
3381 bool found_range = false;
3382
3383 for (int index = 0; index < loops && !found_range; index ++) {
3384 assert(pages > 0, "Nothing to do");
3385 int pages_to_query = (pages >= stripe) ? stripe : pages;
3386 pages -= pages_to_query;
3387
3388 // Get stable read
3389 while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3390
3391 // During shutdown, some memory goes away without properly notifying NMT,
3392 // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3393 // Bailout and return as not committed for now.
3394 if (mincore_return_value == -1 && errno == ENOMEM) {
3395 return false;
3396 }
3397
3398 assert(vec[stripe] == 'X', "overflow guard");
3399 assert(mincore_return_value == 0, "Range must be valid");
3400 // Process this stripe
3401 for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3402 if ((vec[vecIdx] & 0x01) == 0) { // not committed
3403 // End of current contiguous region
3404 if (committed_start != NULL) {
3405 found_range = true;
3406 break;
3407 }
3408 } else { // committed
3409 // Start of region
3410 if (committed_start == NULL) {
3411 committed_start = loop_base + page_sz * vecIdx;
3412 }
3413 committed_pages ++;
3414 }
3415 }
3416
3417 loop_base += pages_to_query * page_sz;
3418 }
3419
3420 if (committed_start != NULL) {
3421 assert(committed_pages > 0, "Must have committed region");
3422 assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3423 assert(committed_start >= start && committed_start < start + size, "Out of range");
3424 committed_size = page_sz * committed_pages;
3425 return true;
3426 } else {
3427 assert(committed_pages == 0, "Should not have committed region");
3428 return false;
3429 }
3430 }
3431
3432
3433 // Linux uses a growable mapping for the stack, and if the mapping for
3434 // the stack guard pages is not removed when we detach a thread the
3435 // stack cannot grow beyond the pages where the stack guard was
3436 // mapped. If at some point later in the process the stack expands to
3437 // that point, the Linux kernel cannot expand the stack any further
3438 // because the guard pages are in the way, and a segfault occurs.
3439 //
3440 // However, it's essential not to split the stack region by unmapping
3441 // a region (leaving a hole) that's already part of the stack mapping,
3442 // so if the stack mapping has already grown beyond the guard pages at
3443 // the time we create them, we have to truncate the stack mapping.
3444 // So, we need to know the extent of the stack mapping when
3445 // create_stack_guard_pages() is called.
3446
3447 // We only need this for stacks that are growable: at the time of
3448 // writing thread stacks don't use growable mappings (i.e. those
3449 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3450 // only applies to the main thread.
3451
3452 // If the (growable) stack mapping already extends beyond the point
3453 // where we're going to put our guard pages, truncate the mapping at
3454 // that point by munmap()ping it. This ensures that when we later
3455 // munmap() the guard pages we don't leave a hole in the stack
3456 // mapping. This only affects the main/primordial thread
3457
3458 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3459 if (os::is_primordial_thread()) {
3460 // As we manually grow stack up to bottom inside create_attached_thread(),
3461 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3462 // we don't need to do anything special.
3463 // Check it first, before calling heavy function.
3464 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3465 unsigned char vec[1];
3466
3467 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3468 // Fallback to slow path on all errors, including EAGAIN
3469 stack_extent = (uintptr_t) get_stack_commited_bottom(
3470 os::Linux::initial_thread_stack_bottom(),
3471 (size_t)addr - stack_extent);
3472 }
3473
3474 if (stack_extent < (uintptr_t)addr) {
3475 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3476 }
3477 }
3478
3479 return os::commit_memory(addr, size, !ExecMem);
3480 }
3481
3482 // If this is a growable mapping, remove the guard pages entirely by
3483 // munmap()ping them. If not, just call uncommit_memory(). This only
3484 // affects the main/primordial thread, but guard against future OS changes.
3485 // It's safe to always unmap guard pages for primordial thread because we
3486 // always place it right after end of the mapped region.
3487
3488 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3489 uintptr_t stack_extent, stack_base;
3490
3491 if (os::is_primordial_thread()) {
3492 return ::munmap(addr, size) == 0;
3493 }
3494
3495 return os::uncommit_memory(addr, size);
3496 }
3497
3498 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3499 // at 'requested_addr'. If there are existing memory mappings at the same
3500 // location, however, they will be overwritten. If 'fixed' is false,
3501 // 'requested_addr' is only treated as a hint, the return value may or
3502 // may not start from the requested address. Unlike Linux mmap(), this
3503 // function returns NULL to indicate failure.
3504 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3505 char * addr;
3506 int flags;
3507
3508 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3509 if (fixed) {
3510 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3511 flags |= MAP_FIXED;
3512 }
3513
3514 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3515 // touch an uncommitted page. Otherwise, the read/write might
3516 // succeed if we have enough swap space to back the physical page.
3517 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3518 flags, -1, 0);
3519
3520 return addr == MAP_FAILED ? NULL : addr;
3521 }
3522
3523 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3524 // (req_addr != NULL) or with a given alignment.
3525 // - bytes shall be a multiple of alignment.
3526 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3527 // - alignment sets the alignment at which memory shall be allocated.
3528 // It must be a multiple of allocation granularity.
3529 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3530 // req_addr or NULL.
3531 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3532
3533 size_t extra_size = bytes;
3534 if (req_addr == NULL && alignment > 0) {
3535 extra_size += alignment;
3536 }
3537
3538 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3539 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3540 -1, 0);
3541 if (start == MAP_FAILED) {
3542 start = NULL;
3543 } else {
3544 if (req_addr != NULL) {
3545 if (start != req_addr) {
3546 ::munmap(start, extra_size);
3547 start = NULL;
3548 }
3549 } else {
3550 char* const start_aligned = align_up(start, alignment);
3551 char* const end_aligned = start_aligned + bytes;
3552 char* const end = start + extra_size;
3553 if (start_aligned > start) {
3554 ::munmap(start, start_aligned - start);
3555 }
3556 if (end_aligned < end) {
3557 ::munmap(end_aligned, end - end_aligned);
3558 }
3559 start = start_aligned;
3560 }
3561 }
3562 return start;
3563 }
3564
3565 static int anon_munmap(char * addr, size_t size) {
3566 return ::munmap(addr, size) == 0;
3567 }
3568
3569 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3570 size_t alignment_hint) {
3571 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3572 }
3573
3574 bool os::pd_release_memory(char* addr, size_t size) {
3575 return anon_munmap(addr, size);
3576 }
3577
3578 static bool linux_mprotect(char* addr, size_t size, int prot) {
3579 // Linux wants the mprotect address argument to be page aligned.
3580 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3581
3582 // According to SUSv3, mprotect() should only be used with mappings
3583 // established by mmap(), and mmap() always maps whole pages. Unaligned
3584 // 'addr' likely indicates problem in the VM (e.g. trying to change
3585 // protection of malloc'ed or statically allocated memory). Check the
3586 // caller if you hit this assert.
3587 assert(addr == bottom, "sanity check");
3588
3589 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3590 Events::log(NULL, "Protecting memory [" INTPTR_FORMAT "," INTPTR_FORMAT "] with protection modes %x", p2i(bottom), p2i(bottom+size), prot);
3591 return ::mprotect(bottom, size, prot) == 0;
3592 }
3593
3594 // Set protections specified
3595 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3596 bool is_committed) {
3597 unsigned int p = 0;
3598 switch (prot) {
3599 case MEM_PROT_NONE: p = PROT_NONE; break;
3600 case MEM_PROT_READ: p = PROT_READ; break;
3601 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3602 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3603 default:
3604 ShouldNotReachHere();
3605 }
3606 // is_committed is unused.
3607 return linux_mprotect(addr, bytes, p);
3608 }
3609
3610 bool os::guard_memory(char* addr, size_t size) {
3611 return linux_mprotect(addr, size, PROT_NONE);
3612 }
3613
3614 bool os::unguard_memory(char* addr, size_t size) {
3615 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3616 }
3617
3618 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3619 size_t page_size) {
3620 bool result = false;
3621 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3622 MAP_ANONYMOUS|MAP_PRIVATE,
3623 -1, 0);
3624 if (p != MAP_FAILED) {
3625 void *aligned_p = align_up(p, page_size);
3626
3627 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3628
3629 munmap(p, page_size * 2);
3630 }
3631
3632 if (warn && !result) {
3633 warning("TransparentHugePages is not supported by the operating system.");
3634 }
3635
3636 return result;
3637 }
3638
3639 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3640 bool result = false;
3641 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3642 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3643 -1, 0);
3644
3645 if (p != MAP_FAILED) {
3646 // We don't know if this really is a huge page or not.
3647 FILE *fp = fopen("/proc/self/maps", "r");
3648 if (fp) {
3649 while (!feof(fp)) {
3650 char chars[257];
3651 long x = 0;
3652 if (fgets(chars, sizeof(chars), fp)) {
3653 if (sscanf(chars, "%lx-%*x", &x) == 1
3654 && x == (long)p) {
3655 if (strstr (chars, "hugepage")) {
3656 result = true;
3657 break;
3658 }
3659 }
3660 }
3661 }
3662 fclose(fp);
3663 }
3664 munmap(p, page_size);
3665 }
3666
3667 if (warn && !result) {
3668 warning("HugeTLBFS is not supported by the operating system.");
3669 }
3670
3671 return result;
3672 }
3673
3674 // From the coredump_filter documentation:
3675 //
3676 // - (bit 0) anonymous private memory
3677 // - (bit 1) anonymous shared memory
3678 // - (bit 2) file-backed private memory
3679 // - (bit 3) file-backed shared memory
3680 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3681 // effective only if the bit 2 is cleared)
3682 // - (bit 5) hugetlb private memory
3683 // - (bit 6) hugetlb shared memory
3684 // - (bit 7) dax private memory
3685 // - (bit 8) dax shared memory
3686 //
3687 static void set_coredump_filter(CoredumpFilterBit bit) {
3688 FILE *f;
3689 long cdm;
3690
3691 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3692 return;
3693 }
3694
3695 if (fscanf(f, "%lx", &cdm) != 1) {
3696 fclose(f);
3697 return;
3698 }
3699
3700 long saved_cdm = cdm;
3701 rewind(f);
3702 cdm |= bit;
3703
3704 if (cdm != saved_cdm) {
3705 fprintf(f, "%#lx", cdm);
3706 }
3707
3708 fclose(f);
3709 }
3710
3711 // Large page support
3712
3713 static size_t _large_page_size = 0;
3714
3715 size_t os::Linux::find_large_page_size() {
3716 size_t large_page_size = 0;
3717
3718 // large_page_size on Linux is used to round up heap size. x86 uses either
3719 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3720 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3721 // page as large as 256M.
3722 //
3723 // Here we try to figure out page size by parsing /proc/meminfo and looking
3724 // for a line with the following format:
3725 // Hugepagesize: 2048 kB
3726 //
3727 // If we can't determine the value (e.g. /proc is not mounted, or the text
3728 // format has been changed), we'll use the largest page size supported by
3729 // the processor.
3730
3731 #ifndef ZERO
3732 large_page_size =
3733 AARCH64_ONLY(2 * M)
3734 AMD64_ONLY(2 * M)
3735 ARM32_ONLY(2 * M)
3736 IA32_ONLY(4 * M)
3737 IA64_ONLY(256 * M)
3738 PPC_ONLY(4 * M)
3739 S390_ONLY(1 * M)
3740 SPARC_ONLY(4 * M);
3741 #endif // ZERO
3742
3743 FILE *fp = fopen("/proc/meminfo", "r");
3744 if (fp) {
3745 while (!feof(fp)) {
3746 int x = 0;
3747 char buf[16];
3748 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3749 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3750 large_page_size = x * K;
3751 break;
3752 }
3753 } else {
3754 // skip to next line
3755 for (;;) {
3756 int ch = fgetc(fp);
3757 if (ch == EOF || ch == (int)'\n') break;
3758 }
3759 }
3760 }
3761 fclose(fp);
3762 }
3763
3764 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3765 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3766 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3767 proper_unit_for_byte_size(large_page_size));
3768 }
3769
3770 return large_page_size;
3771 }
3772
3773 size_t os::Linux::setup_large_page_size() {
3774 _large_page_size = Linux::find_large_page_size();
3775 const size_t default_page_size = (size_t)Linux::page_size();
3776 if (_large_page_size > default_page_size) {
3777 _page_sizes[0] = _large_page_size;
3778 _page_sizes[1] = default_page_size;
3779 _page_sizes[2] = 0;
3780 }
3781
3782 return _large_page_size;
3783 }
3784
3785 bool os::Linux::setup_large_page_type(size_t page_size) {
3786 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3787 FLAG_IS_DEFAULT(UseSHM) &&
3788 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3789
3790 // The type of large pages has not been specified by the user.
3791
3792 // Try UseHugeTLBFS and then UseSHM.
3793 UseHugeTLBFS = UseSHM = true;
3794
3795 // Don't try UseTransparentHugePages since there are known
3796 // performance issues with it turned on. This might change in the future.
3797 UseTransparentHugePages = false;
3798 }
3799
3800 if (UseTransparentHugePages) {
3801 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3802 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3803 UseHugeTLBFS = false;
3804 UseSHM = false;
3805 return true;
3806 }
3807 UseTransparentHugePages = false;
3808 }
3809
3810 if (UseHugeTLBFS) {
3811 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3812 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3813 UseSHM = false;
3814 return true;
3815 }
3816 UseHugeTLBFS = false;
3817 }
3818
3819 return UseSHM;
3820 }
3821
3822 void os::large_page_init() {
3823 if (!UseLargePages &&
3824 !UseTransparentHugePages &&
3825 !UseHugeTLBFS &&
3826 !UseSHM) {
3827 // Not using large pages.
3828 return;
3829 }
3830
3831 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3832 // The user explicitly turned off large pages.
3833 // Ignore the rest of the large pages flags.
3834 UseTransparentHugePages = false;
3835 UseHugeTLBFS = false;
3836 UseSHM = false;
3837 return;
3838 }
3839
3840 size_t large_page_size = Linux::setup_large_page_size();
3841 UseLargePages = Linux::setup_large_page_type(large_page_size);
3842
3843 set_coredump_filter(LARGEPAGES_BIT);
3844 }
3845
3846 #ifndef SHM_HUGETLB
3847 #define SHM_HUGETLB 04000
3848 #endif
3849
3850 #define shm_warning_format(format, ...) \
3851 do { \
3852 if (UseLargePages && \
3853 (!FLAG_IS_DEFAULT(UseLargePages) || \
3854 !FLAG_IS_DEFAULT(UseSHM) || \
3855 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3856 warning(format, __VA_ARGS__); \
3857 } \
3858 } while (0)
3859
3860 #define shm_warning(str) shm_warning_format("%s", str)
3861
3862 #define shm_warning_with_errno(str) \
3863 do { \
3864 int err = errno; \
3865 shm_warning_format(str " (error = %d)", err); \
3866 } while (0)
3867
3868 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3869 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3870
3871 if (!is_aligned(alignment, SHMLBA)) {
3872 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3873 return NULL;
3874 }
3875
3876 // To ensure that we get 'alignment' aligned memory from shmat,
3877 // we pre-reserve aligned virtual memory and then attach to that.
3878
3879 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3880 if (pre_reserved_addr == NULL) {
3881 // Couldn't pre-reserve aligned memory.
3882 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3883 return NULL;
3884 }
3885
3886 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3887 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3888
3889 if ((intptr_t)addr == -1) {
3890 int err = errno;
3891 shm_warning_with_errno("Failed to attach shared memory.");
3892
3893 assert(err != EACCES, "Unexpected error");
3894 assert(err != EIDRM, "Unexpected error");
3895 assert(err != EINVAL, "Unexpected error");
3896
3897 // Since we don't know if the kernel unmapped the pre-reserved memory area
3898 // we can't unmap it, since that would potentially unmap memory that was
3899 // mapped from other threads.
3900 return NULL;
3901 }
3902
3903 return addr;
3904 }
3905
3906 static char* shmat_at_address(int shmid, char* req_addr) {
3907 if (!is_aligned(req_addr, SHMLBA)) {
3908 assert(false, "Requested address needs to be SHMLBA aligned");
3909 return NULL;
3910 }
3911
3912 char* addr = (char*)shmat(shmid, req_addr, 0);
3913
3914 if ((intptr_t)addr == -1) {
3915 shm_warning_with_errno("Failed to attach shared memory.");
3916 return NULL;
3917 }
3918
3919 return addr;
3920 }
3921
3922 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3923 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3924 if (req_addr != NULL) {
3925 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3926 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3927 return shmat_at_address(shmid, req_addr);
3928 }
3929
3930 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3931 // return large page size aligned memory addresses when req_addr == NULL.
3932 // However, if the alignment is larger than the large page size, we have
3933 // to manually ensure that the memory returned is 'alignment' aligned.
3934 if (alignment > os::large_page_size()) {
3935 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3936 return shmat_with_alignment(shmid, bytes, alignment);
3937 } else {
3938 return shmat_at_address(shmid, NULL);
3939 }
3940 }
3941
3942 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3943 char* req_addr, bool exec) {
3944 // "exec" is passed in but not used. Creating the shared image for
3945 // the code cache doesn't have an SHM_X executable permission to check.
3946 assert(UseLargePages && UseSHM, "only for SHM large pages");
3947 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3948 assert(is_aligned(req_addr, alignment), "Unaligned address");
3949
3950 if (!is_aligned(bytes, os::large_page_size())) {
3951 return NULL; // Fallback to small pages.
3952 }
3953
3954 // Create a large shared memory region to attach to based on size.
3955 // Currently, size is the total size of the heap.
3956 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3957 if (shmid == -1) {
3958 // Possible reasons for shmget failure:
3959 // 1. shmmax is too small for Java heap.
3960 // > check shmmax value: cat /proc/sys/kernel/shmmax
3961 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3962 // 2. not enough large page memory.
3963 // > check available large pages: cat /proc/meminfo
3964 // > increase amount of large pages:
3965 // echo new_value > /proc/sys/vm/nr_hugepages
3966 // Note 1: different Linux may use different name for this property,
3967 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3968 // Note 2: it's possible there's enough physical memory available but
3969 // they are so fragmented after a long run that they can't
3970 // coalesce into large pages. Try to reserve large pages when
3971 // the system is still "fresh".
3972 shm_warning_with_errno("Failed to reserve shared memory.");
3973 return NULL;
3974 }
3975
3976 // Attach to the region.
3977 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3978
3979 // Remove shmid. If shmat() is successful, the actual shared memory segment
3980 // will be deleted when it's detached by shmdt() or when the process
3981 // terminates. If shmat() is not successful this will remove the shared
3982 // segment immediately.
3983 shmctl(shmid, IPC_RMID, NULL);
3984
3985 return addr;
3986 }
3987
3988 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3989 int error) {
3990 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3991
3992 bool warn_on_failure = UseLargePages &&
3993 (!FLAG_IS_DEFAULT(UseLargePages) ||
3994 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3995 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3996
3997 if (warn_on_failure) {
3998 char msg[128];
3999 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
4000 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
4001 warning("%s", msg);
4002 }
4003 }
4004
4005 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
4006 char* req_addr,
4007 bool exec) {
4008 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
4009 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
4010 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
4011
4012 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
4013 char* addr = (char*)::mmap(req_addr, bytes, prot,
4014 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
4015 -1, 0);
4016
4017 if (addr == MAP_FAILED) {
4018 warn_on_large_pages_failure(req_addr, bytes, errno);
4019 return NULL;
4020 }
4021
4022 assert(is_aligned(addr, os::large_page_size()), "Must be");
4023
4024 return addr;
4025 }
4026
4027 // Reserve memory using mmap(MAP_HUGETLB).
4028 // - bytes shall be a multiple of alignment.
4029 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
4030 // - alignment sets the alignment at which memory shall be allocated.
4031 // It must be a multiple of allocation granularity.
4032 // Returns address of memory or NULL. If req_addr was not NULL, will only return
4033 // req_addr or NULL.
4034 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
4035 size_t alignment,
4036 char* req_addr,
4037 bool exec) {
4038 size_t large_page_size = os::large_page_size();
4039 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
4040
4041 assert(is_aligned(req_addr, alignment), "Must be");
4042 assert(is_aligned(bytes, alignment), "Must be");
4043
4044 // First reserve - but not commit - the address range in small pages.
4045 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
4046
4047 if (start == NULL) {
4048 return NULL;
4049 }
4050
4051 assert(is_aligned(start, alignment), "Must be");
4052
4053 char* end = start + bytes;
4054
4055 // Find the regions of the allocated chunk that can be promoted to large pages.
4056 char* lp_start = align_up(start, large_page_size);
4057 char* lp_end = align_down(end, large_page_size);
4058
4059 size_t lp_bytes = lp_end - lp_start;
4060
4061 assert(is_aligned(lp_bytes, large_page_size), "Must be");
4062
4063 if (lp_bytes == 0) {
4064 // The mapped region doesn't even span the start and the end of a large page.
4065 // Fall back to allocate a non-special area.
4066 ::munmap(start, end - start);
4067 return NULL;
4068 }
4069
4070 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
4071
4072 void* result;
4073
4074 // Commit small-paged leading area.
4075 if (start != lp_start) {
4076 result = ::mmap(start, lp_start - start, prot,
4077 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
4078 -1, 0);
4079 if (result == MAP_FAILED) {
4080 ::munmap(lp_start, end - lp_start);
4081 return NULL;
4082 }
4083 }
4084
4085 // Commit large-paged area.
4086 result = ::mmap(lp_start, lp_bytes, prot,
4087 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
4088 -1, 0);
4089 if (result == MAP_FAILED) {
4090 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
4091 // If the mmap above fails, the large pages region will be unmapped and we
4092 // have regions before and after with small pages. Release these regions.
4093 //
4094 // | mapped | unmapped | mapped |
4095 // ^ ^ ^ ^
4096 // start lp_start lp_end end
4097 //
4098 ::munmap(start, lp_start - start);
4099 ::munmap(lp_end, end - lp_end);
4100 return NULL;
4101 }
4102
4103 // Commit small-paged trailing area.
4104 if (lp_end != end) {
4105 result = ::mmap(lp_end, end - lp_end, prot,
4106 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
4107 -1, 0);
4108 if (result == MAP_FAILED) {
4109 ::munmap(start, lp_end - start);
4110 return NULL;
4111 }
4112 }
4113
4114 return start;
4115 }
4116
4117 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
4118 size_t alignment,
4119 char* req_addr,
4120 bool exec) {
4121 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
4122 assert(is_aligned(req_addr, alignment), "Must be");
4123 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
4124 assert(is_power_of_2(os::large_page_size()), "Must be");
4125 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
4126
4127 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
4128 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
4129 } else {
4130 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
4131 }
4132 }
4133
4134 char* os::reserve_memory_special(size_t bytes, size_t alignment,
4135 char* req_addr, bool exec) {
4136 assert(UseLargePages, "only for large pages");
4137
4138 char* addr;
4139 if (UseSHM) {
4140 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
4141 } else {
4142 assert(UseHugeTLBFS, "must be");
4143 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
4144 }
4145
4146 if (addr != NULL) {
4147 if (UseNUMAInterleaving) {
4148 numa_make_global(addr, bytes);
4149 }
4150
4151 // The memory is committed
4152 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
4153 }
4154
4155 return addr;
4156 }
4157
4158 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
4159 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
4160 return shmdt(base) == 0;
4161 }
4162
4163 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
4164 return pd_release_memory(base, bytes);
4165 }
4166
4167 bool os::release_memory_special(char* base, size_t bytes) {
4168 bool res;
4169 if (MemTracker::tracking_level() > NMT_minimal) {
4170 Tracker tkr(Tracker::release);
4171 res = os::Linux::release_memory_special_impl(base, bytes);
4172 if (res) {
4173 tkr.record((address)base, bytes);
4174 }
4175
4176 } else {
4177 res = os::Linux::release_memory_special_impl(base, bytes);
4178 }
4179 return res;
4180 }
4181
4182 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
4183 assert(UseLargePages, "only for large pages");
4184 bool res;
4185
4186 if (UseSHM) {
4187 res = os::Linux::release_memory_special_shm(base, bytes);
4188 } else {
4189 assert(UseHugeTLBFS, "must be");
4190 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
4191 }
4192 return res;
4193 }
4194
4195 size_t os::large_page_size() {
4196 return _large_page_size;
4197 }
4198
4199 // With SysV SHM the entire memory region must be allocated as shared
4200 // memory.
4201 // HugeTLBFS allows application to commit large page memory on demand.
4202 // However, when committing memory with HugeTLBFS fails, the region
4203 // that was supposed to be committed will lose the old reservation
4204 // and allow other threads to steal that memory region. Because of this
4205 // behavior we can't commit HugeTLBFS memory.
4206 bool os::can_commit_large_page_memory() {
4207 return UseTransparentHugePages;
4208 }
4209
4210 bool os::can_execute_large_page_memory() {
4211 return UseTransparentHugePages || UseHugeTLBFS;
4212 }
4213
4214 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
4215 assert(file_desc >= 0, "file_desc is not valid");
4216 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
4217 if (result != NULL) {
4218 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
4219 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
4220 }
4221 }
4222 return result;
4223 }
4224
4225 // Reserve memory at an arbitrary address, only if that area is
4226 // available (and not reserved for something else).
4227
4228 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
4229 // Assert only that the size is a multiple of the page size, since
4230 // that's all that mmap requires, and since that's all we really know
4231 // about at this low abstraction level. If we need higher alignment,
4232 // we can either pass an alignment to this method or verify alignment
4233 // in one of the methods further up the call chain. See bug 5044738.
4234 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
4235
4236 // Repeatedly allocate blocks until the block is allocated at the
4237 // right spot.
4238
4239 // Linux mmap allows caller to pass an address as hint; give it a try first,
4240 // if kernel honors the hint then we can return immediately.
4241 char * addr = anon_mmap(requested_addr, bytes, false);
4242 if (addr == requested_addr) {
4243 return requested_addr;
4244 }
4245
4246 if (addr != NULL) {
4247 // mmap() is successful but it fails to reserve at the requested address
4248 anon_munmap(addr, bytes);
4249 }
4250
4251 return NULL;
4252 }
4253
4254 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4255 void os::infinite_sleep() {
4256 while (true) { // sleep forever ...
4257 ::sleep(100); // ... 100 seconds at a time
4258 }
4259 }
4260
4261 // Used to convert frequent JVM_Yield() to nops
4262 bool os::dont_yield() {
4263 return DontYieldALot;
4264 }
4265
4266 // Linux CFS scheduler (since 2.6.23) does not guarantee sched_yield(2) will
4267 // actually give up the CPU. Since skip buddy (v2.6.28):
4268 //
4269 // * Sets the yielding task as skip buddy for current CPU's run queue.
4270 // * Picks next from run queue, if empty, picks a skip buddy (can be the yielding task).
4271 // * Clears skip buddies for this run queue (yielding task no longer a skip buddy).
4272 //
4273 // An alternative is calling os::naked_short_nanosleep with a small number to avoid
4274 // getting re-scheduled immediately.
4275 //
4276 void os::naked_yield() {
4277 sched_yield();
4278 }
4279
4280 ////////////////////////////////////////////////////////////////////////////////
4281 // thread priority support
4282
4283 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4284 // only supports dynamic priority, static priority must be zero. For real-time
4285 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4286 // However, for large multi-threaded applications, SCHED_RR is not only slower
4287 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4288 // of 5 runs - Sep 2005).
4289 //
4290 // The following code actually changes the niceness of kernel-thread/LWP. It
4291 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4292 // not the entire user process, and user level threads are 1:1 mapped to kernel
4293 // threads. It has always been the case, but could change in the future. For
4294 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4295 // It is only used when ThreadPriorityPolicy=1 and may require system level permission
4296 // (e.g., root privilege or CAP_SYS_NICE capability).
4297
4298 int os::java_to_os_priority[CriticalPriority + 1] = {
4299 19, // 0 Entry should never be used
4300
4301 4, // 1 MinPriority
4302 3, // 2
4303 2, // 3
4304
4305 1, // 4
4306 0, // 5 NormPriority
4307 -1, // 6
4308
4309 -2, // 7
4310 -3, // 8
4311 -4, // 9 NearMaxPriority
4312
4313 -5, // 10 MaxPriority
4314
4315 -5 // 11 CriticalPriority
4316 };
4317
4318 static int prio_init() {
4319 if (ThreadPriorityPolicy == 1) {
4320 if (geteuid() != 0) {
4321 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4322 warning("-XX:ThreadPriorityPolicy=1 may require system level permission, " \
4323 "e.g., being the root user. If the necessary permission is not " \
4324 "possessed, changes to priority will be silently ignored.");
4325 }
4326 }
4327 }
4328 if (UseCriticalJavaThreadPriority) {
4329 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4330 }
4331 return 0;
4332 }
4333
4334 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4335 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4336
4337 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4338 return (ret == 0) ? OS_OK : OS_ERR;
4339 }
4340
4341 OSReturn os::get_native_priority(const Thread* const thread,
4342 int *priority_ptr) {
4343 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4344 *priority_ptr = java_to_os_priority[NormPriority];
4345 return OS_OK;
4346 }
4347
4348 errno = 0;
4349 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4350 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4351 }
4352
4353 ////////////////////////////////////////////////////////////////////////////////
4354 // suspend/resume support
4355
4356 // The low-level signal-based suspend/resume support is a remnant from the
4357 // old VM-suspension that used to be for java-suspension, safepoints etc,
4358 // within hotspot. Currently used by JFR's OSThreadSampler
4359 //
4360 // The remaining code is greatly simplified from the more general suspension
4361 // code that used to be used.
4362 //
4363 // The protocol is quite simple:
4364 // - suspend:
4365 // - sends a signal to the target thread
4366 // - polls the suspend state of the osthread using a yield loop
4367 // - target thread signal handler (SR_handler) sets suspend state
4368 // and blocks in sigsuspend until continued
4369 // - resume:
4370 // - sets target osthread state to continue
4371 // - sends signal to end the sigsuspend loop in the SR_handler
4372 //
4373 // Note that the SR_lock plays no role in this suspend/resume protocol,
4374 // but is checked for NULL in SR_handler as a thread termination indicator.
4375 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4376 //
4377 // Note that resume_clear_context() and suspend_save_context() are needed
4378 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4379 // which in part is used by:
4380 // - Forte Analyzer: AsyncGetCallTrace()
4381 // - StackBanging: get_frame_at_stack_banging_point()
4382
4383 static void resume_clear_context(OSThread *osthread) {
4384 osthread->set_ucontext(NULL);
4385 osthread->set_siginfo(NULL);
4386 }
4387
4388 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4389 ucontext_t* context) {
4390 osthread->set_ucontext(context);
4391 osthread->set_siginfo(siginfo);
4392 }
4393
4394 // Handler function invoked when a thread's execution is suspended or
4395 // resumed. We have to be careful that only async-safe functions are
4396 // called here (Note: most pthread functions are not async safe and
4397 // should be avoided.)
4398 //
4399 // Note: sigwait() is a more natural fit than sigsuspend() from an
4400 // interface point of view, but sigwait() prevents the signal hander
4401 // from being run. libpthread would get very confused by not having
4402 // its signal handlers run and prevents sigwait()'s use with the
4403 // mutex granting granting signal.
4404 //
4405 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4406 //
4407 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4408 // Save and restore errno to avoid confusing native code with EINTR
4409 // after sigsuspend.
4410 int old_errno = errno;
4411
4412 Thread* thread = Thread::current_or_null_safe();
4413 assert(thread != NULL, "Missing current thread in SR_handler");
4414
4415 // On some systems we have seen signal delivery get "stuck" until the signal
4416 // mask is changed as part of thread termination. Check that the current thread
4417 // has not already terminated (via SR_lock()) - else the following assertion
4418 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4419 // destructor has completed.
4420
4421 if (thread->SR_lock() == NULL) {
4422 return;
4423 }
4424
4425 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4426
4427 OSThread* osthread = thread->osthread();
4428
4429 os::SuspendResume::State current = osthread->sr.state();
4430 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4431 suspend_save_context(osthread, siginfo, context);
4432
4433 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4434 os::SuspendResume::State state = osthread->sr.suspended();
4435 if (state == os::SuspendResume::SR_SUSPENDED) {
4436 sigset_t suspend_set; // signals for sigsuspend()
4437 sigemptyset(&suspend_set);
4438 // get current set of blocked signals and unblock resume signal
4439 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4440 sigdelset(&suspend_set, SR_signum);
4441
4442 sr_semaphore.signal();
4443 // wait here until we are resumed
4444 while (1) {
4445 sigsuspend(&suspend_set);
4446
4447 os::SuspendResume::State result = osthread->sr.running();
4448 if (result == os::SuspendResume::SR_RUNNING) {
4449 sr_semaphore.signal();
4450 break;
4451 }
4452 }
4453
4454 } else if (state == os::SuspendResume::SR_RUNNING) {
4455 // request was cancelled, continue
4456 } else {
4457 ShouldNotReachHere();
4458 }
4459
4460 resume_clear_context(osthread);
4461 } else if (current == os::SuspendResume::SR_RUNNING) {
4462 // request was cancelled, continue
4463 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4464 // ignore
4465 } else {
4466 // ignore
4467 }
4468
4469 errno = old_errno;
4470 }
4471
4472 static int SR_initialize() {
4473 struct sigaction act;
4474 char *s;
4475
4476 // Get signal number to use for suspend/resume
4477 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4478 int sig = ::strtol(s, 0, 10);
4479 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4480 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4481 SR_signum = sig;
4482 } else {
4483 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4484 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4485 }
4486 }
4487
4488 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4489 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4490
4491 sigemptyset(&SR_sigset);
4492 sigaddset(&SR_sigset, SR_signum);
4493
4494 // Set up signal handler for suspend/resume
4495 act.sa_flags = SA_RESTART|SA_SIGINFO;
4496 act.sa_handler = (void (*)(int)) SR_handler;
4497
4498 // SR_signum is blocked by default.
4499 // 4528190 - We also need to block pthread restart signal (32 on all
4500 // supported Linux platforms). Note that LinuxThreads need to block
4501 // this signal for all threads to work properly. So we don't have
4502 // to use hard-coded signal number when setting up the mask.
4503 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4504
4505 if (sigaction(SR_signum, &act, 0) == -1) {
4506 return -1;
4507 }
4508
4509 // Save signal flag
4510 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4511 return 0;
4512 }
4513
4514 static int sr_notify(OSThread* osthread) {
4515 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4516 assert_status(status == 0, status, "pthread_kill");
4517 return status;
4518 }
4519
4520 // "Randomly" selected value for how long we want to spin
4521 // before bailing out on suspending a thread, also how often
4522 // we send a signal to a thread we want to resume
4523 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4524 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4525
4526 // returns true on success and false on error - really an error is fatal
4527 // but this seems the normal response to library errors
4528 static bool do_suspend(OSThread* osthread) {
4529 assert(osthread->sr.is_running(), "thread should be running");
4530 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4531
4532 // mark as suspended and send signal
4533 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4534 // failed to switch, state wasn't running?
4535 ShouldNotReachHere();
4536 return false;
4537 }
4538
4539 if (sr_notify(osthread) != 0) {
4540 ShouldNotReachHere();
4541 }
4542
4543 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4544 while (true) {
4545 if (sr_semaphore.timedwait(2)) {
4546 break;
4547 } else {
4548 // timeout
4549 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4550 if (cancelled == os::SuspendResume::SR_RUNNING) {
4551 return false;
4552 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4553 // make sure that we consume the signal on the semaphore as well
4554 sr_semaphore.wait();
4555 break;
4556 } else {
4557 ShouldNotReachHere();
4558 return false;
4559 }
4560 }
4561 }
4562
4563 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4564 return true;
4565 }
4566
4567 static void do_resume(OSThread* osthread) {
4568 assert(osthread->sr.is_suspended(), "thread should be suspended");
4569 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4570
4571 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4572 // failed to switch to WAKEUP_REQUEST
4573 ShouldNotReachHere();
4574 return;
4575 }
4576
4577 while (true) {
4578 if (sr_notify(osthread) == 0) {
4579 if (sr_semaphore.timedwait(2)) {
4580 if (osthread->sr.is_running()) {
4581 return;
4582 }
4583 }
4584 } else {
4585 ShouldNotReachHere();
4586 }
4587 }
4588
4589 guarantee(osthread->sr.is_running(), "Must be running!");
4590 }
4591
4592 ///////////////////////////////////////////////////////////////////////////////////
4593 // signal handling (except suspend/resume)
4594
4595 // This routine may be used by user applications as a "hook" to catch signals.
4596 // The user-defined signal handler must pass unrecognized signals to this
4597 // routine, and if it returns true (non-zero), then the signal handler must
4598 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4599 // routine will never retun false (zero), but instead will execute a VM panic
4600 // routine kill the process.
4601 //
4602 // If this routine returns false, it is OK to call it again. This allows
4603 // the user-defined signal handler to perform checks either before or after
4604 // the VM performs its own checks. Naturally, the user code would be making
4605 // a serious error if it tried to handle an exception (such as a null check
4606 // or breakpoint) that the VM was generating for its own correct operation.
4607 //
4608 // This routine may recognize any of the following kinds of signals:
4609 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4610 // It should be consulted by handlers for any of those signals.
4611 //
4612 // The caller of this routine must pass in the three arguments supplied
4613 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4614 // field of the structure passed to sigaction(). This routine assumes that
4615 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4616 //
4617 // Note that the VM will print warnings if it detects conflicting signal
4618 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4619 //
4620 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4621 siginfo_t* siginfo,
4622 void* ucontext,
4623 int abort_if_unrecognized);
4624
4625 static void signalHandler(int sig, siginfo_t* info, void* uc) {
4626 assert(info != NULL && uc != NULL, "it must be old kernel");
4627 int orig_errno = errno; // Preserve errno value over signal handler.
4628 JVM_handle_linux_signal(sig, info, uc, true);
4629 errno = orig_errno;
4630 }
4631
4632
4633 // This boolean allows users to forward their own non-matching signals
4634 // to JVM_handle_linux_signal, harmlessly.
4635 bool os::Linux::signal_handlers_are_installed = false;
4636
4637 // For signal-chaining
4638 bool os::Linux::libjsig_is_loaded = false;
4639 typedef struct sigaction *(*get_signal_t)(int);
4640 get_signal_t os::Linux::get_signal_action = NULL;
4641
4642 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4643 struct sigaction *actp = NULL;
4644
4645 if (libjsig_is_loaded) {
4646 // Retrieve the old signal handler from libjsig
4647 actp = (*get_signal_action)(sig);
4648 }
4649 if (actp == NULL) {
4650 // Retrieve the preinstalled signal handler from jvm
4651 actp = os::Posix::get_preinstalled_handler(sig);
4652 }
4653
4654 return actp;
4655 }
4656
4657 static bool call_chained_handler(struct sigaction *actp, int sig,
4658 siginfo_t *siginfo, void *context) {
4659 // Call the old signal handler
4660 if (actp->sa_handler == SIG_DFL) {
4661 // It's more reasonable to let jvm treat it as an unexpected exception
4662 // instead of taking the default action.
4663 return false;
4664 } else if (actp->sa_handler != SIG_IGN) {
4665 if ((actp->sa_flags & SA_NODEFER) == 0) {
4666 // automaticlly block the signal
4667 sigaddset(&(actp->sa_mask), sig);
4668 }
4669
4670 sa_handler_t hand = NULL;
4671 sa_sigaction_t sa = NULL;
4672 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4673 // retrieve the chained handler
4674 if (siginfo_flag_set) {
4675 sa = actp->sa_sigaction;
4676 } else {
4677 hand = actp->sa_handler;
4678 }
4679
4680 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4681 actp->sa_handler = SIG_DFL;
4682 }
4683
4684 // try to honor the signal mask
4685 sigset_t oset;
4686 sigemptyset(&oset);
4687 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4688
4689 // call into the chained handler
4690 if (siginfo_flag_set) {
4691 (*sa)(sig, siginfo, context);
4692 } else {
4693 (*hand)(sig);
4694 }
4695
4696 // restore the signal mask
4697 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4698 }
4699 // Tell jvm's signal handler the signal is taken care of.
4700 return true;
4701 }
4702
4703 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4704 bool chained = false;
4705 // signal-chaining
4706 if (UseSignalChaining) {
4707 struct sigaction *actp = get_chained_signal_action(sig);
4708 if (actp != NULL) {
4709 chained = call_chained_handler(actp, sig, siginfo, context);
4710 }
4711 }
4712 return chained;
4713 }
4714
4715 // for diagnostic
4716 int sigflags[NSIG];
4717
4718 int os::Linux::get_our_sigflags(int sig) {
4719 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4720 return sigflags[sig];
4721 }
4722
4723 void os::Linux::set_our_sigflags(int sig, int flags) {
4724 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4725 if (sig > 0 && sig < NSIG) {
4726 sigflags[sig] = flags;
4727 }
4728 }
4729
4730 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4731 // Check for overwrite.
4732 struct sigaction oldAct;
4733 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4734
4735 void* oldhand = oldAct.sa_sigaction
4736 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4737 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4738 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4739 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4740 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4741 if (AllowUserSignalHandlers || !set_installed) {
4742 // Do not overwrite; user takes responsibility to forward to us.
4743 return;
4744 } else if (UseSignalChaining) {
4745 // save the old handler in jvm
4746 os::Posix::save_preinstalled_handler(sig, oldAct);
4747 // libjsig also interposes the sigaction() call below and saves the
4748 // old sigaction on it own.
4749 } else {
4750 fatal("Encountered unexpected pre-existing sigaction handler "
4751 "%#lx for signal %d.", (long)oldhand, sig);
4752 }
4753 }
4754
4755 struct sigaction sigAct;
4756 sigfillset(&(sigAct.sa_mask));
4757 sigAct.sa_handler = SIG_DFL;
4758 if (!set_installed) {
4759 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4760 } else {
4761 sigAct.sa_sigaction = signalHandler;
4762 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4763 }
4764 // Save flags, which are set by ours
4765 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4766 sigflags[sig] = sigAct.sa_flags;
4767
4768 int ret = sigaction(sig, &sigAct, &oldAct);
4769 assert(ret == 0, "check");
4770
4771 void* oldhand2 = oldAct.sa_sigaction
4772 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4773 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4774 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4775 }
4776
4777 // install signal handlers for signals that HotSpot needs to
4778 // handle in order to support Java-level exception handling.
4779
4780 void os::Linux::install_signal_handlers() {
4781 if (!signal_handlers_are_installed) {
4782 signal_handlers_are_installed = true;
4783
4784 // signal-chaining
4785 typedef void (*signal_setting_t)();
4786 signal_setting_t begin_signal_setting = NULL;
4787 signal_setting_t end_signal_setting = NULL;
4788 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4789 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4790 if (begin_signal_setting != NULL) {
4791 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4792 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4793 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4794 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4795 libjsig_is_loaded = true;
4796 assert(UseSignalChaining, "should enable signal-chaining");
4797 }
4798 if (libjsig_is_loaded) {
4799 // Tell libjsig jvm is setting signal handlers
4800 (*begin_signal_setting)();
4801 }
4802
4803 set_signal_handler(SIGSEGV, true);
4804 set_signal_handler(SIGPIPE, true);
4805 set_signal_handler(SIGBUS, true);
4806 set_signal_handler(SIGILL, true);
4807 set_signal_handler(SIGFPE, true);
4808 #if defined(PPC64)
4809 set_signal_handler(SIGTRAP, true);
4810 #endif
4811 set_signal_handler(SIGXFSZ, true);
4812
4813 if (libjsig_is_loaded) {
4814 // Tell libjsig jvm finishes setting signal handlers
4815 (*end_signal_setting)();
4816 }
4817
4818 // We don't activate signal checker if libjsig is in place, we trust ourselves
4819 // and if UserSignalHandler is installed all bets are off.
4820 // Log that signal checking is off only if -verbose:jni is specified.
4821 if (CheckJNICalls) {
4822 if (libjsig_is_loaded) {
4823 log_debug(jni, resolve)("Info: libjsig is activated, all active signal checking is disabled");
4824 check_signals = false;
4825 }
4826 if (AllowUserSignalHandlers) {
4827 log_debug(jni, resolve)("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4828 check_signals = false;
4829 }
4830 }
4831 }
4832 }
4833
4834 // This is the fastest way to get thread cpu time on Linux.
4835 // Returns cpu time (user+sys) for any thread, not only for current.
4836 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4837 // It might work on 2.6.10+ with a special kernel/glibc patch.
4838 // For reference, please, see IEEE Std 1003.1-2004:
4839 // http://www.unix.org/single_unix_specification
4840
4841 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4842 struct timespec tp;
4843 int rc = os::Posix::clock_gettime(clockid, &tp);
4844 assert(rc == 0, "clock_gettime is expected to return 0 code");
4845
4846 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4847 }
4848
4849 /////
4850 // glibc on Linux platform uses non-documented flag
4851 // to indicate, that some special sort of signal
4852 // trampoline is used.
4853 // We will never set this flag, and we should
4854 // ignore this flag in our diagnostic
4855 #ifdef SIGNIFICANT_SIGNAL_MASK
4856 #undef SIGNIFICANT_SIGNAL_MASK
4857 #endif
4858 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4859
4860 static const char* get_signal_handler_name(address handler,
4861 char* buf, int buflen) {
4862 int offset = 0;
4863 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4864 if (found) {
4865 // skip directory names
4866 const char *p1, *p2;
4867 p1 = buf;
4868 size_t len = strlen(os::file_separator());
4869 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4870 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4871 } else {
4872 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4873 }
4874 return buf;
4875 }
4876
4877 static void print_signal_handler(outputStream* st, int sig,
4878 char* buf, size_t buflen) {
4879 struct sigaction sa;
4880
4881 sigaction(sig, NULL, &sa);
4882
4883 // See comment for SIGNIFICANT_SIGNAL_MASK define
4884 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4885
4886 st->print("%s: ", os::exception_name(sig, buf, buflen));
4887
4888 address handler = (sa.sa_flags & SA_SIGINFO)
4889 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4890 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4891
4892 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4893 st->print("SIG_DFL");
4894 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4895 st->print("SIG_IGN");
4896 } else {
4897 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4898 }
4899
4900 st->print(", sa_mask[0]=");
4901 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4902
4903 address rh = VMError::get_resetted_sighandler(sig);
4904 // May be, handler was resetted by VMError?
4905 if (rh != NULL) {
4906 handler = rh;
4907 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4908 }
4909
4910 st->print(", sa_flags=");
4911 os::Posix::print_sa_flags(st, sa.sa_flags);
4912
4913 // Check: is it our handler?
4914 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4915 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4916 // It is our signal handler
4917 // check for flags, reset system-used one!
4918 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4919 st->print(
4920 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4921 os::Linux::get_our_sigflags(sig));
4922 }
4923 }
4924 st->cr();
4925 }
4926
4927
4928 #define DO_SIGNAL_CHECK(sig) \
4929 do { \
4930 if (!sigismember(&check_signal_done, sig)) { \
4931 os::Linux::check_signal_handler(sig); \
4932 } \
4933 } while (0)
4934
4935 // This method is a periodic task to check for misbehaving JNI applications
4936 // under CheckJNI, we can add any periodic checks here
4937
4938 void os::run_periodic_checks() {
4939 if (check_signals == false) return;
4940
4941 // SEGV and BUS if overridden could potentially prevent
4942 // generation of hs*.log in the event of a crash, debugging
4943 // such a case can be very challenging, so we absolutely
4944 // check the following for a good measure:
4945 DO_SIGNAL_CHECK(SIGSEGV);
4946 DO_SIGNAL_CHECK(SIGILL);
4947 DO_SIGNAL_CHECK(SIGFPE);
4948 DO_SIGNAL_CHECK(SIGBUS);
4949 DO_SIGNAL_CHECK(SIGPIPE);
4950 DO_SIGNAL_CHECK(SIGXFSZ);
4951 #if defined(PPC64)
4952 DO_SIGNAL_CHECK(SIGTRAP);
4953 #endif
4954
4955 // ReduceSignalUsage allows the user to override these handlers
4956 // see comments at the very top and jvm_md.h
4957 if (!ReduceSignalUsage) {
4958 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4959 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4960 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4961 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4962 }
4963
4964 DO_SIGNAL_CHECK(SR_signum);
4965 }
4966
4967 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4968
4969 static os_sigaction_t os_sigaction = NULL;
4970
4971 void os::Linux::check_signal_handler(int sig) {
4972 char buf[O_BUFLEN];
4973 address jvmHandler = NULL;
4974
4975
4976 struct sigaction act;
4977 if (os_sigaction == NULL) {
4978 // only trust the default sigaction, in case it has been interposed
4979 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4980 if (os_sigaction == NULL) return;
4981 }
4982
4983 os_sigaction(sig, (struct sigaction*)NULL, &act);
4984
4985
4986 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4987
4988 address thisHandler = (act.sa_flags & SA_SIGINFO)
4989 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4990 : CAST_FROM_FN_PTR(address, act.sa_handler);
4991
4992
4993 switch (sig) {
4994 case SIGSEGV:
4995 case SIGBUS:
4996 case SIGFPE:
4997 case SIGPIPE:
4998 case SIGILL:
4999 case SIGXFSZ:
5000 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
5001 break;
5002
5003 case SHUTDOWN1_SIGNAL:
5004 case SHUTDOWN2_SIGNAL:
5005 case SHUTDOWN3_SIGNAL:
5006 case BREAK_SIGNAL:
5007 jvmHandler = (address)user_handler();
5008 break;
5009
5010 default:
5011 if (sig == SR_signum) {
5012 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
5013 } else {
5014 return;
5015 }
5016 break;
5017 }
5018
5019 if (thisHandler != jvmHandler) {
5020 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
5021 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
5022 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
5023 // No need to check this sig any longer
5024 sigaddset(&check_signal_done, sig);
5025 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
5026 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
5027 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
5028 exception_name(sig, buf, O_BUFLEN));
5029 }
5030 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
5031 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
5032 tty->print("expected:");
5033 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
5034 tty->cr();
5035 tty->print(" found:");
5036 os::Posix::print_sa_flags(tty, act.sa_flags);
5037 tty->cr();
5038 // No need to check this sig any longer
5039 sigaddset(&check_signal_done, sig);
5040 }
5041
5042 // Dump all the signal
5043 if (sigismember(&check_signal_done, sig)) {
5044 print_signal_handlers(tty, buf, O_BUFLEN);
5045 }
5046 }
5047
5048 extern void report_error(char* file_name, int line_no, char* title,
5049 char* format, ...);
5050
5051 // this is called _before_ most of the global arguments have been parsed
5052 void os::init(void) {
5053 char dummy; // used to get a guess on initial stack address
5054
5055 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
5056
5057 init_random(1234567);
5058
5059 Linux::set_page_size(sysconf(_SC_PAGESIZE));
5060 if (Linux::page_size() == -1) {
5061 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
5062 os::strerror(errno));
5063 }
5064 init_page_sizes((size_t) Linux::page_size());
5065
5066 Linux::initialize_system_info();
5067
5068 os::Linux::CPUPerfTicks pticks;
5069 bool res = os::Linux::get_tick_information(&pticks, -1);
5070
5071 if (res && pticks.has_steal_ticks) {
5072 has_initial_tick_info = true;
5073 initial_total_ticks = pticks.total;
5074 initial_steal_ticks = pticks.steal;
5075 }
5076
5077 // _main_thread points to the thread that created/loaded the JVM.
5078 Linux::_main_thread = pthread_self();
5079
5080 // retrieve entry point for pthread_setname_np
5081 Linux::_pthread_setname_np =
5082 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
5083
5084 os::Posix::init();
5085
5086 initial_time_count = javaTimeNanos();
5087
5088 // Always warn if no monotonic clock available
5089 if (!os::Posix::supports_monotonic_clock()) {
5090 warning("No monotonic clock was available - timed services may " \
5091 "be adversely affected if the time-of-day clock changes");
5092 }
5093 }
5094
5095 // To install functions for atexit system call
5096 extern "C" {
5097 static void perfMemory_exit_helper() {
5098 perfMemory_exit();
5099 }
5100 }
5101
5102 void os::pd_init_container_support() {
5103 OSContainer::init();
5104 }
5105
5106 void os::Linux::numa_init() {
5107
5108 // Java can be invoked as
5109 // 1. Without numactl and heap will be allocated/configured on all nodes as
5110 // per the system policy.
5111 // 2. With numactl --interleave:
5112 // Use numa_get_interleave_mask(v2) API to get nodes bitmask. The same
5113 // API for membind case bitmask is reset.
5114 // Interleave is only hint and Kernel can fallback to other nodes if
5115 // no memory is available on the target nodes.
5116 // 3. With numactl --membind:
5117 // Use numa_get_membind(v2) API to get nodes bitmask. The same API for
5118 // interleave case returns bitmask of all nodes.
5119 // numa_all_nodes_ptr holds bitmask of all nodes.
5120 // numa_get_interleave_mask(v2) and numa_get_membind(v2) APIs returns correct
5121 // bitmask when externally configured to run on all or fewer nodes.
5122
5123 if (!Linux::libnuma_init()) {
5124 UseNUMA = false;
5125 } else {
5126 if ((Linux::numa_max_node() < 1) || Linux::is_bound_to_single_node()) {
5127 // If there's only one node (they start from 0) or if the process
5128 // is bound explicitly to a single node using membind, disable NUMA.
5129 UseNUMA = false;
5130 } else {
5131
5132 LogTarget(Info,os) log;
5133 LogStream ls(log);
5134
5135 Linux::set_configured_numa_policy(Linux::identify_numa_policy());
5136
5137 struct bitmask* bmp = Linux::_numa_membind_bitmask;
5138 const char* numa_mode = "membind";
5139
5140 if (Linux::is_running_in_interleave_mode()) {
5141 bmp = Linux::_numa_interleave_bitmask;
5142 numa_mode = "interleave";
5143 }
5144
5145 ls.print("UseNUMA is enabled and invoked in '%s' mode."
5146 " Heap will be configured using NUMA memory nodes:", numa_mode);
5147
5148 for (int node = 0; node <= Linux::numa_max_node(); node++) {
5149 if (Linux::_numa_bitmask_isbitset(bmp, node)) {
5150 ls.print(" %d", node);
5151 }
5152 }
5153 }
5154 }
5155
5156 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5157 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5158 // we can make the adaptive lgrp chunk resizing work. If the user specified both
5159 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5160 // and disable adaptive resizing.
5161 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5162 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5163 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5164 UseAdaptiveSizePolicy = false;
5165 UseAdaptiveNUMAChunkSizing = false;
5166 }
5167 }
5168
5169 if (!UseNUMA && ForceNUMA) {
5170 UseNUMA = true;
5171 }
5172 }
5173
5174 // this is called _after_ the global arguments have been parsed
5175 jint os::init_2(void) {
5176
5177 // This could be set after os::Posix::init() but all platforms
5178 // have to set it the same so we have to mirror Solaris.
5179 DEBUG_ONLY(os::set_mutex_init_done();)
5180
5181 os::Posix::init_2();
5182
5183 Linux::fast_thread_clock_init();
5184
5185 // initialize suspend/resume support - must do this before signal_sets_init()
5186 if (SR_initialize() != 0) {
5187 perror("SR_initialize failed");
5188 return JNI_ERR;
5189 }
5190
5191 Linux::signal_sets_init();
5192 Linux::install_signal_handlers();
5193 // Initialize data for jdk.internal.misc.Signal
5194 if (!ReduceSignalUsage) {
5195 jdk_misc_signal_init();
5196 }
5197
5198 if (AdjustStackSizeForTLS) {
5199 get_minstack_init();
5200 }
5201
5202 // Check and sets minimum stack sizes against command line options
5203 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
5204 return JNI_ERR;
5205 }
5206
5207 #if defined(IA32)
5208 // Need to ensure we've determined the process's initial stack to
5209 // perform the workaround
5210 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5211 workaround_expand_exec_shield_cs_limit();
5212 #else
5213 suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
5214 if (!suppress_primordial_thread_resolution) {
5215 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5216 }
5217 #endif
5218
5219 Linux::libpthread_init();
5220 Linux::sched_getcpu_init();
5221 log_info(os)("HotSpot is running with %s, %s",
5222 Linux::glibc_version(), Linux::libpthread_version());
5223
5224 if (UseNUMA) {
5225 Linux::numa_init();
5226 }
5227
5228 if (MaxFDLimit) {
5229 // set the number of file descriptors to max. print out error
5230 // if getrlimit/setrlimit fails but continue regardless.
5231 struct rlimit nbr_files;
5232 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5233 if (status != 0) {
5234 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5235 } else {
5236 nbr_files.rlim_cur = nbr_files.rlim_max;
5237 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5238 if (status != 0) {
5239 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5240 }
5241 }
5242 }
5243
5244 // at-exit methods are called in the reverse order of their registration.
5245 // atexit functions are called on return from main or as a result of a
5246 // call to exit(3C). There can be only 32 of these functions registered
5247 // and atexit() does not set errno.
5248
5249 if (PerfAllowAtExitRegistration) {
5250 // only register atexit functions if PerfAllowAtExitRegistration is set.
5251 // atexit functions can be delayed until process exit time, which
5252 // can be problematic for embedded VM situations. Embedded VMs should
5253 // call DestroyJavaVM() to assure that VM resources are released.
5254
5255 // note: perfMemory_exit_helper atexit function may be removed in
5256 // the future if the appropriate cleanup code can be added to the
5257 // VM_Exit VMOperation's doit method.
5258 if (atexit(perfMemory_exit_helper) != 0) {
5259 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5260 }
5261 }
5262
5263 // initialize thread priority policy
5264 prio_init();
5265
5266 if (!FLAG_IS_DEFAULT(AllocateHeapAt) || !FLAG_IS_DEFAULT(AllocateOldGenAt)) {
5267 set_coredump_filter(DAX_SHARED_BIT);
5268 }
5269
5270 if (DumpPrivateMappingsInCore) {
5271 set_coredump_filter(FILE_BACKED_PVT_BIT);
5272 }
5273
5274 if (DumpSharedMappingsInCore) {
5275 set_coredump_filter(FILE_BACKED_SHARED_BIT);
5276 }
5277
5278 return JNI_OK;
5279 }
5280
5281 // Mark the polling page as unreadable
5282 void os::make_polling_page_unreadable(void) {
5283 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5284 fatal("Could not disable polling page");
5285 }
5286 }
5287
5288 // Mark the polling page as readable
5289 void os::make_polling_page_readable(void) {
5290 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5291 fatal("Could not enable polling page");
5292 }
5293 }
5294
5295 // older glibc versions don't have this macro (which expands to
5296 // an optimized bit-counting function) so we have to roll our own
5297 #ifndef CPU_COUNT
5298
5299 static int _cpu_count(const cpu_set_t* cpus) {
5300 int count = 0;
5301 // only look up to the number of configured processors
5302 for (int i = 0; i < os::processor_count(); i++) {
5303 if (CPU_ISSET(i, cpus)) {
5304 count++;
5305 }
5306 }
5307 return count;
5308 }
5309
5310 #define CPU_COUNT(cpus) _cpu_count(cpus)
5311
5312 #endif // CPU_COUNT
5313
5314 // Get the current number of available processors for this process.
5315 // This value can change at any time during a process's lifetime.
5316 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5317 // If it appears there may be more than 1024 processors then we do a
5318 // dynamic check - see 6515172 for details.
5319 // If anything goes wrong we fallback to returning the number of online
5320 // processors - which can be greater than the number available to the process.
5321 int os::Linux::active_processor_count() {
5322 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5323 cpu_set_t* cpus_p = &cpus;
5324 int cpus_size = sizeof(cpu_set_t);
5325
5326 int configured_cpus = os::processor_count(); // upper bound on available cpus
5327 int cpu_count = 0;
5328
5329 // old build platforms may not support dynamic cpu sets
5330 #ifdef CPU_ALLOC
5331
5332 // To enable easy testing of the dynamic path on different platforms we
5333 // introduce a diagnostic flag: UseCpuAllocPath
5334 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5335 // kernel may use a mask bigger than cpu_set_t
5336 log_trace(os)("active_processor_count: using dynamic path %s"
5337 "- configured processors: %d",
5338 UseCpuAllocPath ? "(forced) " : "",
5339 configured_cpus);
5340 cpus_p = CPU_ALLOC(configured_cpus);
5341 if (cpus_p != NULL) {
5342 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5343 // zero it just to be safe
5344 CPU_ZERO_S(cpus_size, cpus_p);
5345 }
5346 else {
5347 // failed to allocate so fallback to online cpus
5348 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5349 log_trace(os)("active_processor_count: "
5350 "CPU_ALLOC failed (%s) - using "
5351 "online processor count: %d",
5352 os::strerror(errno), online_cpus);
5353 return online_cpus;
5354 }
5355 }
5356 else {
5357 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5358 configured_cpus);
5359 }
5360 #else // CPU_ALLOC
5361 // these stubs won't be executed
5362 #define CPU_COUNT_S(size, cpus) -1
5363 #define CPU_FREE(cpus)
5364
5365 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5366 configured_cpus);
5367 #endif // CPU_ALLOC
5368
5369 // pid 0 means the current thread - which we have to assume represents the process
5370 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5371 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5372 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5373 }
5374 else {
5375 cpu_count = CPU_COUNT(cpus_p);
5376 }
5377 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5378 }
5379 else {
5380 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5381 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5382 "which may exceed available processors", os::strerror(errno), cpu_count);
5383 }
5384
5385 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5386 CPU_FREE(cpus_p);
5387 }
5388
5389 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5390 return cpu_count;
5391 }
5392
5393 // Determine the active processor count from one of
5394 // three different sources:
5395 //
5396 // 1. User option -XX:ActiveProcessorCount
5397 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5398 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5399 //
5400 // Option 1, if specified, will always override.
5401 // If the cgroup subsystem is active and configured, we
5402 // will return the min of the cgroup and option 2 results.
5403 // This is required since tools, such as numactl, that
5404 // alter cpu affinity do not update cgroup subsystem
5405 // cpuset configuration files.
5406 int os::active_processor_count() {
5407 // User has overridden the number of active processors
5408 if (ActiveProcessorCount > 0) {
5409 log_trace(os)("active_processor_count: "
5410 "active processor count set by user : %d",
5411 ActiveProcessorCount);
5412 return ActiveProcessorCount;
5413 }
5414
5415 int active_cpus;
5416 if (OSContainer::is_containerized()) {
5417 active_cpus = OSContainer::active_processor_count();
5418 log_trace(os)("active_processor_count: determined by OSContainer: %d",
5419 active_cpus);
5420 } else {
5421 active_cpus = os::Linux::active_processor_count();
5422 }
5423
5424 return active_cpus;
5425 }
5426
5427 uint os::processor_id() {
5428 const int id = Linux::sched_getcpu();
5429 assert(id >= 0 && id < _processor_count, "Invalid processor id");
5430 return (uint)id;
5431 }
5432
5433 void os::set_native_thread_name(const char *name) {
5434 if (Linux::_pthread_setname_np) {
5435 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5436 snprintf(buf, sizeof(buf), "%s", name);
5437 buf[sizeof(buf) - 1] = '\0';
5438 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5439 // ERANGE should not happen; all other errors should just be ignored.
5440 assert(rc != ERANGE, "pthread_setname_np failed");
5441 }
5442 }
5443
5444 bool os::bind_to_processor(uint processor_id) {
5445 // Not yet implemented.
5446 return false;
5447 }
5448
5449 ///
5450
5451 void os::SuspendedThreadTask::internal_do_task() {
5452 if (do_suspend(_thread->osthread())) {
5453 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5454 do_task(context);
5455 do_resume(_thread->osthread());
5456 }
5457 }
5458
5459 ////////////////////////////////////////////////////////////////////////////////
5460 // debug support
5461
5462 bool os::find(address addr, outputStream* st) {
5463 Dl_info dlinfo;
5464 memset(&dlinfo, 0, sizeof(dlinfo));
5465 if (dladdr(addr, &dlinfo) != 0) {
5466 st->print(PTR_FORMAT ": ", p2i(addr));
5467 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5468 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5469 p2i(addr) - p2i(dlinfo.dli_saddr));
5470 } else if (dlinfo.dli_fbase != NULL) {
5471 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5472 } else {
5473 st->print("<absolute address>");
5474 }
5475 if (dlinfo.dli_fname != NULL) {
5476 st->print(" in %s", dlinfo.dli_fname);
5477 }
5478 if (dlinfo.dli_fbase != NULL) {
5479 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5480 }
5481 st->cr();
5482
5483 if (Verbose) {
5484 // decode some bytes around the PC
5485 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5486 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5487 address lowest = (address) dlinfo.dli_sname;
5488 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5489 if (begin < lowest) begin = lowest;
5490 Dl_info dlinfo2;
5491 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5492 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5493 end = (address) dlinfo2.dli_saddr;
5494 }
5495 Disassembler::decode(begin, end, st);
5496 }
5497 return true;
5498 }
5499 return false;
5500 }
5501
5502 ////////////////////////////////////////////////////////////////////////////////
5503 // misc
5504
5505 // This does not do anything on Linux. This is basically a hook for being
5506 // able to use structured exception handling (thread-local exception filters)
5507 // on, e.g., Win32.
5508 void
5509 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5510 JavaCallArguments* args, Thread* thread) {
5511 f(value, method, args, thread);
5512 }
5513
5514 void os::print_statistics() {
5515 }
5516
5517 bool os::message_box(const char* title, const char* message) {
5518 int i;
5519 fdStream err(defaultStream::error_fd());
5520 for (i = 0; i < 78; i++) err.print_raw("=");
5521 err.cr();
5522 err.print_raw_cr(title);
5523 for (i = 0; i < 78; i++) err.print_raw("-");
5524 err.cr();
5525 err.print_raw_cr(message);
5526 for (i = 0; i < 78; i++) err.print_raw("=");
5527 err.cr();
5528
5529 char buf[16];
5530 // Prevent process from exiting upon "read error" without consuming all CPU
5531 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5532
5533 return buf[0] == 'y' || buf[0] == 'Y';
5534 }
5535
5536 // Is a (classpath) directory empty?
5537 bool os::dir_is_empty(const char* path) {
5538 DIR *dir = NULL;
5539 struct dirent *ptr;
5540
5541 dir = opendir(path);
5542 if (dir == NULL) return true;
5543
5544 // Scan the directory
5545 bool result = true;
5546 while (result && (ptr = readdir(dir)) != NULL) {
5547 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5548 result = false;
5549 }
5550 }
5551 closedir(dir);
5552 return result;
5553 }
5554
5555 // This code originates from JDK's sysOpen and open64_w
5556 // from src/solaris/hpi/src/system_md.c
5557
5558 int os::open(const char *path, int oflag, int mode) {
5559 if (strlen(path) > MAX_PATH - 1) {
5560 errno = ENAMETOOLONG;
5561 return -1;
5562 }
5563
5564 // All file descriptors that are opened in the Java process and not
5565 // specifically destined for a subprocess should have the close-on-exec
5566 // flag set. If we don't set it, then careless 3rd party native code
5567 // might fork and exec without closing all appropriate file descriptors
5568 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5569 // turn might:
5570 //
5571 // - cause end-of-file to fail to be detected on some file
5572 // descriptors, resulting in mysterious hangs, or
5573 //
5574 // - might cause an fopen in the subprocess to fail on a system
5575 // suffering from bug 1085341.
5576 //
5577 // (Yes, the default setting of the close-on-exec flag is a Unix
5578 // design flaw)
5579 //
5580 // See:
5581 // 1085341: 32-bit stdio routines should support file descriptors >255
5582 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5583 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5584 //
5585 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5586 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5587 // because it saves a system call and removes a small window where the flag
5588 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5589 // and we fall back to using FD_CLOEXEC (see below).
5590 #ifdef O_CLOEXEC
5591 oflag |= O_CLOEXEC;
5592 #endif
5593
5594 int fd = ::open64(path, oflag, mode);
5595 if (fd == -1) return -1;
5596
5597 //If the open succeeded, the file might still be a directory
5598 {
5599 struct stat64 buf64;
5600 int ret = ::fstat64(fd, &buf64);
5601 int st_mode = buf64.st_mode;
5602
5603 if (ret != -1) {
5604 if ((st_mode & S_IFMT) == S_IFDIR) {
5605 errno = EISDIR;
5606 ::close(fd);
5607 return -1;
5608 }
5609 } else {
5610 ::close(fd);
5611 return -1;
5612 }
5613 }
5614
5615 #ifdef FD_CLOEXEC
5616 // Validate that the use of the O_CLOEXEC flag on open above worked.
5617 // With recent kernels, we will perform this check exactly once.
5618 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5619 if (!O_CLOEXEC_is_known_to_work) {
5620 int flags = ::fcntl(fd, F_GETFD);
5621 if (flags != -1) {
5622 if ((flags & FD_CLOEXEC) != 0)
5623 O_CLOEXEC_is_known_to_work = 1;
5624 else
5625 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5626 }
5627 }
5628 #endif
5629
5630 return fd;
5631 }
5632
5633
5634 // create binary file, rewriting existing file if required
5635 int os::create_binary_file(const char* path, bool rewrite_existing) {
5636 int oflags = O_WRONLY | O_CREAT;
5637 if (!rewrite_existing) {
5638 oflags |= O_EXCL;
5639 }
5640 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5641 }
5642
5643 // return current position of file pointer
5644 jlong os::current_file_offset(int fd) {
5645 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5646 }
5647
5648 // move file pointer to the specified offset
5649 jlong os::seek_to_file_offset(int fd, jlong offset) {
5650 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5651 }
5652
5653 // This code originates from JDK's sysAvailable
5654 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5655
5656 int os::available(int fd, jlong *bytes) {
5657 jlong cur, end;
5658 int mode;
5659 struct stat64 buf64;
5660
5661 if (::fstat64(fd, &buf64) >= 0) {
5662 mode = buf64.st_mode;
5663 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5664 int n;
5665 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5666 *bytes = n;
5667 return 1;
5668 }
5669 }
5670 }
5671 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5672 return 0;
5673 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5674 return 0;
5675 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5676 return 0;
5677 }
5678 *bytes = end - cur;
5679 return 1;
5680 }
5681
5682 // Map a block of memory.
5683 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5684 char *addr, size_t bytes, bool read_only,
5685 bool allow_exec) {
5686 int prot;
5687 int flags = MAP_PRIVATE;
5688
5689 if (read_only) {
5690 prot = PROT_READ;
5691 } else {
5692 prot = PROT_READ | PROT_WRITE;
5693 }
5694
5695 if (allow_exec) {
5696 prot |= PROT_EXEC;
5697 }
5698
5699 if (addr != NULL) {
5700 flags |= MAP_FIXED;
5701 }
5702
5703 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5704 fd, file_offset);
5705 if (mapped_address == MAP_FAILED) {
5706 return NULL;
5707 }
5708 return mapped_address;
5709 }
5710
5711
5712 // Remap a block of memory.
5713 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5714 char *addr, size_t bytes, bool read_only,
5715 bool allow_exec) {
5716 // same as map_memory() on this OS
5717 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5718 allow_exec);
5719 }
5720
5721
5722 // Unmap a block of memory.
5723 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5724 return munmap(addr, bytes) == 0;
5725 }
5726
5727 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5728
5729 static jlong fast_cpu_time(Thread *thread) {
5730 clockid_t clockid;
5731 int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(),
5732 &clockid);
5733 if (rc == 0) {
5734 return os::Linux::fast_thread_cpu_time(clockid);
5735 } else {
5736 // It's possible to encounter a terminated native thread that failed
5737 // to detach itself from the VM - which should result in ESRCH.
5738 assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed");
5739 return -1;
5740 }
5741 }
5742
5743 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5744 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5745 // of a thread.
5746 //
5747 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5748 // the fast estimate available on the platform.
5749
5750 jlong os::current_thread_cpu_time() {
5751 if (os::Linux::supports_fast_thread_cpu_time()) {
5752 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5753 } else {
5754 // return user + sys since the cost is the same
5755 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5756 }
5757 }
5758
5759 jlong os::thread_cpu_time(Thread* thread) {
5760 // consistent with what current_thread_cpu_time() returns
5761 if (os::Linux::supports_fast_thread_cpu_time()) {
5762 return fast_cpu_time(thread);
5763 } else {
5764 return slow_thread_cpu_time(thread, true /* user + sys */);
5765 }
5766 }
5767
5768 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5769 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5770 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5771 } else {
5772 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5773 }
5774 }
5775
5776 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5777 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5778 return fast_cpu_time(thread);
5779 } else {
5780 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5781 }
5782 }
5783
5784 // -1 on error.
5785 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5786 pid_t tid = thread->osthread()->thread_id();
5787 char *s;
5788 char stat[2048];
5789 int statlen;
5790 char proc_name[64];
5791 int count;
5792 long sys_time, user_time;
5793 char cdummy;
5794 int idummy;
5795 long ldummy;
5796 FILE *fp;
5797
5798 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5799 fp = fopen(proc_name, "r");
5800 if (fp == NULL) return -1;
5801 statlen = fread(stat, 1, 2047, fp);
5802 stat[statlen] = '\0';
5803 fclose(fp);
5804
5805 // Skip pid and the command string. Note that we could be dealing with
5806 // weird command names, e.g. user could decide to rename java launcher
5807 // to "java 1.4.2 :)", then the stat file would look like
5808 // 1234 (java 1.4.2 :)) R ... ...
5809 // We don't really need to know the command string, just find the last
5810 // occurrence of ")" and then start parsing from there. See bug 4726580.
5811 s = strrchr(stat, ')');
5812 if (s == NULL) return -1;
5813
5814 // Skip blank chars
5815 do { s++; } while (s && isspace(*s));
5816
5817 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5818 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5819 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5820 &user_time, &sys_time);
5821 if (count != 13) return -1;
5822 if (user_sys_cpu_time) {
5823 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5824 } else {
5825 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5826 }
5827 }
5828
5829 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5830 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5831 info_ptr->may_skip_backward = false; // elapsed time not wall time
5832 info_ptr->may_skip_forward = false; // elapsed time not wall time
5833 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5834 }
5835
5836 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5837 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5838 info_ptr->may_skip_backward = false; // elapsed time not wall time
5839 info_ptr->may_skip_forward = false; // elapsed time not wall time
5840 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5841 }
5842
5843 bool os::is_thread_cpu_time_supported() {
5844 return true;
5845 }
5846
5847 // System loadavg support. Returns -1 if load average cannot be obtained.
5848 // Linux doesn't yet have a (official) notion of processor sets,
5849 // so just return the system wide load average.
5850 int os::loadavg(double loadavg[], int nelem) {
5851 return ::getloadavg(loadavg, nelem);
5852 }
5853
5854 void os::pause() {
5855 char filename[MAX_PATH];
5856 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5857 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5858 } else {
5859 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5860 }
5861
5862 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5863 if (fd != -1) {
5864 struct stat buf;
5865 ::close(fd);
5866 while (::stat(filename, &buf) == 0) {
5867 (void)::poll(NULL, 0, 100);
5868 }
5869 } else {
5870 jio_fprintf(stderr,
5871 "Could not open pause file '%s', continuing immediately.\n", filename);
5872 }
5873 }
5874
5875 extern char** environ;
5876
5877 // Run the specified command in a separate process. Return its exit value,
5878 // or -1 on failure (e.g. can't fork a new process).
5879 // Unlike system(), this function can be called from signal handler. It
5880 // doesn't block SIGINT et al.
5881 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) {
5882 const char * argv[4] = {"sh", "-c", cmd, NULL};
5883
5884 pid_t pid ;
5885
5886 if (use_vfork_if_available) {
5887 pid = vfork();
5888 } else {
5889 pid = fork();
5890 }
5891
5892 if (pid < 0) {
5893 // fork failed
5894 return -1;
5895
5896 } else if (pid == 0) {
5897 // child process
5898
5899 execve("/bin/sh", (char* const*)argv, environ);
5900
5901 // execve failed
5902 _exit(-1);
5903
5904 } else {
5905 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5906 // care about the actual exit code, for now.
5907
5908 int status;
5909
5910 // Wait for the child process to exit. This returns immediately if
5911 // the child has already exited. */
5912 while (waitpid(pid, &status, 0) < 0) {
5913 switch (errno) {
5914 case ECHILD: return 0;
5915 case EINTR: break;
5916 default: return -1;
5917 }
5918 }
5919
5920 if (WIFEXITED(status)) {
5921 // The child exited normally; get its exit code.
5922 return WEXITSTATUS(status);
5923 } else if (WIFSIGNALED(status)) {
5924 // The child exited because of a signal
5925 // The best value to return is 0x80 + signal number,
5926 // because that is what all Unix shells do, and because
5927 // it allows callers to distinguish between process exit and
5928 // process death by signal.
5929 return 0x80 + WTERMSIG(status);
5930 } else {
5931 // Unknown exit code; pass it through
5932 return status;
5933 }
5934 }
5935 }
5936
5937 // Get the default path to the core file
5938 // Returns the length of the string
5939 int os::get_core_path(char* buffer, size_t bufferSize) {
5940 /*
5941 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5942 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5943 */
5944 const int core_pattern_len = 129;
5945 char core_pattern[core_pattern_len] = {0};
5946
5947 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5948 if (core_pattern_file == -1) {
5949 return -1;
5950 }
5951
5952 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5953 ::close(core_pattern_file);
5954 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5955 return -1;
5956 }
5957 if (core_pattern[ret-1] == '\n') {
5958 core_pattern[ret-1] = '\0';
5959 } else {
5960 core_pattern[ret] = '\0';
5961 }
5962
5963 // Replace the %p in the core pattern with the process id. NOTE: we do this
5964 // only if the pattern doesn't start with "|", and we support only one %p in
5965 // the pattern.
5966 char *pid_pos = strstr(core_pattern, "%p");
5967 const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : ""; // skip over the "%p"
5968 int written;
5969
5970 if (core_pattern[0] == '/') {
5971 if (pid_pos != NULL) {
5972 *pid_pos = '\0';
5973 written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern,
5974 current_process_id(), tail);
5975 } else {
5976 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5977 }
5978 } else {
5979 char cwd[PATH_MAX];
5980
5981 const char* p = get_current_directory(cwd, PATH_MAX);
5982 if (p == NULL) {
5983 return -1;
5984 }
5985
5986 if (core_pattern[0] == '|') {
5987 written = jio_snprintf(buffer, bufferSize,
5988 "\"%s\" (or dumping to %s/core.%d)",
5989 &core_pattern[1], p, current_process_id());
5990 } else if (pid_pos != NULL) {
5991 *pid_pos = '\0';
5992 written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern,
5993 current_process_id(), tail);
5994 } else {
5995 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5996 }
5997 }
5998
5999 if (written < 0) {
6000 return -1;
6001 }
6002
6003 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6004 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6005
6006 if (core_uses_pid_file != -1) {
6007 char core_uses_pid = 0;
6008 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6009 ::close(core_uses_pid_file);
6010
6011 if (core_uses_pid == '1') {
6012 jio_snprintf(buffer + written, bufferSize - written,
6013 ".%d", current_process_id());
6014 }
6015 }
6016 }
6017
6018 return strlen(buffer);
6019 }
6020
6021 bool os::start_debugging(char *buf, int buflen) {
6022 int len = (int)strlen(buf);
6023 char *p = &buf[len];
6024
6025 jio_snprintf(p, buflen-len,
6026 "\n\n"
6027 "Do you want to debug the problem?\n\n"
6028 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
6029 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
6030 "Otherwise, press RETURN to abort...",
6031 os::current_process_id(), os::current_process_id(),
6032 os::current_thread_id(), os::current_thread_id());
6033
6034 bool yes = os::message_box("Unexpected Error", buf);
6035
6036 if (yes) {
6037 // yes, user asked VM to launch debugger
6038 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
6039 os::current_process_id(), os::current_process_id());
6040
6041 os::fork_and_exec(buf);
6042 yes = false;
6043 }
6044 return yes;
6045 }
6046
6047
6048 // Java/Compiler thread:
6049 //
6050 // Low memory addresses
6051 // P0 +------------------------+
6052 // | |\ Java thread created by VM does not have glibc
6053 // | glibc guard page | - guard page, attached Java thread usually has
6054 // | |/ 1 glibc guard page.
6055 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6056 // | |\
6057 // | HotSpot Guard Pages | - red, yellow and reserved pages
6058 // | |/
6059 // +------------------------+ JavaThread::stack_reserved_zone_base()
6060 // | |\
6061 // | Normal Stack | -
6062 // | |/
6063 // P2 +------------------------+ Thread::stack_base()
6064 //
6065 // Non-Java thread:
6066 //
6067 // Low memory addresses
6068 // P0 +------------------------+
6069 // | |\
6070 // | glibc guard page | - usually 1 page
6071 // | |/
6072 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6073 // | |\
6074 // | Normal Stack | -
6075 // | |/
6076 // P2 +------------------------+ Thread::stack_base()
6077 //
6078 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
6079 // returned from pthread_attr_getstack().
6080 // ** Due to NPTL implementation error, linux takes the glibc guard page out
6081 // of the stack size given in pthread_attr. We work around this for
6082 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
6083 //
6084 #ifndef ZERO
6085 static void current_stack_region(address * bottom, size_t * size) {
6086 if (os::is_primordial_thread()) {
6087 // primordial thread needs special handling because pthread_getattr_np()
6088 // may return bogus value.
6089 *bottom = os::Linux::initial_thread_stack_bottom();
6090 *size = os::Linux::initial_thread_stack_size();
6091 } else {
6092 pthread_attr_t attr;
6093
6094 int rslt = pthread_getattr_np(pthread_self(), &attr);
6095
6096 // JVM needs to know exact stack location, abort if it fails
6097 if (rslt != 0) {
6098 if (rslt == ENOMEM) {
6099 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
6100 } else {
6101 fatal("pthread_getattr_np failed with error = %d", rslt);
6102 }
6103 }
6104
6105 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
6106 fatal("Cannot locate current stack attributes!");
6107 }
6108
6109 // Work around NPTL stack guard error.
6110 size_t guard_size = 0;
6111 rslt = pthread_attr_getguardsize(&attr, &guard_size);
6112 if (rslt != 0) {
6113 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
6114 }
6115 *bottom += guard_size;
6116 *size -= guard_size;
6117
6118 pthread_attr_destroy(&attr);
6119
6120 }
6121 assert(os::current_stack_pointer() >= *bottom &&
6122 os::current_stack_pointer() < *bottom + *size, "just checking");
6123 }
6124
6125 address os::current_stack_base() {
6126 address bottom;
6127 size_t size;
6128 current_stack_region(&bottom, &size);
6129 return (bottom + size);
6130 }
6131
6132 size_t os::current_stack_size() {
6133 // This stack size includes the usable stack and HotSpot guard pages
6134 // (for the threads that have Hotspot guard pages).
6135 address bottom;
6136 size_t size;
6137 current_stack_region(&bottom, &size);
6138 return size;
6139 }
6140 #endif
6141
6142 static inline struct timespec get_mtime(const char* filename) {
6143 struct stat st;
6144 int ret = os::stat(filename, &st);
6145 assert(ret == 0, "failed to stat() file '%s': %s", filename, os::strerror(errno));
6146 return st.st_mtim;
6147 }
6148
6149 int os::compare_file_modified_times(const char* file1, const char* file2) {
6150 struct timespec filetime1 = get_mtime(file1);
6151 struct timespec filetime2 = get_mtime(file2);
6152 int diff = filetime1.tv_sec - filetime2.tv_sec;
6153 if (diff == 0) {
6154 return filetime1.tv_nsec - filetime2.tv_nsec;
6155 }
6156 return diff;
6157 }
6158
6159 bool os::supports_map_sync() {
6160 return true;
6161 }
6162
6163 /////////////// Unit tests ///////////////
6164
6165 #ifndef PRODUCT
6166
6167 class TestReserveMemorySpecial : AllStatic {
6168 public:
6169 static void small_page_write(void* addr, size_t size) {
6170 size_t page_size = os::vm_page_size();
6171
6172 char* end = (char*)addr + size;
6173 for (char* p = (char*)addr; p < end; p += page_size) {
6174 *p = 1;
6175 }
6176 }
6177
6178 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6179 if (!UseHugeTLBFS) {
6180 return;
6181 }
6182
6183 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6184
6185 if (addr != NULL) {
6186 small_page_write(addr, size);
6187
6188 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6189 }
6190 }
6191
6192 static void test_reserve_memory_special_huge_tlbfs_only() {
6193 if (!UseHugeTLBFS) {
6194 return;
6195 }
6196
6197 size_t lp = os::large_page_size();
6198
6199 for (size_t size = lp; size <= lp * 10; size += lp) {
6200 test_reserve_memory_special_huge_tlbfs_only(size);
6201 }
6202 }
6203
6204 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6205 size_t lp = os::large_page_size();
6206 size_t ag = os::vm_allocation_granularity();
6207
6208 // sizes to test
6209 const size_t sizes[] = {
6210 lp, lp + ag, lp + lp / 2, lp * 2,
6211 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6212 lp * 10, lp * 10 + lp / 2
6213 };
6214 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6215
6216 // For each size/alignment combination, we test three scenarios:
6217 // 1) with req_addr == NULL
6218 // 2) with a non-null req_addr at which we expect to successfully allocate
6219 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6220 // expect the allocation to either fail or to ignore req_addr
6221
6222 // Pre-allocate two areas; they shall be as large as the largest allocation
6223 // and aligned to the largest alignment we will be testing.
6224 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6225 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6226 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6227 -1, 0);
6228 assert(mapping1 != MAP_FAILED, "should work");
6229
6230 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6231 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6232 -1, 0);
6233 assert(mapping2 != MAP_FAILED, "should work");
6234
6235 // Unmap the first mapping, but leave the second mapping intact: the first
6236 // mapping will serve as a value for a "good" req_addr (case 2). The second
6237 // mapping, still intact, as "bad" req_addr (case 3).
6238 ::munmap(mapping1, mapping_size);
6239
6240 // Case 1
6241 for (int i = 0; i < num_sizes; i++) {
6242 const size_t size = sizes[i];
6243 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6244 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6245 if (p != NULL) {
6246 assert(is_aligned(p, alignment), "must be");
6247 small_page_write(p, size);
6248 os::Linux::release_memory_special_huge_tlbfs(p, size);
6249 }
6250 }
6251 }
6252
6253 // Case 2
6254 for (int i = 0; i < num_sizes; i++) {
6255 const size_t size = sizes[i];
6256 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6257 char* const req_addr = align_up(mapping1, alignment);
6258 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6259 if (p != NULL) {
6260 assert(p == req_addr, "must be");
6261 small_page_write(p, size);
6262 os::Linux::release_memory_special_huge_tlbfs(p, size);
6263 }
6264 }
6265 }
6266
6267 // Case 3
6268 for (int i = 0; i < num_sizes; i++) {
6269 const size_t size = sizes[i];
6270 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6271 char* const req_addr = align_up(mapping2, alignment);
6272 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6273 // as the area around req_addr contains already existing mappings, the API should always
6274 // return NULL (as per contract, it cannot return another address)
6275 assert(p == NULL, "must be");
6276 }
6277 }
6278
6279 ::munmap(mapping2, mapping_size);
6280
6281 }
6282
6283 static void test_reserve_memory_special_huge_tlbfs() {
6284 if (!UseHugeTLBFS) {
6285 return;
6286 }
6287
6288 test_reserve_memory_special_huge_tlbfs_only();
6289 test_reserve_memory_special_huge_tlbfs_mixed();
6290 }
6291
6292 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6293 if (!UseSHM) {
6294 return;
6295 }
6296
6297 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6298
6299 if (addr != NULL) {
6300 assert(is_aligned(addr, alignment), "Check");
6301 assert(is_aligned(addr, os::large_page_size()), "Check");
6302
6303 small_page_write(addr, size);
6304
6305 os::Linux::release_memory_special_shm(addr, size);
6306 }
6307 }
6308
6309 static void test_reserve_memory_special_shm() {
6310 size_t lp = os::large_page_size();
6311 size_t ag = os::vm_allocation_granularity();
6312
6313 for (size_t size = ag; size < lp * 3; size += ag) {
6314 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6315 test_reserve_memory_special_shm(size, alignment);
6316 }
6317 }
6318 }
6319
6320 static void test() {
6321 test_reserve_memory_special_huge_tlbfs();
6322 test_reserve_memory_special_shm();
6323 }
6324 };
6325
6326 void TestReserveMemorySpecial_test() {
6327 TestReserveMemorySpecial::test();
6328 }
6329
6330 #endif